Antimicrobial lipopeptide tridecaptin A₁ selectively binds to Gram-negative lipid II

Significance

The increasing development of antimicrobial resistance is a major global concern, and there is an urgent need for the development of new antibiotics. We show that the antimicrobial lipopeptide tridecaptin A₁ selectively binds to the Gram-negative analogue of peptidoglycan precursor lipid II, disrupting the proton motive force and killing Gram-negative bacteria. We present an example of the selective targeting of Gram-negative lipid II and a binding mode to this peptidoglycan precursor. No persistent resistance develops against tridecaptin A₁ in Escherichia coli cells exposed to subinhibitory concentrations of this peptide during a 1-mo period. This study showcases the excellent antibiotic properties of the tridecaptins in an age where new antibiotics that target Gram-negative bacteria are desperately needed.

Abstract

Tridecaptin A₁ (TriA₁) is a nonribosomal lipopeptide with selective antimicrobial activity against Gram-negative bacteria. Here we show that TriA₁ exerts its bactericidal effect by binding to the bacterial cell-wall precursor lipid II on the inner membrane, disrupting the proton motive force. Biochemical and biophysical assays show that binding to the Gram-negative variant of lipid II is required for membrane disruption and that only the proton gradient is dispersed. The NMR solution structure of TriA₁ in dodecylphosphocholine micelles with lipid II has been determined, and molecular modeling was used to provide a structural model of the TriA₁–lipid II complex. These results suggest that TriA₁ kills Gram-negative bacteria by a mechanism of action using a lipid-II–binding motif.

Recently, a lot of media coverage has been focused on the problem of antimicrobial resistance. A report commissioned by the UK government predicts that by 2050 antimicrobial resistance will have caused 300 million premature deaths and cost the global economy over $100 trillion (1). Even more worrying is the lack of new classes of antibiotics active against Gram-negative bacteria. In the past 50 y, only a few structurally and mechanistically distinct classes of antibiotics have been clinically approved to treat systemic infections (including fidaxomicin, bedaquiline, linezolid, and daptomycin), yet none of these are active against Gram-negative bacteria (2, 3). Two new classes of Gram-negative targeting antibiotics in the clinical pipeline are POL7080 and brilacidin (4, 5). Both of these compounds are modeled on antimicrobial peptides, which are becoming increasingly important in the fight against antibiotic resistance (6). Bacteria produce a wealth of antimicrobial peptides, both ribosomally, including the lantibiotics (7, 8), and nonribosomally, including lipopeptides (9). In particular, lipopeptides are a rich source of antimicrobial compounds, and several examples with activity against Gram-positive (10, 11) and/or Gram-negative bacteria (12) have been recently characterized.

Tridecaptin A₁ (TriA₁) is a member of the tridecaptin family, a group of nonribosomal lipopeptides produced by Bacillus and Paenibacillus species (Fig. 1) (13–15). This acylated tridecapeptide displays strong and selective antimicrobial activity against Gram-negative bacteria, including multidrug-resistant strains of Klebsiella pneumoniae, Acinetobacter baumannii, and Escherichia coli (16). TriA₁ analogs have low cytotoxicity and have been shown to treat K. pneumoniae infections in mice (16, 17). Therefore, we believe that tridecaptin A₁ could be an excellent antibiotic candidate. However, before our investigations little was known about how TriA₁ exerts its selective bactericidal effect against Gram-negative bacteria. A previous structure–activity relationship study by our group suggested that TriA₁, akin to many other lipopeptides, is a membrane-targeting agent. We found that removal of the N-terminal lipid tail abolishes antimicrobial activity; however, the chiral lipid tail could be replaced with an octanoyl chain to give Oct-TriA₁ (Fig. 1), which retains full activity (16). We therefore sought to identify the precise mode and mechanism of action by which TriA₁ kills Gram-negative bacteria.

Fig. 1.

Structures of the tridecaptin analogs TriA₁ and Oct-TriA₁.

Results

Tridecaptin A₁ Targets the Cell Membrane.

