G protein-coupled receptors (GPCRs) constitute the major family of cell surface proteins involved in cell signaling cascades and are the target of ≈50% of clinical drugs (Imming et al., 2006). Studies on ligand-GPCR interactions performed over the last decade have revealed diverse capacities of ligand-GPCR-effector complexes to fine-tune their own signals, broadening its apparent simplicity and highlighting ligands as individual chemical species capable of transmitting messages into cellular function with a versatility unpredicted two decades ago (Kenakin, 2007b; Urban et al., 2007). It is well accepted that agonist (full and partial) ligands and allosteric-positive regulators can invoke different active conformations of GPCRs and that these may allow differential agonist-dependent regulation of signaling pathways. Such effects have been described as “agonist-directed trafficking of receptor stimulus” (Kenakin, 1995), “biased agonism” (Jarpe et al., 1998), “functional selectivity” (Urban et al., 2007), or “collateral efficacy” (Kenakin, 2007a). This recently accumulated experimental evidence has led to the development of novel mathematical representations that attempt to explain the chemical biology of GPCRs and integrate the new knowledge by extending accepted traditional models (De Lean et al., 1980; Kent et al., 1980; Lefkowitz et al., 1993; Samama et al., 1993; Leff, 1995; Weiss et al., 1996; Leff et al., 1997; Scaramellini and Leff, 2002).

Although it is generally believed that antagonists (neutral antagonists and inverse agonists) simply inhibit either agonist-induced or constitutive receptor functions, it is conceptually plausible to envision that certain antagonists could also deactivate GPCR responses in a ligand- and pathway-specific manner. A series of observations, including the ability of certain ligands that are conventionally described as “antagonists” to induce receptor internalization (Baker and Hill, 2007; Kenakin, 2007b), to cause activation of extracellular signal-regulated kinase mitogen-activated protein kinase (Azzi et al., 2003; Wisler et al., 2007), or to promote inverse agonist-specific receptor conformational changes (Vilardaga et al., 2005), are consistent with such a concept. This is particularly relevant because most current drugs that target GPCRs are antagonists.

In the last decade, there has been growing evidence to indicate that GPCR dimerization may be a requisite for function and that binding of small ligands to these receptors occurs on dimeric receptor forms (Ayoub et al., 2002; Milligan, 2004; Herrick-Davis et al., 2005), although monomeric forms of GPCRs have also been shown to be capable of activating G proteins (Bayburt et al., 2007; Whorton et al., 2007), suggesting that functionally active monomeric and dimeric forms of the receptors may coexist in equilibrium. Mechanisms by which ligands differentially regulate signaling pathways mediated by a single receptor generally considered the ability of ligands to differentially stabilize distinct receptor conformations (Hunton et al., 2005). The question is to what extent these different conformations occur at a single or paired receptor.

The 5-HT_2A receptor is a class A GPCR whose antagonists have important applications in the treatment of disorders of the cardiovascular and central nervous systems (Berg et al., 2005), as well as in virology (Elphick et al., 2004). 5-HT_2A receptors regulate IP accumulation mediated by phospholipase C and AA release mediated at least partially by phospholipase A₂, and different 5-HT_2A receptor agonists, for instance serotonin and the hallucinogenic 2,5-dimethoxy-4-iodoamphetamine [(±)DOI], show functional selectivity discriminating between these two signaling pathways (Berg et al., 1998). In a previous study (López-Giménez et al., 2001), in native human brain and in cell lines expressing recombinant human 5-HT_2A receptors lacking constitutive activity, we observed Gpp(NH)p-independent shallow, biphasic, competition binding curves for antagonists competing with agonist radioligands. In light of current knowledge (Wreggett and Wells, 1995; Chidiac et al., 1997; Armstrong and Strange, 2001; El-Asmar et al., 2005; Franco et al., 2005, 2006; Urizar et al., 2005), shallow and steep binding curves in studies using competitive ligands are consistent with receptor dimerization (Albizu et al., 2006), although it requires discarding other pharmacological features such as G-protein stoichiometry (Giraldo, 2008). However, 5-HT_2A receptor dimerization has never been directly demonstrated, although it has been described for the 5-HT_2C receptor, which shares high sequence homology with the 5-HT_2A receptor (Herrick-Davis et al., 2004, 2005, 2006). The aim of the present study was to investigate 5-HT_2A receptors dimerization and to gain insight into its functional relevance.

Materials and Methods

Cell Culture. CHO cells stably expressing human 5-HT_2A receptors at ≈200 fmol/mg protein (CHO-FA4 cells) were maintained in Dulbecco's modified Eagle's medium-F12 supplemented with 10% (v/v) fetal calf serum (FCS), 1% l-glutamine, and 300 μg/ml hygromycin. HEK293 cells were maintained in minimum essential medium (MEM) Eagle supplemented with 10% (v/v) FCS, 1 mM MEM sodium pyruvate, 1% (v/v) MEM nonessential amino acid solution, 100 U/ml penicillin, and 0.1 mg/ml streptomycin. Cells were grown at 37°C in a 5% CO₂ humidified atmosphere.

