|Home | About | Journals | Submit | Contact Us | Français|
Compounds with nominally similar biological activity may exhibit differential toxicity due to differences in their interactions with cell membranes. Many pharmaceutical compounds are amphiphilic and can be taken up by phospholipid bilayers, interacting strongly with the lipid-aqueous interface whether or not subsequent permeation through the bilayer is possible. Bolaamphiphilic compounds, which possess two hydrophilic ends and a hydrophobic linker, can likewise undergo spontaneous uptake by bilayers. While membrane-spanning bolaamphiphiles can stabilize membranes, small molecules with this characteristic have the potential to create membrane defects via disruption of bilayer structure and dynamics. When compared to the amphiphilic therapeutic anticoagulant, warfarin, the bolaamphiphilic analogue, brodifacoum, exhibits heightened toxicity that goes beyond superior inhibition of the pharmacological target enzyme. We explore, herein, the consequences of anticoagulant accumulation in a dipalmitoylphosphatidylcholine (DPPC) bilayer. Coarse-grained molecular dynamics simulations reveal that permeation of phospholipid bilayers by brodifacoum causes a disruption of membrane barrier function that is driven by the bolaamphiphilic nature and size of this molecule. We find that brodifacoum partitioning into bilayers causes membrane thinning and permeabilization and promotes lipid flip-flop – phenomena that are suspected to play a role in triggering cell death. These phenomena are either absent or less pronounced in the case of the less toxic, amphiphilic compound, warfarin.
Bolaform amphiphiles are compounds that possess two hydrophilic end groups joined by a hydrophobic linker. Like traditional amphiphiles, bolaamphiphiles are surface-active agents. In aqueous solutions, they form nanostructures such as micelles, ribbons, and fibers 1-3. Bola species occur naturally in the monolayer membranes of thermophilic microorganisms, conferring the superior stability required for existence in extreme environments 4. Synthetic bolaform species have found application as bilayer membrane stabilizers, synthetic ion channels 5-8, and as additives in drug delivery formulations 1, 9. They are also of considerable interest in monolayer functionalization of solid surfaces 10-12.
While much of the appeal of bolaform additives to bilayer membranes lies in their ability to fortify membrane structure by spanning the full thickness of the bilayer, hydrophobic mismatch and other molecular features can make these two-headed molecules function as disruptors of the lamellar structure formed by phospholipid amphiphiles. By tuning the groups making up the hydrophobic linker, bolaamphiphiles can be made more or less membrane stabilizing 13, 14. Rigid aromatic rings confer more stability than unbranched saturated aliphatic chains 2, 14. Rates of phospholipid flip-flop (transit of phospholipid molecules from one bilayer leaflet to another) have been observed to increase in the presence of membrane-spanning bolaamphiphiles. Flip-flop rates decrease as the stiffness of the aliphatic linker is increased 4, 14. Highly flexible linkers allow bolaform compounds to adopt U-shaped or hairpin conformations in lipid bilayers rather than spanning their thickness 13, 14. In addition to linker rigidity, linker length is critical. As membrane stabilization is a typical objective, few studies have probed the properties of bolaamphiphiles that are shorter than the thickness of typical phospholipid membranes. Short bolaamphiphiles have been reported to have a destabilizing tendency that can drive lateral phase segregation 13. Still fewer studies consider bolaform species that have asymmetric head groups.
In this paper, we consider the interaction of a bolaform anticoagulant compound with a dipalmitoylphosphatidylcholine (DPPC) bilayer. The compound is an anionic, asymmetric molecule with length similar to naturally-occurring phospholipids. Using coarse-grained molecular dynamics simulations, we identify a potential for membrane disruption by defect formation that is a specific consequence of the additive's bolaamphiphilic structure. By contrasting this with the behavior of an amphiphilic analogue, we characterize a potential biophysical mechanism for the heightened cytotoxicity of the bola-species.
