Single-Molecule Resolution of Antimicrobial Peptide Interactions with Supported Lipid A Bilayers

Abstract

The molecular interactions between antimicrobial peptides (AMPs) and lipid A-containing supported lipid bilayers were probed using single-molecule total internal reflection fluorescence microscopy. Hybrid supported lipid bilayers with lipid A outer leaflets and phospholipid (1,2-dioleoyl-sn-glycero-3-phosphoethanolamine (DOPE)) inner leaflets were prepared and characterized, and the spatiotemporal trajectories of individual fluorescently labeled LL37 and Melittin AMPs were determined as they interacted with the bilayer surfaces comprising either monophosphoryl or diphosphoryl lipid A (from Escherichia coli) to determine the impact of electrostatic interactions. Large numbers of trajectories were obtained and analyzed to obtain the distributions of surface residence times and the statistics of the spatial trajectories. Interestingly, the AMP species were sensitive to subtle differences in the charge of the lipid, with both peptides diffusing more slowly and residing longer on the diphosphoryl lipid A. Furthermore, the single-molecule dynamics indicated a qualitative difference between the behavior of AMPs on hybrid Lipid A bilayers and on those composed entirely of DOPE. Whereas AMPs interacting with a DOPE bilayer exhibited two-dimensional Brownian diffusion with a diffusion coefficient of ∼1.7 μm²/s, AMPs adsorbed to the lipid A surface exhibited much slower apparent diffusion (on the order of ∼0.1 μm²/s) and executed intermittent trajectories that alternated between two-dimensional Brownian diffusion and desorption-mediated three-dimensional flights. Overall, these findings suggested that bilayers with lipid A in the outer leaflet, as it is in bacterial outer membranes, are valuable model systems for the study of the initial stage of AMP-bacterium interactions. Furthermore, single-molecule dynamics was sensitive to subtle differences in electrostatic interactions between cationic AMPs and monovalent or divalent anionic lipid A moieties.

Introduction

Antimicrobial peptides (AMP) are a class of biological molecules that are critically important to the innate immune response of most multicellular organisms (1, 2). AMPs have sequences and structures as varied as the species that produce them, and they exhibit diverse mechanisms of antimicrobial action, including pore formation (3, 4, 5), or the assembly of a destructive cell surface “blanket” (i.e., the carpet model) (6, 7). These peptides exhibit antimicrobial activity against a wide variety of foreign cells, including bacteria, fungi, and amebas (8, 9). Of particular interest is the efficacy of AMPs against infections caused by multidrug-resistant Gram-negative bacteria, such as the ESKAPE pathogens (10, 11, 12).

The interaction of AMPs and lipid bilayers is of critical importance to understanding their efficacy against Gram-negative bacteria and for designing engineered peptides as a potential new source of antibiotics. AMP behavior in bilayer environments has been studied using computer simulations (13, 14, 15) and experiments (16, 17) in an effort to understand the interactions between these molecules and how these interactions influence antimicrobial efficacy. The work presented here studies the interactions between AMPs and the lipid A component of the first barrier encountered by AMPs: the bacterial outer membrane (18). Many studies have employed model supported lipid bilayers (SLBs) composed of two-tailed phospholipids to represent the bacterial membrane (7, 19). Although this approach has provided important insights into the mechanisms of cell toxicity, a symmetric phospholipid bilayer is not an accurate model for a Gram-negative bacterial outer membrane. Instead, the actual outer membrane is asymmetric, with the outer leaflet comprising primarily lipid A (and lipid-A-containing lipopolysaccharides), with an inner leaflet composed of a mixture of two-tailed phospholipids, primarily with phosphatidylethanolamine head groups (20, 21).

Lipid A is the primary lipid component of the outer leaflet of Gram-negative bacteria. Also known as endotoxin, this lipid is responsible for septic shock in humans (22) but has medical uses in vaccines as an adjuvant (23, 24). Lipid A has a widely varied structure between bacterial species, consisting of 4–8 hydrophobic tails of varying length and saturation, a sugar head group (typically disaccharide), and 0–2 phosphate groups (25, 26, 27). The phosphate groups are responsible for much of the negative charge associated with the surface of Gram-negative bacteria, particularly in “rough” strains, because of their extremely short or missing O-antigen (28). Because most AMPs are positively charged, it is believed that electrostatic attraction is an important factor in interactions between AMP and lipid A (29, 30, 31). It is also believed that modification of the negatively charged phosphate groups on the lipid A by bacteria results in increased resistance to AMPs (32, 33, 34, 35, 36). Therefore, it is of great interest to study how AMPs interact with this outer-membrane lipid component to understand the role it plays in both AMP resistance and species bias. Hydrophobic interactions also play a significant role in the association between AMP and lipid A as well as the association between AMPs and phospholipid bilayers. In fact, many proposed mechanisms of action for α-helical AMPs rely on the amphiphilic nature of the α-helix to insert themselves into the bilayer, maximizing the exposure of hydrophobic residues to the hydrophobic interior of the bilayer (37, 38).