Receptor Binding Studies at Human 5-HT_2A Receptors. These assays were performed in membranes from CHO-FA4 cells following previously described protocols (López-Giménez et al., 2001). Under the experimental conditions established [5 nM [³H](±)DOB as radioligand, 200-250 μg of protein per tube, and defining nonspecific binding with 10 μM mianserin], specific binding was approximately 75% of total binding.

Measurement of IP Accumulation and AA Release. For these experiments, cells were seeded into 12-well tissue culture plates at a density of 4 × 10⁴ cells/cm². Measurement of IP accumulation and AA release were made simultaneously from the same well (Berg et al., 1998, 1999). In brief, cells were labeled with 1 μCi/ml [myo-³H]inositol (20 Ci/mmol) for 24 h and with 0.1 μCi/ml [³H]AA (200 Ci/mmol) for 4 h at 37°C. After the labeling period, cells were washed and then incubated in 1 ml of experimental medium (Hanks' balanced salt solution, 20 mM LiCl, and 20 mM HEPES) containing vehicle (H₂O) or the indicated concentrations of drugs at 37°C for 10 min. At the end of the incubation period, aliquots (200 μl) of media were added directly to scintillation vials for the measurement of ³H content, which corresponds to AA release (Berg et al., 1998, 1999). The remaining medium was discarded, and 1 ml of 10 mM formic acid (4°C) was added to the wells to extract the accumulated [³H]IP from the cells (IP₁, IP₂, and IP₃, collectively referred to as IP). The released [³H]IP were separated by the anion exchange chromatography method of Berridge et al. (1982) and counted in a liquid scintillation counter (Beckman LS-6000 LL; Beckman Coulter, Fullerton, CA).

cDNA Constructs. The FLAG epitope was introduced at the N terminus of the human 5-HT_2A receptor by PCR with a forward primer containing the sequence of the FLAG epitope (amino acid sequence DYKDDDDK). The c-myc epitope was introduced at the N terminus of the human 5-HT_2A receptor by PCR with a forward primer containing the sequence of the c-myc epitope (amino acid sequence EQKLISEEDL). Fusion proteins of c-myc-5-HT_2A receptor with cyan fluorescent protein (CFP) or enhanced yellow fluorescent protein (YFP) (5-HT_2A R^CFP and 5-HT_2A R^YFP, respectively) were constructed by ligation of two PCR products corresponding to the receptor sequence without stop codon and to each fluorescent protein sequence, amplified from their original plasmids (Clontech, Mountain View, CA), introducing a NotI endonuclease restriction site. The ligation products were subcloned into pcDNA3 plasmid (Invitrogen, Carlsbad, CA) and verified by DNA sequencing.

Transient Transfection of HEK293 Cells for Coimmunoprecipitation and FRET Experiments. For coimmunoprecipitation experiments, HEK293 cells seeded on 100-mm dishes were transiently transfected with 10 μg/dish of total DNA following the calcium phosphate method (Cullen, 1987). For FRET photobleaching experiments, HEK293 cells were grown on poly(d-lysine)-treated glass coverslips in 60-mm dishes to approximately 60 to 80% confluence before transient transfection with the different CFP/enhanced YFP fusion proteins using Effectene transfection reagent (Qiagen GmbH, Hilden, Germany), according to the manufacturer's instructions. The total amount of DNA in the different transfections was held constant with plasmid pcDNA3. Both in coimmunoprecipitation and FRET experiments, FCS was substituted by dialyzed FCS in the corresponding growing media for maintenance of the transiently transfected cells until the time of the experiment (36 h after transfection).