Our initial efforts focused on identifying the mode of action of TriA₁. Measuring the time taken for an antibiotic to exert its antimicrobial effect can provide valuable information on the cellular process targeted by that compound. Bacteriostatic agents halt cell division but do not reduce the number of viable cells, and antibiotics that target protein synthesis and nucleic acid synthesis typically fall within this category. Bactericidal agents reduce the population of viable bacterial cells, and the time taken for a bactericidal agent to kill bacteria provides further information on its target. Therefore, we monitored the growth kinetics of E. coli cells exposed to TriA₁ and a number of other antibiotics (Fig. 2). Lipopeptides like polymyxin B are generally bactericidal within minutes of exposure due to the formation of large, nonspecific pores in the bacterial membrane, whereas many other antibiotics (like ampicillin) exert their killing effect over several hours. Bacteriostatic agents (like chloramphenicol) halt cell growth, but do not kill cells. Optical density measurements showed that cells exposed to TriA₁ grew at a reduced rate for 20 min, at which point a steady decrease in cell count was observed (Fig. 2A). Polymyxin B reduced the cell count immediately, whereas ampicillin displayed an expected 3-h lag and chloramphenicol did not reduce optical density. Complementary results were obtained from a time-kill assay (Fig. 2B), which revealed that cells exposed to TriA₁ showed slightly reduced cell viability after 15 min of exposure, a significant reduction after 30 min, and complete killing after 60 min. Polymyxin B acted faster, significantly reducing the viable cell population after 5 min and killing all cells by 30 min. These results suggested that TriA₁ is a membrane-targeting peptide, but does not act by a generic membrane lysis mechanism like polymyxin B. This is further supported by the selectivity of TriA₁ against Gram-negative bacteria because a peptide that operates through a generic lysis mechanism would also target Gram-positive organisms.

Fig. 2.

(A) Bacterial growth kinetics. Optical densities of E. coli cells exposed to 2× MIC of ampicillin (8 μg/mL), chloramphenicol (32 μg/mL), tridecaptin A₁ (6.25 μg/mL), and polymyxin B (4 μg/mL). Tridecaptin A₁ reduces cell density after 20 min of exposure. (B) Time-kill assays. E. coli cells were treated with 10× MIC of each antibiotic (50 μM), and the number of viable cells was determined at different time points. Tridecaptin A₁ kills cells more slowly than polymyxin B.

TriA₁ Binds to Lipopolysaccharide on the Outer Membrane.

Gram-negative organisms are protected from their environment by an outer membrane, which reduces diffusion of many antibiotics into the cell or periplasmic space. For TriA₁ to attack the bacterial membrane of Gram-negative bacteria, it must first cross this barrier. Polymyxin B traverses the outer membrane by binding to the lipid A portion of lipopolysaccharide (LPS), followed by insertion into the membrane (18). We postulated that tridecaptin A₁ may also cross the outer membrane through an interaction with LPS, given that it possesses several cationic 2,4-diaminobutyric acid residues (Dab) like polymyxin B. Isothermal titration calorimetry (ITC) was therefore used to determine if TriA₁ binds to LPS. Our studies revealed that TriA₁, Oct-TriA₁, the enantiomeric form of TriA₁ (Ent-TriA₁), and unacylated TriA₁ (H-TriA₁) bind to LPS with similar affinities (SI Appendix, Fig. S1). We have previously shown that H-TriA₁ has weak antimicrobial activity, but at sub-minimum inhibitory concentrations (MIC) it can enhance the activity of hydrophobic antibiotics like rifampicin and vancomycin over 100-fold (17). Therefore, it is likely that LPS binding is the mechanistic basis for this disruption of the outer membrane. We have also found that the enantiomer of TriA₁ (Ent-TriA₁) is fourfold less active than the natural peptide, suggesting that TriA₁ interacts with a chiral target (19). Ent-TriA₁ contains the same amino acids and lipid tail as TriA₁ and therefore retains the same amphiphilic properties. It is likely that the observed activity at higher concentrations is due to membrane lysis by a nonspecific detergent-like effect, akin to the mode of action of the polymyxins. This observation that both TriA₁ and its enantiomer bind to LPS with similar affinities suggested that another chiral receptor is involved in the mode of action.

TriA₁ Disrupts the Proton Motive Force.

