Polymyxin B lethality requires energy-dependent outer membrane disruption

Abstract

Polymyxin antibiotics target lipopolysaccharides (LPSs) in both membranes of the bacterial cell envelope, leading to bacterial killing through a poorly defined mechanism. Here we demonstrate that metabolic activity is essential for the lethality of clinically relevant doses of polymyxin B (PmB) and leverage this insight to determine its mode of action. PmB killed exponential-phase Escherichia coli but did not eliminate stationary-phase cells unless a carbon source was available. Antibiotic lethality correlated with surface protrusions visible by atomic force microscopy and LPS loss from the outer membrane via processes that required LPS synthesis and transport but that were blocked by the MCR-1 polymyxin resistance determinant. While energy-dependent outer-membrane disruption was not directly lethal, it facilitated PmB access to the inner membrane, which the antibiotic permeabilized in an energy-independent manner, leading to cell death. This work reveals how metabolic inactivity confers tolerance of an important, membrane-targeting antibiotic.

Fig. 1. PmB lethality requires metabolic activity and is associated with significant morphological changes to the cell surface.

a, Survival of exponential-phase (Exp) or stationary-phase (Stat) E. coli exposed to 4 µg ml⁻¹ PmB in MM ± glucose, as determined by c.f.u. counts. b, AFM phase images showing exponential- or stationary-phase E. coli cells exposed to 2.5 µg ml⁻¹ PmB in MM ± glucose, shown as a function of time. Scale bar, 250 nm. Colour phase scale (scale insert in first row of image at t = 90 min), 6 degrees (row 1), 4 degrees (row 2), 5 degrees (rows 3 and 4). Except for stationary-phase cells without glucose, all bacteria were SYTOX positive by the end of the imaging (Supplementary Fig. 6a). It is worth noting that a lower PmB concentration was used for AFM relative to other experiments, due to the low density of cells used in these assays. The ‘phase’ in degrees represents the shift in the phase of oscillation of the AFM cantilever. c, OM disruption of stationary-phase E. coli cells during the first 20 min of exposure to 4 µg ml⁻¹ PmB, as determined by uptake of NPN fluorescent dye. d, OM disruption of E. coli _PerimCherry as determined by egress of the fluorescent protein mCherry into the culture supernatants of stationary-phase bacteria exposed or not to 4 µg ml⁻¹ PmB in MM ± glucose. e, Combined OM and IM disruption of stationary-phase E. coli exposed to 4 µg ml⁻¹ PmB in MM ± glucose, as determined by uptake of the fluorescent nucleic acid dye SYTOX green. RFU, relative fluorescence units. For c and e, the blank value refers to the relevant fluorophore in medium without bacteria. For c and e, ‘No treat MM’ refers to the absence of PmB in minimal medium; ‘No treat MM + G’ refers to the absence of PmB in minimal medium containing 0.36% glucose. All experiments were replicated in n = 3 independent assays. Error bars show the standard deviation of the mean. Significant differences were determined between stationary-phase MM + G with PmB and each of the other conditions by two-way repeated measures ANOVA. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Source data

Fig. 2. PmB-mediated OM disruption results in LPS loss without detectable disruption of the porin network.

a, High-magnification AFM height and phase images of stationary-phase E. coli exposed to 2.5 µg ml⁻¹ PmB in MM + G, showing the porin network remains largely intact for at least 60 min of antibiotic challenge. Scale bar, 50 nm. Height scale (scale inset in first row of image at t = 90 min), 3 nm. Phase scale, 1 deg. b, Kdo analysis of LPS recovered from filtered supernatants of stationary-phase E. coli exposed, or not, to 4 µg ml⁻¹ PmB in MM ± glucose for 15 min (n = 5). c, Levels of OmpF, GroEL and total LPS in E. coli exposed, or not, to 4 µg ml⁻¹ PmB in MM ± glucose for 15 min. d, Quantification of LPS levels in E. coli exposed, or not, to 4 µg ml⁻¹ PmB in MM ± glucose for 15 min. For b–d ‘No treat MM’ refers to the untreated condition absence of PmB in minimal medium; ‘No treat MM + G’ refers to the untreated condition absence of PmB in minimal medium containing 0.36% glucose. Unless stated otherwise, experiments were replicated in n = 3 independent assays. Error bars show the standard deviation of the mean. Statistical significance of differences was determined by one-way ANOVA; ***P < 0.001; ****P < 0.0001; NS, not significant. Source data

