Colistin requires de novo lipopolysaccharide biosynthesis for activity

Colistin is a last resort antibiotic for infections caused by highly drug-resistant Gram-negative pathogens such as Pseudomonas aeruginosa [1,2]. Unfortunately, treatment failure is common despite low rates of resistance, and efforts to address this are compromised by our poor understanding of colistin’s mode of action [3-6]. Here, we show that colistin causes dose-dependent membrane disruption and killing of P. aeruginosa via a process that requires de novo lipopolysaccharide (LPS) biosynthesis. Colistin binds to LPS in the outer membrane (OM), leading to disruption of the cation-bridges that stabilise the lipopolysaccharide molecules. Due to the weakening of these bridges, de novo synthesis of LPS results in the release of lipopolysaccharide from the OM. The subsequent loss of membrane integrity enables colistin to access the cytoplasmic membrane (CM), leading to permeabilisation and bacterial lysis. Inhibition of LPS biosynthesis in Escherichia coli and Klebsiella pneumoniae also inhibited the bactericidal activity of colistin, suggesting a conserved mechanism across clinically-relevant pathogens. Together, these findings reveal the mechanism by which colistin kills bacteria and provide the foundations for the development of new approaches to enhance treatment outcomes.


3
To test the current model of colistin activity, we employed a panel of assays to investigate 57 membrane damage, cell lysis and bacterial viability during exposure of P. aeruginosa PA14 to clinically-58 relevant concentrations of the antibiotic [17,18]. Colistin caused immediate and dose-dependent 59 disruption of the OM, as determined by uptake of the normally membrane-impermeant dye N-phenyl-60 1-naphthylamine (NPN) (Fig. 1b, Supplementary Fig. 1). Within 30 min of colistin exposure, P. 61 aeruginosa began to release periplasmic β-lactamase into the culture supernatant, indicative of severe 62 OM damage (Fig. 1c). This was followed by disruption of the CM by 60 min, as detected by staining of 63 cells with the membrane-impermeant DNA-reactive fluorophore propidium iodide (Fig. 1d, 64 Supplementary Fig. 2). Subsequently, bacterial lysis induced by colistin was observed with a decrease 65 in OD595nm from 90 min, and there was a corresponding drop in c.f.u. counts by 2 h (Fig. 1ef). These 66 findings provide a timeline for the current model of polymyxin-mediated bacterial killing, whilst the 67 survival of a sub-population of bacteria exposed to the antibiotic confirmed the existence of colistin-68 tolerant persister cells within P. aeruginosa populations (Fig. 1f, Supplementary Fig. 3). 69 To further investigate colistin-mediated membrane disruption, we measured the release of 70 LPS and phospholipids into the extracellular space. P. aeruginosa released large quantities of 71 phospholipids, most likely in the form of OMVs, in the absence of colistin, but this was almost 72 completely blocked by colistin at bactericidal concentrations ( Fig. 1g) [19]. By contrast, colistin 73 exposure resulted in the time-dependent release of LPS into the culture supernatant (Fig. 1h,74 Supplementary Fig. 4). 75 As reported previously, released LPS completely inhibited the antibacterial activity of colistin 76 at 2 µg ml -1 ( Supplementary Fig. 5) [20]. However, we also considered the possibility that colistin-77 induced LPS release contributed to the bactericidal activity of the antibiotic. This is because the loss 78 of LPS from the OM results in lipid asymmetry and a loss of integrity of the OM, which if uncorrected 79 culminates in cell death in most bacteria [21]. To counteract lipid asymmetry, bacteria increase the 80 production and trafficking of LPS to the OM [22]. Therefore, we hypothesised that colistin activity 81 would be enhanced by blocking LPS biosynthesis, thereby preventing P. aeruginosa from restoring the 82 7 requirement for the lipid tail of the polymyxin, the PMB-mediated release of β-lactamase was also 161 inhibited by cerulenin, confirming the synchronous role of de novo LPS biosynthesis in this process 162 (Fig. 4c). These findings are supported by previous work using scanning electron microscopy that 163 demonstrated PMB caused much greater OM damage than PMBN [29]. 164 The severe nature of the OM damage mediated by the polymyxin lipid tail appeared crucial 165 for the bactericidal activity of the antibiotic, since PMB, but not PMBN caused permeabilisation of the 166 CM, cell lysis and bacterial death, via a mechanism that was dependent upon de novo LPS biosynthesis 167 ( Fig. 4efghij). Therefore, whilst the peptide ring of the polymyxin antibiotics weakens the OM 168 sufficiently to allow the uptake of small molecules, the lipid tail is required, in conjunction with de 169 novo LPS biosynthesis, for the severe damage to the OM that enables PMB to exert its bactericidal 170 effects. 171 Combined, the findings described above, together with other studies, enable the construction 172 of a novel model for the mechanism of action of colistin (Fig. 4k). Firstly, colistin interacts with LPS on 173 the bacterial cell surface via the peptide ring of the antibiotic [11]. This displaces the cations that 174 stabilise LPS molecules in the monolayer, and thus results in minor disruption of the OM [27]. As LPS 175 is synthesised and transported to the cell surface, colistin-induced weakening of the lateral forces 176 between LPS molecules causes LPS to be pushed out and released from the outer leaflet of the OM. 177 The loss of LPS leads to severe OM damage, which is additionally dependent upon the lipid tail of the 178 polymyxin antibiotic. This perturbation to the OM enables the release of periplasmic macromolecules 179 and also facilitates access of the relatively large colistin molecules to the phospholipids in the CM, 180 which the antibiotic disrupts via its detergent-like properties [28]. Once the OM and CM are 181 compromised, the bacteria lyse and die. 182 This revised model addresses key gaps in our understanding of the mechanism of action of 183 polymyxin antibiotics by revealing how these antibiotics cross the OM, explaining why the polymyxin 184 acyl tail is required for bactericidal activity and providing a direct link between OM disruption and the 185 bactericidal activity of the antibiotic. Such insight is needed because colistin is an increasingly 186 8 important antibiotic for the treatment of infections caused by multi-drug resistant pathogens, but has 187 a treatment failure rate of >70% [3][4][5][6]. This study provides two explanations for the extremely high 188 rate of treatment failure. Firstly, bacteria release LPS in response to colistin, which sequesters and 189 inactivates the antibiotic [20]. Secondly, we identify LPS biosynthesis as a bacterial process that could, 190 when decreased, confer colistin tolerance to persister cells with reduced metabolic activity [14,15].