With a rationale for how TriA₁ crosses the outer membrane, we next sought to uncover the mechanism by which it acts on the inner membrane. To this end, a series of experiments were performed using fluorescent dyes (Fig. 3). DiBAC₄ is a common dye used to monitor membrane depolarization (20). It enters depolarized cells and binds to proteins or membrane components, exhibiting enhanced fluorescence. Addition of TriA₁ to DiBAC₄-treated E. coli cells resulted in no membrane depolarization, whereas application of polymyxin B led to a rapid increase in fluorescence (Fig. 3A). To further assess the effect of TriA₁ on the inner membrane, its ability to form large pores was assessed using the dye SYTOX Green (Fig. 3B). Upon lysis of the inner membrane, this dye binds to nucleic acids in the cytoplasm, causing an increase in fluorescence (21). Immediate pore formation was not observed when E. coli cells pretreated with SYTOX Green were exposed to TriA₁, whereas addition of the surfactant Triton X-100 caused an instant fluorescence increase. Instead, TriA₁ caused a very gradual increase in fluorescence at a rate consistent with the death of cells based on the time-kill assays, suggesting that this increase is due to cells that are already dead and not an immediate effect of TriA₁. To further probe the action of TriA₁ on the inner membrane, an ortho-nitrophenyl-β-galactosidase (ONPG) assay was performed on E. coli ML-35 cells, which lack the lac permease. The chromophore ortho-nitrophenol is released by the action of cytoplasmic galactosidase on ortho-nitrophenol galactoside only if large pores are formed on the inner membrane. ONPG assays (SI Appendix, Fig. S2) did not show immediate membrane lysis upon TriA₁ addition to E. coli cells.

Fig. 3.

Inner-membrane assays. (A) Membrane depolarization assay. Addition of TriA₁ to DiBAC₄-treated E. coli cells decreases fluorescence by dilution, but no membrane depolarization is observed. Polymyxin B leads to expected fluorescence increase as the inner membrane is depolarized. (B) Membrane disruption assay. Addition of TriA₁ to SYTOX Green-treated E. coli cells does not cause immediate pore formation. Triton X-100 causes rapid inner-membrane lysis. (C) Disruption of proton motive force detected as fluorescence decrease from BCECF-AM–treated E. coli cells. Glucose increases the proton motive force, increasing the cytoplasmic pH and increasing fluorescence. Valinomycin disperses the electrochemical gradient, and addition of TriA₁ rapidly decreases fluorescence. Subsequent addition of nigericin further decreases fluorescence at a faster rate than TriA₁.

The rapid killing of E. coli by TriA₁ stood in stark contrast to the lack of membrane lysis or depolarization observed. The essential nature of the hydrophobic tail for antimicrobial activity and the overall amphiphilic structure of TriA₁ support an interaction with the bacterial membrane; therefore we examined the function of the inner membrane in detail. Bacteria use both protons and potassium ions to generate adenosine triphosphate (ATP) via the proton motive force, a process that is essential for cell growth (22, 23). Other antibiotics are known to affect this vital cellular process. For example, nigericin and valinomycin shuttle a combination of K⁺ and H⁺ across the bacterial membrane (23), whereas the bacteriocin subtilosin forms proton-specific pores in the Gram-positive bacterium Gardnerella vaginalis (24, 25). To determine if TriA₁ affects the proton motive force, we adapted the prodye 2′,7′-bis-(2-carboxyethyl)-5-(and-6)-carboxyfluorescein acetoxymethyl ester (BCECF-AM) for use in Gram-negative bacteria. In the presence of ethylenediaminetetraacetic acid (EDTA), this dye readily crosses the bacterial membrane where it is hydrolyzed by nonspecific esterases to give a pH-sensitive dye, the fluorescence of which decreases as cytoplasmic pH decreases (25). Addition of glucose to BCECF-AM–treated E. coli cells resulted in a fluorescence increase, indicating that they were actively respiring and that BCECF was correctly showing the increase in cytoplasmic pH (Fig. 3C). To assess the effect on the proton gradient, the potassium gradient was first dispersed by addition of valinomycin, which exchanges K⁺ with H⁺, causing a small increase in fluorescence. Subsequent addition of TriA₁ resulted in a significant fluorescence decrease, signifying pore formation and disruption of the proton gradient. Addition of the proton shuttle peptide nigericin also had a similar but more rapid effect, strongly suggesting that TriA₁ disrupts the proton motive force.

TriA₁ Selectively Binds to Gram-Negative Lipid II.