Fig. 3. PmB-triggered LPS loss and bacterial killing require ATP and LPS synthesis and transport.

a, Four-way representation of the effect of various carbon sources, or none, on ATP production and LPS levels in stationary-phase E. coli in the absence of PmB, and bacterial survival and LPS levels after exposure to 4 µg ml⁻¹ PmB. The more yellow the symbol, the more LPS is produced in the absence of PmB; the smaller the symbol, the more LPS that is lost during PmB exposure. b, SDS–PAGE analysis of LPS from stationary-phase E. coli exposed to various carbon sources in MM for 15 min in the absence or presence of 4 µg ml⁻¹ PmB. c–f, Survival of stationary-phase E. coli in MM + G supplemented, or not, with PmB, with or without growth-inhibitory concentrations (1× MIC) LpxC inhibitors ACHN-975 (c) or CHIR-090 (d) or protein synthesis inhibitor tetracycline (e) or MsbA inhibitor G907 (f). g, Survival of E. coli wild type (WT) or LptD4213 strain exposed, or not, to 4 µg ml⁻¹ PmB. h, LPS levels of non-treated E. coli or bacteria exposed to 4 µg ml⁻¹ PmB ± LpxC inhibitors (CHIR-090, ARCN-975) or tetracycline for 15 min (n = 4). Panel includes an SDS–PAGE image of LPS band intensity. For h ‘No treat’ refers to the untreated condition absence of PmB in minimal medium containing 0.36% glucose. It is worth noting that the experiments with G907 used E. coli LptD4213 to enable antibiotic ingress. Unless otherwise stated, all experiments were replicated in n = 3 independent assays. Error bars show the standard deviation of the mean. Statistical significance of differences was determined by one-way (h) or by two-way repeated measures ANOVA between the PmB-treated and PmB + antibiotic-treated groups (c–f) or between wild type with PmB and imp4213 with PmB (g). *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; NS, not significant. Source data

Fig. 4. PmB-mediated LPS loss is necessary for lethality because it provides the antibiotic with access to the IM.

a, Survival (c.f.u.) of stationary-phase E. coli first pre-treated with 4 µg ml⁻¹ PmB in MM + G for 15 min before washing to remove unbound antibiotic, and next (from t = 0) exposed to 4 µg ml⁻¹ PmB, or not, in MM ± glucose. b, LPS levels of untreated E. coli or bacteria exposed to 4 µg ml⁻¹ PmB, 4 µg ml⁻¹ PmBN or 10 mM EDTA for 15 min in MM + G. Note that experiments with EDTA used MM that was not supplemented with MgSO₄ or CaCl₂. Panel includes SDS–PAGE image of LPS band intensity. c, LPS levels of non-treated E. coli or bacteria exposed to 4 µg ml⁻¹ PmB, 4 µg ml⁻¹ PmBN or 10 mM EDTA for 15 min in MM. Panel includes SDS–PAGE image of LPS band intensity. d, Survival of stationary-phase E. coli exposed to a range of PmB concentrations for 2 h in MM + G, following a 15 min pre-treatment, or not, with 4 µg ml⁻¹ PmB, 4 µg ml⁻¹ PmBN or 10 mM EDTA in MM + G. e, Levels of bodipy-vancomycin labelling of E. coli exposed, or not, to 4 µg ml⁻¹ PmB in MM ± glucose for 15 min. f, Survival of E. coli exposed, or not, to PmB with or without 10 mM EDTA in MM + G. g, Survival of E. coli exposed, or not, to PmB with or without 10 mM EDTA in MM. h, AFM phase images showing stationary-phase E. coli exposed to 2.5 µg ml⁻¹ PmB and 10 mM EDTA in MM. Scale bar, 250 nm; phase scale (scale inset in image at t = 90 min), 5 deg. The same cell was imaged during the experiment. All experiments were replicated in n = 3 independent assays. Error bars show the standard deviation of the mean. Statistical significance of differences was determined by one-way (in b, c and e) or two-way repeated measures ANOVA between No treat MM + G and PmB MM + G and between No treat MM and PmB MM (in a) between No treat and pre-treatment conditions (in d) between No treat and antibiotic conditions (in f and g). *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; NS, not significant. Source data

