Drug molecules must cross multiple cell membrane barriers to reach their site of action. We present evidence that one of the largest classes of pharmaceutical drug molecules, the cationic amphiphilic drugs (CADs), does so via a catalytic reaction that degrades the phospholipid fabric of the membrane. We find that CADs partition rapidly to the polar–apolar region of the membrane. At physiological pH, the protonated groups on the CAD catalyse the acid hydrolysis of the ester linkage present in the phospholipid chains, producing a fatty acid and a single-chain lipid. The single-chain lipids rapidly destabilize the membrane, causing membranous fragments to separate and diffuse away from the host. These membrane fragments carry the drug molecules with them. The entire process, from drug adsorption to drug release within micelles, occurs on a time-scale of seconds, compatible with in vivo drug diffusion rates. Given the rate at which the reaction occurs, it is probable that this process is a significant mechanism for drug transport.
Lipinski has shown that most successful pharmaceutical drugs have a common set of chemical properties (Kuroda & Kitamura 1984; Lipinski et al. 1997; Bergstrand & Edwards 2001; Reasor & Kacew 2001; Hukkanen et al. 2005). Many of these properties appear to correspond to factors that enable drug molecules to cross the membrane barriers, which lie between them and their intended target. The work we present here points to a previously unreported mechanism for membrane translocation, requiring properties that appear in Lipinski's rules.
Currently, models of drug translocation across membranes divide the process into two major categories: active and passive translocation. Passive translocation is dominated by diffusion across and out of the membrane, without the expenditure of cellular energy. This process (i) does not involve carrier proteins, (ii) is unsaturable, and (iii) is the main mechanism by which the majority of drugs diffuse through the body. In general, this passive translocation exhibits low structural specificity and is driven by a concentration gradient across the membrane spanning two compartments. In active drug translocation, energy is expended by the cell in processes such as endocytosis, carrier-mediated transport or the use of specific efflux receptors. Active translocation is of particular importance for large drug molecules, whereas the endocytosis and exocytosis pathways are relevant to the movement of even larger systems such as the vitamin B12 complexed with its binding protein across the intestinal wall into blood. Our research shows that for cationic amphiphilic drugs (CADs), there is a third category of translocation mechanism: the chemically activated degradation of the phospholipid matrix that forms the essential fabric of cell membranes.
We have studied the chemical effect of CADs on model membranes formed by mixing a single, well-characterized phospholipid (figure 1a) with the CAD in a physiological buffer.
Usage of this simple model makes the analysis and interpretation of the chemical degradation mechanism straightforward. We have focused our studies on haloperidol and spiperone (figure 2), the two well-known dopamine D2 antagonists, which have to reach their targets in the brain after translocating through many membranes.
Unless otherwise stated, the evidence we present here is from our work on the lipid, 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC). The evidence for the CAD degradation mechanism using this model membrane system is from small-angle X-ray scattering (SAXS), nuclear magnetic resonance (NMR) spectroscopy, transmission microscopy (TM) and fluorescence microscopy (FM).
2. Material and methods
(a) Lipids and drugs
All phospholipids (Avanti Polar Lipids, Inc., Alabaster, AL, USA, 99.0%), i.e. DOPC, 1,2-dilauroyl-sn-glycero-3-phosphocholine (DLPC) and 1,2-di-O-octadecenyl-sn-glycero-3-phosphocholine (diether PC), were lyophilized from cyclohexane (BDH, 99.8%) prior to use. Nitrobenzoxadiazole (NBD)-spiperone and spiperone were purchased from Sigma, Gillingham, UK. Spiperone hydrochloride, haloperidol hydrochloride, flumazenil and raclopride were purchased from Tocris Bioscience, Avonmouth, UK. Diprenorphine and the drugs A–G were kindly supplied by GSK. All chemicals were used as received without further purification.
(b) Sample preparation
Samples were prepared by adding phospholipid to the drug in a mixture of chloroform (BDH, 99.8%):methanol (BDH, 99.8%) in a 3:1 or 1:1 ratio. The majority of the solvent was then removed by using a stream of nitrogen for 1 h and then under vacuum for a period of 1 or 2 days to dry to a constant mass. Bis–Tris buffer (i.e. 0.05 M Bis–Tris (Fluka, 99.0%)) and 0.1 M NaCl (Fluka, 99.0%) at a pH of 7.4 was added (40 wt%) and the mixture combined thoroughly and left to incubate at 37 °C. For FM studies, NBD-spiperone (1%) was added to the membrane vesicles in a saturated solution of spiperone in ethanol.