Both warfarin and brodifacoum are compounds that inhibit blood coagulation by interfering with an essential metabolic cycle, the reduction of vitamin K. This anticoagulant 15-26 action is due to a shared hydroxycoumarin moiety that mimics the structure of vitamin K and enables these compounds to function as inhibitors of the enzyme vitamin K epoxide reductase (VKOR) 27-29. VKOR is an integral membrane protein that resides inside cells of organs such as the liver; consequently, hydroxycoumarin anticoagulants must cross plasma membranes to reach their site of action. Whether compounds traverse membranes via passive or active mechanisms depends on their size and physicochemical characteristics. Passive transport (diffusion across the membrane down a concentration gradient) is facilitated by low molecular weight, high lipid solubility, and absence of charge 30-33. However, even charged amphiphiles of modest size can bind efficiently to lipid interfaces, with slower subsequent transport across the membrane's hydrophobic core mediated by conformational and orientational changes 34, 35. Indeed, liposome partitioning of ionized drugs demonstrates higher bilayer affinity than would be suggested by octanol-water partitioning 36, 37. As bulk phase partitioning is an inadequate predictor of membrane association and experimental approaches are limited with respect to probing the molecular phenomena that govern drug association with and permeation through membranes, computer simulations have become an invaluable tool in elucidating these processes. Several recent publications and reviews demonstrate the utility of molecular simulation methods and transfer free energy calculations for predicting and clarifying biophysical interactions between drug-like molecules and phospholipid membranes 38-43. These approaches have been used to predict drug bioavailability and membrane permeability, with outcomes that correlate well with experimental measures. Interrogation of aggregation and pore-forming behavior provides insight as to mechanisms of toxicity. Furthermore, the molecular level examination of events occurring on femto- to microsecond time-scales elucidates conformational, orientational, and clustering phenomena that underlie structural and dynamical perturbations imposed on bilayers during drug permeation.
Warfarin and brodifacoum are both lipophilic and small enough to plausibly cross membranes by passive permeation. Transmembrane diffusion of coumarin compounds and their derivatives has been the subject of a number of computational and experimental investigations 44-48. Findings suggest that coumarin species do permeate bilayers, with molecular orientation, intramolecular hydrogen bonding and isomerization evolving to accommodate the different molecular environments comprising the bilayer structure36, 44-46. During their transit, we posit that these hydroxycoumarin compounds may disrupt membrane barrier function, creating a potential for cytotoxicity. Warfarin is, in fact, known to cause acute cell damage 49-52. However, its use as a clot-preventing therapeutic ensures conservative dosing and limited tissue exposure. In contrast, brodifacoum is a so-called ‘agent of opportunity’, a commercially available chemical that has the potential to be weaponized 53. The compound is found in consumer-accessible rodenticides. It is odorless and tasteless, and malicious or accidental release could lead to large exposures. A further concern is that brodifacoum can remain sequestered in organs for up to a year 15-17, 20, 24, 54. Hence, the potential for cell damage by brodifacoum is likely higher simply due to degree and duration of exposure. We further posit that the structure of brodifacoum lends itself to greater disruption of bilayer structure. Both warfarin and brodifacoum contain the hydroxycoumarin active group for anticoagulation, which constitutes a polar region in the molecular structure (Figure 1). However, where warfarin possesses a hydrophobic substituent in the form of a phenyl ring, brodifacoum is characterized by an extended hydrophobic region capped by a bromophenyl group whose dipole imbues the molecule with a second polar moiety. Thus, warfarin may be regarded as an amphiphile while brodifacoum, its more toxic counterpart, is an asymmetric bolaamphiphile.
Halogenated hydroxycoumarins of this type have been noted to be amongst the most lethal anticoagulants 55. The work we describe herein suggests that this elevated toxicity may be associated with membrane damage mediated by the unique features of bolaamphiphile accumulation and transit through the phospholipid bilayer. Our molecular simulations reveal differences in mode of entry, molecular orientation and retention within the bilayer, as well as consequent changes in structure of the model membrane. Bilayer thinning and permeabilization due to incorporation of brodifacoum permit water permeation. This, if accompanied by uncontrolled flux of ions and other small molecules, would indicate a dysregulation of the delicate homeostatic balance necessary to maintain cell health. Further, the dynamic orientation of the bolaamphiphilic compound encourages long-lived defects that facilitate phospholipid entry into the hydrophobic core of the bilayer and hence reduce the energy barrier for flip-flop of membrane phospholipids.