Fig. 5. MCR-1 prevents PmB-triggered LPS release.

a, Survival of stationary-phase MG1655, MG1655 mcr-1* or MG1655 mcr-1 exposed to 4 µg ml⁻¹ PmB in MM + G, as determined by c.f.u. counts. b, OM disruption of stationary-phase MG1655, MG1655 mcr-1* or MG1655 mcr-1 cells during the first 20 min of exposure to 4 µg ml⁻¹ PmB in MM + G, as determined by uptake of NPN fluorescent dye. c, OM and IM disruption of stationary-phase MG1655, MG1655 mcr-1* or MG1655 mcr-1 exposed to 4 µg ml⁻¹ PmB in MM + G, as determined by uptake of the fluorescent dye SYTOX green. d, Total LPS in stationary-phase MG1655 mcr-1* or MG1655 mcr-1 exposed, or not, to 4 µg ml⁻¹ PmB in MM + G for 15 min. Graph shows quantification of LPS levels from three independent experiments. e, AFM phase images showing stationary-phase MG1655 mcr-1* and MG1655 mcr-1 exposed to 2.5 µg ml⁻¹ PmB in MM + G. The same cell was imaged during the experiment. Scale bar, 250 nm; phase scale (scale inset in second row of image at t = 90 min), 4 deg. f, Mean surface roughness of MG1655 mcr-1* and MG1655 mcr-1 treated with 2.5 µg ml⁻¹ PmB in MM + G. For b and c, the blank value refers to the relevant fluorophore in medium without bacteria. All experiments were replicated in n = 3 independent assays. Error bars show the standard deviation of the mean. Statistical significance of differences was determined by one-way (in d) or two-way repeated measures ANOVA between MG1655 and mcr-1* or mcr-1 (in a, b and c). *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; NS, not significant. Source data

Fig. 6. Proposed model for how PmB kills bacteria.

a, PmB binds to the OM, causing minor permeabilization, regardless of bacterial metabolic activity (insufficient damage to provide PmB with access to the periplasm). b, In metabolically active bacteria, PmB triggers LPS release via a process that requires ATP-dependent synthesis (Raetz pathway) and transport of LPS (further boosted by novobiocin (Nov)). c, The shedding of LPS compromises the integrity of the OM, resulting in surface protrusions, and enables ingress of the antibiotic into the periplasm where it targets LPS in the IM, leading to permeabilization. d, The loss of integrity of both the OM and IM results in bacterial killing.

Extended Data Fig. 1. Polymyxin B lethality requires metabolic activity.

a, Visual representation of the set-up for all microbiological assays. E. coli MG1655 was grown in 3 ml of MHB for 16 h at 37^oC 180 r.p.m. to stationary phase. The culture was washed three times in MM and resuspended at an inoculum density of 1×10⁸ in 3 ml of MM or MM + G. To test exponential phase E. coli, the overnight culture was diluted 1/1000 in fresh MHB and grown for a further 3 h. b, Survival of stationary phase E. coli exposed to 4 µg ml⁻¹ PmB from t = 0 in the presence of a 1/10 dilution series of 0.36% glucose. A 1/100 dilution of 0.36% glucose supported PmB killing, however, a 1/1000 dilution substantially reduced the rate and degree of PmB killing. c, Survival of stationary phase E. coli exposed to 4 µg ml⁻¹ PmB from t = 0 in MM + G or MM with an equimolar concentration of 2-deoxy-D-glucose, as determined by CFU counts. All experiments were replicated in n = 3 independent assays. Error bars show the standard deviation of the mean. Significant differences were determined by two-way repeated measures ANOVA. P= *<0.05, ***<0.001. Source data

Extended Data Fig. 2. PmB causes surface protrusions on exponential phase E. coli MG1655 and only when glucose is present in stationary phase cells.