Coimmunoprecipitation Studies. Coimmunoprecipitation studies using FLAG- and c-myc-tagged forms of the 5-HT_2A receptor were performed in transiently transfected HEK293 cells. The different coimmunoprecipitation samples corresponded to the following cDNA combinations and transfection conditions: “mock” (10 μg/dish of vector pcDNA3); “FLAG” (10 μg/dish of FLAG-5-HT_2A construct); “myc” (10 μg/dish of c-myc-5-HT_2A construct); and “FLAG + myc” (5 μg of FLAG-5-HT_2A construct + 5 μg/dish of c-myc-5-HT_2A construct/dish). Cells were harvested 36 h after transfection in 10 ml/dish of ice-cold phosphate-buffered saline, and the “mix” sample was prepared at this point by 1:1 mixing of 5 ml of a harvested additional “FLAG”-transfected dish and 5 ml of a harvested additional “myc”-transfected dish. Cells were pelleted by centrifugation at 300g for 10 min, and the pellets were homogenized in 1 ml of 1× RIPA buffer (50 mM HEPES, 150 mM NaCl, 1% Triton X-100, 0.5% sodium deoxycholate, and 0.1% SDS) supplemented with 10 mM NaF, 5 mM EDTA, 10 mM NaH₂PO₄, 5% ethylene glycol, and a protease inhibitor cocktail (Protease Inhibitor Cocktail for general use; Sigma, St. Louis, MO), pH 7.3, and placed on a rotating wheel at 4°C for 1 h. The samples were then centrifuged for 10 min at 14,000g at 4°C, and the supernatants were precleared by incubating them with 50 μl of protein G (Protein G-Sepharose 4B fast flow; Sigma-Aldrich, St. Louis, MO) at 4°C on a rotating wheel for 1 h. After this, the samples were centrifuged at 14,000g at 4°C for 1 min, the cleared supernatant was transferred to a fresh tube, and the protein concentration in the supernatants was determined with a bicinchoninic acid assay protein quantification kit (Uptima, Interchim, France). The protein concentration of individual samples was adjusted to 0.8 mg/ml using 1× RIPA, and 600 μl of each sample was incubated overnight with 40 μl of protein G and 5 μg of anti-FLAG M2 monoclonal antibody (Sigma) at 4°C on a rotating wheel after reserving a 100-μl sample of the supernatants for the assessment of protein expression in the cell lysates. Sixteen hours later, the rotating samples were centrifuged at 14,000g for 1 min at 4°C, and the protein G beads were washed three times with 1 ml of 1× RIPA buffer and resuspended in 40 μl of 2× reducing Laemmli buffer. Reducing Laemmli buffer (6×) was also added to the lysates, and both immunoprecipitated samples and cell lysates were incubated at 37°C for 30 min and resolved by SDS-PAGE in 4 to 15% Tris-HCl polyacrylamide precast gels (Ready Gel precast gels; Bio-Rad, Hercules, CA). After electrophoresis, proteins were transferred to polyvinylidene difluoride membranes (Immun-Blot polyvinylidene difluoride membranes; Bio-Rad), which were blocked for 1 h in TTBS (10 mM Tris-HCl, pH 7.7, 0.9% NaCl, and 0.1% Tween 20) buffer containing 5% dehydrated nonfat milk. Subsequently, membranes were incubated overnight at 4°C with myc-tag rabbit polyclonal antibody (Cell Signaling Technology, Danvers, MA) diluted 1:11,000 in TTBS buffer containing 1% bovine serum albumin and 0.05% sodium azide (immunoprecipitated samples and cell lysates) or with 0.4 μg/ml anti-FLAG M2 monoclonal antibody (Sigma) in TTBS buffer containing 1% bovine serum albumin and 0.05% sodium azide (cell lysates). Goat anti-rabbit (1:5000 in TTBS) or sheep anti-mouse (1:10,000 in TTBS) peroxidase-conjugated secondary antibodies (GE Healthcare, Chalfont St. Giles, Buckinghamshire, UK) were used for detection by enhanced chemiluminescence using ECL Plus Western blotting chemiluminescence detection kit (GE Healthcare) and Hyperfilm ECL films (GE Healthcare).

Microscopic FRET Photobleaching Experiments. FRET between CFP and YFP in HEK293 cells transiently transfected with the different CFP/enhanced YFP constructs was determined in live cells by donor recovery after acceptor photobleaching following previously described protocols (Vilardaga et al., 2003). In brief, HEK293 cells grown on coverslips were maintained in HEPES buffer (137 mM NaCl, 5 mM KCl, 1 mM CaCl₂, 1 mM MgCl₂, and 20 mM HEPES, pH 7.4) at room temperature (22°C) and placed on an Eclipse TE2000-U fluorescence inverted microscope (Nikon) equipped with an oil immersion 100× objective and a dual emission photometric system (TILL Photonics, Gräfelfing, Germany). Samples were excited with a xenon lamp from a polychrome V (TILL Photonics). The emission fluorescence intensities of the fluorescent constructs were determined at 535 ± 15 nm (YFP) and 480 ± 15 nm (CFP) with a beam splitter dichroic long-pass of 505 nm, upon excitation at 436 nm (filter 436 ± 10 nm and a beam splitter dichroic long-pass of 455 nm), resulting the bleed-through of YFP into the 480-nm channel negligible. The emission intensities of CFP were recorded before (CFP_before) and after (CFP_after) 1 min of direct illumination of YFP at 500 nm. FRET efficiency was calculated according to eq. 1:

To ensure that the groups of cells analyzed in the different experiments were similar in terms of their fluorescence characteristics, the levels of YFP expression were determined at the beginning of each experiment as the emission intensity of YFP (recorded at 535 nm) upon direct excitation at 500 nm and the levels of expression of CFP in each cell were determined as the emission intensity of CFP after photobleaching. Fluorescence emission signals detected by avalanche photodiodes were digitalized using an analog-to-digital converter (Digidata1322A; Molecular Devices, Sunnyvale, CA) and stored on personal computer using Clampex 9.0 (Molecular Devices). Data were analyzed using the programs Origin (OriginLab Corp, Northampton, MA) and Prism 4.0 (GraphPad Software Inc., San Diego CA).

