Glucagon-like peptide-1 receptor dimerization differentially regulates agonist signaling but does not affect small molecule allostery

Abstract

The glucagon-like peptide-1 receptor (GLP-1R) is a family B G protein-coupled receptor and an important drug target for the treatment of type II diabetes, with activation of pancreatic GLP-1Rs eliciting glucose-dependent insulin secretion. Currently, approved therapeutics acting at this receptor are peptide based, and there is substantial interest in small molecule modulators for the GLP-1R. Using a variety of resonance energy transfer techniques, we demonstrate that the GLP-1R forms homodimers and that transmembrane helix 4 (TM4) provides the primary dimerization interface. We show that disruption of dimerization using a TM4 peptide, a minigene construct encoding TM4, or by mutation of TM4, eliminates G protein-dependent high-affinity binding to GLP-1(7-36)NH₂ but has selective effects on receptor signaling. There was <10-fold decrease in potency in cAMP accumulation or ERK1/2 phosphorylation assays but marked loss of intracellular calcium mobilization by peptide agonists. In contrast, there was near-complete abrogation of the cAMP response to an allosteric agonist, compound 2, but preservation of ERK phosphorylation. Collectively, this indicates that GLP-1R dimerization is important for control of signal bias. Furthermore, we reveal that two small molecule ligands are unaltered in their ability to allosterically modulate signaling from peptide ligands, demonstrating that these modulators act in cis within a single receptor protomer, and this has important implications for small molecule drug design.

G protein-coupled receptors (GPCRs) are the largest superfamily of cell surface proteins and play crucial roles in virtually every physiological process. Their widespread abundance, yet selective distribution, and ability to couple to a variety of signaling and effector systems make them extremely attractive targets for drug development (1). Recently, there has been increasing interest in the stoichiometry of receptors involved in GPCR signaling complexes and how this may impact on receptor function and drug discovery (2–4).

With the exception of family C GPCRs, where obligate dimerization can occur (4), the role of oligomerization in GPCR function has remained controversial (2–8), and this has been the subject of a number of recent reviews (9–11). Although there is an increasing body of evidence supporting dimerization of GPCRs as a widespread feature of GPCR biology, including numerous studies on family A GPCRs, whether these are stable, transient, constitutive, or ligand dependent, and how they impact on receptor function and drug discovery are less clear, and general rules for oligomeric behavior are not evident (9–16). Even where effects on signaling are studied, these are generally linked to a single pathway and the role of dimerization in the control of receptor engagement and preference for distinct intracellular signaling intermediates (i.e., signal bias) is virtually unstudied.

For family B GPCRs, which encompass many therapeutically important peptide receptors, including those for glucagon, glucagon-like peptides 1 and 2 (GLP-1, GLP-2), parathyroid hormone, and calcitonin, there is consistent evidence for homodimerization (17–25). There is also emerging evidence for functionally significant heterodimerization (25–27). Furthermore, although there is an emerging theme in which dimerization contributes to high-affinity peptide binding and cAMP signaling (17, 18), how dimerization contributes more globally to receptor signaling and whether it plays a role in ligand-directed stimulus bias are unknown.

There is parallel interest in the development of allosteric drugs targeting otherwise intractable GPCRs to enable enhanced selectivity, fine control of receptor function, and/or maintenance of spatial and temporal elements of endogenous (orthosteric) signaling (1, 28). However, little is known on whether such drugs act through cooccupancy of a single receptor protomer (in cis), or asymmetrically across dimers (in trans), with most drug development assumptive on an in cis mechanism of action. Allosteric drugs are likely to be required for targeting of family B GPCRs, given the diffuse orthosteric pharmacophore of their peptide agonists (29) and the difficulty in mimicking this with small molecule compounds. Understanding how such molecules bind and modulate receptor function is crucial for successful optimization of such drugs.

In this study, we demonstrate functionally important homodimerization of the GLP-1 receptor (GLP-1R) that occurs through an interface along transmembrane segment 4 (TM4) and that this dimerization is critical for selective coupling of the receptor to physiologically relevant signaling pathways. Furthermore, dimerization is important for the signal bias of the receptor and discriminates between peptide and nonpeptide-mediated receptor activation, but is not required for small molecule allosteric modulation of the receptor, indicating that rational design of allosteric therapeutics is possible within a single receptor protomer model.

