We demonstrate the effects on membrane of the tubulin-binding chemotherapy drugs: thiocolchicoside and taxol. Electrophysiology recordings across lipid membranes in aqueous phases containing drugs were used to investigate the drug effects on membrane conductance. Molecular dynamics simulation of the chemotherapy drug–lipid complexes was used to elucidate the mechanism at an atomistic level. Both drugs are observed to induce stable ion-flowing pores across membranes. Discrete pore current–time plots exhibit triangular conductance events in contrast to rectangular ones found for ion channels. Molecular dynamics simulations indicate that drugs and lipids experience electrostatic and van der Waals interactions for short periods of time when found within each other’s proximity. The energies from these two interactions are found to be similar to the energies derived theoretically using the screened Coulomb and the van der Waals interactions between peptides and lipids due to mainly their charge properties while forming peptide-induced ion channels in lipid bilayers. Experimental and in silico studies together suggest that the chemotherapy drugs induce ion pores inside lipid membranes due to drug–lipid physical interactions. The findings reveal cytotoxic effects of drugs on the cell membrane, which may aid in novel drug development for treatment of cancer and other diseases.
Cell membrane permeabilization is a very important area in drug discovery research. Electrophysiological recordings of ion currents flowing across membranes doped with various naturally occurring or synthetic membrane proteins (MPs) or antimicrobial peptides (AMPs) may be used to investigate whether a protein–lipid complex creates an ion-flowing pore/channel across a membrane. Using the same technique, we investigate here whether chemotherapy drugs (CDs) induce an event inside membrane in analogy to those due to AMP-induced ion channels. To do so, we have chosen two CDs colchicine and taxol (TXL), as examples, from a long list of several hundred CDs and drug candidates that are either in clinical use for patients, in clinical trials, or under investigation.a To understand the molecular mechanism of drug effects on membranes, we have performed molecular dynamics (MD) simulations of various CD–lipid complexes.
Colchicine, a potential CD candidate, has a long history of use in immune system diseases (1,2). In 2009, colchicine won the USA Food and Drug Administration (FDA) approval as a drug for acute gout and familial Mediterranean fever. It inhibits leukocyte–endothelial cell adhesion (3) and T-cells’ activation (4) by binding to tubulin dimers, which prevents their polymerization into microtubules (MTs) (5). Due to an increased rate of mitosis, cancer cells are more vulnerable to colchicine poisoning than normal cells. The therapeutic potential of colchicine in cancer chemotherapy is limited by its toxicity against normal cells. Colchicine shifts the dynamic equilibrium in MT’s toward disassembly by sequestering most of tubulin (6) and inducing slow disassembly of MT’s (7). This may account for the gradual change in the membrane action potential (decrease) or threshold (increase) and the resting potential, which was found to be altered modestly at colchicine concentrations of ∼2.5 mm. At higher concentrations (10 mm), the resting potential was reduced by up to ∼5 mV (6). Colchicine is also found to bind with nuclear periphery and to disorder the nuclear membrane phospholipid bilayers (8). Since FDA’s approval in 1992, TXL has been used for ovarian, lung, and breast cancer chemotherapy (9) due to its molecular action involving stabilization of MTs (10) making it especially suitable in combination therapy (11). However, TXL’s poor solubility is a serious problem that requires conjugation with cremophor or albumin. A TXL–phospholipid liposome construct increases TXL’s antitumor efficacy by delaying tumor progression compared with the case of free TXL administered in cremophor (12). TXL inhibits endosomal–lysosomal membrane trafficking by favorably inducing small condensed vesicles over large ones (13), which suggests that TXL affects lipid curvature profiles. TXL incorporated into liposomes penetrates into the acyl chain domain, which alters physical properties, for example, phospholipid phase transitions, lipid order parameters, fluidity, etc. (14) of both artificial and biological membranes. Both MT-stabilizing TXL and MT-disrupting colchicine were found not to affect the human platelets’ membrane lipid fluidity (15). The evidence of strong interactions of colchicine with lipids (16) also suggests that it is a poorer candidate as an anticancer drug by being weakly accessible to tubulin. Although the activity of colchicine and TXL has been found to be affected by the presence of lipids or liposomes, there is still lack of concrete information about the effects of these molecules on cell membranes. In particular, little is known about their effects on membrane transport properties as functions of physical, geometrical, and structural characteristics of membranes. Studying these effects will shed important light on off-target interactions of these chemotherapeutic agents and hence may lead to an improved drug design in the future.