The two AMPs (LL37 and melittin) used in this work belong to a class of α-helical AMPs that exhibit an amphipathic α-helical structure when interacting with a lipid bilayer but are unstructured in solution (8, 39, 40, 41). Additionally, both peptides are positively charged under physiological pH conditions, with LL37 exhibiting a +5 charge and melittin having a +6 charge, based on their sequences. LL37 is a human peptide sequence that has antimicrobial properties against Gram-negative bacteria but is relatively ineffective against other types of cells (42). Melittin is a peptide found in bee (Apis mellifera) venom that is known to be toxic to Gram-negative bacteria (43), Gram-positive bacteria (44), fungi (45), and even cancer cells (46). Because the two peptides had similar α-helical structures, overall molecular weight, and hydrophobicity, they were selected for these studies to isolate the effects of electrostatic AMP-lipid interactions without confounding these effects with large differences in hydrophobicity and/or molecular conformation. The conformation of AMPs was not directly measured in these experiments, so it is currently unknown whether the AMPs were disordered or helical or whether multiple populations were present when adsorbed on lipid A bilayers. Although previous experiments and simulations have inferred the effects of electrostatic attraction between the negatively charged bilayer and the positively charged AMP (47, 48), the findings presented here provide direct observations of the ways in which subtle changes in electrostatic interactions influence dynamic molecular phenomena.

In this work, we used Langmuir-Blodgett/Langmuir-Schaefer deposition to create asymmetric SLBs such that the lipid A component experiences a similar environment as it does in a Gram-negative bacterial outer membrane both in terms of chemical functionality as well as membrane dynamics (49, 50). This technique allowed for fine control of lipid concentration and provided a stable experimental platform with very little lipid exchange between the leaflets of the bilayer. Furthermore, deposition in this manner allowed for the creation of bilayers that were more uniform than bilayers formed through vesicle fusion (51).

To study the molecular-level interactions between AMPs and the lipid bilayers, we employed total internal reflection fluorescence microscopy to track large numbers of individual molecules. Particularly important to the study of AMP/Lipid A interactions is the ability to determine diffusive behavior and surface residence time; these phenomena are hypothesized to be relevant to AMP self-assembly processes that are critical for proposed mechanisms of AMP antimicrobial activity (52, 53). Single-molecule imaging has been shown to provide highly sensitive quantitative measurements of the interaction between fluorescently labeled proteins and SLBs (54). In addition to measuring mobility and surface retention, single-molecule tracking allows for the detailed analysis of molecular dynamic heterogeneity that would otherwise be missed in ensemble measurements. In particular, anomalous diffusive behaviors, such as intermittent motion, alternating slow/fast motion, or occasional long “flights,” are distinguishable from simple Brownian motion (random walks) using this approach (55). These anomalous diffusive behaviors have a significant impact on the efficacy of molecular searching, which is related to the efficiency of self-assembly (56, 57). Therefore, it is important to quantify these behaviors when relating diffusion and potential self-assembly to the antimicrobial action of these peptides. In this initial work, we explore AMP dynamics under low surface-coverage conditions in which the AMP molecules are isolated, enabling us to characterize the dynamic behavior of noninteracting AMPs. In future research, we hope to study dynamic under crowded-surface conditions where self-assembled structures may be expected to form.

Materials and Methods

Fluorescently labeled peptides

Peptides were synthesized by GenScript (Piscatawa, NJ) with an unnatural azido-lysine amino acid at the C-terminus. LL37 was provided at 96.2% purity, as reported by GenScript, and melittin was provided at 97.3% purity. Both peptides were soluble in water at concentrations exceeding 4 mg/mL. Fluorescent labeling employed copper-free strained alkyne click reaction chemistry (58). Peptides were dissolved in 1× phosphate-buffered saline (PBS) at room temperature at a concentration of 2 mg/mL. Alexa Fluor 488 dibenzocyclooctyne alkyne (ThermoFisher Scientific, Waltham, MA) was dissolved in dimethyl sulfoxide at a concentration of 2.5 mg/mL. The peptide and fluorophore solutions were mixed at a 3:2 molar ratio of peptide to fluorophore and reacted for 24 h at 4°C. The resultant solution was then passed through a reversed-phase chromatography column (SEC 70 10 × 300 mm column; Bio-Rad, Hercules, CA) at 1 mL/min using PBS as the mobile phase. 1 mL fractions were collected, and fluorophore concentration was measured using a fluorometer (Fluoromax-4 Spectrofluorometer; Horiba Scientific, Irvine, CA). For all experiments, the first fluorescent fraction obtained from the chromatography column was used. Conjugation was further verified through control experiments at equivalent concentrations of unconjugated fluorophore in which no fluorescent molecules were observed to interact with a lipid A bilayer, likely because of electrostatic repulsion from the negative charge on the fluorophore. Representative videos have been included for the fluorophore-only control (Video S1) and melittin on monophosphoryl lipid A (Video S2). The included videos are displayed at a frame rate of 30 frames per second. Peptide/fluorophore conjugates were used in single-molecule experiments at a concentration of 2 × 10⁻¹⁰ M in PBS.