Drugs. [³H](±)DOB (23.1 Ci/mmol) was purchased from PerkinElmer Life and Analytical Sciences (Waltham, MA). [myo-³H]Inositol (20 Ci/mmol) and [³H]arachidonic acid (200 Ci/mmol) were supplied by American Radiolabeled Chemical (St. Louis, MO). Ketanserin, mesulergine, clozapine, and all other drugs and chemicals were reagent-grade products from Sigma/RBI (Alcobendas, Spain). MDL100,907 was a generous gift from Dr. M. Galvan (Marion Merrell Dow, Strasbourg, France).

Data Analysis and Mathematical Modeling. Binding and stimulation response data were fitted by the Hill eq. 2 with Prism software: where y is the response variable, x is the log[A]; [A] is the ligand concentration; Top and Bottom are the maximum and minimum responses, respectively; EC₅₀ is the ligand concentration for half-maximum response; and n_H is the Hill coefficient. Discrimination between one site and two noninterconverting sites for antagonist binding competition curves was performed by comparing the fit provided by eqs. 3 and 4: where x is the log[B] and B is the antagonist, IC₅₀ is the concentration of antagonist that inhibits 50% of the specific radioligand binding for a single binding site receptor, and agonist binding in the absence of antagonist is 100%. In the case of two binding sites (eq. 4), IC₅₀₁ and IC₅₀₂ are the IC₅₀ values for sites 1 and 2, respectively, and f and (1-f) the corresponding fractions of receptor sites.

Curve-Fitting by the Three-State Dimer Receptor Model: Binding and Function. The receptor model shown in Fig. 5 was used for curve-fitting. For the binding of an agonist A in the presence of a varying concentration of an antagonist B, eq. 12 leads to eq. 5, where the constants c₁ to c₅ are combinations of the mechanistic constants included in the model. Equation 5 can be rearranged into eq. 6 (percentages) by making y = 100 for a fixed concentration of ligand A. It can be shown that eqs. 4 and 6 are the same by making a = IC₅₀₁IC₅₀₂, b = f · IC₅₀₁ + (1 - f)IC₅₀₂, and c = IC₅₀₁ + IC₅₀₂. Thus, from a statistical point of view, to state that a two-noninterconvertible sites model fits data better than a one-site model is equivalent to saying that a dimer receptor model fits data better than a one-site receptor model.

The equation for the functional response for either [³H]IP accumulation or [³H]AA release pathway is given by eq. 7 (an empirical relationship resulting from the mechanistic eqs. 13 and 14), where c₁ to c₅ are combinations of the equilibrium constants included in the model (Fig. 5). In the same way as in the binding studies (see above), the percentage of functional response of a fixed concentration of an agonist A in the presence of varying concentrations of a ligand B is given by eq. 8, where a value of 100 is assigned to the activity of the agonist A in the absence of B. To assess whether biphasic curves are present, the goodness of fit for eq. 8 was compared with that for the monophasic eq. 9.

Statistical Comparisons between Models: The F Test. Statistical comparisons between fits provided by eqs. 3 and 4 or between eqs. 8 and 9 were performed by the extra sum-of-squares F test (Giraldo et al., 2002). The F statistic, which allows for the comparison between models if they are nested (one model can be formulated as a particular case of the other), is constructed as where SS is the residual sum of squares, df is the degrees of freedom, and the subscripts 1 and 2 correspond to the model with fewer and greater number of parameters, respectively. Statistical significance was set at P < 0.05.

Assessing Affinity Constants for the Antagonists.K_i values were calculated with the Cheng-Prusoff equation (Cheng and Prusoff, 1973). Calculation of pA₂ values for clozapine was performed as described by Arunlakshana and Schild (1959). After incubation of the antagonist, calculation of the apparent antagonist dissociation constant (K_B) was determined by linear regression with eq. 11: where [B] is the concentration of the antagonist used and dr represents the ratio (dose ratio) of concentrations of the agonist (EC₅₀) that produce identical responses (50% E_max) in the presence and absence of the antagonist. A Schild analysis was performed with three different concentrations of the antagonist, and then the antagonistic potency of clozapine was expressed as pA₂ (the value of [B] for log(dr - 1) = 0) and given as mean ± S.E.M. Affinity constants obtained by Cheng-Prusoff and Schild analyses are considered as exploratory rather than as accurate parameter values. This is because these methods were originally developed for the simplest (A + R = AR) ligand-receptor interaction model. Given the increasing complexity of GPCR models, constant estimates by Schild and Cheng-Prusoff methods should be treated with caution. As recently shown (Giraldo et al., 2007), inclusion of inverse agonism affects the intercept of eq. 11 but not the slope, which remains equal to 1. A slope in eq. 11 different from 1 is expected in receptor dimer models if cooperativity occurs.

