Cholesterol and Lipid Phases Influence the Interactions between Serotonin Receptor Agonists and Lipid Bilayers*

Solid state NMR techniques have been used to investigate the effect that two serotonin receptor 1a agonists (quipazine and LY-165,163) have on the phase behavior of, and interactions within, cholesterol/phosphocholine lipid bilayers. The presence of agonist, and particularly LY-165,163, appears to widen the phase transitions, an effect that is much more pronounced in the presence of cholesterol. It was found that both agonists locate close to the cholesterol, and their interactions with the lipids are modulated by the lipid phases. As the membrane condenses into mixed liquid-ordered/disordered phases, quipazine is pushed up toward the surface of the bilayer, whereas LY-165,163 moves deeper into the lipid chain region. In light of our results, we discuss the role of lipid/drug interactions on drug efficacy.

Solid state NMR techniques have been used to investigate the effect that two serotonin receptor 1a agonists (quipazine and LY-165,163) have on the phase behavior of, and interactions within, cholesterol/phosphocholine lipid bilayers. The presence of agonist, and particularly LY-165,163, appears to widen the phase transitions, an effect that is much more pronounced in the presence of cholesterol. It was found that both agonists locate close to the cholesterol, and their interactions with the lipids are modulated by the lipid phases. As the membrane condenses into mixed liquid-ordered/disordered phases, quipazine is pushed up toward the surface of the bilayer, whereas LY-165,163 moves deeper into the lipid chain region. In light of our results, we discuss the role of lipid/drug interactions on drug efficacy.
The majority of drug targets are membrane proteins. Hence the interaction of a drug with the membrane is crucial for its efficacy. A high location probability in a particular part of a membrane and its orientation with respect to the membrane normal could well be relevant to how the drug is presented to the target protein's binding site (1,2). Cholesterol-rich microdomains ("lipid rafts") add another layer of complexity because the lateral organization of the membrane could well affect the concentration of a drug close to its target protein. This possibility is backed up by the observation that many chemicals, including drugs that target serotonin receptors, preferentially localize into these domains (3)(4)(5). Meanwhile, the location and activity of some membrane proteins, particularly G-protein-coupled receptors, seem to be affected by domains (6). Therefore, it is easy to imagine that a drug's efficacy may be limited by it partitioning into a different membrane domain than its target receptor. Furthermore, it is clear that many membrane proteins are modulated by the properties of the surrounding lipids. Hence, a drug has the potential to regulate a protein by altering these properties (7). This raises the intriguing possibility of a class of drugs that, instead of directly targeting a protein, target the membrane in which the protein resides. This route of action was first proposed as early as 1899 as a possible mechanism for general anesthetics, when it was noted that the potency of an anesthetic correlated very strongly with olive oil/water partitioning (8), yet drugs affecting membrane proteins via lipid interactions, so called "membrane-lipid therapy," have only relatively recently been explored (7,9,10).
Some simple lipid mixtures have been used as models for in vivo microdomains. A particularly well studied mixture consists of 1,2-dipalmitoyl-sn-glycero-3-phosphocholine (DPPC) 2 and cholesterol (11)(12)(13)(14)(15)(16). With 20 mol % cholesterol, Scheidt et al. (14) interpreted the phase behavior as follows. At high temperatures (above 316 K), the whole bilayer is in the liquiddisordered (L d ) phase, characterized by rapid axial rotation and highly disordered aliphatic chains (16). Between 311 and 316 K, a mixed phase of liquid-ordered (L o ) and L d phase lipids exists. Below 311 K, the membrane starts to coalesce into a mixture of L o and gel phase (where the rotation of the lipids slows dramatically) before finally forming a homogenous gel phase. It is the L o domains that are thought to resemble cholesterol-rich microdomains found in biological membranes.