Coarse-grained molecular dynamics simulations of DPPC bilayers in water were employed to study the consequences of warfarin and brodifacoum incorporation in model phospholipid membranes. Equilibrium and steered MD simulations were performed using the LAMMPS simulation package (http://lammps.sandia.gov.; version: 5 Sep. 2014) 56.
The MARTINI coarse-graining approach and force field 57, 58 were employed to create molecule topologies for systems incorporating warfarin, brodifacoum, and a DPPC bilayer fully solvated in water. The MARTINI approach yields coarse-grained representations of each molecule by replacing the constituent atoms with interaction sites consisting of 2-4 heavy atoms. These sites, and their non-bonded interaction properties, are chosen so as to preserve the underlying geometric structure and physicochemical properties of the molecule. The structure of DPPC was taken from an existing topology (http://md.chem.rug.nl/cgmartini/), as was the coarse-grained representation for water. Our coarse-grained representations of the hydroxycoumarin molecules were created by grouping atoms into interaction sites located at the corresponding centers of mass. Warfarin (23 heavy atoms) and brodifacoum (35 heavy atoms) were represented by 8 and 14 coarse-grained particles respectively (Figure 1). Standard MARTINI interaction types were then assigned to each coarse-grained particle and equilibrium bond lengths and angles were determined based on particle connectivity.
Van der Waals interactions between neutral, non-bonded particle pairs were described using a shifted Lennard-Jones 12-6 potential energy function. Non-bonded interactions were cut off at 1.2 nm and the Lennard-Jones potential was shifted from 0.9 nm to the cutoff. For charged non-bonded interactions, the Coulombic electrostatic potential was shifted from 0.0 nm to the cutoff distance of 1.2 nm. Hence, long range electrostatic interactions were neglected, as pairwise interactions beyond 1.2 nm were not considered. A weak harmonic potential was used to describe interactions between covalently bonded sites, and a weak harmonic cosine potential was used to represent chain stiffness for the angles between three consecutive bonded particles.
The selected particle types for coarse-grained models of warfarin and brodifacoum were validated by reproducing the octanol/water partition coefficient using the Bennett Acceptance Ratio 59 (BAR) free energy difference method as implemented in GROMACS. BAR is a free-energy perturbation technique for determining the free-energy difference between two states. BAR has been shown to be the most efficient and robust perturbation method for obtaining free-energy differences 60, 61 and has also been shown to be unbiased statistically as compared to exponential averaging and thermodynamic integration 62. Partition coefficients of 3.1 and 8.65 for warfarin and brodifacoum respectively were estimated by the free energy difference between solvation in water and water-saturated octanol. These values compare well with the reported octanol/water partition coefficients of 2.7 for warfarin and 8.5 for brodifacoum 59.
As brodifacoum and warfarin exist as acid-base pairs and are predominantly found in their charged states at physiological pH, we considered both neutral and deprotonated forms of these species in our simulations. Representation of the deprotonated (ionized) species was achieved by replacing CG interactions sites 3 and 4 in the hydroxycoumarin ring with charge-bearing particle types of charge magnitude -0.5 each.
Each simulation system consisted of a central phospholipid bilayer containing 100 DPPC molecules in each leaflet and surrounded on both sides by water slabs containing 4000 coarsegrained water particles each. The polar headgroups of DPPC were fully solvated in the aqueous phase, with the hydrophobic tails forming the center of the bilayer. In most cases, either the neutral or deprotonated form of the hydroxycoumarin species of interest was initially incorporated within the aqueous phase. In the case of brodifacoum, the poor aqueous solubility of the molecule led to rapid aggregation in the aqueous phase. Thus, simulations incorporating multiple brodifacoum molecules were initiated with the hydroxycoumarin intercalated amongst DPPC molecules in the bilayer. To maintain charge neutrality in simulations containing deprotonated warfarin and brodifacoum, one Na+ counterion was added for each hydroxycoumarin molecule. An additional simulation represented a reference system containing a fully solvated phospholipid bilayer without hydroxycoumarin molecules.