Exponential (Exp) or – (Stat) E. coli cells were exposed to PmB (4 µg ml⁻¹) in MM ± glucose and samples taken at the indicated time points. In keeping with the data from AFM studies, stationary phase cells exposed to PmB in the presence of glucose began to show surface protrusions from ~30 min, which increased over time. Conversely, stationary phase cells exposed to the polymyxin without glucose showed minimal to no surface protrusions. PmB triggered surface protrusions in exponential phase cells whether glucose was present or not. Images for stationary phase cells at 120 min are shown again in an enlarged image to highlight differences in surface appearance. Scale bar represents 400 nm (N = 1).

Extended Data Fig. 3. Tolerance to PmB is not associated with modifications to lipid A.

a, From top to bottom, representative mass spectra showing unmodified lipid A from stationary phase E. coli at time 0 min in MM, stationary phase E. coli at time 15 min in MM, and stationary phase E. coli at time 15 min in MM + G, and stationary phase E. coli at time 15 min in MM + G exposed to 4 µg mL⁻¹ PmB. The peak at ~1800 m/z corresponds to native hexa-acyl diphosphoryl lipid A containing four C14:0 3-OH, one C14:0 and one C12:0. The peaks at ~1710 and ~1800 m/z correspond to the subtraction or addition of phosphate to the native form of lipid A. b, Overlaid mass spectra of stationary phase E. coli at time 15 min in MM + G (black) and stationary phase E. coli at time 15 min in MM + G exposed to 4 µg mL⁻¹ PmB (red), indicating a substantial reduction in native lipid A.

Extended Data Fig. 4. Tolerance to PmB is conserved under nutrient-limiting conditions.

a. Survival of stationary phase Pseudomonas aeruginosa PA14 exposed to 4 µg ml⁻¹ PmB in MM ± glucose, as determined by CFU counts. b, Survival of a panel of E. coli clinical isolates cultured to stationary phase, exposed to 4 µg ml⁻¹ PmB in MM ± glucose for 2 h. c, Survival of C. freundii and E. asburiase clinical isolates cultured to stationary phase, exposed to 4 µg ml⁻¹ PmB in MM ± glucose for 2 h. d, Survival of a K. pneumoniae clinical isolates cultured to stationary phase, exposed to 4 µg ml⁻¹ PmB in MM ± glucose for 2 h. e, Survival of P. aeruginosa clinical isolates cultured to stationary phase, exposed to 4 µg ml⁻¹ PmB in MM ± glucose for 2 h. f, Survival of A. baumannii clinical isolates cultured to stationary phase, exposed to 4 µg ml⁻¹ PmB. g, OM disruption of stationary phase P. aeruginosa cells during the first 20 min of exposure to 4 µg ml⁻¹ PmB, as determined by uptake of the NPN fluorescent dye. h, OM and IM disruption of stationary phase P. aeruginosa exposed to 4 µg ml⁻¹ PmB in MM ± glucose, as determined by uptake of the fluorescent dye SYTOX green. For g, and h, the blank value refers to the relevant fluorophore in medium without bacteria. All experiments were replicated in n = 3 independent assays, error bars show the standard deviation of the mean. Significant differences were determined by two-way (a - h) repeated measures ANOVA. P= *<0.05, **<0.01, ***<0.001, ****<0.0001, ns=not significant. Source data

Extended Data Fig. 5. PmB leads to morphological changes to the OM at the nanoscale.