Results

5-HT_2A Receptors Form Homo-Oligomers in Live Cells. The presence of 5-HT_2A receptor homo-complexes in transfected cell lines was investigated by two sets of experiments. First, coexpression of N-terminally c-myc- or FLAG-tagged 5-HT_2A receptors in HEK293 cells followed by immunoprecipitation of the cell lysates with anti-FLAG antibodies and immunoblotting with anti-c-myc antibodies revealed the coimmunoprecipitation of a major immunoreactive band of approximately 55 kDa, corresponding to a c-myc-tagged protein of molecular weight similar to that described for the 5-HT_2A receptor (Wu et al., 1998) (Fig. 1a, top, “FLAG + myc” line). The anti-c-myc immunoreactivity was not detected when the c-myc- and FLAG-tagged receptors were not expressed in the same cells (Fig. 1a, top, “mix” line), nor when singly expressed FLAG- and myc-receptors were mixed together before immunoprecipitation (Fig. 1a, top, “FLAG” and “myc” lines), ruling out the formation of oligomers or receptor aggregates during the solubilization process and denaturation of the samples before SDS-PAGE. The expression of the differently tagged receptors in these experiments was verified in the cell lysates by immunoblotting with anti-FLAG and anti-c-myc antibodies (Fig. 1a, middle and bottom).

Second, we measured FRET efficiency between 5-HT_2A receptors C-terminally tagged with CFP and YFP (5-HT_2AR^CFP and 5-HT_2AR^YFP, respectively) coexpressed in HEK293 cells by measuring donor recovery after acceptor photobleaching (Fig. 1b). The experiments yielded a FRET efficiency between the two fluorescent 5-HT_2A receptors of 5.97 ± 0.794% (mean ± S.E.M., n = 22) (Fig. 1c). This value was significantly higher (P < 0.001) than the FRET efficiency measured in HEK293 cells showing similar CFP and YFP fluorescent levels but coexpressing a control pair of proteins consisting of 5-HT_2AR^CFP and N-terminally membrane-tagged YFP (YFP^m) for the assessment of nonspecific FRET as a result of random distribution and collision between fluorescent proteins (FRET efficiency = 1.01 ± 0.322%, mean ± S.E.M., n = 15) (Fig. 1c). FRET efficiencies between 5-HT_2AR^YFP coexpressed with other C-terminally CFP-tagged GPCRs, for which interaction with 5-HT_2A receptors is not reported such as the dopamine-1A receptor and the parathyroid hormone receptor type 1 (D_1AR^CFP and PTHR^CFP, respectively) were (mean ± S.E.M.) 1.92 ± 0.407% and 2.11 ± 0.354%, n = 12 and 15, for D_1AR^CFP + 5-HT_2A R^YFP and PTHR^CFP + 5-HT_2AR^YFP, respectively, not significantly different from the nonspecific FRET detected between 5-HT_2AR^CFP and YFP^m among groups of cells displaying similar fluorescence levels (Fig. 1c). Further analysis of the results revealed that the FRET efficiency between 5-HT_2AR^CFP and 5-HT_2AR^YFP increased as a hyperbolic function of the level of acceptor expression and reached a maximal value when most of the donor molecules would be complexed with acceptor molecules (Fig. 1d), a behavior expected for specific protein/protein oligomerization (Mercier et al., 2002; Zacharias et al., 2002). Conversely, FRET efficiency between 5-HT_2AR^CFP and YFP^m increased linearly with the acceptor expression level, as typically expected from random interactions between fluorescent proteins.

Antagonists Differentiate Distinct 5-HT_2A Receptor Signaling Conformations. We compared the capacity of two well known antipsychotic antagonists, clozapine and haloperidol, to compete with the agonist [³H](±)DOB. The atypical antipsychotic clozapine displayed Gpp(NH)p-independent biphasic competition binding curves for (±)DOB-bound 5-HT_2A receptors, whereas the typical antipsychotic haloperidol displayed a monophasic competition binding profile (Fig. 2a). Similar mono- or biphasic profiles also differentiated another series of antagonists, which indicate a selective antagonist-dependent curve shape at the 5-HT_2A receptor (Table 1 and Supplementary Fig. 1a).

We then assessed the signaling properties of the 5-HT_2A receptor by simultaneously measuring the formation of IP and the release of AA after agonist stimulation following a previously described protocol (Berg et al., 1998, 1999). Application of serotonin to CHO-FA4 cells stably expressing the 5-HT_2A receptor stimulated IP formation and AA release in a concentration-dependent manner. Half-maximal activation occurred at a concentration similar to that of serotonin at the two pathways (pEC₅₀ = 6.56 ± 0.37 and 6.60 ± 0.25 for IP formation and AA release, respectively), and with identical Hill coefficients ≈1 (n_H = 0.99 ± 0.06 and 1.08 ± 0.11 for IP formation and AA release, respectively) (Fig. 3a).