Emergence of colistin resistance assay 269
Bacteria in stationary-phase were washed and added to 3 ml MHB containing colistin (2 μg ml -1 ), and 270 survival was determined every 2 h up to 8 h as described above. Following 8 h incubation, the MIC of 271 colistin against the recovered bacterial population was determined by the broth microdilution method 272 as stated above. The recovered bacteria were washed by centrifugation (12,300 x g, 3 min) and 273 resuspension in fresh LB, before being inoculated into 3 ml LB and grown for 18 h to stationary-phase 274 in a shaking incubator (37°C, 180 r.p.m.). This population was then exposed to colistin (2 μg ml -1 ) for a 275 second time, with bacterial viability again measured over 8 h. After the second colistin exposure, the 276 MIC of colistin was again determined, before the bacteria were exposed to colistin for a third time, 277 with associated measurements of survival and the colistin MIC of the recovered population obtained. 278

Determination of membrane lipid release from bacteria 279
The supernatant from bacterial cultures exposed to colistin was recovered every 2 h by centrifugation 280 (12,300 x g, 3 min). Recovered supernatants were mixed with 5 μl of FM-4-64 styryl dye (Thermo Fisher 281 Scientific) at a final concentration of 5 μg ml -1 in the wells of a black-walled microtitre plate. 282 Fluorescence was measured using a Tecan Infinite 200 Pro multiwell plate reader (excitation at 565 283 nm, emission at 600 nm) to quantify the phospholipid released into the supernatant by bacteria 284 exposed, or not, to colistin. 285

Determination of LPS release from bacteria 286
The chromogenic Limulus Amebocyte Lysate (LAL) assay (all reagents from Thermo Fisher Scientific) 287 was used to detect and quantify the LPS released from bacteria into the culture supernatant as 288 described previously [34]. Samples of cell-free culture supernatant (50 μl) were equilibrated to 37°C 289 and loaded into the wells of a microtitre plate at the same temperature. Limulus amebocyte lysate 290 reagent (50 μl) was added to each well, and the mixture incubated at 37°C for 10 min. Chromogenic 291 substrate solution (100 μl, 2 mM) was subsequently added to each well and the microtitre plate was 292 incubated for a further 6 min at 37°C. The enzymatic reaction was stopped by adding 50 μl of 25% 293 acetic acid to each well, and the presence of LPS was determined by measuring absorbance at 405 nm 294 in a Tecan Infinite 200 Pro multiwell plate reader. A standard curve was generated using an E. coli