The initial configuration for each of the simulation systems was generated using the packing optimization program Packmol 63. All simulations were performed at 323 K, which is above the main gel/liquid phase transition temperature of DPPC, thereby maintaining the phospholipid bilayer in the physiologically-relevant liquid-crystalline state 64. A typical simulation duration was 1.12 μs with the final 320 ns used as the production run for data analysis. Equilibration was confirmed by monitoring the average cross-sectional area-per-lipid 65, which was observed to be stable before the production run was initiated. The fluid phase of the bilayer was evidenced by an equilibrium area-per-lipid of 59.5 ± 0.05 Å2 - in agreement with measurements of 57 to 71 Å2 for experimental DPPC systems at similar conditions66. In addition, standard analyses, such as order parameter estimation and determination of mass and electron density distributions were performed to validate the quality of our simulation systems. Results (some described herein) were found to be in close agreement with data from other simulations of DPPC bilayers 57, 67-70.
Simulation boundaries were periodic in all dimensions, and a velocity Verlet method 71 with an integration time step of 20 fs was used to integrate the equations of motion 72-74. An isothermal-isobaric (NPT) ensemble was maintained; volume (and hence the cross sectional area) of each system was allowed to fluctuate freely. Each component of the simulation system was coupled separately to a temperature bath using Nosé-Hoover protocols 75-77 with a coupling constant of 1.5 ps. Pressure coupling at 1 atm was achieved using a Parrinello-Rahman scheme 74 with a coupling constant of 5 ps. To maintain a tensionless bilayer, pressure coupling was implemented as semi-isotropic. Every 10 time steps, a center-of-mass velocity was removed before calculating the temperature of each component. This prevented freezing and translation through space as an artifact of the accumulation of round-off errors by the numerical integration method 78. Graphics were rendered using the Visual Molecular Dynamics program (VMD) 79.
Using a system with hydroxycoumarin molecules initially distributed in the aqueous phase, spontaneous membrane incorporation of each hydroxycoumarin was simulated. Initial aqueous-phase concentrations of warfarin and brodifacoum were 85.7 mM (50 molecules) and 51.4 mM (30 molecules) respectively; the higher concentration of warfarin allowed for similar interfacial concentrations of the two hydroxycoumarins after equilibration. In the case of brodifacoum, for which aqueous solubility is very poor (0.46 μM) 69, this initial concentration led to spontaneous aggregation. The aggregate rapidly associated with the phospholipid bilayer, with subsequent dissociation of the molecules due to favorable interactions with the hydrophobic region of the membrane.
Steered molecular dynamics simulations were employed to estimate the thermodynamic landscape for transit through the bilayer. In each case, one molecule of either warfarin or brodifacoum was steered at a constant velocity through (i) a ‘bare’ DPPC bilayer or (ii) a bilayer already populated with additional warfarin or brodifacoum. The prepopulated systems corresponded to equilibrated systems as described previously, with interfacial concentrations of 11 mol% (12 molecules/leaflet) and 9 mol% (10 molecules/leaflet) for warfarin and brodifacoum respectively. Each simulation consisted of a 200 ns equilibration phase during which the motion of a target warfarin or brodifacoum molecule was constrained in the z-dimension, while the remainder of the system was allowed to freely equilibrate. This was achieved by tethering the molecule to a fixed point in the aqueous phase using a harmonic spring potential. The constraint prevented spontaneous entry into the bilayer during equilibration. A further 10 ns post-equilibration simulation was performed to obtain multiple initial configurations for constant velocity simulations.
The center of mass of the target molecule was then tethered to a moving point. As the tether point was moved at a constant velocity along the z-coordinate from the aqueous phase into the bilayer, the force required was recorded. While traversing this path, the molecule was allowed to move freely in the xy-plane and change conformation without any restraint. The harmonic spring constant and velocity of the tether point were 40 Kcal/mol/Å2 and 1 Å/ns respectively. The tether point velocity was chosen to cause a rate of transit orders of magnitude smaller than the root mean square thermal velocity of ~ 1050 Å/ns for DPPC, thereby ensuring that the bilayer was not needlessly perturbed by solute permeation 80-82. All other simulation conditions for steered simulations were identical to those used for the unconstrained simulations.