Second, we compared the ability of diverse antagonists to block 5-HT_2A receptor signals. Clozapine inhibited serotonin-induced IP accumulation with a monophasic profile and an IC₅₀ value of 89.12 ± 3.10 nM. Conversely, this compound showed an inhibition profile of AA release better described by a biphasic curve (F test, P < 0.01) that yield IC₅₀ values of 2.04 ± 0.51 and 4176 ± 1458 nM for each of the phases and with 56.33% of the sites corresponding to the high-affinity fraction of receptors (Fig. 2b). Increasing concentrations of clozapine induced parallel rightward shifts of the serotonin-stimulated IP response (pA₂ for clozapine-antagonism of 5-HT-induced IP accumulation = 7.98 ± 0.16), and Schild regression analysis of these shifts yielded a slope ≈1 (0.948 ± 0.160) (Fig. 3b). These data correspond well with the predicted property of a competitive antagonist. However, when serotonin-induced AA release was analyzed by Schild analysis, the shifts in the 5-HT concentration-response curves were surmountable, but not parallel, with a slope of 0.34 ± 0.19, indicative of noncompetitive or allosteric antagonism (Fig. 3c). Similar inhibitory effects of clozapine were observed when the 5-HT_2A receptor was activated by another selective agonist, (±)DOI (Supplementary Fig. 2, a and b).

The monophasic versus biphasic inhibitory profiles of clozapine at the serotonin-mediated IP and AA pathways were preserved after inhibition of PLA₂ or PLC, respectively, with selective inhibitors in the cells. Hence, pretreatment of the cells with the PLC inhibitor compound 48/80 (100 μM), which inhibited by 40% the serotonin-induced [³H]IP accumulation, did not alter the biphasic inhibition profile of clozapine at the AA pathway (Fig. 4, a and b, and Supplementary Table 1). In the same way, the monophasic inhibition of IP accumulation by clozapine was not affected by pretreatment of the cells with a PLA₂ inhibitor (10 μM aristolochic acid), which inhibited by 85% the serotonin-induced [³H]AA release (Fig. 4, a and c, and Supplementary Table 1). This eliminates cross-talk between IP and AA pathways as a cause of the clozapine pathway-dependent inhibition profile. Again, similar inhibition patterns for AA release and IP formation were observed for clozapine in the absence or presence of GR55562, a selective antagonist of the 5-HT_1B receptor, discarding a potential contribution of this endogenous receptor subtype to the biphasic antagonist behavior (Fig. 4, a and d, and Supplementary Table 1).

The addition of serotonin to cells pretreated with clozapine for at least 15 min elicited the same profiles of inhibition as when the antagonist was added simultaneously with serotonin, eliminating different binding kinetics of agonists and antagonists as the mechanism responsible for the biphasic curves (Supplementary Fig. 3). These data support that clozapine antagonizes the IP pathway through a competitive mechanism, whereas the mechanism by which clozapine inhibits 5-HT_2A receptor-mediated AA responses would be determined by factors more complex than a simple ligand-receptor interaction.

As in the case of the binding assays, we found that additional antagonists also displayed the ability to inhibit 1 μM 5-HT-activated receptor responses by distinct mechanisms. Like clozapine, ketanserin, risperidone, and MDL100,907 each inhibited 5-HT-stimulated IP accumulation in a monophasic manner but showed biphasic inhibition of AA release (F test, P < 0.05) (Table 1 and Supplementary Fig. 1, b and c). In contrast, haloperidol and mesulergine inhibited 1 μM 5-HT-induced stimulation of both IP accumulation and AA release in a monophasic manner (Fig. 2c, Table 1, and Supplementary Fig. 1, b and c). Again for all of the antagonists, there was a strong correspondence between the negative cooperative binding parameters and the biphasic functional inhibition data obtained at the AA pathway (Table 1). These results therefore provide evidence of a functional antagonist-dependent negative cooperativity in the inhibition of the AA pathway that mirrors the antagonist-dependent negative cooperative binding indicative of 5-HT_2A receptor homodimers.

The Three-State Receptor Dimer Model. Our experimental data are consistent with a dimeric receptor with two active states, one for each biochemical pathway. We propose a model that is based on the existence of the receptor in equilibrium between its inactive dimeric form (R₂) and two distinct active [(R₂)^* and (R₂)^**] receptor states (Fig. 5). The model, which can be considered an extension of both the recently described two-state dimer model (Franco et al., 2005, 2006) (by including an additional active receptor conformation), or the three-state monomer model (Leff et al., 1997) (by assuming that the receptor species are dimers instead of monomers), allows for the differential functional antagonist profiles by assuming a different receptor active conformation for each pathway [e.g., (R₂)^* for IP accumulation and (R₂)^** for AA release].

We aimed to use the model to account for the selectivity of curve shapes both for binding and response curves shown for some antagonists. Ligand-binding and ligand-response equations can be derived from the proposed model by including the proper receptor species.

Thus, the fractional binding of the agonist A in the presence of the antagonist B is defined as

The fractional response of the IP pathway is defined as

The fractional response of the AA pathway is defined as

Inclusion in these expressions of the mechanistic constants of the model (Fig. 5) and algebraic handling of the resulting equations leads to simplified empirical equations appropriate for curve-fitting (see Materials and Methods). This first description of such antagonist behavior in oligomeric receptors establishes a new theoretical model for the homodimeric arrangement of GPCRs, including an inactive (R₂) and two active [(R₂)^* and (R₂)^**] receptor conformations. If we assign the (R₂)^* to the IP accumulation and (R₂)^** to AA release, our data indicate that an apparent negative cooperativity between protomers for some antagonists at the (R₂)^** state, must be present. The model leads to the common pharmacological expressions for binding and function by including the proper receptor species (see Materials and Methods).