Results

GLP-1 Receptor Forms Functionally Important Homodimers.

A number of complementary biochemical, biophysical, and pharmacological approaches were used to demonstrate the oligomeric nature of GLP-1R interactions and the impact on ligand–receptor interactions. As previously observed with other family B GPCRs (17, 18, 25), analysis of receptor–receptor interaction in live cells using a combination of static and saturation bioluminescence resonance energy transfer (BRET) analyses revealed constitutive, specific homooligomerization of the GLP-1R (Fig. 1 A and B). No specific BRET signal was seen with coexpression of the BRET donor GLP-1R-Rluc and the family A cholecystokinin 2 GPCR (CCK2R), or between GLP-1R-YFP and soluble Rluc protein, although a strong BRET signal was observed for the CCK2R homodimer (Fig. S1 A and B), demonstrating specificity of the GLP-1R homooligomeric interaction. This interaction was also not significantly altered by saturating concentrations of either peptide [GLP-1(7-36)NH₂, exendin-4, oxyntomodulin] or nonpeptide (6,7-dichloro2-methylsulfonyl-3-tert-butylaminoquinoxaline, compound 2) (30) agonists of the receptor (Fig. S1C). We also used a combination of bimolecular-fluorescence complementation and BRET to probe whether evidence for higher order oligomers of the GLP-1R could be established. However, coexpression of GLP-1R-Rluc with both the GLP-1R-YN and GLP-1R-YC constructs failed to yield a significant BRET signal (Fig. S1E), despite the generation of functional YFP from the fluorescence complementation established at the level of the dimeric receptor complex (Fig. S1D). All GLP-1R fusion constructs had equivalent binding and cAMP signaling to the wild-type receptor (Fig. S2, Table S1). These data indicate that the GLP-1R dimer is the major oligomeric form of the receptor. Negative cooperativity of peptide binding to the secretin family B GPCR has been described (31), and this was also observed in dissociation kinetic studies of the GLP-1R (Fig. S3A), confirming the oligomeric nature of the GLP-1R.

Fig. 1.

Dimerization of the GLP-1R is important for high-affinity agonist binding and receptor coupling to cAMP production. Saturation (A) and static (B) BRET data from transient coexpression, in Cos-1 cells, of Rluc- and YFP-tagged GLP-1R constructs of wild-type, in the presence or absence of TM4 peptides from the GLP-1R or from the secretin receptor, or dimer disrupting mutants of TM4. In the presence of the GLP-1R TM4 peptide or when TM4 double/triple mutants were present, the static BRET ratio was reduced and was not significantly different from background signal. (C) GLP-1(7-36)NH₂ stimulated cAMP responses in CHO cells stably expressing the GLP-1R in the presence or absence of the GLP-1 receptor TM4 segment peptide. (D) GLP-1(7-36)NH₂ stimulated cAMP responses in Cos-1 cells transiently expressing either the wild-type or L256A, V259A, or G252A, L256A, V259A TM4 mutant GLP-1Rs. The cell surface expression of the G252A, L256A, V259A mutant was 112 ± 10% of the wild-type receptor. (E) Inhibition of ¹²⁵I-GLP-1(7-36)NH₂ binding, to membranes from Cos-1 cells transiently transfected with the wild-type GLP-1R, by unlabeled GLP-1(7-36)NH₂ in the presence or absence of 10 µM GppNHp. (F) Inhibition of ¹²⁵I-GLP-1(7-36)NH₂ by unlabeled GLP-1(7-36)NH₂ in the presence or absence of 10 µM GppNHp, to membranes from Cos-1 cells transiently transfected with the G252A, L256A, V259A mutant GLP-1R. For reference, the curve for the WT receptor is shown with a dashed line. Values are means ± SEM of data from four to five independent experiments performed in duplicate.