Lipid bilayers

Monophosphoryl lipid A and diphosphoryl lipid A, both from Escherichia coli F583 (Rd mutant), were purchased from Millipore Sigma (St. Louis, MO) and dissolved to a concentration of 1 mg/mL in a 73:24:3 (v:v) mixture of chloroform:methanol:water. 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine (DOPE) (Avanti Polar Lipids, Alabaster, AL) was dissolved in chloroform at a concentration of 1 mg/mL. The structures of these molecules are shown in Fig. S1.

Pieces of 25 × 25 mm #1 cover glass (ThermoFisher Scientific) were cleaned with hot piranha solution (30% hydrogen peroxide/70% sulfuric acid by volume) at 50°C for 1 h, followed by thorough rinsing in Milli-Q water (Millipore) and drying under ultrapure nitrogen. The dry wafers were then placed in an ultraviolet (UV)-ozone chamber (PSD series Digital UV Ozone System; Novascan, Ames, IA) for 1 h. Cover glasses cleaned in this way were used within 20 min of removal from the UV-ozone chamber for the deposition of SLBs.

Lipid bilayers were formed using a Langmuir-Blodgett/Langmuir-Schaefer (LB/LS) deposition technique using a NIMA Langmuir trough (KSV NIMA, Espoo, Finland). The LB/LS deposition was adapted from the process that was utilized successfully by Clifton et al (49, 50). Lipids were deposited to the air/water interface by adding 20 μL of lipid A solution or 30 μL of DOPE solution in 2 μL increments. The expanded surface area of the Langmuir trough was 280 cm². The lipids were then compressed at a rate of 20 cm²/min to a surface pressure of 27 mN/m measured via Wilhelmy plate and maintained at room temperature (∼20 ± 1°C). The resultant monolayers formed the basis of the LB/LS deposition, where the inner leaflet of the SLBs was deposited using LB deposition, and the outer leaflet was deposited using LS deposition (see Fig. 1).

Figure 1.

Schematic of Langmuir-Blodgett/Langmuir-Schaefer deposition, as described in the text. To see this figure in color, go online.

To deposit the inner leaflet of the SLB, a clean coverslip was submerged into the Langmuir trough before the addition of any lipids to the interface. Lipids used to form the inner leaflet of the asymmetric bilayer were then deposited at the air/water interface, as described above. The coverslip was then drawn through the lipid monolayer at the air-water interface at a rate of 3 mm/min, depositing a layer of lipid with the head group toward the glass substrate (the inner leaflet). During the deposition of the inner leaflet, the surface pressure was held at a constant 27 mN/m through a feedback loop, compressing the monolayer to maintain a constant pressure as lipids were deposited onto the substrate. The surface pressure of 27 mN/m was chosen because it is above the surface pressure associated with the apparent high-pressure phase transition observed during lipid A compression but well below the observed monolayer collapse pressure for both lipid A and DOPE (∼43 mN/m in both cases). These surface pressures yielded monolayers with estimated areas per molecule of 0.55 nm² for DOPE and 1.2 nm² for both lipid A species in the predeposition monolayer. Previously published results suggest that this is a representative molecular area (and hence, surface pressure) for lipids in biological membranes; moreover, lipid A at a similar value of surface pressure has been used as a model to study interactions between AMP and lipid A at the air-water interface (59, 60). The precise surface pressure chosen is not expected to significantly affect the findings of this study because in the 26–40 mN/m range, both types of lipids do not experience significant phase changes or changes in packing density. Thus, the lipid at the air-water interface is expected to be in a uniform and compressed monolayer state at this surface pressure. The lipid A and DOPE Langmuir isotherms are shown in Fig. S2.

After deposition of the inner leaflet, the remaining lipid monolayer at the air-water interface of the Langmuir trough was removed through pipette aspiration, and removal was verified through measurement of the surface pressure as a function of compression of the now-cleaned air-water interface. The outer leaflet lipid was then deposited and compressed using the Langmuir trough, as described above. The substrate that had already undergone inner leaflet deposition was then rotated to be parallel to the plane of the air/water interface and lowered onto the freshly formed monolayer forming an SLB and passing into the aqueous phase. The resulting SLB was kept submerged in 1× PBS because bilayer disruption is caused by any exposure to air.

