Structure-Interaction Relationship of Polymyxins with the Membrane of Human Kidney Proximal Tubular Cells

Abstract

Multidrug-resistant Gram-negative bacteria are a serious global threat to human health. Polymyxins are increasingly used in patients as a last-line therapy to treat infections caused by these life-threatening ‘superbugs’. Unfortunately, polymyxin-induced nephrotoxicity is the major dose-limiting factor and understanding its mechanism is crucial for the development of novel, safer polymyxins. Here, we undertook the first all-atom molecular dynamics simulations of the interaction between four naturally occurring polymyxins A₁, B₁, M₁ and colistin A (representative structural variations of the polymyxin core structure) and the membrane of human kidney proximal tubular cells. All polymyxins inserted spontaneously into the hydrophobic region of the membrane where they were retained, although their insertion abilities varied. Polymyxin A₁ completely penetrated into the hydrophobic region of the membrane with a unique folded conformation, whereas the other three polymyxins only inserted their fatty acyl tails into this region. Furthermore, local membrane defects and increased water penetration were induced by each polymyxin, which may represent the initial stage of cellular membrane damage. Finally, the structure-interaction relationship of polymyxins was investigated based on atomic interactions at the cell membrane level. The hydrophobicity at positions 6/7 and stereochemistry at position 3 regulated the interactions of polymyxins with the cell membrane. Collectively, our results provide new mechanistic insights into polymyxin-induced nephrotoxicity at the atomic level and will facilitate the development of new-generation polymyxins.

Graphical Abstract

Nephrotoxicity of polymyxins has significantly limited their clinical use. Here we discovered, for the first time, the insertion of polymyxins into the membrane of human renal tubular cells and developed the membrane-based structure-interaction relationship of polymyxins. Our results provide novel insights into the mechanism of polymyxin-induced nephrotoxicity.

The polymyxins (i.e. polymyxin B and colistin) are an ‘old’ class of cyclic lipopeptide antibiotics that are active against Gram-negative bacteria.¹ Subsequent to their introduction into the clinic around sixty years ago, a high incidence of nephrotoxicity and neurotoxicity resulted in a decline in their use beginning in the 1970s.^1-2 However, the emergence of multidrug-resistant (MDR) Gram-negative ‘superbugs’ over the last two decades, combined with a dwindling antibiotic discovery pipeline, has resulted in a resurgence of their use over the last two decades.^2-3 Polymyxins are now increasingly used as a last-line therapy to combat these problematic Gram-negative pathogens, particularly nosocomial infections caused by Pseudomonas aeruginosa, Acinetobacter baumannii and Klebsiella pneumoniae.^4-5 Unfortunately, polymyxin-induced nephrotoxicity occurs in up to 60% of patients following intravenous administration of the currently recommended dosage regimens.^6-9 The nephrotoxicity of polymyxin B and colistin is dose-limiting and as such substantially reduces their therapeutic indices. This, in turn, has significantly limited the ability to optimize their use in the clinic based on their pharmacokinetics/pharmacodynamics.¹⁰ Such suboptimal use may result in treatment failure and drive the emergence of resistance, further limiting treatment options in patients.¹¹ Therefore, there is an urgent need to identify the molecular mechanism of polymyxin-induced nephrotoxicity and the chemical biology information is crucial for the discovery of novel, safer polymyxins.