Due to proven effects of some of the CDs like colchicine and TXL on membrane’s certain electrical and physical properties (6,8,13–16), the understanding of their effects on the membrane’s transport properties and the drug/membrane constituent interactions is important from their cytotoxicity point of view but remains largely unexplored. Certain properties like relatively small size and apparently charge neutrality of these drug molecules raise the possibility that they may act differently on membranes than antimicrobial agents. The latter group members are reported to undergo considerable conformational changes as they cross through hydrophilic/hydrophobic boundary to enter into lipid membrane environment [e.g., see (17,18)], and some of these members cause membrane permeabilization generally through the formation of well-structured long-lived protein-lined (19–21) ion channels. Otherwise, their amphiphile-type effects would modulate the bilayer’s physical properties. For example, comparable effects of amphiphiles, triton X-100 and capsaicin (22), and AMP gramicidin S (GS) at nanomolar concentrations (23) on bilayer mechanical properties or transient disordering defects induced by GS at micromolar concentrations in the lipid layers (24) were observed. The goal of present research was to evaluate the two specific tubulin-binding drug molecules thiocolchicoside (TCC), which is a derivative of colchicine, and TXL (see Figure S1 for their structures) regarding their effects on model lipid membranes using electrophysiology techniques. Our results suggest that these molecules not only permeabilize lipid bilayer membranes but also create certain molecular structures through reorganization between lipid molecules and themselves. We provide evidence for the formation of stable ion pores (25–27) with unconventional characteristics inside model lipid membranes involving these molecules. Using MD simulation we provide computational support for drug-lipid physical interactions which are postulated here to be the mechanisms behind such pore formation. This also sheds new light on the complex interactions between stable structure (e.g., biomolecules) and liquid crystal structure (e.g., membrane), a very important and still unresolved problem in biophysics and soft condensed matter physics. An attempt to better understand these drug–lipid interactions will be made based on screened Coulomb and van der Waals (vdW) interactions that were recently found to be the main mechanisms behind the creation of ion channels due to AMPs gramicidin A (gA) and alamethicin (Alm) (28–31). We postulate the creation of TCC- and TXL-induced ion pores with understanding of the underlying mechanisms to be a novel finding on the mode of action of these CDs with important physiological consequences.
Materials and Methods
Planar lipid bilayers were formed by applying a lipid cocktail of phosphoethanolamine– phosphatydyleserine–phosphatidylcholine (5:3:2, v/v/v)/n-decane using the painting method over a 150-μm septum of a bilayer cuvette. The volume of cuvette was 1 mL (buffer: 0.5 m NaCl, 10 mm HEPES, pH 7.4) in both the cis (recording electrode) and trans (reference electrode) chambers. TCC (from ChemRoutes, Edmonton, AB, Canada) or TXL (from Vector Lab, Burlington, ON, Canada) stocks were prepared in dimethylsulfoxide (DMSO) (4 mg/mL or 7.1 mm), diluted in buffer for further use (1 mg/mL). Every time after vigorously vortexing the stock, TCC or TXL was added to cis chamber buffer while stirring to avoid drug solubility issues. HEPES in buffer was replaced with 2-(N-morpholino) ethanesulfonic acid (MES) and TAPS (2-hydroxy-1,1-bis(hydroxymethyl)ethyl)amino-1-1 propanesulfonic acid)) for solutions at pH 5.7 and 8.4, respectively. Following bilayer formation, we waited for 1 h and tested the bilayer stability by applying transmembrane potential at V = ±400 mV for 3 min. After the addition of drugs, we waited for 20 min before any current was recorded. The ion pore activity at reasonable values of V was recorded as traces. Each experiment was repeated under at least three independent experimental conditions. We used Ussing chamber systems (Physiologic Instruments, San Diego, CA, USA) for this membrane transport study. Clampex 8.2 (Axon Instruments, Union City, CA, USA) was used for data acquisition. Current (pA) through the membrane was recorded at a filter frequency 20 kHz. The current as a function of recorded time was then plotted using Origin 8.5 (OriginLab Corp., Northampton, MA, USA).