Fluorescence recovery after photobleaching

Bilayers were created as above with the addition of 0.5 mol% lissamine rhodamine B head group labeled DOPE (Avanti Polar Lipids). Fluorescence recovery after photobleaching (FRAP) was performed using a custom-built laser focusing device mounted above the microscope sample stage. A 532 nm laser light (Cobolt Samba 50 mW; Cobolt, Solna, Sweden) was directed onto the fluorescent bilayer in a ∼10 μm radius circle at an intensity of ∼13 kW/cm² for 1 s. This intensity and exposure was sufficient to photobleach the illuminated area. Time-lapse images were acquired of the photobleached region at room temperature (∼20 ± 1°C), and the intensity within the photobleached region was measured using ImageJ (software available through National Institutes of Health). The resulting intensity versus time measurements were used to calculate apparent diffusion coefficients and mobile fraction based on the fluorescence recovery over time within the photobleached region. The bleached spot was approximated as circular, and the recovery curves were fitted using the Soumpasis method described by Eq. 1 (61):

$$f\left(t\right)=A{e}^{-\frac{2\tau }{t}}\left[I_0\left(\frac{2\tau }{t}\right)+I_1\left(\frac{2\tau }{t}\right)\right],$$ (1)

where I₀ and I₁ are modified Bessel functions of the first kind of order 0 and 1, respectively, and A is the mobile fraction. The diffusion coefficient was calculated using the formula, D = r²/4τ, where r is the radius of the experimentally measured bleached area, and τ is the fitted recovery time parameter. The recovery curves and fits are shown in Fig. S3.

Total internal reflection fluorescence microscopy single-molecule tracking

Time series of images were captured on an inverted Nikon Eclipse Ti Microscope with a Plan-Apo 100× 1.45 NA oil immersion objective (Nikon, Melville, NY) using a Hamamatsu complementary metal oxide semiconductor camera at 10 frames per second (100 ms exposure time). Excitation light was passed through a motorized total internal reflection illuminator (TI-TIRF-EM; Nikon) to adjust the angle of incidence on the sample to generate an evanescent field at the SLB-water interface. Excitation light for the single-molecule experiments was generated by a 100 mW 491 nm solid-state laser (Cobolt Calypso 100; Cobolt) and calculated to have an intensity incident on the surface of ∼0.75 kW/cm². Individual molecules appeared as diffraction-limited spots in these images. All total internal reflection fluorescence experiments were performed at room temperature (∼20 ± 1°C).

Single-molecule tracking was performed using a custom Mathematica script to identify diffraction-limited spots (fluorescent molecules) and follow them frame-to-frame, as described previously (62, 63). A tracking radius was determined to minimize the number of false linkages in trajectories and maximize the amount of distance a molecule can diffuse between frames before being considered a new molecule. A false linkage would be a trajectory in which a molecule desorbed on frame n and a separate molecule adsorbed on frame n + 1 within the tracking radius. For these experiments, approximately fewer than 1 in 3000 steps was a false linkage due to the sparse coverage of molecules on the surface (see Supporting Materials and Methods for a detailed calculation; Eq. S1). Diffusive steps larger than the tracking radius were considered to be desorption/adsorption events rather than diffusive steps. For the experiments on lipid A/DOPE asymmetric bilayers, a tracking radius of 0.42 μm was set to limit false connections between two separate molecules. Because of the faster motion in symmetric DOPE bilayers, a larger tracking radius of 1.96 μm was required as well as a lower concentration of fluorescent molecules (2 × 10⁻¹¹ M) to minimize the probability of false linkages. For all conditions, a minimum of 50,000 molecular trajectories in each of three separate trials were used in residence time, diffusion, and waiting-time calculations. A molecule’s residence time was calculated as the number of frames during which molecule was visible multiplied by the acquisition time. Diffusive steps (displacements) were calculated as the distance between the centers of intensity for a molecule from one frame to the next.

Statistical analysis of molecular trajectories

Complementary cumulative residence time distributions were generated by calculating the fraction of molecules that remained on the surface t seconds after their initial adsorption. These distributions are conventionally fitted to an exponential mixture model, in which a single exponential would represent first-order desorption kinetics and additional exponentials would represent additional molecular populations. For these experiments, a sum of two exponentials (Eq. 2) was appropriate to fit the residence time distribution as follows:

$$p\left(t\right)=f_1{e}^{-t/t_1}+\left(1-f_1\right)e^{-t/t_2},$$ (2)

where f₁ represented the short residence time population fraction, (1−f₁) was the long residence time population fraction, and t₁ and t₂ represented the short- and long-lived fraction characteristic time constants, respectively. Average residence times (<RT>) were calculated as the weighted average of the characteristic time constants, i.e., <RT> = f₁t₁ + (1–f₁)t₂.

Diffusive steps were observed as changes in the position of the center of intensity for a given molecule from one frame to the next. Complementary cumulative squared step-size distributions were calculated by sorting the squared displacement values in ascending order and ranking at each point where the probability of a step-size R² or longer is given by Eq. 3, as follows:

$$\text{C}\left({\text{R}}_k^2,\mathit{\Delta }\text{t}\right)=1-k/N,$$ (3)

where k is the rank of each step, and N is the total number of steps.