According to the isobaric-isothermal Jarzynski equality (Equation 1), for a system coupled to a constant temperature and pressure bath that is going through a transition between two states at equilibrium, the change in Gibbs free energy (ΔG) between the two states is directly related to the work done (W) on the system as the transition occurs 83-85:
Thus, a running integral of the restoring force over the distance covered by the solute molecule yields the potential of mean force (PMF) 81, 82, 86 for transit from the aqueous phase into the bilayer. This is then equivalent to the work done by the solute molecule during bilayer permeation. In the limiting case of a large spring constant (stiff spring approximation), the potential of mean force has been shown to be approximately equal to the change in free energy 84. In order to reduce systematic errors in estimating the free energy difference from the work done 80, 82, 83, 86, 87, each steered simulation was replicated 10 times using the different initial configurations obtained from the prior z-constrained simulations.
Both neutral and charged forms of the hydroxycoumarin molecules spontaneously accumulated within DPPC bilayers, with a consequent change in structure and dynamics of the membrane. In all cases, incorporation of the deprotonated (ionized) form yielded a similar, but more pronounced effect. As >99% of brodifacoum and warfarin molecules are expected to be deprotonated at physiological pH (pKa ~5 at pH 7.4), we will present herein the results for the charged form of each hydroxycoumarin.
Figure 2 presents snapshots taken from simulations of spontaneous hydroxycoumarin uptake in the presence of a DPPC bilayer. The interfacial concentrations of warfarin (Figure 2A) and brodifacoum (Figure 2B) are similar at 11 mol% and 13 mol% respectively. At these and higher interfacial concentrations (not shown), brodifacoum was not observed to leave the bilayer during the course of the simulation. Warfarin, in contrast, diffuses freely in and out of the DPPC interface, in agreement with the behavior of deprotonated warfarin in atomistic simulations 44. Lateral diffusion of both warfarin and brodifacoum occurs with a diffusivity on the order of 10-6 cm2/s, comparing favorably with the membrane diffusion coefficient value of 1.611 ± 0.164 ×10-6 cm2/s determined by artificial membrane permeation assay for warfarin 45. We are not aware of any similar experimental measure for the much more rarely studied compound, brodifacoum, however, our diffusivity values for both species are comparable to experimental diffusion measurements for other small molecules 88, 89. Neither species exhibits phase segregation or other aggregation within the bilayer. Our observed retention of brodifacoum in simulated bilayers is in line with observations that brodifacoum remains sequestered in tissues for prolonged periods 15-17, 20, 24, 54.
Entry into the bilayer is an activated process for both warfarin and brodifacoum, with transport of each species exhibiting a free energy minimum in the region of the DPPC glycerol backbone (Figure 3). While the activation energy barrier for entry into a bare bilayer is small and similar for the two species (1.8 ± 0.6 kBT and 1.4 ± 0.6 kBT respectively for warfarin and brodifacoum), the intramembrane energy minimum is more pronounced for the more lipophilic molecule (-2.7 ± 0.9 kBT vs. -5.9 ± 1.9 kBT for warfarin and brodifacoum respectively). These global minima occur at 17-20 Å from the bilayer center, lying at the upper end of the 11-17 Å region of residence observed for neutral warfarin in an atomistic simulation of membrane permeation 44. In agreement with our observations, the same atomistic study reported a tendency for the deprotonated (anionic) form of warfarin to reside preferentially at a position further away from the center of the bilayer. Prepopulating the interface with additional hydroxycoumarin molecules augments differences in the thermodynamic landscapes of warfarin and brodifacoum, causing the free energy profile for warfarin to become essentially barrier-free with respect to escape from the interface (shallow minimum of -0.6 ±1.1 kBT, Figure 3A). Indeed the spontaneous simulations demonstrated retention of brodifacoum without aqueous phase re-entry, whereas warfarin molecules were observed to undergo rapid entry into and escape from the bilayer interfacial region. Prepopulation of the bilayer with brodifacoum reduces the depth of the energy well (to 3.0 ± 1.0 kBT), but does not eliminate the minimum. For both species, the activation barrier for bilayer association is essentially unchanged by prepopulation of the interface. This may reflect the association of each molecule with the interface being driven by interactions between the hydroxycoumarin ring and phospholipid headgroups; retention within the bilayer is then mediated by the potential for favorable apolar interactions with the membrane hydrophobic core, which exists for brodifacoum more so than for warfarin.