We used MD methodology to model in silico CD–lipid interactions. Based on the Monte Carlo concept, we considered five different relative locations and orientations randomly generated in each CD–lipid complex created by a CD and a lipid molecule as initial structures for MD simulations to increase sampling size for better statistical analysis. For each location- and orientation-specific complex, a 6-ns explicit water MD simulation at a temperature of 300 K in an aqueous solution at pH 7 was performed. We applied the software package amber 11,b specifically the Amber force field ff03 was used. The explicit water TIP3P model was used to simulate solvent effects. The force field parameters for CDs and lipids (PC and PS) were generated using an Amber module Antechamber (32,33). Note that the force field parameters for CDs are similar to the ones generated for colchicine (34) and TXL (35), respectively. Both studies (34,35) have shown these parameterizations lead to simulations to be consistent with experiments. Therefore, we can expect to observe similar results based on these parameterizations in current studies. Twenty complexes were energy minimized using the steepest descent method for the first ten cycles and then followed by a conjugate gradient for another 1000 cycles. We then applied Langevin dynamics during the process of heating up the system for 200 ps with the energy-minimized complex, in which drug and lipid molecules were being restrained using a harmonic potential with a force constant k = 100 N/m. Afterward, we introduced pressure regulation to equilibrate water molecules around the complex and to reach an equilibrium density for another 200 ps in addition to temperature regulation. The MD production run then was continued for 6 ns. Given five various initial structures for each lipid–drug pair, a total of 30-ns simulation result was analyzed to gain insights of the direct interactions of the corresponding pair. Note that the phospholipid was gently restrained with a harmonic potential with a force constant k = 10 N/m, applied only to the phosphate during the production runs. The purpose of this restraint is to mimic a single phospholipid being ‘restrained’ in membrane, while both head group and two tails of such lipid still possess certain degrees of freedom.
We first present the results concerning the possible changes in the lipid membrane’s electrical conductance properties due to the effects of CDs, TCC and TXL. We then describe MD results that illustrate computational predictions regarding drug–lipid interactions between drug molecules and lipid membranes, once they find themselves in each other’s close proximity.
Electrophysiological results presented in Figure 1 (long-time traces are also presented in Figure S2.A) show considerable conductance events in membranes doped with TCC or TXL (36). An intact lipid membrane in aqueous phases is generally non-conducting to ions (see Figure S2.B), but in the presence of TCC/TXL, different current levels were observed. Long-time records indicate that time-independent/random appearance of current events was induced by both compounds. The stability of current levels varied significantly. The point count plot (presented in Figure S2.A) shows that the conductance events do not correspond to discrete current levels with discrete conductance values. These current levels appeared with various possible conductance values, similarly to events observed due to GS (24). Although no major qualitative difference between TCC- and TXL-induced membrane permeabilization was observed, TXL-induced current levels covered a relatively lower conductance range than those for TCC. It is also worth mentioning that after breaking a membrane doped with drugs and recording the current traces on a reconstructed membrane, we observed a reduced appearance of current events (traces not shown here and were not used in analysis).
In Figure 1, short-time records show independent conductance events. Any current event or conductance burst (see Figure S2.A) can be due to an independent drug-induced conductance event or a combination thereof. We observed no current events across lipid membranes without being doped with drugs. To demonstrate that we have presented a 60-s recorded current trace in Figure S2.B. A current event does not represent any control or a reproducible parameter, which could elucidate the nature of the conducting pore induced by a drug. Note that the independent conductance events appear with triangular shapes, which means that the conductance in a single event is not constant but increases/decreases spontaneously over the time interval comparable with the ‘lifetime’ of any specific conductance event. To the best of our knowledge, the appearance of triangular conductance events with time-dependent change in conductance is a new and unique phenomenon. The amplitudes of these events are also different from typical values. We observed random spontaneous transitions between different current levels within a discrete conductance event during its lifetime. These discrete events were found to be approximately characterized by conductance values in the range ∼0.01 to 0.1 pA/mV and lifetimes in the range ∼5 to 30 ms based on randomly chosen 20 discrete conductance events. The lowest possible value for both conductance and lifetime should ideally approach 0, but events with lower than the above-mentioned values are masked by the noise. For comparison, the conductance events of gA and Alm channels are also shown. Similar to the conductance events in channels formed by integral membrane proteins or AMPs, both gA and Alm events have rectangular shapes. This means that transitions between discrete current levels are transient or extremely abrupt. In gA channels, only a back-and-forth transition exists between two current levels representing dimer/monomer states. In Alm channels, there exist many transitions between different conductance states consisting of various numbers of monomers. All transitions are transient that means the current fluctuations at the transition take almost no time. A time-dependent current fluctuation between random current levels in any TCC/TXL-induced conductance event is therefore an important novel finding reported here.