These complementary cumulative squared step-size distributions were fitted to a mixture of two Gaussian distributions given by Eq. 4, as follows:

$$\text{C}\left(R^2,\mathit{\Delta }\text{t}\right)=m_1{e}^{-R^2/4D_1\mathit{\Delta }t}+\left(1-m_1\right)e^{-R^2/4D_2\mathit{\Delta }t},$$ (4)

where m₁ is the fraction of steps with characteristic diffusion coefficient D₁, and (1 − m₁) is the fraction of steps with a characteristic diffusion coefficient of D₂. Δt is the time between frames, which was 0.1 s for these experiments.

As for residence times, the average diffusion coefficient was calculated as the weighted average of the two diffusion coefficients (D₁ and D₂), weighted by the relative fraction of steps in each diffusive mode. For diffusion calculations, the fraction of steps was used rather than the fraction of molecules because a single-molecular trajectory could exhibit both short and long diffusive steps during different parts of the trajectory.

For both residence time and diffusion steps, cumulative distributions are preferable because experimental data can be presented without artifacts from binning. Average value calculations were performed using the fits rather than measured values because of artifacts such as the finite minimal observable residence time as well as the diffusive step length maximum. For these experiments, the difference between the measured average values and model-determined average values was negligible.

Surface coverage (θ₀) is defined as the steady-state number of adsorbed molecules per unit area. This value was determined simply by counting the number of fluorescent molecules per frame and dividing by the area of the field of view. Assuming a proportional dependence of surface coverage on bulk concentration (appropriate in the low coverage limit), the surface coverage in mass/area units can be estimated by Eq. 5, where [SM] is the bulk concentration of fluorescent molecules used to determine θ₀, MW_AMP is the mass per molecule of the AMP, and [bulk] is the concentration at which one wishes to estimate a new surface coverage:

$${\theta }_1=\frac{{\theta }_0\ast \left[bulk\right]\ast M{W}_{AMP}}{\left[SM\right]}.$$ (5)

Waiting-time distributions were calculated as the time between diffusive hops with lengths longer than a predetermined cutoff length. For the purposes of these experiments, a step length of 200 nm was chosen as the cutoff between short diffusive steps and long flights. This 200 nm step size was significantly larger than the experimental localization uncertainty (∼25 nm) and more than twice as large as the expected diffusive step size for a molecule inserted into the bilayer (∼90 nm in 100 ms). Only waiting times during which a molecule had one or more short (<200 nm) diffusive steps followed by a long (>200 nm) diffusive step were included in these calculations. A stretched exponential distribution (Eq. 6, where A is the preexponential factor, τ is the characteristic relaxation time, and β is the stretching exponent) was used to fit the waiting-time distributions as follows:

$$P\left(t\right)=A\ast e^{-{\left(\frac{t}{\tau }\right)}^{\beta }}.$$ (6)

That the data were well described by a stretched exponential is understandable given that stretched exponential functions are conventionally used to represent dynamics in situations with a heterogeneous energy barrier landscape (64).

The first moment of this equation is often termed the mean relaxation time, <τ>. The first moment is given by Eq. 7, where τ is the characteristic relaxation time, β is the stretching exponent, and Γ is the gamma function:

$$\langle \tau \rangle =\frac{\tau }{\beta }\mathit{\Gamma }\left(\frac{1}{\beta }\right).$$ (7)

Results

Hybrid supported bilayers with DOPE inner leaflets and outer leaflets composed of monophosphoryl or diphosphoryl lipid A (from E. coli) containing 0.5 wt% liss-rhodamine labeled DOPE were prepared using LB/LS deposition and characterized with FRAP and fluorescence imaging to ensure bilayer quality. Based on the analysis of FRAP data, the immobile fraction of lipids in the outer leaflets of these films was calculated to be 0.04 ± 0.01, which is in line with other reports using this method of bilayer formation (65). The apparent mean diffusion coefficients for lipids within these bilayers, as measured using FRAP, were 0.023 ± 0.006 and 0.022 ± 0.005 μm²/s for diphosphoryl and monophosphoryl lipid A asymmetric bilayers and 1.9 ± 0.2 μm²/s for symmetric DOPE bilayers. Fig. S3 shows the FRAP recovery curves, and Table S2 shows the fit parameters for these measurements.

AMPs were added to the PBS (pH 7.4) buffer in contact with nonfluorescent SLBs at a concentration of ∼2 × 10⁻¹⁰ M and then tracked using the single-molecule tracking method described above. As described above, the spatiotemporal trajectories of the peptides were analyzed to characterize the motion of the molecules and determine the overall surface residence time of each molecule.

Complementary cumulative residence time probability distributions for the AMP molecules are shown in Fig. 2. In this semilogarithmic format, first-order desorption kinetics would appear as a straight line, and processes involving multiple independent modes would be modeled using an exponential mixture. For these experiments, two exponentials were required to fit the data (solid lines in Fig. 2), which suggested that, instead of a single desorption rate, there were two distinct populations of molecules at the lipid A/aqueous interface.