Proton NMR and specifically magic angle spinning-assisted nuclear Overhauser enhancement spectroscopy (MAS-NOESY) experiments have proven to be an excellent tool for investigating, with atomic resolution, the location of small molecules embedded in lipid membranes (17)(18)(19)(20). However, these studies have been limited to binary mixtures because anything more complicated results in serious spectral overlap. To circumvent this problem, we have used ternary mixtures of agonists, cholesterol, and chain deuterated DPPC lipids (DPPC-d 62 ), thus removing the signals that would otherwise swamp the cholesterol resonances. The system allows interactions between cholesterol and agonist to be observed. Furthermore, the rapid axial motion of the lipids in the L o and L d phases allows high resolution 1 H spectra to be collected in both of these phases (14).
The work presented here follows from a study in which it was noticed that several of the agonists induced significant chemical shifts in the cholesterol signals of a brain lipid extract (17). To gain further insight into the interactions between bilayers containing cholesterol and the agonists, we now present a solid state NMR-based approach that utilizes DPPC-d 62 to reveal NOESY interactions within the L o and L d * This work was supported by a Royal Society project grant and an Engineering and Physical Sciences Research Council doctoral training account. □ S The on-line version of this article (available at http://www.jbc.org) contains supplemental Fig. 1

EXPERIMENTAL PROCEDURES
Materials-DPPC, DPPC-d 62 , and cholesterol were purchased from Avanti polar lipids (Alabaster, AL). All other chemicals, including quipazine as a maleate salt and LY-165,163, were purchased from Sigma.
Sample Preparation-Samples of DPPC/cholesterol, DPPC/ agonist, and DPPC/cholesterol/agonist were prepared at mole ratios of 8:2, 9:1, and 7:2:1, respectively. Typically, 80 mg of DPPC was used per sample. The lipids and agonists were codissolved in chloroform/methanol (1:1, v/v). The solvents were evaporated under vacuum, and the resulting lipid cake was suspended in 1 ml of doubly distilled water, frozen in liquid nitrogen, and then lyophilized overnight under high vacuum. The dry lipid mixtures were hydrated with ϳ200 l of D 2 O for proton NMR or H 2 O for deuterium NMR experiments and then subjected to three freeze/thaw cycles. The mixtures were centrifuged to remove excess water. 4-mm MAS rotors were filled with the resulting pellets. Samples were kept in the rotors for both 1 H and 2 H measurements.
NMR Measurements-All NMR experiments were carried out on a Bruker Avance II 500-MHz spectrometer using a 4-mm MAS probe operating at a frequency of 500.1013 MHz ( 1 H) and a wide line probe at 76.7685 MHz ( 2 H). 1 H experiments were carried out with an MAS speed of 10 kHz. 2 H measurements were conducted without sample spinning. 1 H spectra were externally referenced to tetramethylsilane at 0 ppm. 1 H experiments were conducted with a typical /2 pulse length of 7 s and a relaxation delay of 4 s. Two-dimensional NOESY experiments had 256 or 512 increments and up to 64 scans/increment. NOESY build-up curves were acquired using mixing times between 10 and 800 ms. Samples were heated to 333 K before cooling to 273 K and then heating to the desired measuring temperature. Experiments were conducted at 298, 308, and 316 K (according to the thermocouple in the probe head). MAS at 10 kHz was found to heat the sample by 2 K (derived from the relative positions of methanol peaks (21)). All subsequent references to the sample temperature will take this into account.
2 H quadrupole echo experiments (22,23) were acquired with a spectral width of 100 kHz, a recycle delay of 1 s, 30-s echo delay, 10-ms acquisition time, /2 pulses of 5.5 s, and between 2048 and 4096 scans. All samples were heated to 333 K before cooling down to 273 K prior to temperature scans. Measurements were then taken at 1 K intervals between 273 and 333 K (according to the thermocouple in the probe head).
NMR data were processed using Topspin version 1.3 (Bruker Instruments, Karlsruhe, Germany). NOESY peak volumes were obtained by peak fitting and integration using CARA (24). Data Analysis-NOESY data were used to calculate the location of a given nucleus in the lipid membrane using the "full matrix rate analysis," described in detail by Huster et al. (25). In short, experimentally measured NOESY peak volumes, represented by the matrix A, at the mixing time t m , and the cross relaxation rates R are linked by the matrix equation as follows.
The relaxation rate matrix R is calculated by rewriting Equation 1 as follows.