(c) Small-angle X-ray scattering
Samples for X-ray measurements were prepared by transferring the required amounts of co-lyophilized lipid–drug–buffer to Lindemann glass X-ray capillary tubes with a diameter of 1.5 mm (Gulmay Medical, TBD). The capillaries were flame sealed and then a silicon sealant plug (Dow Corning Corp.) was applied to ensure airtight sealing required to maintain the excess water hydration level. SAXS patterns were obtained using a high-intensity laboratory point source. The Cu Kα X-rays from a Bruker AXS FR591 rotating anode generator (Bruker AXS, Congleton, UK) were focused to a point, 300 μm in diameter, by torroidal optics. Photometrically accurate images were recorded using a XIDIS two-dimensional optoelectronic image-intensified X-ray detector (Photonic Science, Mountfield, UK). Sample temperature was controlled within ±0.03°C by a servo-controlled thermoelectric heater. The X-ray experiments were computer controlled using the software originally developed by E. F. Eikenberry (Paul Scherrer Institute, Switzerland) and S. M. Gruner (Cornell University, New York) and now customized for use at Imperial College. Image analysis was performed using the AXcess software package developed by Andrew Heron (Imperial College, London). The data shown was recorded at 25°C.
(d) Nuclear magnetic resonance spectroscopy
Solid-state NMR spectra were recorded on a Bruker 600 MHz NMR spectrometer by placing the samples (i.e. lipid–drug–buffer) in 4 mm zirconia rotors, which were then sealed with a Kel-F end cap. Single pulse and proton-decoupled experiments were acquired using standard Bruker pulse programs. The MAS experiments were performed at a speed of 3–4000 Hz. The data shown was recorded at 25°C.
(e) Transmission and fluorescence microscopy
The microscopic experiments were done using giant unilamellar vesicles (GUVs), electroformed at room temperature (Angelova & Dimitrov 1986), with an in-house designed and built device. For the microscopic studies, the GUVs were prepared by dissolving the lipid–CAD mixture (for transmission experiments) or only the lipid (for the florescence experiments) in a chloroform : methanol (9:1) mixture and then depositing the solution onto platinum electrodes set 5 mm apart. Subsequently, the solvent was removed slowly under vacuum. Then, the electrodes were hydrated using a solution of 0.1 M sucrose and 0.01 M Bis–Tris (pH 7.4) at 25°C. An a.c. field was slowly applied in the range of 0.2–1.0 V at 10 Hz for over 10 min. The vesicles were allowed to grow for 1 h before the field was slowly reduced to zero. The GUVs obtained with this method were stable for over 6 h, a period that exceeds the total duration of our experiments by an order of magnitude. The GUVs were imaged on an inverted Nikon TE2000 microscope using differential interference contrast (Nikon UK Limited, Surrey) and a CFI Plan Fluor 20X optic. For fluorescence imaging, a cube fitted with a 465 nm excitation filter and 505 nm emission filters (Chroma Technology, Rockingham, VT, USA) was used. An ND4 filter was used to minimize photo-bleaching effects. The images were acquired with a Hamamatsu Orca ER camera, which was set to maximum gain for the fluorescence experiments. The fluorescence experiments used exposure times of 0.150 s, with a 5.0 s window between each frame. The total time period for the video was 500 frames (i.e. 2500 s).
3. Small-angle X-ray scattering
The time-dependent SAXS data for haloperidol and spiperone effects on the buffer–DOPC system are shown in figure 3.
The presence of CADs in the model membrane system is characterized by a marked swelling of the phospholipid bilayer when compared to the control sample, the buffer–DOPC system without the CAD. This is indicative of a charged membrane interface, which causes opposing bilayers to repel and therefore leads to an increase in the repeat spacing. This phenomenon is time-dependent with the SAXS studies revealing that at an early stage of investigation, the buffer–DOPC–CAD system produces a broad diffuse diffraction pattern with peaks occurring in the ratio of 1:2:3, characteristic of a disrupted fluid lamellar Lα phase. This is in contrast to the sharp and well-defined diffraction pattern of the control system in the absence of CAD (Suwalsky et al. 1988).