There is an important feature observed in the CD-induced triangular conductance events which is that the time for the conductance to rise and fall for each individual event is rather consistent. From many such events, the rise and fall times can be used to characterize the conductance. These times may be influenced by the applied voltage or by the concentration of drug present and thus can manifest the kinetics of channel growth and decay. The time-dependent increase and decrease of the pore conductance, which can be considered as pore conductance changes over time, were therefore analyzed under experimental conditions stated in Figure 1 and found to be 1.61 ± 0.46 and 1.71 ± 0.47 pA/mV.s for TCC-induced pore and 0.76 ± 0.27 and 0.75 ± 0.23 pA/mV.s for TXL-induced pore, respectively. These values were derived by investigating 20 randomly chosen discrete conductance events for both TCC and TXL. The corresponding values for peptide-induced pores/channels (e.g., gA and Alm channels) approach to infinity, as in both cases the transitions between different conductance levels take no measurable time.
A linear dependence was observed for values of pore activity (A) (below 1.0) on V and TCC/TXL concentration (cD) for up to very high values of both (see Figure S3). Pore activity can be defined as the duration of time when the membrane was found to be permeable to ions relative to the total current record time. It is not feasible to state lower/upper cutoff values for V and cD due to their interdependence. We also observed conductance events (data not shown) at a much lower range and for more biologically relevant (to cancer therapy) concentrations, for example, at cD = 10 μm for both TCC and TXL, but at V = 200 mV or higher. Although we rigorously vortexed and then stirred the stock while adding into the aqueous phase, we cannot rule out the possibility of solubility issues. We therefore predict that by overcoming the possible solubility issues of these drug molecules with concomitant use of other solubility-enhancing agents (which is a possible choice in therapy), it should be possible to observe membrane conductance events induced by TCC and TXL at even lower concentrations, which are physiologically relevant in the context of cancer treatment.
No considerable change in pore activity in the range of pH values between 5.7 and 8.5 was observed (see Figure S4).
After observing the effects of drugs on membrane’s conductance, we performed MD simulations to investigate CD–lipid interactions in molecular level. Figure 2 shows the MD results using five CD–lipid complexes as initial structures as shown in the inset. It plots the separation distance of centers of mass of CD and lipid molecules, ddrug–lipid, against simulation time t (ns). Note that ddrug–lipid was used as the simplest property to quantify the effects of CD–lipid interactions. It shows that ddrug–lipid fluctuates around 10 Å, similarly to the initial setting in two to three simulations in TXL-PC, TCC-PC and TXL-PS, TCC-PS. Drugs and lipids were observed to be gradually separated in most of the simulations. Figure 3 plots snapshots of CD–lipid complex at the beginning (left) and at 6th ns (right) of the simulations. Note that it only presents the case that a CD likely to bind with the single lipid. Although TXL in both blue and red cases shown in Figure 2 starts with different orientations, the simulations indicate TXL is likely to bind to similar location that is near phosphate group domain shown in Figure 3A. We also observe TCC (cyan case in Figure 2) tends to bind to similar location (Figure 3B). Yet, Figure 3A,B shows TXL and TCC (green case in Figure 2) to likely insert into the cavity between two short tails of PS, respectively.
The solvent-accessible (SA) area of the complex in the MD simulations was used to investigate whether the hydrophobic effects contribute to CD–lipid binding. Figure 4 shows SA areas in all four cases against ddrug–lipid. When both drug and lipid molecules are completely separated, we can expect them to be entirely exposed to solvent, that is, the corresponding SA areas are at a maximum. The figure shows that the SA areas in all four cases are unchanged between the start and the investigated 20 Å length. This suggests that within this range, the drug–lipid complexes stay at an equilibrium solvation condition.