Bilayer permeation subsequent to association with the interface is also an activated process. A thermodynamic barrier of ~21 kBT makes the bilayer crossing event kinetically inaccessible in the spontaneous permeation simulations. Given the magnitude of this barrier, simulation times on the order of 10 μs would be required to have a ~50% chance of observing a crossing event 46. Fluctuations in system energy occur with a magnitude of ~213 kBT, indicating that the crossing event should be possible 90, even given the ionized state of the hydroxycoumarin. We note that the predicted thermodynamic barrier for membrane crossing compares reasonably well with atomistic simulations of neutral warfarin traversing a phospholipid bilayer. Karlsson et al. report free energy barriers of 9 kBT and 11 kBT respectively for open and closed (hemiketal) isomers of warfarin respectively 44. The deprotonated form of warfarin represented in our own study would be expected to experience a larger thermodynamic barrier in traversing the hydrophobic core of the membrane both due to its negative charge and the steric hindrance of DPPC close-packing, which is greater than that of the DOPC employed in the atomistic study. The free energy profile for deprotonated warfarin also exhibits a global minimum further from the center of the bilayer, in agreement with atomistic simulations 44.
Incorporation of either hydroxycoumarin at constant surface pressure causes the bilayer to expand laterally. At an interfacial warfarin mole fraction xw = 0.11, an 11% expansion of the cross-sectional area was observed (i.e., from 59.5 ± 0.05 Å2/lipid to 65.9 ± 0.03 Å2/lipid). This yields an apparent area of 53 Å2 per warfarin molecule when ideal mixing behavior is assumed. Brodifacoum, at a mole fraction xb = 0.13, caused an expansion of 16%, with a corresponding apparent area per brodifacoum molecule of 62 Å2. The lateral expansion of the DPPC bilayer is accompanied by overall thinning in the direction normal to the water/lipid interface. The thickness of the bilayer, measured as the distance between the phosphate groups of DPPC in each leaflet, was reduced by 2 Å (4%) in the case of warfarin. A similar concentration of brodifacoum reduced the bilayer thickness by 4.2 Å (9%).
Expansion of the membrane also impacts the dynamics of the acyl chains forming the hydrophobic core of the bilayer. The presence of extraneous molecules increases the disorder of the DPPC tail groups, as indicated by the orientational order parameter, P2. P2 is defined in Equation 2, and is evaluated as a time and ensemble average for each coarse-grained inter-particle bond, with θ as the time-varying angle between the bond and the bilayer normal.
Ranging between -0.5 and 1 (for anti-alignment and perfect alignment with the normal respectively), P2 gives an indication of both the orientation of the bond in question and the variability of that alignment across time and space within the bilayer. A P2 value of zero indicates disorder, with no preferred orientation in the system. We find that both warfarin and brodifacoum reduce the orientational order of bonds in the phospholipid acyl chains, and they do so in a manner that increases with distance from the glycerol backbone (Figure 4A). The effect is of similar magnitude for the two hydroxycoumarin species, with the exception of the first bond in the sn-1 chain for the case of brodifacoum, the chain farthest from the phosphate group. Here, the bond displays a greater degree of alignment with the bilayer normal than in the case of warfarin uptake. The reduction of P2 is indicative of an increase in fluidity of the bilayer, a logical consequence of the increased headgroup spacing induced by insertion of the hydroxycoumarin species in this region. Fluidization in the form of melting point depression has likewise been observed in experimental studies of warfarin uptake by artificial membranes 91.
Although both molecules possess an amphiphilic character, brodifacoum has a longer hydrophobic region compared to the phenyl ring of warfarin. The tetralin and biphenyl ring moieties of brodifacoum provide favorable interactions with phospholipid acyl chains. In Figure 2B, several brodifacoum molecules are observed to be incorporated within the DPPC bilayer such that their lipophilic segments penetrate the hydrophobic core of the bilayer. The oxygen atoms in the hydroxycoumarin moiety lend a polar character to the headgroup common to both anticoagulants. In the case of brodifacoum, a second polar moiety arises. For brodifacoum, electronic effects are transmitted to the brominated group through conjugation by delocalized electrons in the phenyl rings, elevating the polarity of that group. Thus, rather than simply functioning as an amphiphile, brodifacoum has a bolaamphiphilic character. This allows for more than one preferred orientation in the bilayer. Indeed, simulations indicate a number of dynamic orientation effects.