From the time-dependent two-dimensional X-ray diffraction pattern, it is possible to show that, after a period of three weeks, a different liquid crystalline phase, the inverse hexagonal (HII) phase, predominates in both spiperone and haloperidol systems. This indicates that there is a continuous action of the CAD upon the phospholipid bilayer.
In contrast to the one-dimensional stacking of bilayers in the fluid lamellar phase, the inverse hexagonal phase consists of micellar cylinders of indefinite length arranged in a two-dimensional hexagonal format. The existence of the HII phase is not evident in the control system.
4. Nuclear magnetic resonance spectroscopy
Further investigations of CAD–membrane interactions were carried out via solid-state NMR spectroscopy. Solid-state 31P NMR spectroscopy of anisotropic, liquid crystalline systems is highly revealing (Seelig 1978), as the residual chemical shift anisotropic tensors (powder patterns) from the 31P nucleus in the phospholipids readily indicate the presence and the nature of different liquid crystalline phases. These investigations, performed in parallel with the SAXS experiments, show a significant change in the NMR powder patterns characteristic of the fluid lamellar Lα phase (figure 4) between the buffer–DOPC–CAD and the control, the buffer–DOPC system.
The presence of CADs in the phospholipid bilayer causes the appearance of a sharp isotropic peak at ca −3 p.p.m. and a second distinct narrow fluid lamellar powder pattern superimposed on the original (figure 4a(ii),b(ii,iii)). The sharp isotropic peak is characteristic of a micellar phase coexisting with a fluid Lα lamellar phase. The sharp isotropic line arises from rapid averaging, on the NMR time-scale that occurs in micelles in solution. This averaging mitigates the line-broadening interactions found in an anisotropic phase, e.g. fluid Lα lamellar phase. This confirms the observation of the presence of a micellar phase in the SAXS data. The second narrower powder pattern (figure 4a(ii),b(ii)), superimposed on a broader one, arises from the presence of a single-chain mono-oleoyl phospholipid or lysophosphatidylcholine (figure 1d; 1-oleoyl-2-hydroxy-sn-glycero-3-phosphocholine) and has been reported previously (Bhamidipati & Hamilton 1995). This was also confirmed by 31P NMR studies on mixtures of DOPC and lysophosphatidylcholine (data not shown), which showed identical 31P powder patterns. Furthermore, after a period of three weeks, the 31P NMR spectra of both the spiperone–buffer–DOPC and haloperidol–buffer–DOPC systems showed the characteristic pattern for an inverse hexagonal phase, also confirmed by the SAXS data.
Although anisotropic solid-state NMR spectra contain valuable information, it is also very useful to remove the line-broadening interaction to give the isotropic high-resolution solid-state NMR spectra. This was carried out by MAS NMR experiments, which involve rapid (3–4 kHz) physical rotation of the sample at the magic angle (θ=54.74°) to the magnetic field. This gives high-resolution information about the framework and the structure of the phospholipid bilayer. Solid-state 13C MAS NMR shows clear signals from the 13C nuclei in the structure of the phospholipid. Therefore, when the buffer–DOPC–CAD was examined, there were two major and one minor peaks of interest (figure 5b). The major peaks arise at 181 and 177 p.p.m. and the minor peak at 178 p.p.m.
The chemical shifts of these two peaks are indicative of the presence of the carbonyl group of oleic acid (181 p.p.m.; figure 1e) and ester (177 p.p.m.; figure 1a) moiety in the phospholipid, DOPC. The intensity of the oleic acid signal increased with time and that of the DOPC decreased with time. The minor peak can be attributed to the ester carbonyl group from lysophosphatidylcholine (figure 1d; Bhamidipati & Hamilton 1995).
From both SAXS and NMR studies, it is evident that in the presence of the CAD, the fluid Lα lamellar phase initially coexists with a micellar phase before ultimately breaking up to form a new phase, the HII phase.
5. Fluorescence and transmission microscopy
Direct visualization of the membrane degradation process should also be possible. This was carried out by injecting a solution of fluorescent-labelled CAD (figure 6) and CAD to a buffer–lipid mixture, where the lipids were formed into GUVs by electroformation.