Here, X is the matrix of eigenvectors, and D is the diagonal matrix of eigenvalues of the normalized peak volume matrix a(t m ) ϭ A(t m )((A(0) Ϫ1 ). The relaxation rates, contained in R, were taken as indicators of the relative location probabilities and are plotted to give a location profile of the agonists. All calculations were carried out with the help of Python (Python Software Foundation, Wolfeboro Falls, NH), specifically with the packages matplotlib (available from the sourceforge Web site) and scipy (26). 2 H spectra were de-Paked, and first spectral moments were calculated using NMR-Depaker software (available from the Launchpad Web site). The de-Paking procedure was performed according to the fast Fourier transform-based fast deconvolution algorithm (27).
Individual C-2 H bond order parameters (S CD i ) were calculated from quadrupolar splittings ( q i ), as described previously (23,28,29), where A Q is the static deuterium quadrupolar constant (167 kHz for C-2 H bonds (30). S CD i profiles for the sn-1 chain of DPPC-d 62 were constructed according to the published assignments (28,29) and comparisons with specifically labeled phospholipids (31,32).
First moment (M 1 ) calculation was performed using a sign reversal for negative frequencies to avoid zero values due to 2 H NMR spectrum symmetrical relative to the origin, where ϭ 0 is the center of the spectrum.

RESULTS
The low agonist/lipid ratio (1:10 mole ratio) used in this study is desirable to minimize the effect of the agonist on the phase behavior of the lipids.
1 H MAS-Proton MAS spectra of DPPC and DPPC-d 62 / cholesterol lipid membranes in the presence and absence of 5HT 1a receptor agonists are shown in Fig. 2. Spectra were collected with 10-kHz sample spinning. At this speed, lipid mixtures in the gel phase yield extremely broad lines, which render individual resonance unresolvable ( Fig. 2A, right). However, lipids in the L o or L d phases exhibit rapid axial rotations. This motion combined with that of the MAS results in well resolved proton spectra. All NOESY measurements, described below, take place at a temperature where spectra are well resolved and hence most of the gel phase has melted.
Assignments of the agonist signals were performed previously (17). However, for the sake of simplicity, only two peaks for each will be used for the analysis. Assignments for DPPC have been determined elsewhere (14). Proton cholesterol assignments are derived from published MAS 13 C measurements of cholesterol/DMPC mixtures (33) and confirmed with DPPC/cholesterol using 13 C-1 H correlation experiments (data not shown).
In DPPC/agonist mixtures, above the gel-L d phase transition at 310 K, most DPPC proton resonances are resolved ( Fig. 2A). The addition of cholesterol to the DPPC results in loss of resolution in the aliphatic region (below 2.5 ppm) as the cholesterol signals overlap with those from the chains of DPPC. Replacing protonated DPPC with DPPC-d 62 resolves this issue by removing 99% of the proton signals from the DPPC chains, thus revealing cholesterol chain signals (Fig. 2B).
Many of the cholesterol signals overlap with one another, meaning that it is not always possible to unambiguously as-  (17), Scheidt et al. (14), and Soubias et al. (33), respectively, and correspond to labels on the molecules in Fig. 2. Superscript c denotes a cholesterol peak, and superscript p denotes a phospholipid peak. For DPPC, peaks are as fol- sign a cholesterol peak to a particular part of the molecule. However, broadly speaking, it is possible to group the peaks into structural categories. The terminal CH 3 groups (C26/27 c ) are clearly resolved from the CH 3 groups protruding up from the plane of the sterol rings (C18 c , C19 c , and C21 c ), whereas the protons on the sterol rings give another set of peaks. 1  Phase Behavior-Phase diagrams of binary mixtures of DPPC and cholesterol are well documented. However, the phase behavior of the DPPC/cholesterol/agonist mixtures are unknown. To gain some insight into the properties of these mixtures, 2 H NMR measurements were performed at 1 K intervals between 273 and 333 K on DPPC-d 62 with and without cholesterol and agonists ( Fig. 3 and supplemental Fig. 1). Fig.  4 shows the temperature dependence of the first spectral moment (M 1 ) for the 2 H spectra of DPPC-d 62 plus agonist (without cholesterol). M 1 is proportional to the average quadrupolar splitting and thus gives a measure of the width of all spectral components (29). The spectra of pure DPPC-d 62 show a small decrease in the M 1 at 301 K as the lipid enters a ripple phase (34). This is followed by a sudden decrease between 309 and 310 K as the lipid goes through the gel to L d transition. The addition of agonists to the system appears to abolish the spectral changes characteristic of the gel to ripple phase transition. Furthermore, in the presence of agonist and prior to chain melting, M 1 is significantly lower than for DPPC-d 62 alone. The transition from the gel to L d phase is also slightly lowered and broadened, now occurring between 304 and 307 K for DPPC-d 62 /quipazine and between 304 and 309 K for DPPC-d 62 /LY-165,163.