Having identified a probable mode of action for tridecaptin A₁, we next sought to identify the chiral receptor involved in its mechanism of action. Given the lower antimicrobial activity of Ent-TriA₁, and the fact that TriA₁ acts on the bacterial membrane, we postulated that binding of TriA₁ to a chiral receptor on the surface of the inner membrane is important. Lipid II is the final monomeric intermediate in peptidoglycan biosynthesis and a target of several antimicrobial peptides, including nisin (26, 27), plectasin (28), and teixobactin (10). It is synthesized in the cytoplasm and anchored to the interior of the inner membrane by an undecaprenyl chain. Lipid II is subsequently flipped to the exterior of the inner membrane, where it is then elongated into peptidoglycan. The structure of lipid II differs between Gram-positive and Gram-negative bacteria, namely on residue 3 of the pentapeptide portion. In most Gram-negative bacteria, this amino acid is meso-diaminopimelic acid (DAP), whereas in most Gram-positive bacteria it is lysine (Fig. 4A). Using ITC, we evaluated the binding affinity of TriA₁ to Gram-negative (Fig. 4B) and Gram-positive lipid II (SI Appendix, Fig. S3). TriA₁ binds to Gram-negative lipid II in a 1:1 stoichiometry and with a K_d of 4 μM, which is close to the MIC of TriA₁ against E. coli (∼3 μM). We were surprised to find that, although TriA₁ binds strongly to Gram-negative lipid II, it has a much weaker binding affinity to the Gram-positive variant. This observation would explain why TriA₁ is ∼50-fold less active against Gram-positive bacteria, and therefore we sought to further investigate this phenomenon. Gratifyingly, Ent-TriA₁ does not bind to Gram-negative lipid II (SI Appendix, Fig. S4), confirming that this is an important chiral receptor for TriA₁. When TriA₁ was premixed with one equivalent of Gram-negative lipid II, its activity was completely abolished in a spot-on-lawn assay against E. coli (Fig. 4C). In contrast, premixing TriA₁ with Gram-positive lipid II had very little effect on the antimicrobial activity of the peptide, further evidencing this selective interaction. The same experiment with Ent-TriA₁ and Gram-negative lipid II had no effect on activity (SI Appendix, Fig. S4). This is a rare instance of an antimicrobial compound binding selectively to lipid II from Gram-negative bacteria.

Fig. 4.

(A) Structure of Gram-negative and Gram-positive lipid II. (B) ITC of TriA₁ + Gram-negative lipid II. (C) Spot-on-lawn assay with E. coli cells. (Left) TriA₁ (50 μM). (Middle) 1:1 TriA₁ (100 μM):G⁺LII (100 μM). (Right) 1:1 TriA₁ (100 μM):G⁻LII (100 μM). TriA₁ is active, and premixing with Gram-positive lipid II (G+LII) slightly reduces the zone of inhibition. Premixing with Gram-negative lipid II (G-LII) abolishes activity.

In Vitro Assays Link Lipid II Binding to Membrane Disruption.

To corroborate our theory that TriA₁ binds to lipid II on the inner membrane and disrupts the proton motive force, we developed an in vitro assay using large unilamellar vesicles (LUV) and BCECF acid to measure intravesicle pH changes. BCECF was encapsulated in 50-nm LUVs, such that the internal pH was set to pH 8.0 (Fig. 5A). Transferring these vesicles into a pH 6 buffer did not lead to an observable change in fluorescence over 30 min of incubation, whereas an immediate decrease was observed upon the addition of Triton X-100, confirming the integrity of the vesicles. Oct-TriA₁, which has identical antimicrobial activity to TriA₁ and also binds to Gram-negative lipid II, was used in these studies. Addition of Oct-TriA₁ to LUVs containing 1 mol% Gram-negative lipid II at concentrations mimicking the bacterial MIC resulted in a rapid decrease in fluorescence, signifying pore formation (Fig. 5B). A much weaker decrease in fluorescence occurred with LUVs containing 1 mol% Gram-positive lipid II, which was comparable to the effect observed against vesicles lacking any lipid II. These studies provide further evidence that TriA₁ selectively recognizes Gram-negative lipid II and link this recognition to the observed bactericidal activity.

Fig. 5.

(A) In vitro assay measures formation of small proton pores. BCECF is encapsulated in LUVs with an internal pH of 8, and the external buffer is pH 6. Pore formation results in a proton gradient and decrease in fluorescence. (B) BCECF LUVs with no lipid II, 1 mol% Gram-negative lipid II, or 1 mol% Gram-positive lipid II are treated with 1.8 μM Oct-TriA₁. Gram-negative lipid II significantly accelerates pore formation.

Synthesis of Gram-Negative Lipid II Analog for NMR Studies.