As seen with the related calcitonin and secretin family B GPCRs (17, 18), disruption of the TM4 interface by either GLP-1R TM4 peptides or by mutation of the hydrophobic face of TM4 (L256A, V259A, or G252A, L256A, V259A; Fig. S4) abolished the BRET signal (Fig. 1 A and B), consistent with TM4 constituting the principal dimer interface for GLP-1Rs as well as other family B GPCRs (17, 18, 31). Disruption of GLP-1R dimerization either by coincubation of GLP-1R BRET constructs with GLP-1R TM4 (Fig. 1C) or mutation of TM4 (Fig. 1D), led to a ∼10-fold decrease in GLP-1(7-36)NH₂ potency to form cAMP, indicating that dimerization is important for efficient coupling of the receptor to the Gs/AC/cAMP signaling cascade. The decrease in potency at the mutant receptors was not due to changes in cell surface expression because whole-cell binding measurements revealed that expression of these mutant receptors was not significantly different from the wild-type receptor. Analysis of GLP-1(7-36)NH₂ peptide inhibition binding, in GLP-1R–expressing Cos-1 membranes, revealed a complex pattern of binding that was best fit with a three-site mass action model (Fig. 1E, Table 1). Disruption of dimerization via the TM4 mutants led to complete loss of the very-high-affinity state (K₁) and marked reduction in the proportion of high-affinity (K₂) sites (Table 1, Fig. 1F, Fig. S5A), and a similar effect is observed with coincubation of the wild-type receptor with the TM4 peptide (Fig. S5B). This effect was mimicked by incubation of membranes with the nonhydrolyzable GTP analog, GppNHp (Fig. 1E, Table 1), suggesting that the principal effect of dimerization is to maintain a high-affinity G protein-complexed state. Coincubation of GLP-1R TM mutant membranes with GppNHp had minimal additional effect on agonist peptide binding (Table 1, Fig. 1F), whereas the negative cooperativity observed in dissociation kinetic studies (Fig. S3A) was abolished in the G252A, L256A, V259A GLP-1R mutant (Fig. S3B).

Table 1.

Disruption of GLP-1R dimerization attenuates high-affinity GLP-1 binding

Receptor construct	K1	K2	K3
Binding affinity (pIC50)*	10.29 ± 0.60	8.40 ± 0.13	6.99 ± 0.12
Fraction
−GppNHp
GLP-1R	0.14 ± 0.06	0.76 ± 0.08	0.10
GLP-1R(L256A, V259A)	0.00 ± 0.04	0.46 ± 0.08	0.54
GLP-1R(G252A, L256A, V259A)	0.00 ± 0.03	0.31 ± 0.08	0.69
+GppNHp
GLP-1R	0.00 ± 0.05	0.38 ± 0.09	0.62
GLP-1R(G252A, L256A, V259A)	0.00 ± 0.06	0.14 ± 0.10	0.86

Binding data were analyzed with a three-site logistic fit and presented as –logM estimates.

*Binding affinity parameters were shared across the data set.

Dimerization Plays Distinct Roles in GLP-1R Peptide and Small Molecule Agonist Signaling.

To date, investigation of the role of homodimerization on family B receptor signaling has been confined to examination of cAMP formation (17, 18). Family B receptors, however, are pleiotropically coupled, with multiple signaling pathways important for receptor function (32). In addition to cAMP formation, both phosphorylation of ERK and mobilization of intracellular calcium are physiologically important in GLP-1–mediated control of pancreatic β-cell function (33). We have used the dimer-disrupting triple mutant (GLP-1R G252A, L256A, V259A) to probe the extent to which dimerization contributes to GLP-1R signaling across these three pathways. A similar (<10-fold) decrease in GLP-1(7-36)NH₂ potency was observed in cAMP accumulation and ERK phosphorylation assays (Fig. 2 A and B), with no change in the time course for pERK (Fig. S6). However, there was almost complete abrogation of the intracellular calcium response (Fig. 2C), demonstrating distinct consequences of dimerization across pathways. Oxyntomodulin is an endogenous agonist of the GLP-1R that exhibits signal bias relative to that of GLP-1 peptides for at least cAMP versus pERK (34), whereas exendin-4 is clinically used as a GLP-1 mimetic (35). Disruption of dimerization had similar effect on the responses to all three peptides (Fig. 2 D and E, Fig. S7), with loss of calcium signaling maintained even at 10⁻⁵ M. In contrast, there was marked loss of cAMP signaling by the small molecule agonist, compound 2, but only a relatively small reduction in pERK signaling (Fig. 2 G and H), the latter likely because of decreased affinity of that compound (Fig. 3). Equivalent effects were seen in CHO-FlpIn cells transfected with the mutant receptors (Fig. S8), demonstrating that the effects were independent of cellular background. To further confirm that the differential effects on signaling were due to disruption of dimerization, we have also used a minigene encoding TMs 3 and 4 of the GLP-1R, with TM3 required to enable correct orientation of the TM4 segment. Transfection of this construct into CHOflpIn cells stably expressing the GLP-1R yielded qualitatively similar results to the mutant receptor (Fig. S9), although the magnitude of effect for all three pathways was dependent on transfection efficiency. Collectively, these data indicate distinct modes of receptor activation for peptide versus small molecule agonists of the GLP-1R. The differential effects on signaling pathways is unlikely to be due to simply changing coupling efficiency across the board, because the relative potency of ligands for pERK and calcium signaling are similar at the wild-type receptor, with dramatically larger effects of disrupting dimerization on calcium mobilization.