DPPC-d 62 /cholesterol mixtures behave as expected (11). There is no sudden phase transition; instead, there is a shallow decrease in line widths and first spectral moments indicative of gradual transitions between gel, L o , and L d phases, all of which may co-exist (Fig. 5). The presence of agonists lowers the transition temperatures but not the overall trend.
However, at low temperature, the samples containing agonist and cholesterol displayed unusual behavior (Fig. 5). Between 278 and 293 K, the first spectral moments increased before peaking prior to the onset of gel phase melting.
Closer inspection of the 2 H spectra of DPPC-d 62 reveals that at some temperatures two sets of peaks arise from the terminal methyl groups; this can be clearly seen in the insets of Fig. 3. This splitting is the result of the non-equivalency of the two lipid chains (11) and can be used as a reporter for the presence of L o phase (11,35,36). Methyl peak splittings are observed between 293 and 317 K for DPPC-d 62 /cholesterol and over a similar temperature range, just shifted down by a few K (between 287 and 313 K) for DPPC-d 62 /cholesterol/ quipazine. In the case of DPPC-d 62 /cholesterol/LY-165,163, methyl group splittings were observed over a much larger range, between 280 and 327 K. Interestingly, methyl peak splitting is also seen at 306 K in the DPPC-d 62 /LY-165,163 sample (supplemental Fig. 1d). The magnitudes of these splittings are shown in Figs. 4 and 5.
2 H spectra can also be used to report on the gel phase where it is manifested as large quadrupole splittings and poorly resolved signals. Using these criteria, it is possible to define the upper limit of the gel phase as being 310 K (in agreement with the literature (11,14)), 300 K, and 298 K for DPPC-d 62 samples containing cholesterol, cholesterol plus quipazine, and cholesterol plus LY-165,163, respectively.
Order Parameters-Individual bond C-2 H order parameter (S CD i ) profiles derived from 2 H NMR on DPPC-d 62 at 323 K are shown in Fig. 6. The non-equivalency of the two lipid chains leads to a separate profile for each chain (29). However, for the sake of clarity, only data from the sn-1 chain are shown.
It is immediately noticeable that the S CD i value of bilayers containing cholesterol is, as expected, appreciably higher than  NOESY measurements on DPPC-d 62 /cholesterol/agonist were conducted at the lowest temperature at which individual resonances could still be resolved (300 K) with mixing times of 10, 200, 300, and 400 ms. The experiments were repeated at 318 K with mixing times increased to 20, 400, 500, and 600 ms to take into account the decrease in correlation times at the higher temperatures (37). At 318 K, DPPC-d 62 /cholesterol/ quipazine is in a pure L d phase, but DPPC-d 62 /cholesterol/ LY-165,163 still has traces of L o phase (Fig. 5). However, increasing the temperature further resulted in correlation times being too short to practically measure cross-peak intensities (data not shown).
A typical MAS-NOESY spectrum of LY-165,163/DPPC-d 62 and cholesterol membranes is depicted in Fig. 7. Cross-peaks are visible between all aromatic LY-165,163, glycerol and   headgroup DPPC, and cholesterol resonances. The build-up rates of the cross-peaks with increased mixing times give an indication of the proximity of two protons. Consequently, the location probability of a particular agonist proton in the lipid membrane can be determined relative to other lipid protons (Fig. 8). If multiple agonist proton signals are resolved (see Fig. 2), the orientation of the molecule relative to other components in the bilayer can also be determined.