In an effort to understand the interaction between TriA₁ and Gram-negative lipid II on a molecular level, we sought to elucidate the NMR solution structure of TriA₁ in dodecylphosphocholine (DPC) micelles containing Gram-negative lipid II. Natural lipid II, which contains a fatty undecaprenyl chain, is not very amenable to NMR studies. It is prone to micelle formation in aqueous solvent, and the multiple methyl and methylene signals from the C₅₅ chain can drown out important signals in its binding partner needed for complete structural characterization. Breukink and coworkers have previously shown that a Gram-positive lipid II analog with an (E,E)-farnesyl (C₁₅) chain retains full binding affinity to nisin (29), and this analog was used in the elucidation of the NMR solution structure of the nisin-lipid II complex (27). Although Breukink and coworkers found that the stereochemistry of the isoprene units in the chain were not important for nisin binding, it was unclear if this region of lipid II was required for interactions with TriA₁. Therefore, we embarked on the synthesis of (Z,Z)-farnesyl Gram-negative lipid II (1) (Fig. 6). The total synthesis of Gram-positive lipid II was previously reported by VanNieuwenhze and coworkers (30, 31), whereas a semisynthesis of Gram-negative lipid II has been reported by Walker and Kahne (32). By modification of these literature procedures, we performed a total synthesis of Gram-negative lipid II analog (1) (Fig. 6 and SI Appendix). The disaccharide core was constructed by a glycosylation between acetimidate 2 (SI Appendix, Scheme S1) and glycol 3 (SI Appendix, Schemes S2 and S3). After protecting group manipulation and phosphorylation of the anomeric position of alanyl disaccharide 4, the appropriate tetrapeptide (5) (SI Appendix, Schemes S4 and S5) was coupled to the alanine carboxylate to yield pentapeptidyl disaccharide 6. Deprotection of the benzyl phosphate, followed by coupling to carbonyldiimidazole (CDI)-activated (Z,Z)-farnesyl phosphate and global deprotection, gave the desired Gram-negative lipid II analog 1.

Fig. 6.

Total synthesis of (Z,Z)-farnesyl Gram-negative lipid II (1). (A) TMSOTf, 4 Å MS, CH₂Cl₂, rt, 18 h, 61%. (B) (i) ZnCl₂, AcOH/Ac₂O, rt, 24 h (ii) Zn, THF/AcOH/Ac₂O, rt, 24 h, 63% (two steps). (C) (i) H₂, Pd/C, MeOH, rt, 3 h, (ii) (ⁱPr)₂NP(OBn)₂, tetrazole, CH₂Cl₂, rt, 2 h, (iii) 30% H₂O₂/THF, −78 °C, 2 h, 84% (three steps). (D) DBU, CH₂Cl₂, rt, 0.5h, quant. (E) Tetrapeptide 5, TFA/CH₂Cl₂, 2h; HATU, DIPEA, DMF, rt, 24 h, 78%. (F) (i) H₂, Pd/C, MeOH, rt, 2.5 h, (ii) CDI-activated (Z,Z)-farnesyl phosphate, DMF, rt, 4d, (iii) NaOH, H₂O/dioxane, 37 °C, 2 h, 25% (three steps).

Characterization of TriA₁–Lipid II Complex by NMR and Molecular Modeling.

With Gram-negative lipid II analog 1 in hand, we then proceeded to elucidate the NMR solution structure of Oct-TriA₁ in DPC micelles doped with Gram-negative lipid II. First, the chemical shifts (SI Appendix, Table S1) and Nuclear Overhauser Effect (NOE) correlations of Oct-TriA₁ in DPC micelles were assigned, and the solution structure was calculated using CYANA (SI Appendix, Fig. S5) (33). Without lipid II, Oct-TriA₁ adopts a looped structure with all hydrophobic residues on one face. Upon addition of one equivalent of lipid II analog 1 (SI Appendix, Table S2) to Oct-TriA₁ in DPC micelles, significant amide chemical shift changes occurred on d-Val1, d-Dab2, and d-Dab8 in TriA₁ (SI Appendix, Fig. S6); and on d-γ-Glu2, Ala4, and Ala5 in lipid II (SI Appendix, Fig. S7). This is indicative of a conformational change in both molecules and confirms that they interact (SI Appendix, Tables S3 and S4). With lipid II present, Oct-TriA₁ adopts a more open and amphiphilic structure with an apparent π-stacking interaction between d-Trp5 and Phe9 (Fig. 7A). Interestingly, d-Dab8, which we have previously shown is the critical residue for Oct-TriA₁ activity (19), is located at the base of a possible binding pocket.

Fig. 7.