Histograms of ddrug–lipid from all five 6-ns simulations and the corresponding energy contributions from two non-bonded interactions, van der Waals (EvdW), and electrostatic forces (EES) versus ddrug–lipid are shown in Figure 5. The histogram of ddrug–lipid shows that both TXL and TCC spent more than 2 ns within 6 Å <ddrug–lipid <10 Å and away from lipids most of the time (see the upper panels in Figure 4). It suggests the possibility for drugs to briefly bind with lipids. Figure 5 (top panel) indicates that TXL likely favors the interaction with PC over PS, while TCC shows no significant lipid-specific preference. Both EES and EvdW for TXL and TCC interacting with PC are inversely proportional to ddrug–lipid, while there are no such trends in either TXL-PS or TCC-PS cases. Below ddrug–lipid <6 Å, which is on the order of the lipid head group dimension, the CD–lipid binding stability drastically decreased (see the upper panel in Figure 5), and below ddrug–lipid <4 Å, no stability was observed because there was no structure found at this low distance value (see Figure 2). It is shown that both EES and EvdW are strongly effective within 12 Å (EvdW slightly dominant as shown in the inset plots of Figure 5).
The evidence of chemotherapy drug-induced ion pore formation in lipid membranes
The electrophysiology results indicate the formation of special structures inside membranes resulting from the action of the CD molecules TCC and TXL. Erdal et al. (37) have recently reported that a large number of potential anticancer drugs induce p53 protein–independent apoptosis and that lysosomal membrane permeabilization is a mediator of many such responses. Our direct observation of both membrane permeabilization and the formation of ion pores due to TCC and TXL sheds new light on the cytotoxic effects and off-target interactions exhibited by these drugs. Stimulation of conductance events due to CDs may also lead to the discovery of a novel type of ion pore formed by compounds whose primary mode of action involves binding to cytoplasmic proteins. All membrane-permeabilizing proteins or AMPs form ion channels in which transitions between different current levels happen instantaneously. So far, only TCC and TXL have been shown to induce single conductance events with a current versus time plot that has a triangular shape. This indicates that the induced pore radius continuously changes back-and-forth, giving a range of cross-sectional areas of the pores. The time-dependent change in the TCC/TXL pore conductance rules out the possibility of channel formation of the types represented by linear β-helix gA or barrel-stave Alm channels (19,20). Although the point count plots (see the right-hand panel of Figure S2.A) resemble random events observed for GS, which represent defects (24), fine tuning of the conductance events (see Figure 1) suggests that they are long-lived, not just transient (24) but with no fixed conductance. No such time-dependent transitions between non-zero current levels as observed in CD-induced discrete conductance events (see Figure 1) are found in GS-induced defects. The latter shows transient conductance events between zero current levels only. TCC/TXL events superficially resemble Alm events with important exceptions that there is an absence of stable and fixed Alm-like discrete conductance values corresponding to different discrete channel cross-sections representing different Alm conductance states. Also, unlike instantaneous transitions (apparently time independent) between Alm states, the CD-induced pore current transitions are time dependent (see Figure 1). The CD-induced pore conductance’ change (increasing and decreasing) over time was therefore found (presented in the Results section) to be finite while the peptide (gA or Alm)-induced pore conductance changes’ over time appear with infinite values. Our results, however, can neither rule out nor support a mechanism similar to Bechinger’s in-plane diffusion model (38,39). Here, the bilayer-disrupting molecules disorder the hydrocarbon chains of the adjacent phospholipid molecules and create a local bilayer thinning. This leads to local disturbances in bilayer packing and an eventual formation of toroidal-type pores (like those shown in Figure 6). It is worth mentioning that after bilayer formation with a combination of three different classes of lipids (phosphoethanolamine, phosphatydyleserine, phosphatidylcholine), we waited for 1 h and tested the bilayer stability under at least 20 independent experimental conditions without the presence of drugs by applying a very high transmembrane potential ∼400 mV for about 2–3 min. We found no instability or conductance events in the bilayer. Therefore, the creation of an electric field–induced or any specific lipid–induced conductance events can be ruled out, which would be characterized as pre-pore or metastable states in lipid bilayers reported by, for example, Melikov et al. (40) using a different lipid composition. The observed conductance events were also determined not to be due to DMSO, as the maximum amount of it was <0.8% of the volume of the aqueous solution at which it does not induce any bilayer-disrupting effects (confirmed in many of our studies and in ref. (41)). It is worth discussing whether these CD-induced pores resemble the channels formed by ceramides in lipid membranes (42). Ceramides, despite being lipids, form channels where the molecules are considered to be held together in a column spanning the hydrophobic portion of the membrane with the columns being held together along the channel lumen that makes a cylindrical pore (pore diameter ∼0.8 to 11 nm) (42). CD pore conductance changes spontaneously back-and-forth (see Figure 1) between lower and higher values. This indicates that no stable conductance levels exist as those observed in ceramide channel currents (42). That means a number of ceramide molecule complexes created so-called ‘lipidic’ pores whose stable cross-sections for a certain value of conductance resemble those in Alm channels. This rules out the possibility of any resemblance between CD-induced pores and ceramide channels.