The locations of quipazine and LY-165,163 in DPPC broadly agree with our previous studies conducted in dioleoyl-sn-glycero-3-phosphocholine (DOPC) (17); the aromatic ring at one end of the quipazine molecule has a maximum location probability in the chain region of the bilayer, whereas the other end locates in the interface region, indicating that it has a net orientation parallel to the lipids. In the case of LY-165,163, the maximum location probability is the same as reported previously in DOPC, but in DPPC, it is much more tightly distributed around the chain/glycerol regions.
It should be noted at this point that the reported quipazine resonances are the same as we used previously (17). However, here, different LY-165,163 peaks are used because they can be more clearly resolved at 300 K. One of these peaks is derived from protons situated in the center of the molecule, whereas the other is at one end of the LY-165,163 molecules (Fig. 1); thus, some spatial resolution has been lost. The NOESY measurements on DPPC-d 62 /cholesterol/agonist samples give  DECEMBER 31, 2010 • VOLUME 285 • NUMBER 53 information on the interactions between agonist and cholesterol at the expense of information on interactions between the chain region of DPPC and the agonist.

Cholesterol Modulates Drug/Phospholipid Interactions
At 318 K (Fig. 8, E and F), both agonists have weak interactions with the terminal methyl groups of the cholesterol chain, consistent with the previous observations that the agonists do not accumulate deep within the chain region of the bilayer. Beyond this similarity, the agonists behave quite differently. Quipazine shows large cross relaxation rates with all of the remaining cholesterol resonances. The aromatic ring of quipazine interacts more strongly with the C18 c than the piperazine ring at the other end of the molecule, whereas the piperazine ring has stronger interactions with the phospholipid headgroup. Interestingly, the presence of cholesterol also seems to change the interactions between quipazine and the phospholipid. Cross-relaxation rates between quipazine and DPPC ␣ resonance are larger in the ternary DPPC-d 62 /cholesterol sample than in the binary DPPC sample.
In the case of LY-165,163, the center of the molecule has strong cross-peaks with C9/19/21 c but little contact with anything else. Meanwhile, the aromatic at the end of the molecule has significant interactions with all reported cholesterol peaks with the exception of the C14/17/24 c peak. Furthermore, the presence of cholesterol appears to increase the interactions between LY-165,163 and the phospholipid headgroup.
The second set of NOESYs were conducted at 300 K (Fig. 8,  C and D). Under these conditions the orientational bias of quipazine is even more apparent. As before the aromatic ring interacts more strongly with C18 c but now, unlike in the binary DPPC or DOPC mixtures, the piperazine ring has strong interactions with all three sets of phospholipid headgroup protons. In the case of LY-165,163, cooling the sample causes an increase in the interactions between the center of the LY-165,163 molecule and the various CH 3 groups in cholesterol (with the exception of the chain terminal methyls), whereas the aromatic ring appears to have a preference for the cholesterol ring protons. Furthermore, the interactions between LY-165,163 and the phospholipid headgroups are lost.

DISCUSSION
This work investigates the effect of cholesterol and lipid phase behavior on the interactions between two serotonin receptor agonists and the lipid bilayer components. Our prior study showed that the agonists induced a chemical shift of cholesterol signals in brain lipid bilayers (17). To further investigate this phenomenon, we have simplified the system so that it now contains just phospholipid, cholesterol, and an agonist. This allows the composition of the system to be manipulated to include chain-deuterated lipids, resulting in increased 1 H spectral resolution and allowing 2 H measurements to be performed on the same sample. Together, this allows NOESY cross-peaks to be observed between all components, giving information on the direct interactions between membrane components, whereas analysis of the 2 H spectra provides data on the agonists' bulk effects.