Figure 2.

Complementary cumulative probability distributions of residence times for AMPs on lipid A bilayers. The lines represent the best fits to the sum of two exponentials. To see this figure in color, go online.

The population fractions and characteristic residence times for each fraction, as defined in Eq. 2, are shown in Table 1. The short-lived fraction exhibited characteristic residence times in the range of 0.1–0.2 s and comprised the majority of adsorbed molecules (typically 70–75%). The characteristic residence times associated with the longer-lived population were generally in the range of 0.5–0.9 s. This desorption behavior, in which there were two characteristic time constants and a majority of molecules exhibited the shorter time constant, was common to all AMPs tested on all lipid A surfaces. Given the extremely low concentrations used and the fact that no statistically significant differences in fluorescence intensity were observed between the observed populations, the molecules in these experiments were assumed to be monomeric, and the kinetic behavior was not due to populations associated with the aggregation state. Photobleaching was not an important factor in these calculations because the characteristic time constant for bleaching was on the order of hundreds of seconds under these conditions (66). The most notable feature that emerged from this population analysis was the fact that the characteristic timescale, t₂, for the longer-lived population was significantly longer on bilayers containing diphosphoryl lipid A than on bilayers containing monophosphoryl lipid A. Also, the longer-lived population exhibited longer characteristic residence times for LL37 than for melittin on both lipid A surfaces. The differences in timescales associated with the short-lived populations were generally insignificant.

Table 1.

Parameters for Double-Exponential Fits to Residence-Time Distributions

	f1	t1 (s)	t2 (s)
LL37-di	0.69(2)	0.11(1)	0.88(3)
LL37-mono	0.73(2)	0.105(7)	0.73(3)
Melittin-di	0.75(4)	0.17(1)	0.60(4)
Melittin-mono	0.76(5)	0.15(1)	0.47(3)

f₁ is the fraction associated with the shorter characteristic time t_1. The remaining fraction is associated with residence time t_2. The numbers in parentheses represent uncertainty in the least significant digit (95% confidence interval).

These surface residence time data were analyzed to determine the average surface residence times, which were generally in the range of 0.3–0.4 s. Interestingly, both AMP species exhibited significantly different average residence times on the two lipid A bilayers (p < 0.015 and <0.1 for LL37 and melittin, respectively), as shown in Fig. 3 (and Table S1). Specifically, the average residence times of both peptides were determined to be ∼20% longer on bilayers with diphosphoryl lipid A in the outer leaflet compared with bilayers containing monophosphoryl lipid A, again consistent with stronger electrostatic interactions between the AMPs and the diphosphoryl head groups. It was not possible to determine AMP residence times on symmetric DOPE bilayers because of the difficulties associated with reliably tracking quickly moving molecules for an extended period of time without the occasional false “breakage” of a trajectory. Qualitatively, the molecular trajectories on DOPE had a similar characteristic timescale as that of the trajectories on lipid A bilayers (<1 s). This difficulty in tracking AMPs on DOPE bilayers did not significantly affect our ability to characterize diffusion, however.

Figure 3.

Average residence times for fluorescently labeled LL37 and melittin on diphosphoryl E. coli lipid A/DOPE asymmetric bilayers (blue) and monophosphoryl asymmetric bilayers (orange). The error bars represent the SD for three separate experimental trials. To see this figure in color, go online.

In addition to residence times, the overall surface affinity for each peptide was determined by calculating the average number of molecules per unit area on the surface. This measurement allowed for the extrapolation of the measured surface coverage at single-molecule concentrations to more physiologically relevant concentrations, as described by Eq. 5. In particular, the measured affinity for AMPs to lipid A in our measurements extrapolates to surface-coverage values of ∼0.06 and 0.025 mg/m² for LL37 and melittin, respectively (this represents ∼0.4% surface coverage by area), at a concentration of 5 μM, which is approximately the minimal inhibitory concentration for both peptides on E. coli (67, 68).

As a measure of peptide mobility, the complementary cumulative distributions of squared displacements were analyzed as described above to determine the average apparent diffusion coefficient of fluorescently labeled LL37 and melittin peptides in contact with the bilayers. The full set of complementary cumulative step-size distributions and statistical fits are shown in Fig. S4. In general, when in contact with outer leaflets containing lipid A, the peptides exhibited apparent diffusion coefficients on the order of ∼0.1 μm²/s, with diffusion constants more than an order of magnitude faster on symmetric DOPE bilayers. The difference in magnitude of the diffusion coefficient was a direct consequence of the increased fluidity of the DOPE bilayer. The specific values are shown in Fig. 4 (and Table S3) along with the apparent diffusion coefficients of outer-leaflet lipids in those bilayers, as measured by single-molecule trajectories of liss-rhodamine labeled DOPE inserted into the layer during deposition. Importantly, for both peptides, the diffusion was significantly faster (by ∼20%) on the bilayers with outer leaflets containing monophosphoryl lipid A compared to the bilayers with diphosphoryl lipid A (p < 0.005 for both AMPs), again suggesting stronger interactions between the cationic AMPs and the more highly charged diphosphoryl head groups. Interestingly, the observed diffusion coefficients for AMPs were also much faster than those observed for in-plane lipid diffusion in both FRAP and single-molecule observations on bilayers containing lipid A. This suggested that the molecular motion of the AMPs was not satisfactorily explained by the fluidity of the bilayer. For example, the Saffman-Delbruck model, which suggests that protein diffusion in bilayers is dominated by bilayer drag, is relevant for the DOPE bilayers in which in-plane diffusion is the dominant regime. However, this model does not consider the possibility of desorption and readsorption (i.e., “flights”), which dominate surface mass transport as the interfacial viscosity becomes large (69). This transition to desorption-mediated diffusion at high liquid-liquid interfacial viscosity has also been observed at the oil-water interface (70).