a, Combined brightfield and fluorescence (SYTOX) images of the AFM scan region for the experiments in Fig. 1b with exponential (Exp) and stationary (Stat) phase E. coli at 90 min post PmB treatment (the circled cells are those chosen for the image sequences shown here and were all SYTOX-positive by 90 mins, except for stationary phase E. coli in MM). b, Higher-magnification AFM height and phase scans of stationary phase E. coli. The surface of untreated cells is covered in a network of pores, here best visible in the phase images. The resolution of pores (trimeric OMPs) is ultimately compromised by the progressive roughening of the OM after PmB is added in the presence of glucose (MM + G). In MM + G, PmB caused disruption to the OM in the form of protrusions. Without glucose the appearance of the OM at the nanoscale did not change. Scalebars: a, 4 µm, b, 100 nm; height scales (scale inset in first row of image at t = 90 min): 20 nm; phase scale: 2 deg (row 2) and 1 deg (row 4). c, Quantification of protrusion height (left) and OM mean roughness (right). Protrusion height data are presented as the mean ± SD of median values of the height of features above 50% of the maximum height of each image. Mean roughness values were computed by Gwyddion and data are presented as the mean ± SD. Measurements were taken from 3 different 500 nm scans of 3 different E. coli MG1655 cells imaged in separate experiments. Source data

Extended Data Fig. 6. PmB-induced LPS loss is maximal at 15 minutes and is a conserved phenotype.

a. Levels of total LPS in E. coli exposed, or not, to 4 µg ml⁻¹ PmB for 0-, 15-, and 30 min in MM + G. b, Bar graph of LPS levels according to densitometry analysis of a. Densitometric values were interpolated according to the standard curve in Supplementary Fig. S7C and subsequently normalized to the No treatment controls, to give ‘Relative abundance LPS’. c, Levels of total LPS in P. aeruginosa exposed, or not, to 4 µg ml⁻¹ PmB in MM ± glucose. d, Bar graph of LPS levels according to densitometry analysis of a. Experiments were replicated in n = 4 a, b or n = 3 c, d independent assays. Error bars show the standard deviation of the mean. Significant differences were determined by one-way ANOVA P= **<0.01, ****<0.0001, ns=not significant. Source data

Extended Data Fig. 7. PmB killing requires ATP.

a, Survival of stationary phase E. coli exposed to 4 µg ml⁻¹ PmB in MM ± equimolar concentrations of different sugars, as determined by CFU counts. b, Standard curve plotting Log₁₀ ATP concentration (nM) against Log₁₀ luminescence counts. Simple linear regression was performed using prism version 10.4.1, where the blue line represents the line of best fit, and the red dotted line the 95% confidence interval. c, ATP concentration according to standard curve interpolation, of stationary phase E. coli incubated in equimolar concentrations of different sugars across a 30-minute time course. d, Four-way correlation matrix showing correlation coefficients of PmB susceptibility, ATP concentration and densitometric analysis pre- and post-PmB treatment. All experiments were replicated in n = 3 independent assays. Error bars show the standard deviation of the mean. Source data

Extended Data Fig. 8. PmB killing requires LPS synthesis.

a, b, c Survival of stationary phase P. aeruginosa exposed, or not, to 4 µg ml⁻¹ PmB in MM + G with or without 1X MIC of LpxC inhibitors CHIR-090 (a) or ACHN-975 (b) or the protein synthesis inhibitor tetracycline (c). (d) AFM phase images showing stationary phase E. coli MG1655 cells exposed to 2.5 µg ml⁻¹ PmB in the presence or not of 0.125 µg ml⁻¹ CHIR-090 in MM + G, shown as a function of time. (e) AFM phase images showing stationary phase E. coli imp4231 cells exposed to 2.5 µg ml⁻¹ PmB in the presence or not of 0.5 µg ml⁻¹ G907 in MM + G, shown as a function of time. Scalebar: d, e 250 nm. Phase scale (scale inset in first row of d and e, at t = 90 min): d, 2.5 deg, e 4 deg (top), 6 deg (bottom) All experiments were replicated in n = 3 independent assays. Error bars show the standard deviation of the mean. Significant differences were determined by two-way repeated measures ANOVA between PmB-treated and PmB + antibiotic-treated conditions. P= **<0.01, ns=not significant. Source data

Extended Data Fig. 9. Novobiocin promotes PmB-mediated LPS loss and killing.