In the absence of cholesterol, the agonists have no noticeable effect on structural characteristics of the L o phase (derived from 2 H data; see Table 1) and a small but significant effect on the lipid phase behavior (Fig. 4). Most notably, they appear to abolish the gel to ripple phase transition while lowering M 1 prior to the onset of chain melting. Together, these data suggest that either the agonists induce formation of the ripple phase at a much lower temperature (so its onset is not seen within the measured temperature range) or the incidence of the ripple phase is much reduced by the presence of agonist. Both agonists also reduce the gel-L d phase transition temperature, which now occurs 3 and 5 K lower in the presence of quipazine and LY-165,163 respectively. The phase transition also occurs over a wider temperature. This implies that there may be a small range in which a mixture of phases exists, a supposition that is backed up by the observation that at 306 K and in the presence of LY-165,163, the 2 H terminal methyl peaks are split (supplemental Fig. 1d and Fig. 4). These spectral features are consistent with the presence of L o phase (see below). As far as we are aware, prior to this, L o phases in DPPC have only ever been reported in the presence of sterols.
The combination of agonists and cholesterol generates some unusual phase diagrams. First spectral moments as a function of temperature behave as expected for mixtures of DPPC-d 62 and cholesterol (Fig. 5); as temperature increases, M 1 slowly decreases before lipids begin to change phase and M 1 drops more swiftly. At high temperatures, DPPC-d 62 /cholesterol/agonist mixtures behave in an analogous fashion to DPPC-d 62 /cholesterol. However, quite different behavior is seen at low temperatures. The limit of the gel phase is a good 10 K lower in the presence of agonist. However, more interestingly, immediately prior to the onset of gel melting M 1 increases, reaching a peak for both agonists at about 293 K and then dropping as expected (Fig. 5, a and b). As far as we are aware, this phenomenon has not been reported previously and may represent an unknown sub-gel phase. However, the scope of this study does not include any further analysis of this part of the phase diagram, and we have not investigated it further.
The methyl peak splitting seen in the 2 H spectra of DPPCd 62 is thought to reveal the onset of the L o phase. The terminal methyl group peaks splitting into doublets (Fig. 3 and supplemental Fig. 1) arises from non-equivalence of the two aliphatic chains in the L o phase (11,35,36). This occurs be-  greatly extends the temperature range in which the methyl peak splittings are observed. However, at low temperatures (e.g. Fig. 3, 283 K), these splittings are not accompanied by any other sharp spectra features that would be expected for lipids undergoing rapid axial motion associated within a L o phase. These sharp features do not begin to emerge until 299 K (DPPC-d 62 /cholesterol), 286 K (DPPC-d 62 /cholestero/LY-165,163), and 293 K (DPPC-d 62 /cholesterol/quipazine). Nevertheless, wherever the exact onset of the L o phase is, the conclusions remain the same. Together, the presence of the unknown sub-gel phase, the extended range of the L o phase, and the order parameter measurements all seem to suggest that LY-165,163 has a marked effect on the bulk bilayer properties when in the presence of cholesterol. By contrast, quipazine induces the same sub-gel phase, but at higher temperatures it does little more than shift the phase transitions by a few degrees. These differences between the agonists effects on the phase behavior are mirrored in their effect on bilayer order parameters. Neither agonist causes any change in the order parameters when mixed with DPPC alone. However, when cholesterol is added into the mix the further addition of LY-165,163 then causes a significant increase in order parameter, whereas once again quipazine elicits no change. These changes may be more intuitive when expressed in terms of structural characteristics (Table 1), derived from the order parameter measurements as described by Petrache et al. (38). Now it becomes apparent that, as expected, cholesterol causes an increase in the hydrophobic thickness by ϳ2 Å. LY-165,163 has no effect by itself but together with cholesterol causes a further thickening of 0.7 Å. NOESY interactions can be used to shed light on specific contacts made between agonists and lipids (Fig. 8). The magnitude of the NOESY cross-peaks may be a function of not just the internuclear distances but also the mobility of interacting nuclei and spin diffusion. Huster and Gawrisch (25,39,40) have conducted several studies to better understand the contribution that all of these factors have on the interpretation of cross-relaxation rates. They have concluded that spin diffusion can be ruled out in lipid bilayer systems under MAS conditions and that "cross relaxation rates report true statistics of nearest-neighbor contacts in the bilayer" (25,39,40).