(A) NMR solution structure of TriA₁ in DPC micelles containing Gram-negative lipid II. Orange: hydrophobic residues; purple: d-Dab8; and cyan: other residues. (B) Lipid II analog 1 docked into TriA₁. Hydrophobic residues interact with the lipid II terpene tail, and the pentapeptide occupies the binding pocket. (C) Modeled interaction shows H-bonding between d-Dab8 and DAP3.

To obtain an experimentally derived model of the Oct-TriA₁-Lipid II complex, the structure of lipid II analog 1 in the presence of Oct-TriA₁ was also calculated using CYANA and then docked into the NMR solution structure of Oct-TriA₁ (Fig. 6A) using AutoDock Vina (Fig. 7B) (34). The model suggests that the N-terminal lipid tail and d-Trp5 are in close proximity to the lipid II terpene chain. This would be expected as the terpene tail of lipid II anchors it to the exterior of the inner membrane and would associate with the more hydrophobic residues of TriA₁. The model also suggests that there is an H-bonding interaction between the γ-amino group of d-Dab8 and the ε-carboxylate on DAP3 in lipid II (Fig. 7C). d-Dab8 is essential for the antimicrobial activity of the tridecaptins, and DAP3 in lipid II is essential for TriA₁ binding. Notably, the model does not show interactions between Oct-TriA₁ and the pyrophosphate moiety of lipid II. No shift in the ³¹P-NMR of lipid II is observed on addition of Oct-TriA₁, confirming that the pyrophosphate is not involved in Oct-TriA₁ binding (SI Appendix, Fig. S8). This is an instance of an antimicrobial peptide that does not make use of a pyrophosphate cage.

TriA₁ Displays Low Levels of Resistance Development.

The results from the mode of action and mechanism of actions studies suggested that tridecaptin A₁ exerts its bactericidal effect by binding to lipid II on the inner membrane and disrupting the proton motive force. Lipid II is a late-stage intermediate in peptidoglycan biosynthesis, and the proposed binding site for TriA₁ would be difficult for bacteria to modify without affecting its subsequent processing by enzymes in the peptidoglycan biosynthesis pathway. One would therefore expect that resistance development against TriA₁ would be limited. An in vitro evolution study was performed in which E. coli cells were continuously exposed to sub-MIC concentrations of Oct-TriA₁ or ciprofloxacin for 1 mo. During this experiment, the activity of the nucleic acid synthesis inhibitor ciprofloxacin decreased eightfold, whereas no persistent resistance developed against Oct-TriA₁ (Fig. 8).

Fig. 8.

Resistance study with E. coli cells.

Discussion

We have shown that tridecaptin A₁ exerts its bactericidal effect on Gram-negative bacteria by binding to lipid II on the surface of the inner membrane and by disrupting the proton motive force. Monitoring the time taken for TriA₁ to kill E. coli cells by growth kinetic measurements and time-kill assays suggested that TriA₁ targets the cell membrane (Fig. 2). However, TriA₁ does not act as quickly as polymyxin B, which kills bacteria by forming large nonspecific pores on the inner membrane. A study of the effect that TriA₁ has on the bacterial membrane using fluorescent dyes (Fig. 3) revealed that TriA₁ does not depolarize the membrane or kill Gram-negative bacteria by the formation of large pores. Initially, this result was quite surprising as several pieces of evidence suggested that it interacts with the cell membrane. First, the lipid tail of TriA₁ is essential to its antimicrobial activity. This is similar for many other lipopeptides, which must insert their lipid tail into the cell membrane for membrane disruption. Also, the time taken by TriA₁ to kill Gram-negative bacteria is consistent with membrane-acting antibiotics, and its overall amphiphilic structure further suggested a membrane interaction. This led us to consider that TriA₁ targets the bacterial membrane by a mode of action that is not membrane lysis. The proton motive force is a vital process that occurs across the bacterial membrane because it is the primary method by which bacteria produce ATP. The observation that other peptide antibiotics like nigericin, valinomycin, and subtilosin can disrupt the proton motive force led us to develop an in vivo assay using BCECF-AM to assess the impact that TriA₁ has on this important cellular function. This experiment showed that addition of TriA₁ rapidly decreases the cytoplasmic pH of E. coli cells, which was visualized as a rapid decrease in fluorescence (Fig. 3C). This signifies the formation of pores, allowing the transport of protons from the more acidic extracellular buffer into the cytoplasm. Combining this result with the observation that TriA₁ does not depolarize the bacterial membrane or form large pores led us to conclude that TriA₁ forms proton-specific pores. This will ultimately block the synthesis of ATP and kill bacteria.