Discussion

GPCR oligomerization has been described both in cultured cells and native tissues (Milligan, 2008). Although there are examples of different GPCRs fully functional as monomers or as oligomers (Meyer et al., 2006; Ernst et al., 2007; Whorton et al., 2007; Milligan, 2008), GPCR oligomerization may constitute a mechanism that facilitates receptor transport, regulates G protein coupling (González-Maeso et al., 2008; Vilardaga et al., 2008), and, as recently shown in heterodimeric receptors, permits direct conformational cross-talk between receptors (González-Maeso et al., 2008; Vilardaga et al., 2008). The possibility that homo-oligomers can also adopt multiple active states extends the capacity of ligands to distinctly modulate different signaling pathways driven by the same receptor. Particularly, antagonist ligands constitute the main source of drugs available, and if they may stabilize particular receptor conformations, it should imply therapeutical consequences. Here we present data supporting the existence of 5-HT_2A receptor homodimers in live cells. Our results from coimmunoprecipitation studies are consistent with dimerization/oligomerization of 5-HT_2A receptors by noncovalent interactions not resistant to the SDS-PAGE reducing conditions used. In addition, FRET experiments detected a specific FRET signal between 5-HT_2A R^CFP and 5-HT_2A R^YFP consistent with the presence of 5-HT_2A receptor homodimers in the live cells studied.

5-HT_2A receptors are targets of antipsychotics, and it was recently shown that the antipsychotic clozapine but not haloperidol down-regulates the heterodimer mGlu2R-5-HT_2A receptor through a 5-HT_2A-dependent process (González-Maeso et al., 2008). We report now that these two drugs also differentiate themselves by blocking 5-HT_2A responses by two distinct mechanisms. When we studied the binding of different drugs at these receptors, some antagonists, like the atypical antipsychotic drugs clozapine and risperidone, differentiate from others, like the typical antipsychotic drug haloperidol, because they displayed biphasic binding competition curves for agonist [³H](±)DOB-labeled human 5-HT_2A receptors, whereas haloperidol displayed monophasic competition curves. These differences were also observed for other 5-HT_2A antagonists (ketanserin and MDL100,907 are biphasic inhibitors, whereas mesulergine is a monophasic one).

Although the two enantiomers of [³H](±)DOB may have different affinities for 5-HT_2A receptors, in a previous study (López-Giménez et al., 2001), we reported saturation studies in which in the presence of Gpp(NH)p, [³H](±)DOB bound to a unique population of receptors in the same cell line as used in the present study. We assumed that if the two enantiomers had different affinities for 5-HT_2A receptors in our system, this would have been reflected by two sites in the latter study. Furthermore, a behavior similar to that of [³H](±)DOB in the cell line was observed by autoradiography in brain sections at human native 5-HT_2A receptors labeled with [¹²⁵I](±)DOI, which shows no differences in the affinities of its enantiomers for 5-HT_2A receptors (Knight et al., 2004).

Biphasic competition-binding curves can be described by three mechanistic explanations: the presence of different monomeric receptor species, the formation of a receptor-G protein complex in a limited G protein concentration, or the occurrence of receptor dimerization with negative cooperativity between the binding sites of the protomers. In our system, the possible presence of different monomeric receptors had previously been discarded by detailed pharmacological analysis (López-Giménez et al., 2001).

In addition, the absence of Gpp(NH)p effect on the biphasic profile of the binding curves observed for some antagonists indicated that the two phases did not reflect G protein-dependent high- and low-affinity states of the receptor and eliminates the Ternary Complex Model (De Lean A. et al., 1980) and the Extended Ternary Complex Model (Lefkowitz et al., 1993; Samama et al., 1993) as frameworks to interpret the data. Allosteric models (Bosier and Hermans, 2007; Kenakin, 2007b) were also considered inappropriate because complete displacement of radioligands from their specific binding sites was observed for all of the antagonists tested. In this context, negative cooperativity within receptor dimerization seems to be a necessary condition to account for biphasic curves. Indeed, the occurrence of biphasic competition binding curves is currently accepted as indicative of GPCR dimerization, in which binding of a single molecule of agonist to a GPCR dimer will produce negative cooperative effects on the propensity of a second molecule to bind (Chinault et al., 2004; Albizu et al., 2006; Sartania et al., 2007).

A captivating challenge in GPCR homodimerization is to understand its functional consequences in pharmacology, in what extent homodimerization between identical receptors represent the possibility of new pharmacology different from the individual parent identical receptors? To gain insight into the functional consequences of the observed antagonist-dependent selective cooperativity at homodimeric 5-HT_2A receptors, we first examined the functional behavior on second-messenger signaling pathways of both the atypical antipsychotic clozapine and the typical antipsychotic haloperidol, which showed negative cooperative and noncooperative binding, respectively.