Observations under the fluorescence microscope demonstrated that the degradation of the GUV occurs on time-scales consistent with normally observed pharmacokinetics and uses pharmaceutical concentrations of drug molecules. This has been difficult to detect by the less sensitive X-ray and NMR techniques. The fluorescent-labelled CAD (i.e. NBD-spiperone; figure 6) was injected as a saturated solution in ethanol in the aqueous environment adjacent to the electrodes that support the electroformed DOPC GUVs (see electronic supplementary material). Once injected, the fluorescent-labelled CAD binds rapidly (within 150 ms) to the membrane interface. Even by the first frame, there is evidence of membrane degradation and within 5 min, the GUVs are visibly distorted with many membranous fragments formed on both sides of the membrane (figure 7a), which float away from the host GUVs' membrane carrying CAD molecules. The formation of membrane fragments is a result of drug-induced ester hydrolysis at the membrane level. The degradation of the GUVs continues until they are completely disintegrated; the duration of this process being 40 min. Formation of co-lyophilized CAD–DOPC small unilamellar vesicles (SUVs) (figure 7b) was difficult, as these appeared both unstable and short-lived. Results typically consisted of severely distorted vesicles, in which there are a myriad of different smaller vesicles evident.
6. Phospholipid degradation model
We have examined the physical nature of the degradation fragments at the molecular- and meso-scale using SAXS and NMR spectroscopy and FM. Buffer–lipid–CAD mixtures (not formed into GUVs, but in the fluid lamellar phase) were observed by SAXS and solid-state 31P and 13C NMR spectroscopy. We observed the emergence, over time, of signals that were characteristic of micelle formation (figure 4a(ii),b(ii)). At even longer times (where there was extensive degradation), we recorded signals using NMR and SAXS that were indicative of the presence of the inverse hexagonal phase (figure 3d).
Micelle formation has been observed previously in CAD–lipid systems, and has been attributed to the formation of CAD–lipid conjugates (Schreier et al. 2000). However, this cannot explain the time-dependent formation of the inverse hexagonal phase as well as micelles. The simplest hypothesis to explain the appearance of both these phases and the earlier visual observations is that an acyl chain is being formed from the hydrolysis of DOPC, resulting in the formation of single-chain mono-oleoylphosphatidylcholine (MOPC) and oleic acid. It is known that MOPC will phase separate from the parent membrane and form micelles (Sheetz & Singer 1974; Bergstrand & Edwards 2001). This is represented schematically in figure 8.
This will leave behind a membrane enriched in oleic acid and DOPC, which will ultimately drive the remaining membrane mixture into an inverse hexagonal phase (Templer et al. 1998).
To directly test the hypothesis, solution-state NMR was used to determine the main degradation products in this process. Measurements were carried out by dissolving CAD-degraded lipid membranes in an organic solvent mixture (chloroform : methanol; data not shown). This again confirmed the presence of MOPC, oleic acid and residual DOPC. Further corroborating evidence was gathered employing TM (figure 7), using an identical co-lyophilization technique.
We propose that the chain hydrolysis is the result of acid-catalysed ester cleavage. The hydrolysis must occur via the presence of CAD at the interface, which is able to protonate the ester carbonyl group and therefore catalyse the hydrolysis of ester bond by water. There is already strong evidence in the literature that CADs reside at the interface when absorbed into phospholipid bilayers. Our SAXS observations of the fluid lamellar phase of CADs mixed with DOPC reflect an increase in repulsions between bilayers, consistent with ionized CADs residing at water–membrane interface as would be required for the suggested reaction mechanism. To test this hypothesis, we have made GUVs from the ether-linked phospholipid, dioctadecenylphosphidylcholine (figure 1c). In the absence of an ester linkage, the proposed reaction mechanism should be incapable of causing chain hydrolysis and indeed, this is observed with no observable membrane degradation (using the techniques previously described, SAXS and NMR) over a period of two weeks (figure 9). FM studies (figure 7c), in which GUVs were made from the drug dioctadecenylphosphidylcholine, also showed no evidence of degradation with time.
In the previous work (Attard et al. 2000), we have demonstrated that association with, and chemical activity of, amphipathic molecules at membrane interfaces are sensitive to the degree of elastic stress stored in a membrane. We find this to be the case for CADs as well as with CADs exhibiting lower membrane-binding affinity and slower reaction rates in membranes with reduced stored curvature elastic stress with respect to DOPC, e.g. DLPC (figure 10).