Thiocolchicoside/taxol-induced conductance events therefore do not appear to be similar to GS-induced defects or to other protein-lined channels (e.g., gA’s linear β-helix, Alm’s barrel-stave pore, etc., see Figure 6) or so-called ‘lipidic’ ceramide channels (42). We have also ruled out the possibility of any observed conductance events to be due to membrane conditions such as those caused by an electric field across the membrane, any characteristic of membrane-constituting lipids, DMSO, etc. The CD-induced events therefore correspond to a special structure induced inside membranes by forming a drug/lipid molecular complex. We hypothesize that lipids are forced to line across pore openings (see Figure 6), and TCC/TXL molecules are localized behind the head group region near the hydrocarbon chains of the lipids (14). The moderate dependence (linear) of A on cD and V (see Figure S3) in comparison with the 2.6 power dependence on the identical parameters reported in the cases of protein-lined channels (43,44) also provides support for the proposed model of the TCC/TXL pore (see Figure 6). This structure allows the membrane to be doped with higher amounts of molecules at a moderate energy cost as the molecules just penetrate into the lipid head group–hydrocarbon chain interface. Only this type of broken membrane structure can ensure a back-and-forth spontaneous change in the pore’s geometric cross-sectional area requiring no addition or release of pore-inducing agents, which leads to the formation of triangular conductance events with a possibility of spontaneous change of pore conductance between any values. Admittedly, further studies are badly needed to confirm the proposed less defined stoichiometry and structure of the toroidal-type channel, which can possibly be induced by CDs in membranes. Work is currently underway in that direction using various experimental techniques and computer model studies. In the case of protein-lined channels, the membrane’s resting thickness near the channels is considered to be changed but may not vanish completely. This could indicate bilayer deformation as predicted in a mechanosensitive channel study (45). Unlike in gA channels (41) and Alm channels (submitted), which exhibit substantial phenomenological pH effects, we have not observed considerable pH effects on the drug-induced pore activity (pH 5.7-8.5). This suggests that: (i) the phenomenological function of drug-induced pores is different from conventional protein-lined channels, for example, gA, Alm, etc. (see the model diagrams in Figure 6), and (ii) CD molecules are equally effective in both normal cell membranes and cancer cell membranes, which exist in different pH environments (46). Both of these points make sense considering that these molecules are electrostatic (ES) charge neutral, while other channel-forming peptides are charge bearing (pH sensitive) with specific (helical, beta sheet, etc.) structures.
The drug–lipid physical interaction as a possible cause of chemotherapy drug-induced pore formation
The main condition for any ion-transporting pore/channel formation is that the channel-forming agents and lipids must physically organize to form certain type(s) of stable structure(s). The purpose of performing MD simulations presented in this article was to discover the physical energies that play primary role(s) to ensure the time-dependent physical coexistence between CDs and lipids in membranes. MD simulations were meant not to calculate the stability of the toroidal pores; rather as mentioned here, these simulations were designed to investigate the stability of the pairwise coexistence (due to physical interactions) of the channel-forming agents and lipids. The stability of experimentally discovered toroidal pores is predicted to be due to resultant effects of many such CD–lipid physical interactions.
Molecular dynamics results suggest that the CD–lipid complex fluctuates within a separation over a period of time. These results suggest that both TXL and TCC likely bind with PC and PS, given appropriate initial conditions. Through our in-depth analysis, we found evidence suggesting that the hydrophobic effect is unlikely to contribute to the distance-dependent CD–lipid binding. The analysis of energy contributions from two non-bonded interactions, EvdW and EES versus ddrug–lipid, revealed crucial insights into the cause of the observed stability of the CD–lipid complexes. Both EvdW and EES appear to be the main contributors to the energetic CD–lipid binding, and vdW interactions contribute slightly more than ES interactions as the drug and lipid approach closer. Binding stability generally is found to decrease quickly with increasing ddrug–lipid within 6 Å <ddrug–lipid <16 Å at which considerable structures were observed (see Figure 2). Both vdW and ES interactions contribute comparably with both energies decreasing with increasing ddrug–lipid. Beyond a 16-Å separation, the interactions become negligible. The CD–lipid stability is still observed due to the effects of a combination of possible sustained presence of long range EES (although EvdW ≈ 0, data not shown) and other energies possibly induced by the surrounding environment. Large standard deviations (see Figure 5) are suggestive of the conformational space of the CD–lipid complexes not being completely explored in MD simulations. Nonetheless, this incompleteness does not preclude our proposed interpretation. Importantly, the CD–lipid interactions resemble the protein–lipid vdW and ES interactions found in MD simulations of a gA channel in phospholipid membranes (47). This suggests that the CD–lipid interactions resemble these mechanisms, namely the vdW and ES interactions as found in the channel-forming AMP–lipid interactions.