Correlation times certainly influence observed cross-peak intensities however, these have been shown (at least for ethanol/lipid interactions) (37) to be consistent throughout an L d bilayer. But in mixed phase systems the lipids will be in at least two different motional regimes, with the lipids in the L o phase likely to experience longer correlation times than the same lipids in an L d phase. This could lead to the NOESY data overrepresenting interactions that occur in the L o phase. The aim of the NOESY measurements conducted at 300 K was to investigate interactions in the L o phase; therefore, the results may be skewed toward L o phase interactions. However, this does not affect the interpretation of the results. At the higher temperature (318 K), the bilayer is almost entirely L d phase, and therefore interactions in the L o phase will represent a tiny fraction of the overall signal.
Given these factors, it is then reasonable to interpret the cross-relaxation rates in Fig. 8 as an indication of the distribu-tion of an agonist in the bilayer. Therefore, the locations of the agonists in the DPPC lipid bilayer are in broad agreement with those published on DOPC; the agonists tend to locate in the interface region, and quipazine is oriented parallel with the lipid axis, although the distribution of LY-165,163 does seem to be narrower than in DOPC (17). Similar differences between drug interactions with DPPC and DOPC have been reported before with ␤2-adrenoreceptor agonists (5). Possibly, interactions between the -electrons of the alkene groups in the DOPC chain and the agonist's aromatics may have served to pull the agonist deeper into the DOPC bilayer. Alternatively, here we have used lower concentrations of agonist as compared with our previous MAS-NOESY study (17); consequently, the bilayer may not be saturated with agonist and thus is able to accommodate the majority of the agonist in its preferred location.
NOESY measurements on the samples containing DPPCd 62 cholesterol and agonists were collected at 318 K where the mixtures are dominated by the L d phase, and at 300 K, where the bilayers are a potential mix of gel, L d , and L o phases. Any gel phase components will give very broad lines, rendering the individual resonances unresolvable. Therefore, the interactions being observed will be taking place in the L d and/or L o phases.
At 318 K quipazine interacts strongly with all of the resolved cholesterol peaks except the terminal methyls (Fig. 8F). The piperazine ring of quipazine interacts more strongly with the phospholipid headgroups, whereas the opposite end of the agonist has stronger interactions with C18 c at the bottom of the cholesterol ring structure. These observations demonstrate that quipazine's orientational preference is maintained in the presence of cholesterol. In fact, cholesterol seems to increase the interaction between the piperzine group and the protons in the DPPC headgroup. This trend is even more apparent at 300 K (Fig. 8D), where the piperzine ring now interacts with all of the headgroup protons. This seems to imply that the ordering of the chains that occurs due to the presence of cholesterol, and then again in the L o phase, causes the quipazine to be displaced and pushed up toward the surface of the bilayer. It should be noted that differences between actual cross-relaxation rates at different temperatures and samples are difficult to interpret due to the drastically different motilities experienced in each case. The increase in temperature is expected to lead to shorter correlation times and a concomitant reduction in cross-peak intensity and cross-relaxation rate. This is observed for all lipid/agonist interactions with the exception of the piperazine resonances on the quipazine (Fig.  8, D and F). Here the cross-relaxation rates are higher at the elevated temperature. This suggests that either the nuclei are in very close proximity to the lipids or, at the higher temperature, the piperazine ring is in a motional regime that leads to an enhanced NOE. We can speculate that this may be a consequence of a strong interaction between cholesterol and the piperzine ring, leading to a longer correlation time and stronger NOE. Meanwhile, the other end of the quipazine ring would be left free to rotate around the single bond in the center of the molecule and hence is less affected by piperazine/ cholesterol interaction.
In contrast, to quipazine, LY-165,163 does not show the same degree of interactions with cholesterol (Fig. 8, C and D). There are only very weak interactions with the bottom of the cholesterol rings (C14/17/24 c ), which seems surprising given that the rest of the ring structure is accessible to LY-165,163. Possibly, the C18 c and C21 c methyl groups that protrude out from the plane of the cholesterol rings may shield this region of the sterol, but more interestingly, and in contrast to quipazine, LY-165,163 appears to be pushed deeper into the bilayer at the cooler temperature. This is evidenced by the fact that interactions between LY-165,163 and the phospholipid headgroup are weaker at lower temperatures, whereas there is a concomitant increase in interactions with C18 c .