Our observations indicate a simple membrane translocation and transport pathway for CAD molecules (figure 8). CADs will bind rapidly to membrane surfaces, although with some selectivity for the mechanical state of the membrane. Once at the membrane–water interface, the CAD rapidly catalyses chain hydrolysis, although once again, the rates for this depend on the state of membrane stress and of course the chemistry of the CAD itself. The MOPC that is produced will form phase-separated micelles and other membrane fragments on either side of the parent membrane. CADs bound to these fragments are then carried in the micelle to the next membrane interface to undergo the same process again.
CADs have a well-known thematic structure, which is typified by the presence of a secondary and/or tertiary nitrogen (Reasor & Kacew 2001), a flat, planar structure, modest molecular mass and high levels of hydrophobicity. We believe that these empirically determined structural motifs, which are very common in CADs, facilitate the phospholipid degradation mechanism. The hydrophobicity of the CAD is necessary for its absorption onto the phospholipid bilayer. The CADs' planar shape and small size (relative to the lipid chain) mean that their incorporation between the phospholipids is easy and results in a drop in stored membrane curvature elastic stress by allowing lipid chain splay to occur. At physiological pH, the secondary and/or tertiary nitrogen is protonated and therefore charged. This anchors the CAD to the polar–apolar interface on the lipid membrane, placing it in proximity to the carbonyl ester. The protonation of nitrogen is an absolute requirement for acid hydrolysis to occur. We would propose that our observations may therefore provide some chemical and physical reasoning to Lipinski's rules for drug design.
Therefore, to test this hypothesis further, we carried out studies on 12 more CADs (figure 11). The results show a strong variation in the degradation rate of the CADs; nonetheless, these preliminary data indicate that this is not a specific but a general phenomenon (figure 12) for this class of compounds.
In this paper, we have shown evidence of a new type of CAD transport mechanism through the membrane (figure 8). We observe that once in the close vicinity of the membrane system, CADs will partition rapidly to the polar–apolar interface of the membrane initiating acid-catalysed ester hydrolysis of the phospholipid fabric in the membrane. This will generate MOPC that will form phase-separated micelles and membranous fragments on either side of the parent membrane. CADs resident in these fragments are then transported to another membrane interface to undergo the same process again.
The rate at which this process occurs is influenced by two key factors: the mechanical state of the membrane and the chemistry of CAD molecule. It is shown that CADs will have different reaction rates for membranes with different curvature elastic stresses. Thus, we show that the ester hydrolysis will occur at slower rates in a DLPC-based system with respect to a DOPC system. Key structural features of CAD molecules facilitate the phospholipid degradation mechanism. The amphiphilic nature of the CAD molecules will drive membrane partition with their planar shape and relatively small size, facilitating the process via a drop in the stored membrane stress by favouring chain splay. Protonation and charge of the CAD at physiological pH ensure that these molecules will reside in close proximity to the carbonyl ester group within the membrane. The protonation of the CAD molecule is essential for acid hydrolysis. These features show a strong correlation with Lipinski's rules for drug design. In addition to explaining some parts of the rules of drug design, we believe that our observations may form interesting lines of enquiry into related areas of membrane research.
The first of these is the CAD-induced cytotoxic effect, phospholipidosis. Here, the administration of CADs is found to induce the development of excessive membrane growth within cells, resulting in abnormal vesicular bodies and high levels of degraded phosphatidylcholine fragments. We postulate that these are products of the chemical degradation of lipids by CADs (Mortuza et al. 2003). In addition, it has been shown that phosphatidylcholine (PC) production is dependent on the stored curvature elastic stress in membranes, and we know that these levels are altered by the changes in lipid composition brought about by CADs. It would be intriguing to determine if the exogenous, chemical alteration of lipid composition, and hence stored membrane curvature elastic stress, in cells are pushing the rate of PC production to abnormally high levels.