Fig. 2.

Disruption of GLP-1R dimerization differentially modifies signal bias and orthosteric versus allosteric agonist function. Concentration–response curves were generated for GLP-1(7-36)NH₂ (A–C), oxyntomodulin (D–F), and compound 2 (G and H) in three different functional assays, cAMP accumulation (A, D, and G), ERK1/2 phosphorylation (B, E, H), and calcium mobilization (C and F) using cells transiently transfected with either wild-type GLP-1R or G252A, L256A, V259A GLP-1R. Cell surface expression of the mutant was not significantly different from WT (112 ± 10%). cAMP and pERK data are from transiently transfected Cos-1 cells and calcium mobilization data from transfected CHOflpIn cells due to the limited calcium response in Cos-1 cells. Comparative data for all pathways and all peptide ligands in CHOflpIn cells are shown in Fig. S8. Data are normalized to maximal peptide response observed at the wild-type receptor. Data points represent the mean ± SEM of four individual experiments performed in duplicate.

Fig. 3.

Allosteric modulation of the GLP-1R occurs within a single receptor protomer. Concentration–response curves were generated for GLP-1(7-36)NH₂ (A, C, and E) and oxyntomodulin (B, D, and F) in the presence and absence of increasing concentrations of compound 2 (A–D) or BETP (E and F) in cAMP accumulation assays using Cos-1 cells expressing wild-type GLP-1R (Left) or G252A, L256A, V259A GLP-1R (Right). Data are normalized to maximal peptide response, are fitted with an operational model of allosterism, and are representative of the mean ± SEM of four independent experiments performed in duplicate.

Allosteric Modulation of the GLP-1R Occurs Within a Single Receptor Protomer.

Currently, very little is known with respect to how allosteric modulators exert their cooperativity, including whether this requires dimerization of receptors or is due to coincident binding to a single monomeric unit. As demonstrated previously (34), compound 2 displays weak positive cooperativity with GLP-1(7-36)NH₂ and strong positive cooperativity with oxyntomodulin (Fig. 3 A and C, Fig. S10, and Table 2). This cooperativity was not significantly altered by the dimer-disrupting mutant, despite the loss of agonism seen with compound 2 (Fig. 3 B and D, Fig. S10, and Table 2). An equivalent preservation of allosteric cooperativity was seen with the structurally distinct modulator, 4-(3-benzyloxy-phenyl)-2-ethylsulfinyl-6(trifluoromethyl)pyrimidine (BETP) (Fig. 3 E and F). To confirm that this cooperativity was not probe dependent, we also examined the allosteric modulation of the GLP-1 metabolite, GLP-1(9-36)NH₂ (Fig. S11) (36). These data demonstrate that the allosteric modulation of the GLP-1R occurs within a single monomeric unit and does not require dimerization of the receptor.

Table 2.

Allosteric parameters for WT and G252A, L256A, V259A GLP-1Rs

	GLP-1R wild type		GLP-1R(G252A, L256A, V259A)
Interacting ligands	pKB	Logαβ (αβ)	pKB	Logαβ (αβ)
Compound 2	5.56 ± 0.13		4.71 ± 0.21
GLP-1(7-36)NH2		0.59 ± 0.11 (3.9)		0.13 ± 0.15 (1.3)
Oxyntomodulin		1.46 ± 0.15 (29)		1.74 ± 0.12 (55)
BETP	5.20 ± 0.20		4.78 ± 0.15
Oxyntomodulin		1.14 ± 0.14 (14)		1.62 ± 0.11 (42)

Quantitative parameters for the allosteric interaction of compound 2 or BETP with GLP-1(7-36)NH₂ or oxyntomodulin. cAMP data were analyzed with an operational model of allosterism as defined in Methods. pK_B values represent the negative logarithm of the affinity for the allosteric ligands derived from application of the operational model of allosterism. Logαβ values represent the composite cooperativity between the allosteric modulator and the orthosteric ligand. Antilogarithms are shown in parentheses.