Figure 4.

Average diffusion coefficients determined from single-molecule tracking for fluorescent DOPE inserted into the outer leaflet and peptides adsorbed to diphosphoryl lipid A/DOPE asymmetric bilayers (blue), monophosphoryl asymmetric bilayers (orange), and DOPE symmetric bilayers (green). The error bars represent the SD for three separate experimental trials. To see this figure in color, go online.

In addition, we analyzed the statistics of trajectories in detail to obtain mechanistic information about the nature of the interfacial diffusion. Of particular interest was the question of whether peptides and lipids exhibited trajectories consistent with two-dimensional (2D) Brownian motion or whether the diffusion was anomalous. Fig. 5 A shows representative step-size distributions (i.e., the self-part of the van Hove correlation function), specifically the projection of diffusive step lengths along the x axis. In this format, simple Brownian motion appears as a Gaussian distribution, whereas non-Brownian motion (or a mixture of diffusive modes) is non-Gaussian. This presentation also allows for the observation of heavy tails, which are indicative of three-dimensional mediated flights (55). Fig. 5 B shows a representative subset of molecular trajectories for melittin on DOPE bilayers and monophosphoryl lipid A asymmetric bilayers. These trajectories illustrate the non-Brownian nature of steps on lipid A compared to those on DOPE. Some trajectories remain within a very localized area, whereas others undergo intermittent large motions (flights). Also evident is the difference in diffusive length scales between DOPE and lipid A.

Figure 5.

(A) Probability density distributions of diffusive steps along the x axis for melittin on DOPE (magenta triangles) and monophosphoryl lipid A (black squares). The lines represent fits of the central peaks to a single Gaussian distribution. (B) Representative trajectories for melittin on DOPE (magenta) and monophosphoryl lipid A (black) are shown. To see this figure in color, go online.

AMPs on a DOPE SLB exhibited mostly 2D Brownian motion, as indicated by the excellent fit of a single Gaussian function to the distribution for melittin in Fig. 5 A. However, the trajectories of AMPs on bilayers containing lipid A were not simply Brownian, and the distributions were not well described by single Gaussian functions, as indicated by the representative step-size distribution for melittin on a bilayer containing monophosphoryl lipid A shown in Fig. 5 A. In particular, these distributions exhibited a narrow central Gaussian peak, with heavy tails that could be asymptotically represented with an exponential or power law decay (55, 71). Step-size distributions for all adsorbates on all bilayers are shown in Fig. S5. Qualitatively similar behavior has previously been observed for the diffusion of various molecular adsorbates at solid/liquid interfaces (52, 55) and suggests that AMP molecules interacting with lipid A SLBs exhibited intermittent diffusive motion, in which trajectories alternated between slow Brownian “crawling” motion and long, nearly-instantaneous “flights,” as shown by the representative lipid A trajectories (black) in Fig. 5 B. In previous work, this particular type of motion was observed on both solid surfaces and highly viscous liquid-liquid interfaces and interpreted as the alternation of 2D diffusion- and desorption-mediated “hopping” through the adjacent liquid (55). This “hopping” mechanism of interfacial mass transport was recently confirmed explicitly using three-dimensional single-molecule tracking of human serum albumin at the solid-liquid interface (72). Such motion has been quantitatively described using a generalized continuous time random walk model, which comprises variable-length waiting times interspersed with variable-length flights. The waiting times and flight lengths are assumed to be stochastic and associated with statistical distributions that are often heavy-tailed (73, 74).