Extraction of LPS and visualisation by silver staining 300
LPS was extracted from the supernatant of bacterial cultures as previously described, and the ultrafast 301 silver staining method was used for visualising LPS on 12% SDS-PAGE gels [35,36]. Stationary-phase 302 bacteria (1 ml) were washed and inoculated to a final concentration of 10 8 c.f.u. ml -1 into 9 ml MHB 303 containing the relevant antibiotics. Bacterial cultures were incubated with shaking (37°C, 180 r.p.m.) 304 for up to 8 h, and 9 ml of supernatant was recovered following centrifugation (3270 x g, 30 min) and 305 filter sterilisation (0.2 µm filter) to remove bacterial cells. Absolute ethanol (30 ml) was added to the 306 supernatant and the samples were stored at -20°C for 30 min. The precipitate was pelleted by 307 centrifugation at 3270 x g for 30 min and resuspended in 500 µl Laemmli buffer. Precipitated protein 308 was removed by digesting with proteinase K (50 µg) overnight, before 25 µl was run on an acrylamide 309 mini-gel system. For certain experiments, the concentration of LPS in these samples was determined 310 using the LAL assay described above. Electrophoresis was carried out at 12 mA in the stacking gel and 311 25 mA in the separating gel. 312 LPS preparations subjected to SDS-PAGE were visualised by first oxidising the gel with 200 ml 40% 313 ethanol-5% acetic acid containing 0.7% sodium periodate for 20 min at room temperature. The gel 314 was washed three times with distilled water for 5 min each time. Silver staining solution was freshly 315 prepared by first adding concentrated ammonium hydroxide (4 ml) to 0.1 M sodium hydroxide (56 ml) 316 before adding 200 ml of distilled water. Silver nitrate (20% w/v, 10 ml) was added in drops whilst 317 stirring, before the addition of 30 ml of distilled water. This solution was used to stain the gel for 10 318 min, which was subsequently washed with distilled water for 5 min three times. The silver stain was 319 developed with 200 ml of water containing formaldehyde (100 µl of a 37% w/v solution) and 10 mg 320 citric acid. Development was stopped using 10% acetic acid before the gel was washed in distilled 321 water and then imaged using a Bio-Rad Gel Doc EZ Imager (Bio-Rad Laboratories). 322 323 324

Determination of colistin activity 325
The activity of colistin during incubation with P. aeruginosa was determined using an established zone 326 of inhibition assay [37]. Bacterial cultures exposed to colistin were centrifuged (12,300 x g, 3 min), and 327 the supernatant recovered. Stationary-phase P. aeruginosa was diluted in MHB to a concentration of 328 10 6 c.f.u. ml -1 , before 60 μl of this bacterial culture was spread across the surface of an MHA plate and 329 allowed to air dry. Wells measuring 10 mm in diameter were made in the agar plate and filled with 330 130 μl of the culture supernatant recovered from colistin-treated bacterial populations. Agar plates 331 were incubated in air statically at 37°C for 16 h and the zone of growth inhibition around each well 332 was measured at four perpendicular points. A standard plot was generated showing a linear 333 relationship between the size of the zone of inhibition and colistin concentration, which enabled 334 inhibitory zone size to be converted to percentage colistin activity. Colistin activity was also measured 335 after 4 h incubation (37°C, end-over-end rotation) in the presence of purified LPS, or in MHB alone 336 over the course of 8 h in a shaking incubator (37°C, 180 r.p.m.). 337

Determination of colistin binding to bacterial cells 338
Colistin was labelled with the fluorophore BoDipy FL SE D2184 (Thermo Fisher Scientific) by incubating 339 100 μl of the BoDipy NHS ester compound (10 mg ml -1 in dimethyl sulfoxide, DMSO) with 250 μl colistin 340 (10 mg ml -1 in water) and 650 μl sodium bicarbonate (0.2 M, pH 8.5) for 2 h at 37°C. BoDipy molecules 341 that had not bound to colistin were removed by dialysis using a Float-A-Lyser G2 dialysis device 342 (Spectrum Laboratories, USA) that had a molecular weight cut-off of 0.5 kDa. Dialysis was carried out 343 at 4°C against sterile distilled water, which was changed four times during the course of the 24 h 344 dialysis period. Time of flight mass spectrometry analysis confirmed successful labelling with the 345 fluorophore, and the antibiotic activity of BoDipy-labelled colistin was assessed using MIC and 346 bacterial survival assays, as described above.

Statistical analyses 374
Experiments were performed on at least three independent occasions, and resulting data are 375 presented as the arithmetic mean of these biological repeats, unless stated otherwise. Error bars 376 represent the standard deviation of the mean. For single comparisons, a two-tailed unpaired Student's 377 t-test was used to analyse the data. For multiple comparisons at a single time point or concentration, 378 data were analysed using either a one-way analysis of variance (ANOVA) for parametric data, or a 379 Kruskal-Wallis test for non-parametric data. Where data were obtained at several different time points 380 or concentrations, a two-way ANOVA was used for statistical analysis. Appropriate post-hoc tests 381 (Dunnett's, Welch's, Sidak's, Holm-Sidak) were carried out to correct for multiple comparisons, with 382 details provided in the figure legends. Asterisks on graphs indicate significant differences between 383 data, and the corresponding p-values are reported in the figure legend. All statistical analyses were 384 performed using GraphPad Prism 7 software (GraphPad Software Inc., USA). 385

Data availability 386
The data that support the findings of this study are available from the corresponding author upon 387 reasonable request. 388 389 390