It is clear from both the spontaneous simulations and the transfer free energy that both hydroxycoumarin species will preferentially partition into the lipid bilayer. Closer examination indicates initial association and subsequent retention of brodifacoum are mediated by directed interactions between the bolaamphiphilic anticoagulant and the amphiphilic interface. For both warfarin and brodifacoum, association with the bilayer is facilitated by interactions between the hydroxycoumarin ring and the charge-bearing DPPC headgroup. Both species then reorient as they enter the interface to maintain this favorable interaction. Given that the length and apolar tetralin-biphenyl region of brodifacoum mimic, to some extent, the amphiphilic structure of the lipid layer, the molecule might be expected to intercalate with DPPC. However, the brominated moiety allows for other possibilities. Normalized density profiles in Figure 5 demonstrate the location of the brominated interaction site relative to the hydroxycoumarin moiety of brodifacoum (the hydroxycoumarin moiety of warfarin exhibits a single peak located within 1 Å of the brodifacoum hydroxycoumarin peak - data not shown). The brominated site is predominantly located in close proximity to the hydroxycoumarin rings, suggesting that the brodifacoum molecule orients parallel to the bilayer interface rather than the perpendicular orientation that would indicate intercalation. Both the hydroxycoumarin rings and the brominated group are primarily localized in the polar regions of the bilayer.
Both warfarin and brodifacoum undergo dynamic conformational changes in the bilayer. In the case of warfarin, rotations are observed about the bond connecting the hydroxycoumarin and phenyl groups. When situated in the bilayer, brodifacoum has the potential to sweep out a large volume as it folds and extends about the tetralin structure (rendered in purple in Figure 6). These conformational changes alter the shape and effective size of brodifacoum such that the molecule's radius of gyration ranges from 3.7 Å (folded) to 6.7 Å (elongated). In contrast, warfarin is characterized by a radius of gyration measuring 3.4 ± 0.1 Å. The folded configuration of brodifacoum predominates, occurring 14 times as frequently as the elongated form. Thus, in its predominant conformation, this bolaamphiphile does not intercalate into the bilayer in an extended form. The larger size and dynamically changing conformation of brodifacoum likely contribute to the pronounced structural changes observed within the bilayer following brodifacoum incorporation. The predominance of the folded conformation of brodifacoum at the interface is due to association of the molecule with the bilayer. In aqueous solution, none of the conformations of brodifacoum predominates.
The presence of either warfarin or brodifacoum increases hydration of the simulated bilayer. This is particularly evident in steered simulations, where water solvates the hydroxycoumarin ring and is carried into the hydrophobic region along with the anticoagulant molecule (Figure 7). Much like a detergent molecule, upon association with the bilayer, both warfarin and brodifacoum cause transient defects in the membrane structure 92, 93. Phosphate and choline moieties in close proximity to the hydroxycoumarin are drawn inward, thinning the membrane and allowing water contact with the hydrophobic core. The effect is relatively brief for warfarin, with rebounding of the phospholipid headgroup to restore the structure of the membrane leaflet as warfarin enters the tail region of the first leaflet. The effect is prolonged for brodifacoum, for which hydroxycoumarin ring solvation continues even as the molecule crosses the center of the membrane. Quantifying the extent of water permeation via the radial distribution function for water around the DPPC tail ends (not shown), we find that brodifacoum admits to the bilayer more than twice the amount of water that associates with warfarin. Furthermore, the slight polarity of the brominated group at the opposite extreme of the brodifacoum structure allows for water ingress from the opposing leaflet in order to solvate that moiety (Figure 7C). The bolaamphiphilic nature of brodifacoum thus facilitates the creation of a transient water pore. This occurs even before the molecule's center of mass reaches the second leaflet. This phenomenon does not occur in simulations involving warfarin. Similar structural defects and water pores have been shown to aid ion transport and translocation of lipids from one leaflet of a bilayer to the other (known as flip-flop) 94-96. The water channel creates a favorable hydrophilic pathway for transit of ions and lipids through the hydrophobic bilayer.