Having identified a probable mode of action, we next became interested in the mechanism of action. ITC revealed that, like polymyxin B, TriA₁ binds to LPS on the outer membrane of Gram-negative bacteria. Removal of the lipid tail from TriA₁ does not affect LPS binding, but the unacylated derivative of TriA₁ has substantially lower activity than the natural peptide. Therefore, the lipid tail is important for either outer- or inner-membrane penetration. The observation that the enantiomer of TriA₁, which is less active than TriA₁, binds to LPS with a similar binding affinity shows that another chiral receptor is involved in the mechanism of action. As several other antimicrobial peptides bind to the peptidoglycan precursor lipid II, which is presented on the surface of the inner membrane, we investigated if lipid II was also a target for TriA₁. A combination of ITC (Fig. 4B and SI Appendix, Fig. S1), inhibition assays (Fig. 4C), and in vitro assays (Fig. 5) provided compelling evidence that TriA₁ selectively binds to the Gram-negative variant of lipid II, which contains DAP rather than lysine on its pentapeptide chain. TriA₁ has strong activity against Gram-negative bacteria and strongly binds to lipid II from these organisms. Its activity is sequestered through complex formation with this analog, and the presence of Gram-negative lipid II in model membranes substantially increases the pore-forming ability of TriA₁. Because TriA₁ has much lower activity against Gram-positive bacteria, one would expect it to not bind very strongly to lipid II from these organisms, to not lose activity when mixed with Gram-positive lipid II, and to not show increased activity against model membranes doped with Gram-positive lipid II. This is exactly what our experiments show.

We synthesized an analog of Gram-negative lipid II and used this to calculate the NMR solution structure of tridecaptin A₁ in DPC micelles containing one equivalent of Gram-negative lipid II (Fig. 7A). Molecular docking studies were then used to produce a model of the TriA₁-lipid II complex (Fig. 7 B and C), which is supported by several lines of experimental evidence. First, the amide chemical shifts of lipid II that underwent the largest change upon addition of Oct-TriA₁ are on the pentapeptide portion. Second, no interaction is predicted with the pyrophosphate of lipid II, and no change in the ³¹P-NMR upon addition of Oct-TriA₁ to lipid II analog 1 was observed. Third, and most importantly, a key hydrogen bond is predicted between the γ-amino group of d-Dab8 and the ε-carboxylate on DAP3 in lipid II. A previous structure–activity relationship study showed that d-Dab8 is absolutely essential for activity (19), and its substitution to d-Ala decreases its activity against Gram-negative bacteria to the level observed against Gram-positive organisms. If residue 3 on the pentapeptide portion of lipid II is lysine rather than DAP, the binding affinity of TriA₁ is significantly reduced. Therefore, this predicted interaction perfectly describes these experimental observations and led us to conclude that it is the presence of DAP on lipid II that gives rise to the remarkable selectivity of tridecaptin A₁ against Gram-negative bacteria. The observation that resistance development against Oct-TriA₁ is limited (Fig. 8) is also important, given the lack of new antibiotics that target Gram-negative bacteria, and we believe that the tridecaptins are an attractive class of future antibiotic candidates.

Materials and Methods

Whole-Cell Studies.

See SI Appendix.

BCECF-Containing LUV Preparation.

1,2-Dioleoyl-sn-glycero-3-phosphoethanolamine (16 mg) and 1,2-dioleoyl-sn-glycero-3-phospho-(1′-rac-glycerol) (4 mg) were dissolved in chloroform (2 mL). If necessary, G+LII or G-LII were then added as a solution in 2:3:1 CHCl₃:MeOH:H₂O to 1 mol%. The solution was thoroughly mixed, the solvent was removed under reduced pressure, and the film was dried in the dark under high vacuum overnight. The desiccated lipids were rehydrated with potassium phosphate buffer (50 mM, pH 8, 2 mL) and BCECF acid (2 mM, 10 µL). In dim light the solution was shaken thoroughly, vortexed, and transferred to a 5-mL cryovial. The vial was frozen in liquid nitrogen and thawed at 37 °C. The lipids were shaken thoroughly until finely suspended and refrozen. This process was repeated five times in total. The finely dispersed vesicles were then extruded 21 times (back and forth 10.5 times) through a lipid extruder (Avanti Polar Lipids) containing a 50-nm pore. Nonencapsulated dye was removed by passing the pale yellow solution through a Sephadex G-50 size exclusion column (50 mM potassium phosphate buffer as eluent, pH 8). The vesicles were then stored in the dark on ice at 4 °C for further use. Phosphate concentration was determined using the Stewart method (35).

