Potent activity of polymyxin B is associated with long-lived super-stoichiometric accumulation mediated by weak-affinity binding to lipid A

Abstract

Polymyxins are gram-negative antibiotics that target lipid A, the conserved membrane anchor of lipopolysaccharide in the outer membrane. Despite their clinical importance, the molecular mechanisms underpinning polymyxin activity remain unresolved. Here, we use surface plasmon resonance to kinetically interrogate interactions between polymyxins and lipid A and derive a phenomenological model. Our analyses suggest a lipid A-catalyzed, three-state mechanism for polymyxins: transient binding, membrane insertion, and super-stoichiometric cluster accumulation with a long residence time. Accumulation also occurs for brevicidine, another lipid A-targeting antibacterial molecule. Lipid A modifications that impart polymyxin resistance and a non-bactericidal polymyxin derivative exhibit binding that does not evolve into long-lived species. We propose that transient binding to lipid A permeabilizes the outer membrane and cluster accumulation enables the bactericidal activity of polymyxins. These findings could establish a blueprint for discovery of lipid A-targeting antibiotics and provide a generalizable approach to study interactions with the gram-negative outer membrane.

Fig. 1. SPR-detection of polymyxin B binding to the gram-negative outer membrane.

A Cartoon of SPR chip surface with bound whole bacterial cells or outer membrane vesicles (OMVs) (created with BioRender.com released under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International license). B Transmission electron micrograph of wild-type OMVs (wt-OMVs) exposed to buffer (left) or polymyxin B (right). Representative images based on n = 3 biological replicates shown. Scale bar = 100 nm. C Scanning electron micrograph of wt-OMVs immobilized via amine-coupled polymyxin B to the surface of a C1 chip. Representative images based on n = 3 biological replicates shown. Scale bar = 300 nm. D SPR binding curves showing a stepwise increase in response (RU) upon exposure of wt-OMVs to eight serial-doubling concentrations of polymyxin B from 39 nM–5 µM in PBS plus 0.0005% tween-80. The tight binding (gray arrow) and reversible binding (red arrow) fractions are indicated and are observed in an approximately 2:1 ratio. SPR curve is representative of n > 3 independent replicates.

Fig. 2. SPR curves for binding of antibiotics to whole bacterial cells or OMVs.

Each serial injection profile was recorded as outlined in Fig. 1 with eight serial doubling concentrations with a maximum concentration of 625 nM. Representative SPR sensorgram of polymyxin B binding to (A) wt-OMVs (253 RU loaded) or (B) resistant-OMVs isolated from E. coli PmrA^G53E cells (250 RU loaded). Representative SPR sensorgrams of whole cell binding by polymyxin B with (C) wild-type E. coli cells (169 RU loaded) or (D) polymyxin B-resistant E. coli cells expressing mcr-1 (102 RU loaded). Representative SPR sensorgrams of PMBN binding with (E) wt-OMVs (512 RU loaded) and (F) resistant-OMVs isolated from E. coli PmrA^G53E cells (681 RU loaded). G Representative SPR sensorgrams of polymyxin B (solid line, 245 RU loaded) or PMBN (dashed line, 206 RU loaded) binding to wt-OMVs in the presence of 32 mM Mg²⁺. H Representative SPR sensorgrams of brevicidine binding to wt-OMVs (black line, 462 RU loaded) or resistant-OMVs isolated from E. coli PmrA^G53E cells (blue line, 422 RU loaded). All traces are representatives of n ≥ 3 independent SPR runs.

Fig. 3. Binding of polymyxin B and PMBN to E. coli cells, OMVs, and LPS and fitting to approximate kinetic models.

Affinity analysis of PMBN binding to (A) E. coli cells and (B) OMVs using a 1:1 kinetic boundary model fit (Eq. (S3)). C PMBN binding to LPS (black curves) and replicated (red curves) after pre-saturation of the surface with polymyxin B, as shown in (D). Both curve sets were fit to Eq. (S3) to estimate affinity. D Saturation of the LPS surface with polymyxin B repeated at two different LPS densities and fit to a simple 1:1 kinetic binding model. Binding of polymyxin B to (E) E. coli cells, and (F) OMVs fit to a 1:1 two-compartment binding model (Eq. (S2)). G Upper curve: binding of polymyxin B to LPS and fit to a two-state kinetic binding model (Eq. (S6)). Lower curve: replicate of upper curve where the LPS surface was pre-saturated with polymyxin B as shown in (D). H Polymyxin B dissociation curves for cells and OMVs pre-saturated with polymyxin B. Polymyxin B occupancy was obtained by chaser SPR analysis, where repeated PMBN injections report changes in polymyxin B occupancy allowing dissociation to be estimated at multiple timepoint readings that were then fit to a two-state dissociation model (Eq. (S7) and Supplementary Information Sections A and D).

Fig. 4. Reaction mechanism for LPS-targeting antibiotics.

A proposed structural model that describes LPS-catalyzed super-stoichiometric accumulation of polymyxin B clusters, cPMB mediated by transition state intermediates is shown. PMBi is injected polymyxin B; PMB is polymyxin B (gray circle); L is LPS (black circle), divalent metal ions are red circles; PMBL is a transient polymyxin B-LPS complex; n indicates available membrane insertion sites (open circles); nPMBL is the membrane-inserted polymyxin B-LPS species; tPMB is a transient nPMBL dimer and LL is a transient LPS dimer; and cPMB is a polymyxin B monomer within a cluster. k_t is a mass transport rate; K_D1 is an affinity constant for mass transport-limited binding. Rate constants in red fonts were used when fitting this model to polymyxin B binding to wt-OMVs in Fig. 5A and rate constants in blue fonts were repeating rate constants that appear in other reaction steps.

Fig. 5. Three-state model fitted to experimental SPR binding curves.

Titration of polymyxin B to a maximum concentration of 625 nM with (A) wt-OMVs and (B) resistant-OMVs. The fitted SPR curves are shown (left, upper panel - SPR data (black), modeled (red)) together with decomposition of one of these binding curves into component species (left, lower panels - PMBL (pink), nPMBL (turquoise), tPMB (dark purple), cPMB (light purple), composite (black)). The fitted model is near superimposable upon the experimental SPR binding curves. C, D 2D fitspace analysis associated with each fitted data set are shown (right panels). Binding constants were constrained to global values per curve set and the resulting parameter values, standard error associated with the fit, confidence intervals, and χ2 values are summarized in Table 2.