The behavior of the agonists with respect to cholesterol can be easily reconciled with the agonists' effects on the bulk lipid properties. Quipazine is pushed out of the chain region by cholesterol and the increase in order associated with the L o phase. Consequently, it has a relatively small effect on the order parameters and phase transition barriers. However, LY-165,163 appears to locate deeper in the membrane as the fraction of L o increases, so it is in a position where it can have a greater influence on the phase behavior and cause the observed increase in hydrophobic thickness.
Drug/lipid interactions have the potential to modulate the efficacy of the drug in several ways. First, the drug must be able to negotiate membranes on the way to its target site. This may involve crossing the gut epithelium and the blood brain barrier. If this occurs passively then the drug's ability to partition into cellular membranes is vital. There is plenty of evidence to show that the composition of a membrane alters the partitioning coefficient of drugs (10), but until now, the systems used to study these effects have not been accessed with MAS-NOESY measurements. Hence, although we can be sure that, for example, the addition of cholesterol to a lipid bilayer can have dramatic affects on a drug's ability to partition into a membrane (5), we have not until now been able to see how this behavior is manifested on an atomic level.
A related factor is the propensity of agonists to preferentially partition into particular membrane domains. Serotonin receptors have been shown to be activated by and accumulate in lipid raft domains (3)(4)(5). Therefore, for maximum bioefficiency it would clearly be advantageous for a drug to accumulate in the same domain as its target. Because we have shown that quipazine and LY-165,163 both interact strongly with cholesterol in the L o phase, it would be reasonable to assume that these agonists accumulate in this phase and hence also partition into cholesterol-rich raft domains where they can more readily access their target protein.
Furthermore, the recent structure of the ␤2-adrenoreceptor suggests that the protein binds cholesterol via a 4-amino acid cholesterol binding motif (41). Most serotonin receptors have the same motif, which appears to be an evolutionarily conserved cholesterol binding site (42), so it is easy to imagine how an interaction between cholesterol and a drug may affect its efficacy, either by interfering with the cholesterol/protein contacts or by the cholesterol "delivering" the drug to the protein.
This then leads to a further factor; the ability of an agonist to access the binding sites on its target protein. To do this most efficiently the drug must accumulate in the membrane close to the binding site and be presented to the protein in the correct orientation. This is of particular importance for serotonin receptors given that its binding sites are thought to be accessible via the membrane as well as the solvent (1). Our data show that the tightening of the bilayer associated with the presence of cholesterol and L o phases results in a change in the location of the agonists. Quipazine is pushed up and out of the hydrophobic region, whereas LY-165,163 is pushed deeper into the membrane. Therefore, it is quite possible that cholesterol and the L o phase serve to retard quipazine binding to some serotonin receptors. Interestingly, quipazine binds relatively weakly to serotonin 1a receptors (43) (with its membrane-embedded binding site) compared with LY-165,163 (44), but it binds much more strongly to the serotonin 3 receptor, which is a ligand-gated ion channel and not a G-protein-coupled receptor and so is thought to have a more exposed binding site.
Finally, it is apparent from our data that the agonists can affect the phase behavior of the lipid bilayer. This is particularly true for LY-165,163, which promotes the L o phase both in the presence and absence of cholesterol and also decreases the fluidity of the bilayer. If this phenomenon holds true at physiological concentrations (and other G-protein-coupled receptor agonists have been shown to alter the bulk bilayer properties at very low concentrations (5)), then the drug may cause condensation of L o phase in vivo. This in turn could serve to modulate the activity of the target protein. There is plenty of evidence that drug molecules can affect the bulk properties of membranes (reviewed by Lucio et al. (10) and references therein). However, the possibility of designing drugs to modulate protein activity via drug/lipid interactions is only just beginning to be explored.