Following filtration in the glomeruli, polymyxins undergo tubular reabsorption and accumulate extensively in kidney tubular cells.^12-14 Previous studies reported that two major transporters, megalin and PEPT2, play a key role in the reabsorption of polymyxins in kidney tubular cells.^15-16 We reported that the significant intracellular accumulation of polymyxins results in mitochondrial damage, release of reactive oxygen species and cellular apoptosis.^17-18 Additionally, in a manner that is similar to the permeabilization of the bacterial outer membrane, previous molecular dynamics (MD) simulations have shown that polymyxins are able to insert their hydrophobic tail into other mammalian-like phospholipid bilayers, thereby enhancing membrane permeability.^19-20 Polymyxin B has also been shown to induce selective lipid exchange by interacting with the monoanionic phospholipid molecules in model membranes consisting of phosphatidylglycerol and phosphatidic acid;²¹ binding of polymyxins to the phospholipid bilayer may change the electron density profile of the membrane.²² Together, the current literature suggests that direct interactions between polymyxins and the phospholipid membrane of kidney tubular cells very likely contributes to polymyxin-induced nephrotoxicity. To date, the atomic-level mechanism of this interaction remains unclear, which has significantly limited our understanding of the structure-toxicity relationship of polymyxins and the discovery of new-generation polymyxins.^{4, 6, 23} In the present study, we simulated a membrane model of human kidney proximal tubular (HK-2) cells²⁴ and employed all-atom MD simulations to systematically investigate the interactions of four naturally occurring polymyxins A₁, B₁, M₁ and colistin A. We demonstrated that upon exposure to polymyxins, the structural and dynamic properties of the membrane were perturbed, which increased the permeability of the membrane to water. Finally, the atomic-scale structure-interaction relationship was elucidated for polymyxins with the kidney cell membrane.

Results

Polymyxins bind to the membrane of human kidney proximal tubular cells

In the initial state of each simulation system, polymyxin A₁ (PMA₁), B₁ (PMB₁), M₁ (PMM₁) and colistin A molecules (Figure 1) rapidly bound to the membrane surface after approximately 2-ns of simulation time, through favourable interactions with the phospholipid headgroups in the outer leaflet of the HK-2 cell membrane (Figure 2A). To examine the role of each polymyxin residue in the initial binding process, the number of contacts between each residue and the membrane components was calculated (Figure 2B). For PMB₁, the fatty acyl tail and five Dab residues at positions 1, 3, 5, 8 and 9 formed the majority of the overall contacts with the membrane. In contrast, the residues playing a key role in binding to the membrane were Dab3, Dab5 and Dab9 for colistin A; Dab1 and D-Dab3 for PMA₁; and Dab1, Thr7 and Dab9 for PMM₁. These results indicate that the positively charged Dab side chains were largely responsible for the initial binding of polymyxins to the membrane surface. In our simulations the polymyxin molecules did not show obvious aggregation on the phospholipid membrane of human kidney tubular cells within the initial binding phase (Figure S1).

Figure 1. Chemical structures of four naturally occurring polymyxins A₁, B₁, M₁ and colistin A.

Polymyxins are amphipathic lipopeptides consisting of a cyclic heptapeptide ring, a linking tripeptide, and a fatty acyl tail. Five positively charged α,γ-diaminobutyric acid (Dab) residues are located at positions 1, 3, 5, 8, and 9. Positions 3, 6 and 7 represent variable regions within the polymyxin core structure and contain residues with different stereochemistry at position 3 and hydrophobicity at positions 6/7 in the chosen polymyxins (shown in red). L/D-Dab = L/D-2,4-diaminobutyric acid, L-Thr = L-threonine, D-Phe = D-phenylalanine, L-Leu = L-leucine.

Figure 2. Simulation of the binding of polymyxin A₁, B₁, M₁ and colistin A to the human kidney proximal tubular cell membrane.

(A) The initial configuration of the simulation system. The phosphate atoms of phospholipids are shown in orange spheres and the hydrocarbon tails are shown in grey lines. The polymyxin molecules are shown in grey sticks with red oxygen atoms and blue nitrogen atoms. Water molecules are depicted in lines with red oxygen atoms and white hydrogen atoms. For clarity, the ions are not shown. (B) The number of contacts between polymyxin residues and the membrane during the initial binding process (i.e. the first 2 ns). Contact was considered to be made if the distance between the heavy atom of the polymyxin residue and any heavy atom of membrane lipids was shorter than 5 Å. The contact number was counted with the precision of 10 ps. (C) Lipid composition of the polymyxin-binding regions of the membrane. PC, phosphatidylcholine; PE, phosphatidylethanolamine; PS, phosphatidylserine.