Discussion

The occurrence and significance of dimerization remains controversial outside of family C GPCRs, where there is evidence for obligate oligomerization. For family B receptors, there is consistent evidence for homooligomerization, together with the potential for functionally important heterooligomerization (18–23, 25–27). Within this theme, there is an emerging paradigm of behavior in which homodimerization of receptors occurs via a TM4/TM4 interface and that this interaction is required for optimal function of the receptor, generation of high-affinity agonist binding, and signaling via formation of cAMP (17, 18, 31); this paradigm is true also for the GLP-1R.

A major finding of the current study is that disruption to the dimer interface had differential impact on signaling engaged by the GLP-1R, with attenuation of cAMP formation and phosphorylation of ERK, but almost complete abrogation of the intracellular calcium mobilization response. Although differential effects on the signal pathways could be due to different strength of stimulus coupling, these results cannot be interpreted as simply a consequence of strength of stimulus coupling as pERK1/2 (in addition to intracellular calcium mobilization) is also less well coupled than cAMP and this pathway is only minimally (less so than cAMP) affected by disruption of dimerization. This suggests that fine control of GLP-1R signaling, and the ability to engage with its full range of signaling intermediates is linked to the ability of the receptor to form dimers. As receptor-mediated intracellular calcium mobilization is less well coupled than Gs-mediated cAMP formation, it may require both receptor dimerization and occupancy of both receptor protomers within the dimer to elicit the response, providing a mechanism through which receptors and ligands achieve the conformational complexes associated with signal bias of the receptor. However, this is not the case for coupling to ERK1/2 phosphorylation. The GLP-1R has been shown to signal downstream of coupling to Gs, Gi/o, and Gq proteins as well as β-arrestins (33, 37), with at least part of the intracellular calcium signal being Gi dependent (38) and at least part of the pERK response dependent on β-arrestin (37). It is likely that the disruption to dimerization seen in the current study differentially alters the coupling efficiency of the receptor to these different effectors. Furthermore, it provides a potential mechanistic basis through which changes to G protein-coupling preferences occur at higher concentrations of agonist.

A second major finding of the current study was the distinct contribution of dimerization to orthosteric versus allosteric activation of cAMP signaling through the GLP-1R, with effective abrogation of cAMP response (>30-fold) to compound 2, whereas there was only limited (<8-fold) loss of peptide-mediated cAMP formation. This implies that compound 2 has limited capacity to activate Gs protein via monomeric GLP-1R and that the allosteric agonist has a distinct mode of receptor activation to the peptide agonists. This further supports a role for dimerization in controlling signal bias (rather than a phenomenon related to strength of coupling) as disruption of dimerization has a stronger effect on cAMP formation than on phosphorylation of ERK, despite greater relative efficacy for compound 2 in formation of cAMP. A differential mode of receptor activation by compound 2 is consistent with findings of distinct effects of receptor polymorphism (T149M) or mutation (ECL2 mutants) on peptide versus allosteric agonist activation of the receptor, where the allosteric agonism is preserved at receptors with impaired peptide response (38, 39). Furthermore, it demonstrates that dimerization can contribute to ligand-directed signal bias.

A third significant finding of the current study was that allosteric modulation of orthosteric peptide binding and signaling was maintained after disruption of the dimer, despite loss of allosteric agonism, demonstrating that the allosteric modulation was occurring in cis within a single protomer. The use of two chemically distinct modulators, compound 2 from NovoNordisk (30) and BETP from Eli Lilly (40), confirmed that this is likely to be a generalized feature of allosteric modulation of the GLP-1R. Understanding how allosteric drugs alter the function of orthosteric ligands is important for drug development, particularly for rational design of ligands in which exclusion from the orthosteric binding site of a single receptor protomer is generally assumed.