The second area that is related to our observations is that certain peptide sequences can also bind to membranes and then promote acid-catalysed ester hydrolysis of the phospholipid ester linkages. Evidence (Naito et al. 2002) for this is from studies that show that the dynorphin peptide sequence actively binds to the phospholipids and causes cleavage of the phospholipid ester bonds. This is shown by the evidence of the isotropic peak in the static 31P NMR spectra that are reported when micelle-forming lysophospholipids are generated. It is also possible that other peptide sequences, e.g. the arginine-rich stretches on proteins that form the so-called protein transduction domains, can operate with the same mode of action. An example of this is the TAT-4 sequence found in HIV.
Finally, it has been generally recognized in the area of gene therapy (Pack et al. 2005) that cationic lipids and polymers used for delivery of nucleic acids can also be cytotoxic. We have observed membrane degradation for the commonly used gene delivery agent, polyethyleneimine. This suggests that ester hydrolysis may be a common cause of cytotoxicity in gene delivery agents.
E. T. Samulski (Department of Chemistry, University of North Carolina, USA). There are many ineffective drugs. If efficiency is due to the ability to catalyse membrane degradation, perhaps ineffective drugs could be improved by combining them with a catalyst for membrane degradation.
R. Templer. Our measurements have been made on simplified membrane systems, so we have to be careful about extrapolating our results to in vivo conditions. With that caveat in mind, it does seem to me to be reasonable to do exactly what you suggest, i.e. to physically separate the region of a molecule that imparts drug function from that which causes lipid hydrolysis. I suspect that if we look at natural systems, we would find that this has already been explored by nature. In particular, the delivery of toxins and the like may use this approach.
A. A. Kornyshev (Imperial College London, UK). How many tails can be eaten by one drug molecule?
R. Templer. We have observed both hydrocarbon chains being lysed in our experiments with cationic amphiphilic drugs. First, we observe single-chain lysis. This produces micelles and leaves a fatty acid behind in the bilayer membrane, which is (in the initial stages of the reaction) rich in double-chain lipid. As the reaction progresses and more micelles are formed, the evidence shows that the proportion of fatty acid increases to such a level in the membranes that the inverse hexagonal phase forms. If we wait even longer, we then observe free phosphorylcholine and lot of fatty acids. This is a direct evidence that both tails have been hydrolysed and that the drug molecules must have been in micelles.
H. F. Gleeson (School of Physics and Astronomy, University of Manchester, UK). Why do the micelles form ‘into’ the cell and why do they selectively carry the drug rather than it remaining in the lamellar structure?
R. Templer. Micelles form on both sides of the membrane and, therefore, both inside and outside of the vesicle. In our video recording you only notice those going inwards because they are not free to flow out of the field of view. Therefore, you can clearly observe an increase in fluorescence intensity inside the vesicle, but not outside. It is well known that lyso-PC will phase separate from double-chained PC at very low compositions and our fluorescence data and analysis of reaction products show that the drug is in micelles as well as in vesicle walls. I suspect that in fact the cationic amphiphilic drug sits rather comfortably at the polar–apolar interface of a highly curved interface, such as is found in a micelle, since it will provide some ‘shielding’ of the hydrophobic core of the micelle.
N. W. Roberts (School of Physics and Astronomy, University of Manchester, UK). Using the mechanism you described, is there any way, without being toxic, that you could improve the drug delivery efficiency?
R. Templer. I think that I will have to engage in some speculation here to give you any sort of answer. Let us assume that phospholipidosis is indeed caused by the continuous elevation of curvature elastic stress due to the production of fatty acid in the cell membrane. The important word in the last sentence is continuous, since the cell can (and does) cope with short bursts of chain lysis. In the model we are proposing for drug transport, the rate at which the drug crosses a membrane depends on the catalytic rate. As this rate is increased so is the rate at which drugs diffuse through the body, which in effect means the drug will not stick around on membranes for very long and will find its target rather rapidly. Once it finds its target we assume it no longer consumes lipids. This seems to imply that (in the absence of other effects that remove drug from your body) if we increase drug delivery efficiency we should reduce the chances of phospholipidosis. This would be so, because the residency time on a cell membrane will be short and the drug is more likely to be rapidly taken up by its receptor and hence removed from damaging lipid membrane composition. So it seems to me that with this simple model we would expect to find only improvements in drug delivery and reductions in toxicity by increasing the efficiency of the acid-catalysed hydrolysis of lipid chains.
This work has been generously supported by GSK plc and the EPSRC Platform grant GR/S77721.