Next, we examined the phospholipids that interacted with each of the four polymyxins to define the binding selectivity. The lipid composition of the polymyxin-binding regions of the membrane are shown in Figure 2C. Of all the phospholipids interacting with each of the four different polymyxins, phosphatidylcholine (PC) was involved in approximately 40% (range: 34 - 44%) whereas phosphatidylethanolamine (PE) was involved in approximately 20% or less (range: 12 - 21%). In contrast, phosphatidylserine (PS) was involved in over 40% of the interactions with PMB₁, PMA₁ and PMM₁ (range: 44 - 50%); and 33% for colistin A. It appears that long-range electrostatic interactions between the Dab residues of each polymyxin molecule and the negatively charged phosphate and carboxyl groups of phospholipids play a major role in the initial binding of polymyxin molecules onto the membrane surface.

Polymyxins insert into the membrane of human kidney proximal tubular cells

The spontaneous insertion of polymyxin molecules into the hydrophobic region of the HK-2 cell membrane was observed across all four types of polymyxins (Figure 3A). Six molecules of polymyxin A₁, B₁, M₁ or colistin A were included in each simulation system and only one polymyxin molecule eventually inserted into membrane. We tracked the relative positions of the inserted polymyxin molecules in the membrane and visualized their insertion kinetics (Figure 3B and S2). After binding to the membrane surface, PMB₁, PMM₁ and colistin A only inserted their fatty acyl tails into the hydrophobic layer of the membrane, while leaving the cyclic peptide rings above the head groups of the HK-2 membrane lipids. Surprisingly, PMA₁ completely penetrated into the hydrophobic core of the membrane after approximately 60-ns of simulation time and adopted a folded conformation therein. The PMA₁ folded conformation involved distortion around its D-Dab3 with its hydrophobic segments (i.e. fatty acyl tail and D-Leu6) oriented toward the membrane center (Figure 3A).

Figure 3. Polymyxin A₁, B₁, M₁ and colistin A insert into the hydrophobic region of the kidney cell membrane.

(A) Snapshots show the insertion of polymyxin A₁ B₁ M₁ and colistin A into the hydrophobic region of the membrane of human kidney proximal tubular cells. The averaged shell of phosphate atoms of the membrane is depicted to show the positions of each polymyxin relative to the headgroups of the membrane lipids. Atoms #1 and #2 represent the last and first carbon atoms of the fatty acyl tail of polymyxins, respectively, and Atom #3 indicates the C_α of D-Phe6 on polymyxin B₁ (D-Leu6 in the other three polymyxins). (B) The time-evolved coordinates of the selected carbon atoms along the Z-axis. The line colours correspond to that of the atomic spheres in Fig 3 A.

Polymyxins affect the physical properties of the membrane of human kidney proximal tubular cells

To further understand the effect of polymyxin insertion upon the membrane we examined a number of physical properties of the membrane in the absence and presence of each polymyxin. To this end, the lateral diffusion and order parameters of the phospholipids in the membrane were measured to characterize the dynamic properties of the membrane. In the absence of polymyxin molecules, the average lateral diffusion of the membrane was 7.83 (+/− 0.69) × 10⁻⁸ cm² s⁻¹. In contrast, the average lateral diffusion in the presence of polymyxins decreased to 3.69 (+/− 0.69) × 10⁻⁸ cm² s⁻¹ with PMB₁, 5.15 (+/− 0.57) × 10⁻⁸ cm² s⁻¹ with colistin A, 4.11 (+/− 0.65) × 10⁻⁸ cm² s⁻¹ with PMA₁ and 4.37 (+/− 0.24) × 10⁻⁸ cm² s⁻¹ with PMM₁ (Figure S3); the order parameters of the phospholipid carbon tails were also decreased (Figure S4). These results indicate that polymyxin treatments reduced the membrane fluidity by forming non-covalent contacts that cross-linked neighboring phospholipid molecules. Furthermore, we measured the thickness of the bilayer and the area per lipid to investigate the potential effect of polymyxins on the membrane structure (Table S1). Although the overall membrane thickness and area per lipid molecule did not change significantly during polymyxin treatment, interestingly, the region of the membrane around the polymyxin insertion site was significantly thinned (Figure S5). This indicates that polymyxins primarily caused a localized disorganization of the membrane structure. Furthermore, a considerable number of water molecules entered the hydrophobic region of the membrane as a consequence of the insertion of PMB₁ and PMA₁; whereas no observable water penetration occurred in the membrane systems treated with colistin A and PMM₁ (Figure 4). These results indicate that the insertion of certain polymyxins into the hydrophobic layer of the membrane might enhance water permeability in human kidney proximal tubular cells.

Figure 4. Penetration of water molecules into the membrane due to polymyxin treatment.

Water molecules are shown with sphere models with red oxygen atoms and white hydrogen atoms. Polymyxins are shown with green stick models and membrane lipids with grey line models. The averaged phosphate shell is shown to indicate the boundary between the membrane and bulk water.

Atomic force microscopy (AFM) was undertaken to measure the membrane roughness of the HK-2 cells with and without polymyxin B treatment (Figure 5). Compared to the control samples, the membrane roughness of HK-2 cells significantly increased following treatment with 25 μM polymyxin B for 60 min, with further increases (a more uneven distribution) with the higher polymyxin B concentration (100 μM). Consistent with the simulation observations, our AFM results indicate the disorganization effect of polymyxins on the HK-2 cell membrane.

Figure 5. Membrane roughness of HK-2 cells following the treatment with polymyxin B for 60 min.

R_a: averaged roughness; R_q: root-mean-square roughness. Data are presented as means ± S.D. for the measured R_a and R_q of each experiment.

Structure-interaction relationship of polymyxins with the membrane of human kidney proximal tubular cells

To decipher the structure-interaction relationship of polymyxins with the membrane of HK-2 cells, the interaction complex was analyzed at the atomic level and the interaction energy (IE) for each polymyxin residue was determined (Figure 6). PMB₁ inserted into the membrane and its fatty acyl tail was embedded into the hydrophobic region of the membrane, where it interacted with the hydrocarbon tails of surrounding phospholipids (IE = −64.4 KJ mol⁻¹). Dab1, Dab3 and Dab5 interacted strongly with the phospholipid head groups in the membrane, with an averaged IE of −264.1 KJ mol⁻¹, −102.9 KJ mol⁻¹ and −219.8 KJ mol⁻¹, respectively. In contrast, the interactions contributed by Dab8 gradually diminished when D-Phe6 started to form hydrophobic contacts with the membrane phospholipids (approximately after 20-ns simulations). Dab9 made only marginal contacts (IE = −7.1 KJ mol⁻¹) with the membrane during the entire penetration trajectory.

Figure 6. Interaction of polymyxin A₁, B₁, M₁ and colistin A with the membrane of human kidney proximal tubular cells.

The polymyxin molecules are shown in blue sticks; PE, PC and PS are shown in cyan, orange and green sticks, respectively. The key residues of polymyxins responsible for the interaction and their averaged interaction energy are labelled. The color spectrum from yellow to blue represents the increase of interaction intensity. The time points of polymyxin penetration are shown with red arrows.

For colistin A, the fatty acyl tail was also surrounded by and interacted with the hydrocarbon tails of phospholipids (IE = −28.4 KJ mol⁻¹). The positively charged side chains of Dab1 and Dab3 formed electrostatic interactions with the head groups of adjacent phospholipid molecules, albeit, their interactions were not as stable as Dab5 during the insertion phase (i.e. after 40-ns of simulation). Notably, for colistin A, the interaction of D-Leu6 was less than its counterpart D-Phe6 in PMB₁ (d-Leu6 IE = −1.9 KJ mol⁻¹ vs. D-Phe6 IE = −44.3 KJ mol⁻¹), whereas Dab8 (IE = −162.1 KJ mol⁻¹) and Dab9 (IE = −241.1 KJ mol⁻¹) of colistin A strongly bound to the head group region of the membrane.

During the membrane insertion of PMA₁, both the fatty acyl tail and D-Leu6 formed hydrophobic contacts with the carbon tails of the neighboring phospholipid molecules. The Dab1, Dab5 and Dab8 residues of PMA₁ bound to the lipid head groups strongly during the initial phase of membrane penetration (i.e. 40 – 100 ns of simulation), while Dab9 strongly interacted with phospholipid head groups only after PMA₁ inserted into the membrane (i.e. 60 – 100 ns of simulation). Interestingly, Thr7 formed two hydrogen bonds with the carbonyl groups of the phospholipid fatty acyl chains. The side chain of D-Dab3 adopted an outward orientation and hardly interacted with the head groups of any membrane phospholipids.

With PMM₁, its fatty acyl tail formed hydrophobic interactions with the hydrocarbon tails of membrane phospholipids. Dab1, Dab3, Dab5 and Dab9 made intensive electrostatic interactions with the phosphate and/or carboxyl groups of phospholipids, while the interaction from Dab8 was not very stable when PMM₁ inserted into the membrane (i.e. after 25-ns of simulation). Unlike PMA₁, Thr7 of PMM₁ did not form any hydrogen bonds with any membrane phospholipids due to the longer distance with the carbonyl groups of the phospholipid fatty acyl chains.

Given that the Dab residues of all four polymyxins examined played a dominant role in the overall interaction with the membrane, we mapped their interaction targets onto the membrane phospholipid structures (Figure S6). It is evident that the negatively charged moieties on phospholipid molecules (e.g. phosphate groups of PC, PE and PS; and the carboxyl group of PS) were the main binding sites for polymyxin Dab residues. However, each polymyxin analogue displayed distinct binding patterns. For PMB₁ and PMA₁, Dab1 predominately bound to the phosphate group of PC. Whereas in colistin A and PMM₁ systems Dab1 was found to mainly bind to the carboxyl group of PS (Figure S6). These results indicate that the interaction between the polymyxins and the membrane of human kidney proximal tubular cells was primarily dependent on the electrostatic interactions between the positively charged Dab residues of polymyxins and negatively charged moieties of phospholipids.

Discussion

To date, very little is known on the structure-toxicity relationship of polymyxins. Understanding the mechanism of polymyxin-induced nephrotoxicity is of great importance for the development of novel, safer polymyxins to treat infections caused by MDR Gram-negative pathogens. Here, we employed all-atom MD simulations with an HK-2 cell membrane model to examine the interactions of four naturally occurring polymyxins which contain structural variations at the variable regions within the polymyxin core structure. For the first time, we provide important evidence that polymyxins can spontaneously insert into the hydrophobic region of the HK-2 cytoplasmic membrane. Importantly, the insertion abilities of each polymyxin varied considerably, which was consistent with the nephrotoxicity of each polymyxins in human and mice.^25-26 This in turn suggests that such differences may result from stereo-specific structural differences that affect the ability of the polymyxin to insert into, and locally disorganize, the membrane of kidney tubular cells.

Polymyxin-induced nephrotoxicity is eventually a consequence of their significant accumulation within kidney tubular cells wherein they cause mitochondrial damage, release of reactive oxygen species and eventually apoptosis followed by cell death.^{6, 17-18} However, the initial interaction between the positively charged polymyxins and the negatively charged membrane of kidney tubular cells has never been investigated. Our all-atom molecular dynamics simulation results revealed that, prior to uptake into the tubular cells, polymyxin molecules interact with the outer leaflet of the cell membrane, a process primarily driven by electrostatic interactions between the positively charged Dab residues of polymyxins and the negatively charged phosphate and carboxyl groups of phospholipids (Figure 2). Following this initial interaction, the fatty acyl tails of the polymyxins (and for PMA₁, the entire molecule) spontaneously penetrate into the hydrophobic region of the membrane (Figure 3), altering the membrane structural and dynamic properties (Table S1, Figure 4 and S5). Additionally, the AFM measurements experimentally confirmed that polymyxin B treatment disorganized the membrane of HK-2 cells and significantly changed the membrane roughness (Figure 5). Similarly, our simulation findings demonstrated that polymyxins disorganized and penetrated into the membrane of human kidney proximal tubular cells, which can subsequently interfere with cell function and cause toxicity.

A previous investigation utilizing molecular dynamics simulations and the same membrane phospholipid compositions as used in the present study reported that polymyxin B molecules clustered within the kidney cell membrane, cooperatively disorganizing the membrane structure.²² However, in that study the polymyxin molecules were manually placed into the membrane, a situation that may not reflect the real interaction scenario between polymyxin molecules and the kidney cell membrane. In contrast, the insertion of polymyxins in our study occurred spontaneously. In the inserted state, the polar cyclic ring of polymyxins formed a complicated interaction network with the headgroups of phospholipids in the membrane (Figure 6), consistent with previous simulation studies of polymyxin interactions with phospholipid bilayers. ^19-20 These polar and electrostatic interactions likely impede further penetration of polymyxins across the membrane, resulting in their accumulation within the outer leaflet of the membrane. Unlike the aggregation of polymyxins which has been observed on the surface of the bacterial outer membrane,²⁰ our results in the present study indicated that the polymyxin molecules inserted into the kidney cell membrane as monomers. This difference is likely due to the structural and chemical differences between lipopolysaccharide (the major component of bacterial outer membrane) and phospholipid molecules (the major component of mammalian cell membrane). Lipopolysaccharide has more negatively charged groups (e.g. typically two phosphate groups per lipid A) and a larger molecular size than the common membrane phospholipids. Therefore, the bacterial outer membrane provides specific binding sites for polymyxins on its surface.

To investigate the structure-interaction relationship of polymyxins with the kidney tubular cell membrane, we utilized four naturally occurring polymyxins that contained representative structural variations (stereochemistry at position 3 and hydrophobicity at positions 6/7) at the variable regions within the polymyxin core structure (Figure 1). Interestingly, minor structural changes at positions 3, 6 and 7 significantly impacted their ability to insert into the tubular cell membrane (Figure 3). PMB₁, colistin A and PMM₁ only inserted their fatty acyl tails into the hydrophobic region of the membrane, with the cyclic heads of colistin A and PMM₁ more favourably bound to the negatively charged membrane surface compared to PMB₁ (Figure 3). Considering the only structural differences between colistin A, PMM₁ and PMB₁ occurred at positions 6 and 7, our results indicate that the structural characteristics at these two positions are key to the binding of the polymyxin cyclic head to the membrane surface. In contrast to d-Leu6-l-Leu7 in colistin A and d-Leu6-l-Thr7 in PMM₁, the longer side chain of d-Phe6-l-Leu7 in PMB₁ exerted larger steric hindrance, preventing Dab8/Dab9 from interacting with the membrane and thereby inhibiting the surface-binding of the cyclic head of PMB₁. Additionally, the repulsively electrostatic interactions between the positively charged side chain of polymyxin Dab residues and the choline group of PC or ethanolamine group of PE is likely another factor in preventing the cyclic ring of PMB₁ from approaching the membrane surface. Interestingly, the change in the stereochemistry of Dab3 significantly affected the insertion dynamics of polymyxins. The side chain of l-Dab3 adopted an inward orientation in PMB₁, colistin A and PMM₁, whereas the side chain of d-Dab3 in PMA₁ protruded outwards (Figure 6). Due to the difference in the conformational orientation, d-Dab3 of polymyxin A1 formed little interaction with the membrane lipids, whereas l-Dab3 in PMB₁, colistin A and PMM₁ interacted with the membrane lipids intensively (Figure 6). Another difference is that the hydroxyl group of l-Thr7 of PMA₁ formed two hydrogen bonds with the carbonyl groups of the neighboring phospholipid molecule, whereas l-Leu7 in PMB₁ and colistin and l-Thr7 in PMM₁ did not form such interactions with the phospholipids. The different interaction patterns resulted in the formation of different conformations when different polymyxins interact with the membrane. Within the folded conformation, the fatty acyl tail and d-Leu6 of PMA₁ help to shield a part of structural polarity of the polymyxin molecule, thereby enhancing the penetration of PMA₁ into the hydrophobic layer of the kidney tubular cell membrane. PMB₁, colistin A, PMM₁, PMA₁ interacted differently with the various phospholipid molecules of the renal tubular cell membrane, and their different insertion abilities may in part explain their different abilities to induce nephrotoxicity. For example, polymyxin A was shown more nephrotoxic in patients than polymyxin B and colistin.²⁵ In our study PMA₁ completely inserted into the membrane whereas PMB₁, colistin A and PMM₁ only partially inserted; therefore, PMA₁ caused more severe membrane damage, which is consistent with the order of their different in vivo nephrotoxicities (Table S2).

Our results clearly show that differences in the structural characteristics of different polymyxins substantially affect their interactions with the membrane of human kidney proximal tubular cells, and it may be possible to minimize the nephrotoxicity of polymyxins by modifying the core structure. Unfortunately, most structural modifications to date that sought to minimize toxicity have also compromised the antimicrobial activity.²⁷ Structure-activity relationship studies of polymyxins have shown that both the Dab residues and the fatty acyl tail are essential for antimicrobial activity, contributing to the electrostatic and hydrophobic interactions, respectively, when binding to the bacterial outer membrane.^{4, 28} Unfortunately, the key polymyxin residues involved in the interaction with the kidney cell membrane, namely most Dab residues and the hydrophobic tail, are also those which are indispensable for their antibacterial activity. This situation poses a significant challenge when considering how to minimize toxicity while simultaneously retaining activity. Our MD results indicated that shortening the side chain of the Dab residues might reduce the electrostatic interactions between polymyxins and the phospholipids in kidney cell membrane, thereby minimizing the nephrotoxicity. We subsequently conducted MD simulations using our newly developed polymyxin analogue FADDI-287 (Figure S7) which is structurally similar to colistin A but shows much less nephrotoxicity in mice.²⁹ Interestingly, under the same simulation conditions as the other four polymyxins, FADDI-287 molecules only bound to the membrane surface and did not insert into the kidney cell membrane during the simulations (Figure S7). This result further indicates that a polymyxin molecule which cannot insert the kidney cell membrane in our MD systems is very likely less nephrotoxic. More chemical biology research is required to elucidate the complex structure-activity-toxicity relationships of polymyxins in order to uncouple their nephrotoxicity from the antibacterial activity.

To the best of our knowledge, this is the first systematic study to characterize the interactions of structurally different polymyxins with the membrane of human kidney proximal tubular cells using all-atom MD simulations. Different polymyxins can spontaneously insert into the hydrophobic region of the kidney tubular cell membrane with different degrees, perturbing the structure and dynamics of the membrane, and increasing the permeability to water. Importantly, novel structure-interaction relationships of polymyxins were illustrated at the membrane and atomic level. These important findings improve our understanding of the mechanism underlying polymyxin nephrotoxicity and will assist in the development of novel, safer polymyxins to treat life-threatening infections caused by Gram-negative ‘superbugs’.

Methods

System preparation

The lipid composition of kidney epithelial cell membranes reported by Marquez et. al.,²⁴ was used to construct membranes with a composition of 50% PC, 25% PE and 25% PS. In total, 128 phospholipid molecules were employed to construct the membrane model which provided sufficient surface area for the interaction of six polymyxin molecules. There was less than 5% cholesterol in the membrane of kidney epithelial cells,²⁴ and our preliminary simulations revealed that cholesterol did not exert any significant effect on the membrane insertion of polymyxin B₁ (Figure S8). Therefore, cholesterol was not included in our membrane model. PC and PE contain a cationic choline or ethanolamine group linked to their phosphate groups, whereas PS contains an anionic serine group linked to its phosphate group. Within the membrane model PS is negatively charged whereas PC/PE are neutrally charged at physiological pH. CHARMM-GUI Membrane Builder was used to assemble the symmetric all-atom phospholipid bilayers model in a tetragonal box.³⁰ TIP3P water molecules and sodium ions were added to hydrate and neutralize the simulation system, respectively.

Four naturally occurring polymyxins, namely polymyxin B₁ (PMB₁), colistin A, polymyxin A₁ (PMA₁) and polymyxin M₁ (PMM₁) were investigated (Figure 1). The 3D structures of polymyxins were constructed in Chem3D and energy minimization was performed on the built structures to relieve any potential steric clashes. The topology parameters of polymyxins were generated in SwissParam server compatible with CHARMM all atom force field.³¹ Six polymyxin molecules (polymyxin:phospholipid = 1:10) were randomly placed above the bilayer so that the minimum distance between the centre of mass of polymyxin molecules and the surface of the bilayer was larger than 1.5 nm. Correspondingly, the overlapping water molecules and an appropriate amount of sodium ions were removed to maintain the charge neutralization of the whole system. Details of the simulation systems are provided in the Supporting Information (Table S3).

Molecular dynamics simulations

All MD simulations were performed using GROMACS 5.1.2 (www.gromacs.org) with CHARMM36 all-atom lipid parameters.³² For each system, energy minimization was performed using the steepest descent algorithm with the tolerance of 1000 kJ mol^−1.nm⁻¹. The standard Membrane Builder six-step equilibration process was carried out by gradually turning off the restraints on the lipid molecules. Then, the simulations were run for 100 ns with three replicates. During the production simulation the temperature was maintained at 310 K using the Nose-Hoover method and the pressure was maintained at 1 bar using the semi-isotropic pressure coupling approach with Parrinello-Rhaman barostat.^33-35 Electrostatic interactions were treated using the Particle Mesh Ewald (PME) method with a short-range cut-off of 1.2 nm.³⁶ The cut-off value for van der Waals interactions was set to 1.2 nm. The neighbour list was updated every ten steps. All bonds involving hydrogen atoms were constrained using the LINCS algorithm.³⁷ The time step was set to 2-fs during the production run.

Data analysis and visualization

To define the geometry of each polymyxin molecule, three atoms in the polymyxin core structure were selected. These were the first and last carbon atoms of the fatty acyl tail, and the C_α atom of the residue in position 6 (i.e. d-Phe-6 in polymyxin B₁ and d-Leu in the remaining polymyxins A₁, M₁ and colistin A). The z-axis coordinates of these three atoms were tracked using gmx traj utility to display the insertion kinetics of polymyxins along the simulations. Membrane plugin analysis tool MEMPLUG in VMD was used to calculate the averaged membrane thickness, membrane thickness map and area per lipid molecule.³⁸ All structural snapshots were visualized using PyMOL.³⁹

Measurement of membrane roughness using AFM

Human kidney proximal tubular cells (HK-2) were cultured as we previously described in detail.¹⁸ Cells were treated for 60 min with polymyxin B at 25 μM or 100 μM, with another set of untreated cells served as the control group. AFM imaging was conducted by the PicoSPM II AFM (Molecular Imaging) and Dimension Icon AFM (Bruker) in the atmosphere environment using tapping mode. The cantilever length was 115-135 μm, the thickness 2.65-4.15 μm and the resonance frequency 230-410 kHz. The Single Crystal Si tip’s (Bruker Co., US) height was 10-15 μm and the radius 7-10 nm. Samples from all three groups (untreated control, polymyxin B 25 μM and 100 μM) were scanned. Cells in these samples were randomly selected from the glass slide to measure the membrane roughness of the cell. For each group nine roughness measurements were collected and the average roughness (R_a) and root-mean-square roughness (R_q) were used to represent the roughness of the cell membrane. These parameters have been widely used to characterize damages to the cell membrane from environmental toxins.^40-41 The wyddion software was employed to analyze the experimental data.⁴²