In the context of the continuous time random walk model, the distribution of waiting times can be determined empirically, as shown in Fig. 6 A. These distributions included only waiting times longer than 0.2 s because of the susceptibility of the first time point to experimental artifacts. When included, the first time point does not significantly change the observed trends, but the apparent quality of the fit is reduced. The stretched exponential fits to these waiting time distributions are shown as solid lines through the data points. For these data, all conditions exhibited essentially identical stretched exponential exponents (β) in the range of 0.33–0.35, whereas the characteristic relaxation time constant <τ> was strongly dependent on the identity of the bilayer. The exact values for these parameters are shown in Table S4. For each of these fits, the mean relaxation time, <τ>, was determined by calculating the first moment of the fitting function. The magnitude of the mean relaxation time is the theoretical mean for the overall waiting-time distribution and not the observed mean waiting time. The theoretical value was used because the observed mean waiting time is limited by the image acquisition time. Fig. 6 B shows the values of the mean relaxation time, as calculated via Eq. 6. Importantly, both peptides exhibited mean relaxation times that were significantly longer on diphosphoryl lipid A asymmetric SLBs than on monophosphoryl lipid A asymmetric SLBs (p < 0.005 for both AMPs), which was again consistent with stronger electrostatic interactions between the cationic AMPs and the SLBs with greater negative charge.

Figure 6.

(A) Waiting-time probability distribution and (B) characteristic timescale for AMPs on asymmetric lipid A SLBs. The waiting-time probabilities are offset by factors of 10 for clarity, with the LL37-diphosphoryl lipid A series having no offset. The lines are best fits of a stretched exponential to the data. To see this figure in color, go online.

Discussion

The results discussed above represented an advance in experimental capabilities for studying AMP/lipid A interactions and also showed promise for understanding the mechanisms by which AMPs act on biological membranes. Notably, individual AMP molecules were imaged and tracked on a bilayer that contains lipid A in a similar environment to that of the lipid A found in the Gram-negative bacterial outer membrane. This advance allowed for model-independent characterization of molecular residence times, diffusive mechanisms, and apparent diffusion coefficients.

Although studying AMP interactions with phospholipid membranes provides essential insights into the mechanisms of cell toxicity, this work demonstrates the utility of using membranes that comprise asymmetric layers of lipid A and a phospholipid like the bilayers of the outer membrane of Gram-negative bacteria when studying the initial interaction of AMPs with the cell. The diffusive dynamics were much slower compared to conventional phospholipid bilayers, and there was also a qualitative difference in the diffusive mechanism observed between DOPE and the lipid A bilayers. Furthermore, it was shown that the effective diffusion coefficient for AMP on asymmetric lipid A SLBs (∼0.1 μm²/s) was significantly faster than the observed diffusion of lipids within the cell membrane (∼0.02 μm²/s) (75), which indicated that simple models for AMP diffusion are not sufficient when discussing their interaction with lipid A.

Diffusion is of critical importance to many molecular interactions, particularly with respect to searching mechanisms. It is well known that an efficient method of target searching involves slow, methodical searching of a small area and then long-distance jumps to another area to be searched (57). This mechanism was modeled well by the behavior exhibited by AMPs on lipid A but absent in AMP diffusion on DOPE. Studying these behaviors can yield insight into how AMP molecules locate and interact with both each other and for hypothetical defects or heterogeneities at biological interfaces.

In addition to the differences observed between phospholipid bilayers and lipid A asymmetric bilayers, dynamic single-molecule tracking has proven to be capable of detecting differences in AMP interactions with two subtly different lipid A moieties. Both melittin and LL-37 showed longer residence times and slower diffusion on diphosphoryl E. coli lipid A than on monophosphoryl E. coli lipid A. These two lipids differ only by a single phosphate group, but both AMPs exhibited a ∼20% longer residence time and ∼15% slower diffusion on the diphosphoryl lipid A bilayer. This demonstrated the effectiveness of single-molecule microscopy in its ability to study molecular behavior with enough precision to potentially reveal the changes in AMP interactions with surfaces that contain lipid A from different species of bacteria as well as strains that are resistant to the antimicrobial effects of the AMPs. All of these molecular behaviors are expected to be related to differences in lipid A structure among species and to lipid A modifications associated with AMP resistance. Therefore, analyzing the molecular dynamics on the outer membrane lipid A is of critical importance to understanding these phenomena.

Conclusions

The results presented here demonstrate that AMPs exhibit behavior at the molecular level that is distinctive to the identity of the substrate with which they are interacting. A more negatively charged surface increases the residence times of the positively charged peptides and also slows their diffusion. These data also suggested that hybrid bilayers containing lipid A offer a superior model to simple DOPE bilayers for studying the interactions between AMPs and membranes containing lipid A; these interactions likely influence the observed macroscopic phenomena of AMP species bias and bacterial resistance to AMPs. The dynamics were much slower on the hybrid bilayers, and the peptides actually exhibited fundamentally different diffusive mechanisms on bilayers containing lipid A compared to DOPE bilayers. These findings lay the groundwork for further work aimed at characterizing the impact of various types of phosphate modification on interactions between lipid A and AMP (and between AMPs in the lipid A bilayer environment), enhancing understanding of the underlying mechanisms of AMP activity and of bacterial resistance.

Author Contributions

N.N. performed the experiments and prepared the figures. N.N. and D.K.S. analyzed the data and contributed to writing the manuscript.

Editor: Arne Gericke.