In Vitro Assay Using BCECF LUVs.

Excitation and emission wavelengths were set to 500 nm and 522 nm, respectively. Freshly prepared lipid vesicles (10 µL) were added to potassium phosphate buffer (50 mM, pH 6, 2 mL). The fluorescence was monitored, with stirring, for ∼100 s to establish a baseline. Oct-TriA₁ was added, and fluorescence was monitored until stable. Triton X-100 (1%, 50 µL) was added to quench fluorescence. Experiments shown are representative of results from three technical replicates.

NMR Spectroscopy.

Oct-TriA₁ and lipid II were dissolved separately in 600 μL of a 180-mM DPC micelle solution [10% (vol/vol) D₂O/90% H₂O in 10 mM sodium phosphate buffer at pH 6] to a final concentration of 4 mM. All spectra were referenced to the methylene protons in DPC at 1.52 ppm. One-dimensional ¹H-NMR and 2D homonuclear ¹H¹-H Total Correlation Spectroscopy (TOCSY) and Nuclear Overhauser Effect Spectroscopy (NOESY) experiments were acquired at 27 °C on a four-channel 600-MHz Varian VNMRS spectrometer with a horizontal center nexus (HCN) z-axis pulsed-field gradient probe. The acquisition software used was VNMRJ 4.2A. TOCSY and NOESY experiments used a 8,000-Hz spectral window in both the directly and indirectly detected dimensions. A total of 512 experiments were used to define the indirectly detected dimension, with 32 and 64 scans for each experiment for the TOCSY and NOESY spectra, respectively, with a total of 4,882 real and imaginary points acquired in the directly detected dimension. A spin lock mix time of 250 ms was used for TOCSY spectra and a mix time of 150 ms was used for NOESY spectra. Water suppression was achieved by presaturation during the relaxation delay. For NOESY, saturation of the water peak was also applied during the mix time. NMRPipe and NMRView were used for data processing. Chemical shift assignments were performed manually. After complete spectral assignment of Oct-TriA₁ and the Gram-negative lipid II analog, the solutions were mixed, and the TOCSY (70-ms mix time) and NOESY (125-ms mix time) data were acquired for the TriA₁-Lipid II complex at 20 °C on a triple resonance HCN cryoprobe-equipped Varian VNMRS 700 MHz spectrometer with z-axis pulsed-field gradients and VNMRJ 4.2A as host control. Spectral width for both experiments was 10,000 Hz in both the directly and indirectly detected dimensions. A total of 256 acquisitions were used to define the indirectly detected dimension, with 96 and 128 scans for each experiment for the TOCSY and NOESY spectra, respectively, and a total of 8,192 real and imaginary points acquired in the directly detected dimension.

Structure Calculations.

CYANA 2.1 (33) was used to calculate the structure of Oct-TriA₁ in DPC micelles with and without lipid II using automatically assigned NOE crosspeaks. Custom library files for noncanonical amino acids were created based on energy minimized structures drawn with Avogadro. For Oct-TriA₁ in DPC micelles without lipid II, 211 crosspeak NOEs were selected and 134 of these were automatically assigned by CYANA (101 short-range, 13 medium-range, and 20 long-range) were used in the structure calculation. For Oct-TriA₁ with lipid II, 188 crosspeak NOEs (168 short-range, 14 medium-range, and 6 long-range) were used in the structure calculation. Seven cycles were done with 10,000 steps per cycle giving 20 structures. Structures were generated from the CYANA output.pdb files using MacPyMOL. The structure with the lowest target function was used in subsequent docking studies.

Docking Studies with Lipid II.

CYANA was used to calculate the structure of lipid II in the presence of Oct-TriA₁ using 67 NOE restraints (64 short-range and 3 medium-range). Of the 20 calculated structures, the first structure had the lowest target function value and was used as the input ligand structure for docking studies. AutoDockTools (v. 1.5.6) was used to convert the lipid II and Oct-TriA₁ .pdb files to .pdbqt files. The exhaustiveness was set to 8. Then Z,Z-Farnesyl Gram-negative lipid II was docked into Oct-TriA₁ using AutoDock Vina (34).

Lipid II Analog Synthesis.

See SI Appendix.