The time-sequenced series of snapshots in Figure 8 demonstrate how a DPPC molecule becomes displaced from its usual position in a leaflet and gets drawn into the hydrophobic core as it interacts with brodifacoum. The headgroup of DPPC experiences strong van der Waals interactions with the polar hydroxycoumarin ring and the brominated site in brodifacoum. Electrostatic interactions also exist between the charge-bearing sites of the hydroxycoumarin ring and the oppositely charged DPPC choline site. Having been displaced from the leaflet of origin, the DPPC molecule can undergo end-over-end flipping that facilitates insertion into the opposing leaflet. Although lipid flip-flop would not alter the bilayer composition in our model system (where the leaflets are identical), real cell membranes have asymmetric compositions and undergo regulated transverse lipid motion 93, 97-100. In asymmetric membranes, increased lipid flip-flop rates may result in function-altering changes in leaflet composition, with consequences such as membrane curvature and disruption of cell signaling mechanisms that can lead to cell death96, 101, 102.
Warfarin and brodifacoum are compounds with nominally similar biological activity but significantly different levels of toxicity. Both are anticoagulants whose shared hydroxycoumarin structure allows inhibition of vitamin K reduction – a process necessary for normal blood coagulation. While the heightened toxicity of brodifacoum and similar halogenated coumarin derivatives has long been attributed to their prolonged retention in tissues, we propose an alternative mechanism that may contribute to cell death. The results presented herein indicate that the bolaamphiphilic structure of brodifacoum provides a possible biophysical mechanism for heightened cytotoxicity.
Brodifacoum is an anionic, asymmetric, bolaamphiphilic molecule with length similar to naturally-occurring phospholipids. Its hydrophobic section is flexible, which permits the molecule to adopt a U-conformation when incorporated into the bilayer. The conformation is dynamic, as the relatively weak polarity of one headgroup allows the molecule to stretch and intercalate with the hydrophobic tails of DPPC. The sweeping motion that accompanies alternation between folded and stretched conformations prolongs the existence of defects in the bilayer, through which water can permeate into the hydrophobic core. This facilitates entry of phospholipid headgroups by creating a polar environment that lowers the associated free energy barrier. Headgroup entry is a precursor to lipid flip-flop, a phenomenon that would upset the closely regulated asymmetry of plasma membranes. Hence, our simulations suggest that brodifacoum incorporation in a phospholipid bilayer can drive two phenomena capable of mediating cell death – water penetration and lipid flip-flop 93, 96, 101, 102. While increased lipid flip-flop has been observed to result from incorporation of amphiphilic drug molecules 93, 94, 98-100 and membrane-spanning bolaamphiphiles 4, 14, we demonstrate herein a potential mechanism for flip-flop driven by a non-membrane spanning bolaamphiphile.
Halogenated coumarin anticoagulants like brodifacoum have long been noted to be more lethal pesticides than others, but without a clear mechanism to explain enhanced mortality. There has been little study of these compounds, as they function primarily as potent rodenticides and not as therapeutics like their well-studied, anticoagulant counterpart, warfarin. However, identification of brodifacoum as a potential chemical threat motivates a better understanding of its detrimental effects; this will inform the search for relevant countermeasures for cases of mass exposure.
Warfarin and brodifacoum are just two members in a family of hydroxycoumarin compounds of varying lipophilicity and toxicity. Some members of this family exhibit cell toxicity that has been harnessed in anti-tumor applications 103-105, while others do not. A modeling approach similar to that presented herein would shed light on the molecular mechanisms underlying differential membrane-mediated toxicity of these compounds. Our observations also motivate a detailed study of the role of linker length, hydrophobicity, and flexibility in determining the interfacial properties of non-membrane spanning bolaamphiphiles. As our findings originate from a simple, homogeneous, model membrane, a further systematic study probing the differential effects of membrane lipid composition, pH, and ion concentration is clearly warranted.
We acknowledge financial support from National Science Foundation (Arlington, VA) grant 1228035 and National Institutes of Health CounterACT grant U01NS083457. We are also grateful to Professor C. Jameson for comments on the manuscript.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.