Both CDs are found to permeabilize lipid bilayer membranes by creating ion pores, a mechanism that is usually exerted on lipid membranes by various AMPs. The MD simulation–based discovery of common types of primary physical interactions (vdW and ES) between pore-inducing agents and lipids for both CDs and AMPs (will be published elsewhere) also suggest that these two different classes of biologically active molecules may act on membranes using common molecular actions. Therefore, CDs are likely to generate certain AMP-type mechanisms.
A mechanistic understanding of the drug–phospholipid interaction can be gained from the novel picture of interaction energetics between membrane-active agents (MAAs) and a lipid bilayer membrane that has recently been reported (28–31). In this work, a screened Coulomb interaction model has been used to represent the interaction of MAAs with lipids in the bilayer, due to the presence of the distributions of the localized charges on both MAAs and phospholipids of all types. When any MAA and phospholipid approach each other to form a MAA–phospholipid complex (e.g., the case of CD–PS/PC interaction in this study), the localized charges present in the complex experience each other’s electric fields. In this ES interaction scenario, the charges interact with each other not only directly (via Coulomb interactions) but also indirectly through other charges in the vicinity. Hence, the interaction between any two localized charges becomes screened, so the interaction takes the traditional form of the screened Coulomb interaction.c Furthermore, the interaction energy may change due to many other parameters such as variations in the membrane’s electrical conditions, the presence of hydrocarbons, lipid density, etc.
In general, the binding energy between CD and phospholipid in a membrane is due to the sum of the traditional Lennard-Jones (31) and screened Coulomb potentials (28–31).c A change in drug–phospholipid binding stability, which appears through the distance-dependent probability (see Figure 5 in our MD simulation), is mainly due to the change in drug–phospholipid coupling energy that varies mainly according to the effects of the hydrophobic coupling to be filled with localized single, double, etc. charges, respectively, that are present within the interaction field in a drug–phospholipid complex. In the case of the interaction of any CD with a phospholipid bilayer, the screened Coulomb interaction extends beyond the nearest neighbor phospholipids to other phospholipids residing in the vicinity, exactly following the protocol presented in references (28–31).c
To better understand the generalized drug–PS/PC interactions, more MD studies are needed to focus on CDs and phospholipid membrane rather than a single phospholipid. This deeper understanding of the interactions will provide clear insights into the membrane effects of CDs.
Chemotherapy drugs TCC and TXL have been shown to permeabilize lipid bilayer membranes by forming ion pores. A possible cause of pore creation may be due to the resultant effects of many physical drug–lipid interactions that were discussed in our article. The reported results and our computational modeling have contributed to an improved understanding of the general mechanism of ion pore formation in membranes. It has also shed light on the underlying mechanisms of complex interactions between structurally stable (pore/channel forming) peptides or small biomolecules and liquid crystal structure type of lipid membranes. This study has concluded that the molecular mechanisms of CDs include both the commonly investigated interactions in the cell’s cytoplasm and the hitherto largely neglected interactions across the cell’s membrane. This new insight will be useful in finding novel drugs to treat cancer and other diseases by focusing both on the cytoplasmic and membrane regions.
M.A. acknowledges support from the Deanship of Scientific Research at King Saud University for funding the work through the research group project no. rgp-vpp-151. J.A.T. acknowledges funding from NSERC, the Allard Foundation, the Canadian Breast Cancer Foundation, and the Alberta Cancer Foundation. We are grateful to D. Nahirney, K. Barakat, P. Winter, J. Mane, and Prof. Olaf Sparre Andersen. Pharma Matrix high-throughput virtual screening cluster and the facilities of the Shared Hierarchical Academic Research Computing Network (SHARCNET: http://www.sharcnet.ca) and Compute/Calcul Canada were used for simulations.