We established an experimental design mimicking that of the binding competition studies and generated concentration-response curves for both ligands at two noninterlinked 5-HT_2A-mediated signaling pathways, IP accumulation and AA release. Although haloperidol showed a monophasic inhibition of the 5-HT-dependent activation of both effector pathways, clozapine inhibited the 5-HT-stimulated IP accumulation in a monophasic manner, but intriguingly, its inhibition of the 5-HT-stimulated AA release was biphasic with a strong parallelism with the negative cooperative binding observed for this ligand in radioligand binding assays, even in the fraction of high and low populations and again better described by a dimer receptor model than by a monomer receptor model. We discarded that this behavior could be due to endogenously expressed 5-HT_1B receptors or to different kinetics between 5-HT and clozapine in interacting with the receptors. Classic Schild analysis of the concentration-response curves of 5-HT in the presence of clozapine at both effector pathways showed a surmountable and parallel shift of 5-HT concentration-response curve at the IP pathway and a surmountable but noncompetitive antagonism at the AA pathway, indicating that clozapine antagonizes the IP pathway without apparent cooperativity, whereas it shows apparent negative cooperativity at the AA pathway.

We found that all of the other antagonists included in our study showing negative cooperativity in binding assays (ketanserin, risperidone, and MDL100,907) antagonized 5-HT-induced AA release in a biphasic manner differentiating themselves from the monophasic inhibition showed by haloperidol and mesulergine at the same pathway. There was again a strong quantitative correspondence between the negative cooperative binding parameters and the biphasic functional inhibition data obtained at the AA pathway for these ligands. All of the antagonists displayed monophasic inhibition profiles at the IP pathway.

These results therefore provide evidence of a functional antagonist-dependent negative cooperativity in the inhibition of the AA pathway that mirrors the antagonist-dependent negative cooperative binding indicative of 5-HT_2A receptor homodimers. To our knowledge, this is the first evidence that antagonists are capable of differently inhibiting signaling pathways at GPCRs.

Although our data are also compatible with a dynamic equilibrium between monomeric and dimeric 5-HT_2A receptors showing different functionality (i.e., monomeric receptors signaling through the IP pathway and dimeric receptors signaling through the AA pathway showing the latter ligand-dependent cooperativity), a dimeric form would always be a necessary component of the system, because at least a dimeric form showing cooperativity between the binding sites of the protomers is required for explaining the biphasic antagonism at the AA effector pathway. As described recently (Rovira et al., 2009), the inclusion of monomer species would increase the complexity of the model without providing extra information. Thus, monomer receptor species were not included in the model, and only interconvertible dimeric receptor conformations were considered. To mathematically fit these interconvertible conformations, we propose here a model based on the existence of an equilibrium between the receptor in its inactive dimeric form (R₂) and two distinct active [(R₂)^* and (R₂)^**] receptor states. This model allows for the differential functional antagonist profiles experimentally observed by assuming a different receptor active conformation for each pathway [e.g., (R₂)^* for IP accumulation and (R₂)^** for AA release].

We aimed to use this model to account for the specific binding and response curve shapes shown for the different antagonists. Ligand-binding and ligand-response equations can be derived from the proposed model by including the proper receptor species.

A computer simulation using eqs. 12 to 14 from the model depicted in Fig. 5 resulted in profiles that closely match the singular properties found for clozapine in our experiments: namely, monophasic for inhibitory IP response, and biphasic for both binding and inhibitory AA responses (Fig. 6). In brief, the simulation, for a varying concentration of an antagonist B in the presence of a fixed concentration of an agonist A, assumed the following: 1) the agonist displayed null cooperativity at the inactive conformation, (R₂), and positive cooperativity for both the IP and AA pathways; 2) the antagonist displayed null cooperativity for both the inactive conformation and the IP pathway and negative cooperativity for the AA pathway; and 3) the affinity of the antagonist for the second site of the receptor when the first site is occupied by the agonist is greater for the AA pathway conformation than for the IP and inactive conformations. This simulation illustrates the complexity and versatility of a system composed of a dimer receptor with multiple active conformations, in which subtle changes in the ligand-receptor set of interactions may profoundly affect the curve profile.

In summary, both from direct biophysical and indirect binding and functional studies, a dimer arrangement for the 5-HT_2A receptor was proposed. The pathway-dependent pharmacological profile shown by some antagonists was explained by a three-state dimer receptor model, including one inactive and two active receptor conformations. The model accounted for the functional selectivity displayed by the antagonist ligands by the cross-talk between protomers through the dimer interface. This specific dimer-effector antagonism opens up a new avenue for understanding and reinterpreting the functional meaning of the cooperative binding of many antagonists of GPCRs in general and of 5-HT_2A receptors in particular.

Appendix

Equations governing the three-state dimer receptor model (Figure 5): L stands for a ligand in general. We use the symbol A for the agonist and B for the antagonist. Thus the above model represents the binding of either A or B to the receptor. In addition to “pure” agonist- or antagonist-receptor complexes, we must consider the presence of the mixed species AB(R₂), AB(R₂)^*, and AB(R₂)^**. This can be done by using the following chemical equilibria: