Melittin-Induced Permeabilization, Re-sealing, and Re-permeabilization of E. coli Membranes

Abstract

The permeabilization of model lipid bilayers by cationic peptides has been studied extensively over decades, with the bee-sting toxin melittin perhaps serving as the canonical example. However, the relevance of these studies to the permeabilization of real bacterial membranes by antimicrobial peptides remains uncertain. Here, we employ single-cell fluorescence microscopy in a detailed study of the interactions of melittin with the outer membrane (OM) and the cytoplasmic membrane (CM) of live Escherichia coli. Using periplasmic green fluorescent protein (GFP) as a probe, we find that melittin at twice the minimum inhibitory concentration first induces abrupt cell shrinkage and permeabilization of the OM to GFP. Within ∼4 s of OM permeabilization, the CM invaginates to form inward facing “periplasmic bubbles.” Seconds later the bubbles begin to leak periplasmic GFP into the cytoplasm. Permeabilization is localized, consistent with possible formation of toroidal pores. Within ∼20 s, first the OM and then the CM re-seals to GFP. Some 2–20 min later, both CM and OM are re-permeabilized to GFP. We invoke a mechanism based on curvature stress concepts derived from model bilayer studies. The permeabilization and re-sealing events involve sequential, time-dependent build-up of melittin density within the outer and inner leaflets of each bilayer. We also propose a mechanical explanation for the early cell shrinkage event induced by melittin and a variety of other cationic peptides. As peptides gain access to the periplasm, they bind to the anionic peptido-crosslinks of the lipopolysaccharide layer, increasing its longitudinal elastic modulus. The cell wall shrinks because it can withstand the same turgor pressure with smaller overall extension. Shrinkage in turn induces invagination of the CM, preserving its surface area. We conclude by comparing the behavior of different peptides.

Introduction

Melittin, a toxic component of bee venom, is a 26-residue cationic peptide of +6 charge that can lyse eukaryotic cells and kill bacterial cells (1). In the crystal structure of melittin, the peptide adopts a bent, amphipathic helical structure. Residues 1–10 and 13–26 form helices whose axes lie 120° apart due to the presence of Pro-14 (2). Melittin has served as an important model for understanding the mechanisms by which cationic antimicrobial peptides (AMPs) permeabilize lipid vesicles. Previous work includes studies of small, large, and giant unilamellar vesicles (SUVs, LUVs, and GUVs), both zwitterionic and anionic (3, 4, 5, 6, 7). At peptide/lipid (P/L) ratios of ∼1:100 or lower, melittin induces permeabilization sites (usually interpreted as pores) that pass small ions and organic dye molecules (5, 6, 8). Even after a long observation time, not all of the dye is released (9), consistent with formation of transient pores with lifetimes on the order of 10 ms (3). At P/L ratios of ∼1:50 or higher, melittin evidently induces permanent equilibrium permeabilization sites in GUVs (7). Oriented circular dichroism of melittin interacting with hydrated multilayers shows that at these higher concentrations, most melittin molecules change orientation from parallel to perpendicular to the bilayer plane (7). This is consistent with formation of toroidal pores, although other permeabilization models are possible. Accordingly, neutron diffraction experiments on multilayers at high P/L ratios have detected D₂O-filled pores of 4.4 nm inside diameter (10). Most recently, stable permeabilization sites have been induced in cytoplasmic membrane (CM) spheroplasts derived from Escherichia coli (11).

The relevance of studies of AMP interactions with model lipid bilayers to the mechanisms by which the same peptides kill real bacterial cells remains uncertain (12). In an effort to provide more information, we and others have used single-cell, time-resolved fluorescence microscopy to directly observe membrane permeabilization events in live bacterial cells (13, 14, 15, 16, 17, 18, 19, 20). Here, we present a detailed study of melittin interactions with the membranes of live E. coli. Surprisingly, using GFP as the probe species, we observe sequential permeabilization, re-sealing, and re-permeabilization of both the outer membrane (OM) and the CM. Seconds after the initiation of OM permeabilization, we detect short-lived, inward facing invaginations in the CM. For septating cells, these “periplasmic bubbles” evidently provide the conduit through which GFP, the dye Sytox Orange, and presumably melittin itself, gain access to the cytoplasm. We suggest that when melittin crosses the OM and enters the periplasm, it binds to anionic cross-links of the peptidoglycan (PG) layer (21). This causes the cell wall to stiffen and contract. Cell-wall shrinkage in turn forces the CM to shrink in length by invagination. Our proposed mechanism for the permeabilization and re-sealing events invokes the same curvature stress arguments originally used to explain AMP permeabilization and re-sealing of lipid vesicles (5, 7, 22).

Materials and Methods

We have developed a variety of real-time, single-cell imaging protocols that reveal the sequence of events in the attack of AMPs on live E. coli at an unusual level of detail. The protocols differ in imaging modality, labeling methods, and camera frame rate. In all experiments reported here, cells are plated in a microfluidics chamber and are growing in a continuous flow of aerated, 30°C EZ-rich defined medium (EZRDM). The flow of 10 μM melittin (twice the aerobic minimal inhibitory concentration (MIC)) in EZRDM begins at t = 0 and continues throughout the duration of the imaging experiment. This maintains a constant peptide concentration in the cell surround.

Bacterial strains, materials, and growth conditions

The background (“WT”) strain is MG1655 (K12) in all cases. Experiments on periplasmic GFP used strain JCW10, in which TorA-GFP is expressed from plasmid pJW1 as previously described (23). TorA-GFP is transported to the periplasm by the twin-arginine transport system and the TorA signal peptide is cleaved, leaving free GFP in the periplasm. Under our induction conditions, at least 90% of the green fluorescence comes from GFP that has been transported to the periplasm. This was shown in studies of LL-37 (20) and Cecropin A (18), both of which induced loss of at least 90% of the total green fluorescence after permeabilization of the OM and before permeabilization of the CM to Sytox dye. Efficient export of GFP to the periplasm is corroborated in this work by inspection of transverse intensity linescans, which exhibit two distinct peaks (see below). In the strain JCW1, cytoplasmic GFPmut2 was expressed using the lac promoter on the plasmid pMGS053, as previously described (24).

Melittin was purchased from Sigma-Aldrich (St. Louis, MO; cat. no. M2272). LL-37 was purchased from Anaspec (Fremont, CA; cat. no. 61302). The hybrid synthetic peptide CM15 with C-terminal amidation was received from Jimmy Feix (Medical College of Wisconsin, Milwaukee, WI). The DNA stains Sytox Green (cat. no. S7020) and Sytox Orange (cat. no. S11368), and the membrane dye FM4-64 (cat. no. T13320) were purchased from Thermo-Fisher Scientific (Waltham, MA). The Sytox dyes are essentially non-fluorescent in solution phase but exhibit strong fluorescence on binding to chromosomal DNA.

Bulk cultures were grown in EZRDM (25), which is a MOPS-buffered solution at pH 7.4 supplemented with metal ions (cat. no. M2130, Teknova, Hollister, CA), glucose (2 mg/mL), amino acids, and vitamins (cat. no. M2104, Teknova), nitrogenous bases (cat. no. M2103, Teknova), 1.32 mM K₂HPO₄, and 76 mM NaCl. Cultures were grown from glycerol frozen stock to stationary phase overnight at 30°C. Subcultures were grown to exponential phase (OD = 0.2–0.6 at 600 nm) at 30°C before sampling for the microscopy experiments.

MIC assay

The aerobic MIC values for the various AMPs (Table 1) were determined using the broth microdilution method as previously described (20). Twofold serial dilutions of melittin in EZRDM were performed in separate rows of a polystyrene 96-well plate, with each plate containing an inoculum of E. coli MG1655. The inoculum was a 1:20 dilution from a bulk culture at midlog phase (OD₆₀₀ = 0.5) grown at 30°C. The plate was incubated at 30°C and shaken at 200 Rpm in a Lab-Line Orbital Environ shaker (model 3527, Lab-Line Instruments, Melrose Park, IL) for 6 h. The MIC value was taken as the lowest concentration for which no growth was discernible (OD₆₀₀ < 0.05) after 6 h.

Table 1.

Antimicrobial Agents Compared in This Work

Antimicrobial Agent	Sequence	Net Charge	MIC (μM)a
LL-37	LLGDFFRKSKEKIGKEFKRI-VQRIKDFLRNLVPRTES	+6	4
Cecropin A	KWKLFKKIEKVGQNIRDGII-KAGPAVAVVGQATQIAK-NH2	+7	0.9
Melittin	GIGAVLKVLTTGLPALISWI-KRKRQQ-NH2	+6	5
CM15	KWKLFKKIGAVLKVL-NH2	+6	5
β-peptide copolymer MM63:CHx37		63% cationic side chains	25 μg/mL

^aMinimum inhibitory concentration (in micromolar) over 6-h period in aerated EZRDM medium at 30°C, determined by OD for successive twofold dilutions in 96-well plates. Copolymer MM₆₃:CHx₃₇ lacks a defined molar mass, so MIC is in micrograms per milliliter.

Microscopy

As previously described (17), imaging of individual cells was carried out at 30°C in a microfluidics chamber consisting of a single rectilinear channel of uniform height of 50 μm and width of 6 mm, with a channel length of 11 mm. The total chamber volume is ∼10 μL. After bonding of the PDMS chamber to the glass coverslip, 0.01% poly-L-lysine (molecular weight >150,000 Da) was injected through the chamber for 30 min and rinsed thoroughly with Millipore water. E. coli cells are immobilized on the coverslip but grow normally. During imaging experiments, the chamber was maintained at 30°C with an automatic temperature controller.

Single-cell imaging was performed on two different microscopes, a Nikon TE300 inverted microscope with a 100×, 1.3 NA phase-contrast objective and a Nikon Eclipse Ti inverted microscope with a 100×, 1.45 NA phase-contrast objective. For the TE300, images were further magnified 1.45× in a home-built magnification box. GFP, Sytox Green, and FM4-64 were imaged using 488 nm excitation (sapphire laser, Coherent, Santa Clara, CA), expanded to illuminate the field of view uniformly. Sytox Orange was imaged using 561 nm excitation (sapphire laser, Coherent). Laser intensities at the sample were typically ∼5 W/cm² at 488 nm and ∼2.5 W/cm² at 561 nm. Fluorescence images were obtained with an electron-multiplying charge-coupled device camera, either iXon 897 or iXon 887 from Andor (Belfast, United Kingdom). In both cases, the pixel size corresponds to 110 ± 10 nm at the sample.

The slower, one-color time-lapse movies were obtained with 50-ms exposure time in each channel, with fluorescence and phase-contrast images interleaved at 6-s intervals (12 s per complete cycle). The emission filter was HQ525/50 (Chroma Technology, Bellows Falls, VT) for GFP and Sytox Green and D675/50 (Chroma Technology) for FM4-64. For fast one-color movies, fluorescence images only were acquired at 0.5 s/cycle with 50-ms exposure time. For fast two-color experiments, μManager was used to obtain the data and switch filters between frames using a LB10-NW filter wheel (Sutter, Sacramento, CA). The time-lapse movies were obtained with 50-ms exposure time for each color, with green fluorescence (488 nm excitation) and red fluorescence (561 nm excitation) interleaved (overall 2 s/cycle). To minimize spectral bleed-through in the two-color experiments, we utilized the narrow filters, HQ510/20, for the green channel and HQ600/50M for the red channel.

For the FM4-64 membrane staining assay, MG1655 cells were incubated with 1 μg/mL FM4-64 for 10 min before plating in the microfluidics chamber. Fresh, pre-warmed, aerated EZRDM was used to wash away unbound cells. After the wash, 10 μM melittin plus 1 μg/mL FM4-64 was injected when taking the movies. To maintain good aeration and steady bulk concentrations, the medium with melittin and FM4-64 was flowed continuously at 0.3 mL/min for one minute first, and then at 0.3 mL/h for the rest of the experiment.

Results

Timing overview

The series of membrane-related events we are able to detect is complex. Details will follow. For the sake of clarity, in Fig. 1, we present a composite average timeline showing seven membrane-related events observed after the onset of flow of 10 μM melittin at t = 0. We label the times at which specific membrane permeabilization or re-sealing events occur by the sequence t₁, t₂, …, t₇, with the index corresponding to the typical order of events in time. The events include the onset of OM permeabilization to periplasmic GFP (t₁), the onset of “periplasmic bubble” formation (t₂), the onset of CM permeabilization to Sytox Orange (t₃), the onset of CM permeabilization to periplasmic GFP (t₄), the re-sealing of the OM to GFP (t₅), the re-sealing of the CM to GFP (t₆), and the re-permeabilization of both CM and OM to GFP (t₇). Fig. 1 is an amalgamation of results from the different experiments described below—no single experiment detects all seven events. The times in Fig. 1 are averages across cells measured relative to t₁, the widely variable time of the onset of OM permeabilization and cell shrinkage. Quantitative details of the timing of specific events relative to each other are provided as histograms in Fig. S1. Time data are presented as the mean ± SD in Table 2.

Figure 1.

Time line summarizing the seven membrane-related events observed after initiation at t = 0 of flow of 10 μM melittin (2× MIC) over plated E. coli cells expressing periplasmic GFP. Time lags are measured relative to the initial cell shrinkage and OM permeabilization event at t₁. See Table 2 and Fig. S1 for mean values and histograms of each quantity across cells. To see this figure in color, go online.

Table 2.

Summary of Timing Measurements

Experiment	Measurement	N	Mean ± SD	Range
ppGFP + phase contrast, 12 s/cycle	t1	40	3.7 ± 6.4 (min)	0.2–25.2 (min)
	t7–t1	26	7.5 ± 5.4 (min)	1.8–20.4 (min)
ppGFP + Sytox Orange, 2 s/cycle	t2–t1	25	4 ± 2	2–8
	t3–t2	25	1 ± 1	0–6
ppGFP only, 0.5 s/cycle	t4–t2	33	4.3 ± 2.3	1.5–12.5
	t5–t4	24	3.8 ± 1.4	1–6.5
	t5–t1	36	13.3 ± 3.9	6.5–23
	t6–t4	23	13.1 ± 8.3	2.5–30.5

Notation for events is as follows: t₁, onset of OM permeabilization to GFP; t₂, onset of periplasmic GFP bubble formation; t₃, onset of CM permeabilization to Sytox Orange; t₄, onset of CM permeabilization to GFP; t₅, re-sealing of OM to GFP; t₆, re-sealing of CM to GFP; and t₇, re-permeabilization of both CM and OM to GFP. See Fig. 1 for average timeline of events. See Fig. S1 for histograms of the various timing distributions. N is the number of individual cells in each sample. The ± values are one standard deviation of single measurements. Values are in seconds except as noted.

Transient disruption of the E. coli membrane barrier by melittin

The first experiments use the E. coli strain JCW10, which expresses GFP that is transported to the periplasm by the Tat system (23). On excitation at 488 nm, cells exhibit a halo of green fluorescence (Fig. 2, A and B), indicating a predominantly periplasmic spatial distribution of GFP. Typically, ∼90% or more of total GFP has been transferred to the periplasm before the melittin experiment begins (18, 20), as confirmed by the double-peaked transverse intensity linescan (Fig. 2 B). Fluorescence images are interleaved with phase-contrast images that monitor cell length versus time to a precision of ±50 nm. One full imaging cycle is completed every 12 s. At least 90% of the 25 cells in a typical field of view exhibit a similar sequence of membrane-related events, but each cell has its own timing.

Figure 2.

Example of effects of 10 μM melittin (2× MIC) on a single, representative E. coli cell expressing periplasmic GFP in aerobic growth conditions at 30°C. The frame rate is 12 s/cycle, and flow of melittin begins at t = 0. (A) Phase-contrast and fluorescence snapshots. (B) Transverse intensity profiles along the yellow line shown in (A). The profile is periplasmic before adding melittin and cytoplasmic shortly after adding melittin. (C) Time dependence of cell length (from phase-contrast images) and total GFP fluorescence intensity. The abrupt 37% decrease in GFP fluorescence intensity coincides with cell shrinkage. Most remaining GFP is trapped inside the cell envelope until t = 22 min. Scale: the width of an E. coli cell under our growth conditions is 900 nm. To see this figure in color, go online.

A representative example is shown in Fig. 2 and Movie S1. At t₁ = 36 s, the OM is permeabilized to GFP and the cell shrinks in length (Fig. 2 C). This cell loses 37% of its GFP intensity and 20% of its length in less than two camera frames (<24 s). Each cell is different. The fractional GFP loss at t₁ ranges from 36 to 64%, with a mean across cells of ∼50%. The spatial distribution of the remaining GFP quickly changes from the periplasmic (halo) distribution to that of a filled cytoplasm (Fig. 2 B), indicating CM permeabilization and influx of periplasmic GFP into the larger cytoplasmic volume. Beginning at time t₅ = 60 s, the rate of GFP loss decreases abruptly. (The intervening timing events t₂, t₃, and t₄ will be determined from other experiments.) After a transition period, ∼10 times slower leakage of GFP out of the cell envelope continues, as inferred from the subtle decrease in total GFP intensity from t = 5–21 min. Evidently the breach in the cell envelope has largely re-sealed to GFP. At t₇ = 21 min, fast leakage resumes, and by t = 23 min, all GFP has been lost to the cell surround. It is not clear from these experiments which membrane(s) (OM or CM or both) have re-sealed over the 21-min interval (t₅–t₁) to contain most of the remaining GFP. Nor is it clear which membrane(s) have “re-permeabilized” to enable complete loss of GFP beginning at t₇ = 21 min.

The distribution of times t₁ between the onset of melittin flow at t = 0 and the onset of GFP loss and cell shrinkage is shown in Fig. S1. Across 40 cells, the mean is <t₁> = 3.7 ± 6.4 min (± 1 SD). From 12 s/cycle imaging, we can only infer that the time interval (t₅–t₁) during which the OM leaks periplasmic GFP rapidly to the surround is shorter than ∼24 s. Sampling at 12 s/cycle is too slow to capture this difference accurately.

Transient formation of invaginations in the CM and pooling of periplasmic GFP

During the same 12 s/cycle movies of periplasmic GFP, in ∼30% of cells, we observe formation of bright, transient puncta of GFP occurring at essentially the same time as cell shrinkage and loss of ∼50% of the GFP intensity. When observed, the puncta typically last only one or two frames, or ∼12 s. The bright puncta are evidently caused by pooling of GFP within the periplasm; as soon as a punctum appears, the periplasmic halo of the remaining GFP becomes much dimmer. To test whether such short-lived puncta occur in all cells, we repeated the experiment with periplasmic GFP imaging at the much faster camera rate of 0.5 s/cycle. These faster movies reveal that all cells exhibit transient puncta of GFP fluorescence. Three examples are shown in Fig. 3 and Movie S2. In 12 of 18 septating cells, the puncta form in pairs on opposite sides of the septal region (Fig. 3 B), suggesting that GFP may sometimes be pooling in a circumferential “donut” structure surrounding the septal region. In non-septating cells, single puncta may form anywhere along the cell periphery. The puncta are typically larger than the diffraction limit in size. Intensity linescans across the puncta have cross sections of ∼400–700 nm FWHM. The puncta evidently have sufficient volume to cause pooling of much of the remaining periplasmic GFP within several seconds.

Figure 3.

Fluorescence snapshots of single E. coli cells expressing periplasmic GFP in aerobic growth conditions after addition of 10 μM melittin at t = 0. Images were acquired at 0.5 s/cycle. (A) A septating cell that forms one periplasmic GFP bubble at the septal region. (B) A septating cell that forms a pair of periplasmic GFP bubbles at the septal region, possibly indicating an annular (donut-like) invagination. (C) An apparently non-septating cell that forms one periplasmic GFP bubble. Scale: the width of an E. coli cell under our growth conditions is 900 nm.

The bright puncta are apparently invaginations in the CM (inward-facing periplasmic volumes) rather than blebs in the OM (outward-facing periplasmic volumes). The intensity peak of a punctum always moves inward (toward the long cell axis) as the bubble expands. In addition, we use evidence from the higher signal/noise images of the membrane stain FM4-64 during addition of melittin (Fig. S2). In those images, the excess membrane always faces inward rather than outward. Furthermore, we see no evidence of outward facing blebs in the phase contrast images. Finally, for cells expressing cytoplasmic rather than periplasmic GFP (Fig. S3), we observe no transient outward-facing GFP bubbles. Evidence below will indicate that the GFP puncta are short-lived because the periplasmic bubbles soon leak GFP into the much larger cytoplasmic volume as well as the cell surround.

Timing of localized disruption of the CM to Sytox Orange

In the 12 s/cycle movies of periplasmic GFP, the periplasmic halo image evolves to a filled cytoplasmic image shortly after the initial loss of GFP intensity (Fig. 2, A and B). This indicates CM permeabilization to GFP and concomitant loss of periplasmic GFP to the much larger cytoplasmic volume. To gain more insight into the nature and timing of CM disruption, we obtained a set of two-color, 2 s/cycle movies imaging periplasmic GFP and the DNA stain Sytox Orange. Sytox Orange is non-fluorescent in solution but becomes highly fluorescent on binding to the chromosomal DNA after membranes are compromised. The onset of Sytox Orange fluorescence in a nucleoid spatial pattern marks the time of permeabilization of the OM and CM to small molecules. Such two-color experiments enable direct observation within single cells of the relative timing of OM permeabilization to GFP at t₁, formation of the periplasmic GFP bubble(s) at t₂, CM permeabilization to Sytox Orange at t₃, and nearly complete re-sealing of the cell envelope to GFP at t₅. In the representative septating cell of Fig. 4 and Movie S3, the onset of OM permeabilization to GFP precedes the onset of formation of the periplasmic GFP bubble by (t₃–t₁) = 4 s. The onset of OM and CM permeabilization to Sytox Orange is simultaneous with bubble formation within one camera frame: (t₃–t₂) ≤ 2 s. The permeability to GFP decreases at t₅, but the permeability of the CM to Sytox Orange persists. The cell envelope leaks GFP rapidly over an interval (t₅–t₁) = 32 s.

Figure 4.

Timing of total fluorescence intensity of GFP and Sytox Orange from two-color imaging of the same cell. The camera alternates acquisition of green and orange fluorescence images at 2 s/cycle. (A) Fluorescence snapshots of a single E. coli cell expressing periplasmic GFP in aerobic growth conditions after addition of 10 μM melittin and 5 nM Sytox Orange beginning at t = 0. Left: GFP. Right: Sytox Orange. Image brightness for snapshots before 144 s was enhanced 14× to enable visualization of weak signals. (B) Time dependence of total fluorescence intensity in each channel for the cell shown in (A). (C) Axial intensity distribution of Sytox Orange fluorescence at different time points, as shown. Sytox Orange first appears near the septal region, then spreads gradually to the entire cell. The four-peaked distribution at long times is characteristic of the axial distribution of the nucleoids (chromosomal DNA). The width of an E. coli cell under our growth conditions is 900 nm. To see this figure in color, go online.

Averaged across cells (Table 2), the periplasmic bubble forms at <(t₂–t₁)> = 4 ± 2 s after the onset of OM permeabilization. Sytox Orange fluorescence begins to rise within <(t₃–t₂)> = 1 ± 1 s of the onset of bubble formation. In all septating cells, the bubble already begins to leak Sytox Orange across the CM as it gathers GFP from the rest of the periplasm. Bubble formation and leakage of Sytox Orange into the cytoplasm occur in parallel with leakage of GFP to the cell surround. The Sytox Orange signal continues to rise for at least 2 min after the cell envelope has re-sealed to GFP.

For septating cells, Sytox Orange always enters the cytoplasm at the septal location. This is seen clearly in the axial linescans of Sytox Orange intensity versus time, which show a staining pattern that evolves outward from punctal to the characteristic lobal pattern of the nucleoids (example in Fig. 4 C) (26). The pattern is consistent with the suggestion that Sytox Orange is leaking across the CM and into the cytoplasm through the periplasmic bubble. Melittin itself likely enters the cytoplasm by the same pathway. In sharp contrast, for non-septating cells, Sytox Orange always enters the cytoplasm at one endcap, after which the signal slowly spreads across the entire nucleoid (example in Fig. S4). For non-septating cells, the location of the periplasmic GFP bubble seems unrelated to the position of localized entry of Sytox Orange into the cytoplasm.

Timing of OM and CM permeabilization and re-sealing to GFP

Finally, we return to the 0.5 s/cycle, one-color movies of periplasmic GFP in an effort to understand the timing of permeabilization and re-sealing events of the OM and CM with respect to GFP. Detailed analysis of the same septating cell shown in Fig. 3 A and Movie S2 follows. We are attempting to dissect three signals: the loss of total GFP from the cell envelope, the growth and decay of the periplasmic bubble, and the transport of GFP through the bubble into the cytoplasm. Accordingly, we measure intensity versus time within three different regions of interest (ROIs), as shown in Fig. 5 A. The first ROI measures total fluorescence from the entire cell, shown by the T(t) trace in Fig. 5 C; T stands for total. The second ROI is a small box that covers the location where the periplasmic GFP bubble will grow and shrink, shown by the trace B(t) in Fig. 5 C; B stands for bubble. This ROI includes some cytoplasm, some periplasm, and some extracellular space. The third ROI comprises the sum of intensities within two small boxes that lie within the cell body far from the bubble location, shown by the C(t) trace in Fig. 5 C. C(t) stands for cytoplasm; it is intended to report primarily on leakage of GFP into and out of the cytoplasm. However, it includes ∼10% periplasmic volume. As shown in transverse linescans of the two-dimensional projected images of periplasmic and cytoplasmic GFP (Figs. 2 B and 5 B), C(t) inevitably responds to changes in GFP intensity within the periplasm as well. The intensities B(t), C(t), and T(t) are background-corrected, mean intensities over each ROI. They are not normalized to each ROI area. The absolute magnitudes are not to be compared.

Figure 5.

Time evolution of total GFP fluorescence intensity, T(t), GFP intensity in the region where the periplasmic bubble forms, B(t), and GFP intensity in a region sampling primarily cytoplasmic GFP, C(t). The cell under study is the same one shown in Fig. 3A. Image acquisition occurs at 0.5 s/cycle. (A) Regions of interest 1, 2, and 3 are used for total GFP intensity, GFP bubble intensity, and cytoplasmic GFP intensity. (B) Left: Fluorescence snapshots of the representative cell before and after melittin injection. Right: Transverse intensity profiles along the yellow lines in the fluorescence snapshots. (C) Time dependence of total GFP intensity in the three regions of interest. Permeabilization and re-sealing events marked as t₁, t₂, t₄, t₅, and t₆ denote the onset of OM permeabilization to GFP, the onset of periplasmic GFP bubble formation, the onset of CM permeabilization to GFP, the time of re-sealing of the OM to GFP, and the time of re-sealing of the CM to GFP, respectively. See text. Scale: the width of an E. coli cell under our growth conditions is 900 nm.

For the cell in Fig. 5, total intensity, T(t), shows rapid loss of GFP to the surround beginning at t₁ = 42 s after addition of melittin. This rapid loss continues until t₅ = 53 s, when the rate of loss slows down by about a factor of 4. The slower loss of intensity after t = 53 s is real. The rate of photobleaching, as judged by the slope of T(t) at t < 40 s, is much slower. The intensity in the bubble ROI, B(t), shows a dip in the range t = 40–43 s. This is due to loss of periplasmic GFP to the surround before bubble formation, the same loss detected less sensitively by T(t). We assign the onset of bubble formation to t₂ = 43 s, when B(t) reaches a minimum and begins to rise. From t = 43–49 s, the bubble is inflating. Its intensity peaks at t = 49 s and then decreases, first rapidly and then more slowly. The intensity C(t) remains constant until the same t = 43 s, when it begins to decrease. We attribute this to loss of periplasmic intensity as the bubble grows. Recall that C(t) is a mixture of periplasmic and cytoplasmic signals, with the periplasmic component dominant at early times. C(t) reaches a minimum at t₄ = 49 s, when it begins to rise. This time point marks the onset of leakage of GFP from the bubble, across the CM, and into the cytoplasm. Accordingly, the intensity in the central part of the transverse linescan begins to increase (Fig. 5 B). Notice that t₄ coincides with the peak in B(t). We attribute the subsequent rapid decrease in B(t) to simultaneous, parallel loss of bubble intensity to both the cytoplasm and the cell surround. When the rate of loss of B(t) slows down at t₅, C(t) is still rising. Evidently it is the OM, not the CM, that has mostly re-sealed against GFP. The distribution of intervals (t₅–t₁) during which the OM leaks GFP rapidly has mean <(t₅–t₁)> = 13.3 ± 3.9 s (Fig. S1; Table 2). Loss of intensity of B(t) slows down even further at t > 65 s, when entry of GFP into the cytoplasm has effectively ceased. We designate t₆, at ∼65 s, as the time at which the CM has re-sealed to GFP. Beyond t₆, C(t) remains constant. At the same t₆, the rate of loss of bubble intensity, B(t), decreases, but B(t) continues to lose intensity very slowly. We attribute the very slow loss at t > 65 s to a continuing slow leak through the OM. Accordingly, for t > 65 s, the decay rate of T(t) appears similar to that of B(t).

Notice that the CM remains permeable to GFP for some 13 s (t₆–t₅) after the OM has mostly re-sealed to GFP. These seem to be independent events. Both CM and OM are simultaneously quite permeable to GFP only for the short time period (t₅–t₄) = 4 s. This short interval explains why experiments imaging cytoplasmic GFP after melittin addition (Fig. S3 A) show little or no loss of GFP on the several-minute timescale of the early OM and CM permeabilization events. Finally, at much longer times, t₇ ∼ 2–20 min (not shown in Fig. 5), both membranes have become permeable to GFP once again, and the cell drains completely.

Our measurements provide no information about the molecular-level nature of the permeabilization sites. However, the timescale of GFP leakage during different events provides some indication of the degree of permeability of the OM or CM to GFP at each stage of the attack. These timescales vary significantly. The initial loss of GFP through the OM is rapid, occurring on average in <(t₅–t₁)> ∼ 13 s. Transfer of GFP from the bubble through the CM and into the cytoplasm is comparably rapid, <(t₆–t₄)> ∼ 13 s. In contrast, when melittin attacks cells expressing cytoplasmic GFP (Fig. S3 A), the final leakage from the cytoplasm to the cell surround beginning at t₇ is very slow, typically occurring over ∼4 min. This loss is apparently limited by the combined CM and OM permeability, not by GFP diffusion in the cytoplasm (27, 28). For comparison, in the attack of LL-37 on cells expressing cytoplasmic GFP, the cytoplasm typically drains entirely in ∼20 s or less after membrane permeabilization (example in Fig. S3 B). The underlying cause of the bottleneck to the final stage of melittin-induced GFP leakage is not clear.

Discussion

Melittin interactions with model lipid bilayers

There is an extensive literature describing interactions of AMPs with model lipid bilayers in the form of small or large unilamellar vesicles (SUVs or LUVs), giant unilamellar vesicles (GUVs), cushioned planar bilayers, and, most recently, E. coli cytoplasmic membranes in the form of spheroplasts (3, 4, 5, 6, 7, 11). Melittin is probably the peptide most often studied by these methods. A persistent question addresses the relevance of such studies of model membranes to the mechanisms by which AMPs interact with real bacterial membranes. Because this work provides the most detailed account thus far of melittin interactions with bacterial membranes, an attempt to find connections seems worthwhile.

In experiments on model lipid bilayers, melittin apparently exhibits different behaviors at low versus high peptide/lipid (P/L) ratios, and also for zwitterionic vesicles (typically 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC)) as compared to vesicles containing 30–100% anionic lipids (typically 2-oleoyl-1-palmitoyl-sn-glycero-3-phosphocholine (POPG)). In 1998, Schwarz et al. (3) observed graded (partial) release of dye from zwitterionic vesicles induced by melittin at low P/L ratios on the order of 1/1000. Graded release indicates that vesicles were transiently permeabilized to the dye, but re-sealed on the timescale of complete release of dye content. They attributed the partial release to formation of transient pores. Quantitative modeling yielded a very short estimated pore lifetime, below ∼10 ms. Transient permeabilization by melittin at low P/L ratios was corroborated more recently in studies by Wimley and Hristova (5, 6) of both zwitterionic and mixed zwitterionic/anionic vesicles. Melittin interactions with bilayers may also depend on the degree of hydration of the sample. Stable equilibrium pores were observed in low-hydration multilayers (7), whereas similar P/L ratios caused transient permeabilization at high hydration (5, 6).

An appealing mechanism for transient pore formation at low P/L ratios invokes the build-up of curvature stress (the “wedge effect”) as melittin binds initially to the outer leaflet of the vesicle (5, 7). Eventually, the stress induces vesicle rupture (attributed to pore formation) and dye release. The same rupture enables translocation of melittin to the vesicle inner leaflet. Equilibration of melittin density across the two leaflets alleviates the asymmetric curvature stress and the pore re-seals. For melittin at low P/L ratios, at equilibrium, the bilayer is once again impermeable to dye. There are no persistent pores.

There is also evidence that melittin can induce membrane disruptions (perhaps toroidal pores) large enough to pass a small globular protein such as the 27 kDa GFP. In 1982, DeGrado et al. (29) showed that melittin transiently permeabilized erythrocytes to hemoglobin. They attributed the re-sealing process to translocation of melittin across the membrane, similar to the mechanism invoked here. In a 2001 study of purely anionic vesicles (100% POPG), Ladokhin and White (4) found that melittin at high P/L ratios >1:35 induced membrane disruptions that leaked both small and larger fluorescently labeled dextrans (4 and 50 kDa) with comparable efficiency. Although the disruptions were ascribed to a “detergent-like” mechanism, they could equally well have been pores of diameter ∼3 nm. In 2001, such large, melittin-induced pores were detected quite directly by Huang and co-workers (10) in fully hydrated, oriented multilayers using in-plane neutron scattering. For melittin at P/L >1:30 on POPC multilayers, D₂O-filled pores of inside diameter 4.4 nm were inferred. At the same high P/L ratios, oriented circular dichroism showed that most of the melittin helices were inserted into the bilayer (perpendicular orientation). The inferred structure was that of a toroidal pore lined by a mixture of melittin and lipid molecules. A recent study showed that melittin at a P/L ratio of 1:50 enabled passage of 10 kDa dextran across lipid vesicle bilayers. A synthetic melittin variant “MelP5” showed permeabilization of vesicles to both 10 kDa dextran and 24 kDa chymotrypsin at P/L ratios as low as 1:500 (30).

In 2013, the Huang laboratory studied melittin interactions with anionic GUVs made of 7:3 POPC/POPG (7). Stable permeabilization to small dye molecules occurred only at high P/L ratios in excess of 1:45. The permeabilization persisted on a timescale of hours, strongly suggesting an equilibrium state. Recently, the same laboratory has studied the interaction of melittin with large spheroplasts made of E. coli cytoplasmic membrane (11, 31). Comparably stable permeabilization was observed in the spheroplasts (11). The proposed mechanism invokes initial formation of transient pores due to the same outer-leaflet curvature stress invoked before to explain graded dye release from vesicles. Such transient pores would enable melittin translocation, which leads to build-up of positive curvature stress in both leaflets. At sufficiently high P/L, the equilibrium state has equal melittin concentration in both GUV leaflets and includes stable permeabilization sites, possibly due to toroidal pores, that incorporate excess melittin in an orientation perpendicular to the plane of the membrane. The stable pores would thus relieve membrane tension caused by apposition of two leaflets, both with positive curvature stress, in a planar geometry.

Proposed mechanism of membrane permeabilization and re-sealing steps

This work provides strong evidence of melittin-induced transient permeabilization of both the OM and CM of live E. coli to GFP. Membranes permeable to GFP will enable melittin translocation and will leak other small proteins as well as a host of small ions and neutral molecules. The GFP permeabilization sites tend to occur at curved membrane surfaces (both septa and endcaps), where the anionic phospholipids cardiolipin (CL) and phosphatidylglycerol (PG) are known to concentrate. The sites are localized, but we cannot determine whether the membrane disruptions are pore-like or less organized. The transient OM and CM disruptions last ∼10 s, after which they re-seal to GFP. Some 2–20 min later, both membranes have re-permeabilized to GFP.

In contrast to most vesicle experiments, our flow experiments hold the external melittin concentration constant at 10 μM, providing an unlimited supply of the AMP to the plated bacterial cells. As time progresses, the total concentration of melittin within a cell builds up from the outside in. The mean total concentration of a cationic AMP in an E. coli cell, averaged over all internal components, can become very high (32, 33). For the synthetic peptide ARVA, Wimley showed binding of ∼10⁷–10⁸ peptides per cell, corresponding to a mean concentration in the low-millimolar range (32). In effect, our experiments likely sweep the membrane-bound melittin concentration over a very wide range of P/L ratios during the 30-min observation time.

The real E. coli cell envelope is structurally and compositionally much more complex than a unilamellar membrane comprising pure lipids. Comparisons with model lipid bilayers are necessarily speculative. Nevertheless, we can paint an appealing mechanistic picture that incorporates ideas first generated to explain the vesicle experiments. We imagine an initial buildup of melittin in the anionic lipopolysaccharide (LPS) layer; melittin gradually penetrates to the outer leaflet of the OM, generating curvature stress. This leads to membrane disruption and the initial leakage of periplasmic GFP to the cell surround (t₁). The disruption enables rapid translocation of melittin into the periplasm, where it binds to the PG layer, inducing cell shrinkage. The envelope shrinkage forces shrinkage of the CM, which invaginates to form inward-facing periplasmic bubbles (t₂). (See the proposed explanation below.) As melittin concentration builds up within the periplasm, it binds to the inner leaflet of the OM, relieving the curvature stress and causing nearly complete re-sealing of the OM to GFP (t₅).

Meanwhile, the same process of differential curvature stress repeats itself on the inside leaflet of the periplasmic bubble (the outer leaflet of the CM), which soon bursts and begins to leak Sytox Orange (t₃), GFP (t₄), and presumably melittin into the cytoplasmic space. Melittin density on the outside bubble leaflet (the inner leaflet of the CM) then builds up to re-seal the bubble to GFP (t₆). Evidently, the CM remains permeable to Sytox Orange and, by inference, to melittin, throughout OM and CM permeabilization and re-sealing to GFP. We suggest that melittin continues to translocate across both OM and CM, causing a gradual buildup of curvature stress in both leaflets of both membranes. That buildup occurs slowly, over ∼8 min, in part because binding of melittin to both membranes is competing with binding to a variety of anionic bacterial components. These include the LPS layer of the OM; within the periplasmic space, the PG layer and anionic glycopolymers, proteins, and lipids; and, within the cytoplasmic space, the chromosomal DNA, ribosomes, lipids, and a host of soluble anionic proteins and RNAs. These anionic components evidently act as highly absorbent sinks for cationic AMPs (32, 33). Eventually, both OM and CM re-permeabilize to GFP and all residual GFP is lost to the surround (t₇).

The timescale of re-sealing of the OM and CM to GFP is ∼10 s, much longer than the lifetime of ∼10 ms for the pores that leak dye from vesicles in the low, P/L ∼ 1:1000 regime. It seems unlikely that the pores in these two different types of experiment are closely related. Our speculative mechanism requires GFP-sized pores that open at much higher P/L ratios and then subsequently close. Such large, transient pores have not been observed in vesicle experiments, which have focused on dye efflux.

Proposed mechanism of cell shrinkage and periplasmic bubble formation

How does melittin cause the initial cell-shrinkage event and concomitant formation of the inward-facing periplasmic bubbles? Here, we propose, to our knowledge, a novel mechanical concept that couples binding of melittin to the PG layer with cell shrinkage and bubble formation. The PG layer is a single gigantic molecule comprising a meshwork of circumferential glycan strands cross-linked to each other by short, flexible peptide chains oriented longitudinally (21). The peptide linkers contain two anionic amino acids each. The PG layer is longitudinally elastic but circumferentially stiff. It is covalently bound to lipoproteins whose hydrocarbon tails are embedded in the inner leaflet of the outer membrane. The high concentration of neutral and ionic solute molecules in the cytoplasm causes influx of excess water, resulting in a large turgor pressure that drops across the PG/OM layer (34). Mechanical equilibrium is attained when the turgor pressure is balanced by the restoring force of the stretched PG layer, which expands beyond its equilibrium size. This is analogous to blowing up a balloon.

We propose that binding of cationic AMPs to the anionic peptide cross-links changes the physical properties of the PG layer, in effect increasing its longitudinal stiffness (elastic modulus). If the turgor pressure remains essentially constant while the elastic modulus of the PG layer increases, the PG layer will contract. Mechanical equilibrium will be re-established at a shorter overall length of the cell envelope. This is a plausible explanation for the shrinkage of cell length as cationic AMPs enter the periplasm. If it is correct, we would expect the timescale of shrinkage of cell length to mimic the timescale of build-up of the cationic AMP concentration in the periplasmic space.

Our original study of LL-37 (20) used a rhodamine fluorescent label on the AMP itself, fortuitously enabling direct observation of the build-up of Rh-LL-37 in the periplasm as a wave of dequenched red fluorescence that slowly expanded from septum to cell tips. Independent experiments demonstrated that Rh-LL-37 indeed binds to purified PG. In the example of Fig. 2 of (20), the shrinkage of cell length and the build-up of periplasmic Rh-LL-37 intensity both occur gradually on the same timescale, over ∼8 min. This behavior is consistent with our suggestion that AMP binding leads to compression of the PG layer and longitudinal cell shrinkage.

What happens to the OM and the CM as the PG layer contracts in length? Both membranes suddenly have excess surface area. The OM is covalently bound to the PG layer by proteolipids. It could form ruffles or blebs (35), it could mechanically de-couple from the PG layer by releasing lipoprotein copies, or it could bud off LPS vesicles to the surround. Our methods described here do not discern blebs and would be insensitive to ruffles or small vesicles. In contrast, the CM is only weakly mechanically coupled to the PG layer, if at all, and the shrinking PG layer encloses the CM. As the PG layer contracts axially, the CM must somehow shrink its effective length. It could ruffle, it could bud off internal vesicles, or it could invaginate. Our observation of the inward-facing septal periplasmic bubbles induced by melittin indicates that the CM in fact invaginates. The surface area of a bubble can account for the cylindrical surface area lost in a 15% shrinkage event. The surface area of a periplasmic bubble of radius r_bubble ∼ 300 nm is S_bubble = 4πr_bubble² ∼ 1.1 μm². This is indeed 15% of the cylindrical surface area of a cell of radius R_cell ∼ 400 nm and cylindrical length L ∼ 3 μm: S_cylinder = 2πR_cellL ∼ 7.5 μm².

This is the only mechanism we can imagine that explains large invaginations in the CM, i.e., inward-facing bubbles. As melittin accumulates in the periplasm, it first binds differentially to the outer leaflet of the CM, part of which forms the “inside leaflet” of the bubble. This creates positive curvature stress on the CM. In unconstrained space, this type of stress would induce outward-facing blebs, not inward-facing bubbles. Our model instead supposes that the mechanical constraint imposed by the shrinking PG layer forces the bubble to grow inward, taking up extra surface area and relieving the axial stress due to PG-layer shrinkage. As melittin continues to accumulate preferentially in the inside leaflet of the bubble bilayer (the outer leaflet of the CM), the resulting curvature stress causes the bubble to rupture in a few seconds. This connects the bubble volume to the cytoplasmic volume and enables Sytox Orange and periplasmic GFP to enter the cytoplasm. The re-sealing of the bubble to GFP may occur as sufficient melittin translocates to the cytoplasm, where it gains access to the outside leaflet of the bubble (inner CM leaflet), thus relieving the curvature stress.

Septal periplasmic GFP bubble formation evidently occurs for both melittin and CM15. For melittin, the septal bubbles remain connected to the periplasmic space and soon become leaky to the cytoplasm as well. For CM15, the bubbles evidently isolate GFP from both the periplasmic space and the cytoplasm, as if a stable, internal GUV has formed (Fig. S6; Movie S4). Although appealing as an explanation of bubble behavior in the case of melittin, the proposed mechanism does not explain the observation that CM15 induces stable “internal GUVs.”

Comparisons across antimicrobial peptides

By now we have used single-cell imaging methods to directly observe the attack of a variety of AMPs on live E. coli, including LL-37 (20), cecropin A (18), melittin, and CM15 (a synthetic cecropin A/melittin hybrid) (17). All are positively charged, but they vary in sequence, length, and charge density (Table 1). The observed membrane phenomena and the sequence of events vary significantly across these peptides.

For the longer peptides LL-37 and cecropin A, the initially observed events were abrupt shrinkage of cell length by 15–20% and essentially complete loss of periplasmic GFP to the cell surround. No periplasmic GFP bubbles were observed, perhaps because the rapid loss of GFP to the cell surround precluded their observation. No membrane re-sealing events were observed, perhaps for the same reason. The two shorter peptides CM15 and melittin behave similarly to each other. In both cases, part of the periplasmic GFP intensity is lost almost simultaneously with cell shrinkage. The OM evidently re-seals, and much of the remaining periplasmic GFP then moves inward into the cytoplasm. For septating cells, both peptides induce periplasmic bubbles of GFP intensity at the septal region. See Fig. S6 and Movie S4 for the CM15 data. Only much later do both CM and OM re-permeabilize, enabling complete loss of GFP to the cell surround. For melittin, the periplasmic GFP bubbles quickly drain. For CM15, the GFP bubbles persist, even after complete loss of GFP from both cytoplasm and periplasm.

For all four AMPs, septating cells are attacked earlier than non-septating cells, and curved membrane surfaces are preferentially permeabilized (septal region and endcaps; see Fig. S5). We speculate that the sensitivity of curved membranes to permeabilization by cationic AMPs may be due in part to the higher concentration of the anionic lipids cardiolipin and phosphatidyl glycerol at those locations. Localized permeabilization (consistent with pore formation) seems quite general across AMPs. There are subtle differences in the patterns of OM and CM attack. For example, for septating cells, cecropin A permeabilizes the OM at the septal region but the CM at one endcap. LL-37 and melittin permeabilize both the OM and the CM in the septal region (Fig. S5).

Finally, we contrast the effects on E. coli of MM₆₃:CHx₃₇, a highly cationic random copolymer of β-peptide subunits with a mean length of ∼35 subunits and a mean charge of ∼+22. Like melittin and CM15, the random copolymer induced inward movement of periplasmic GFP, indicating translocation across the OM without complete permeabilization to GFP (16). However, the first observation was cell shrinkage (by only ∼9%) and formation of what we called “endcap periplasmic GFP bubbles.” The copolymer-induced bubbles localized at both endcaps; they are highly reminiscent of the plasmolysis spaces induced by abrupt external osmotic upshift. We believe that the osmotic effects due to translocation of MM₆₃:CHx₃₇ and a host of accompanying anions into the periplasm cause the endcap bubbles. As a result, the osmolality of the periplasm increases, much as it does after an abrupt external osmotic upshift. The result is loss of cytoplasmic water and eruption of endcap plasmolysis spaces. In contrast, the osmotic effects are much smaller when a more moderately charged +6 CM15 or melittin molecule enters the periplasm and binds to the PG layer. For these reasons, we believe the endcap periplasmic bubbles induced by MM₆₃:CHx₃₇ are plasmolysis spaces caused by an effective osmotic upshift in the periplasm, whereas the septal periplasmic bubbles induced by melittin and CM15 are not.

Conclusions

It is increasingly clear that introduction of a high concentration of polycationic peptide into an E. coli cell, most of whose biopolymer content is polyanionic, wreaks havoc in a variety of ways. However, it all begins with permeabilization of the OM and CM to the peptide itself. Here, we have characterized the spatiotemporal effects of melittin on the E. coli membranes in great detail. For both the OM and the CM, we have observed membrane permeabilization, nearly complete re-sealing, and subsequent re-permeabilization to the globular protein GFP. We speculate that a mechanism invoking transient permeabilization and permanent re-permeabilization, both due to the build-up of membrane curvature stress, might prove to be common to both bacterial membranes and model lipid bilayers. However, the variety and specificity of the effects of different AMPs on E. coli membranes remain surprising. The spatiotemporal complexity observed here goes well beyond the effects observed thus far in model lipid bilayers.

Author Contributions

Z.Y., H.C., and J.C.W. all contributed to the design of the project. Z.Y. and H.C. performed the experiments. Z.Y. and J.C.W. analyzed the data and wrote the manuscript. H.C. helped edit the manuscript.

Editor: Tommy Nylander.

Supporting Material

Movie S1. Interleaved Images of Periplasmic GFP and Phase Contrast, 12 s/cycle

Flow of 10 μM melittin begins at t = 0. Same cell as in Fig. 2.

Movie S2. Interleaved Images of Periplasmic GFP and Phase Contrast, 0.5 s/cycle

Flow of 10 μM melittin begins at t = 0. Same three cells as in Fig. 3.

Movie S3. Interleaved Images of Periplasmic GFP and the DNA Stain Sytox Orange, 2 s/cycle

Flow of 10 μM melittin begins at t = 0. Same cell as in Fig. 4.

Movie S4. Interleaved Images of Periplasmic GFP and Phase Contrast, 12 s/cycle

Flow of 10 μM CM15 begins at t = 0. Same cell as in Fig. S6 A.