Previous work has shown that the antibiotic novobiocin promotes the rate of LPS transport from IM to OM via an interaction with LptB. In turn, this leads to increased susceptibility to polymyxin B, although the mechanism was not established. Based on the findings described in this manuscript, we hypothesised that increased LPS transport to the OM would promote PmB-mediated LPS loss, which correlates with bacterial killing by the antibiotic. a, b, Checkerboard broth microdilution assay showing the synergistic growth-inhibitory interaction between novobiocin and PmB (a), or PmBN (b) against E. coli, as determined OD_595nm after 18 hr incubation and in line with previous findings. Note: PmBN is an inactive polymyxin analogue that served as a useful control for increased entry of novobiocin caused by OM disruption. (c, d) Survival of E. coli exposed, or not, to 1 µg ml⁻¹ PmB (c) or 1 µg ml⁻¹ PmBN (d) with or without 32 µg ml⁻¹ novobiocin. These concentrations were chosen because they showed maximal synergy in the PmBN checkerboard assay. For (c), there was no killing of E. coli by the polymyxin or novobiocin alone, but a significant reduction in the viability of E. coli in the presence of both antibiotics. By contrast, there was no reduction in bacterial viability in the presence of PmBN with novobiocin. Therefore, bacterial killing in these assays appears to be PmB-mediated, rather than increased ingress of novobiocin into cells caused by OM disruption. e, f, Combined OM and IM disruption of stationary phase E. coli exposed to 1 µg ml⁻¹ PmB (e) or 1 µg ml⁻¹ PmBN (f) in MHB with and without 32 µg ml⁻¹ novobiocin, as determined by uptake of the fluorescent dye SYTOX green. Crucially, in the presence of PmBN, novobiocin did not cause membrane disruption, even at concentrations well above those required to inhibit growth in the checkerboard assay. However, novobiocin promoted PmB-mediated membrane disruption in a dose-responsive manner, confirming that novobiocin promotes the activity of PmB by increasing IM disruption, which is the key step for lethality^,. Next, we wanted to examine the impact of novobiocin on PmB-mediated LPS loss. (g) Representative SDS-PAGE image of LPS band intensity of stationary phase E. coli exposed, or not, to 1 µg ml⁻¹ PmB with and without 32 µg/ml novobiocin for 15 min in MHB. (h) Densitometric analysis of the LPS gel in (g), showing that PmB or novobiocin exposure alone at 1 µg ml⁻¹ or 32 µg ml⁻¹ respectively, had minimal impact on LPS abundance after 15 mins incubation (N = 4). However, there was a significant drop in LPS levels when PmB and novobiocin were used in combination at these concentrations. Therefore, we concluded that novobiocin enhances PmB activity by promoting LPS loss, most likely via its reported effect on increasing LPS transport from the IM to OM^,. Unless otherwise stated, all experiments were replicated in n = 3 independent assays. Error bars show the standard deviation of the mean. Significant differences were determined by one- (h) or two-way repeated measures ANOVA (c, d, e, f). P= *<0.05, **<0.01, ***<0.001, ****<0.0001, ns=not significant. Source data

Extended Data Fig. 10. Low-dose PmBN and EDTA exposure has little effect on the OM at the nanoscale.

AFM low and high magnification scans showing stationary phase E. coli MG1655 exposed to 2.5 µg ml⁻¹ PMBN (a) or 10 mM EDTA (b) in the presence of glucose for up to 90 minutes. Scalebars: large scans 250 nm, higher-magnification scans 100 nm; phase scale (scale inset in third row of a and b, at t = 90 min) (from top to bottom): a, 4 deg (row 1) and 0.7 deg (row 3), b, 5 deg (row 1) and 1 deg (row 3). Height scale (scale inset in third row of a and b, at t = 90 min): a, 3 nm, b, 10 nm. (representative images shown of N = 1).