The base paradigm illustrated in the current study, that of critical importance of dimerization for efficient coupling to Gs/AC/cAMP, maintenance of very high-affinity agonist peptide binding (presumably linked to occupancy of a single protomer in the dimer), and negative cooperativity between agonist-occupied receptors is similar to that described for the dopamine D₂ receptor, in which optimal physiological signaling is predicted to occur through activation of a single protomer within the dimeric complex (7). Interestingly, although multiple domains (TM4/5 and TM1) of the D₂ receptor have been implicated in oligomerization (41–43), there is an activation-related conformational change that occurs via the TM4 dimer interface (42). With the current study, a functionally important TM4 dimeric interface has now been demonstrated in the three family B GPCRs that have been studied to date (17, 18), and may imply a more general role for allosteric conformational transition across receptor protomers for this TM domain. In this vein, it is interesting to note that heterodimerization of GLP-1 and GIP receptors led to altered GLP-1–induced arrestin recruitment and intracellular calcium mobilization (27), and although the dimer interface for family B GPCR heterodimers has yet to be empirically established, it is possible that allosteric regulation of receptors across TM4 is relevant for signaling even for heteromeric complexes.

It is unclear whether the TM4 peptide or the TM4 mutation fully abrogates dimerization or changes the interaction from stable to transient, such that a BRET signal is not observed. Nonetheless, it is clear that stable dimerization is required for critical elements of GLP-1R function. The content of monomeric versus dimeric receptors in the membrane has been difficult to study, and this is not known for any family B GPCR. Recent work with family A receptors suggests that many GPCRs exist in a dynamic equilibrium between monomeric and dimeric states, with the stability of dimeric complexes differing for individual receptor subtypes (12–15). Nonetheless, even for highly transient interactions, such as those seen for FPR receptors, the mean residence time for interaction (∼90 ms) (15) would allow for coupling to and activation of G proteins (40-70 ms) (44, 45), consistent with dimerization playing an important role in the efficiency of G-protein coupling and, indeed, in differentiating strength of stimulus and pleiotropic receptor coupling.

In conclusion, our work provides mechanistic insight into the role of the oligomeric status of the GLP-1R for biased agonism and allosteric modulation of the receptor, and this has important implications for drug discovery and development at the therapeutically attractive B family of GPCRs.

Methods

GLP-1R Constructs.

Human GLP-1R constructs tagged with Renilla luciferase (Rlu) or yellow fluorescent protein (YFP) inserted in frame at the carboxyl terminus of the mature protein were prepared, as described previously (26). GLP-1R alanine-replacement mutants were generated using the QuikChange site-directed mutagenesis kit (Stratagene), replacing residues 252, 256, and 259 within the predicted lipid-exposed face of TM4 separately and in combination. Additional GLP-1R constructs were prepared by inserting amino-terminal or carboxyl-terminal portions of YFP, representing residues 1–158 (YN) or residues 159–238 (YC) just before the TGA stop codon. The sequences of all receptor constructs were verified by direct DNA sequencing. All constructs had equivalent binding affinity and potency in cAMP accumulation assays (Table S1, Fig. S2).

Transfections and Cell Culture.

Cos-1, CHO, or FlpInCHO cells used for transient transfections were maintained in DMEM supplemented with 5% (vol/vol) heat-inactivated FBS and incubated in a humidified environment at 37 °C in 5% (vol/vol) CO₂. GLP-1R receptor constructs were transiently transfected using either the DEAE-dextran method as previously described or metafectene (Invitrogen) following the manufacturer’s protocol. CHO cells stably expressing the human GLP-1 receptor (CHO-GLP-1R) were propagated in Ham’s F-12 medium supplemented with 5% (vol/vol) FBS (46). For all whole-cell assays, cells were seeded into 96-well culture plates 24 h after transfection at a density of 15,000–20,000 cells/well for Cos-1 cells or 30,000 cells/well for CHOFlpIn cells and incubated overnight at 37 °C in 5% (vol/vol) CO₂ before assaying. The level of cell surface-expressed mutant and wild-type receptors was not significantly different. For robustness, experiments were performed in multiple cellular backgrounds, with a subset of control experiments performed in both the American and Australian laboratories. All signaling assays performed in Cos-1 cells transiently expressing the wild-type or mutant receptors were repeated in transiently transfected CHOFlpIn cells (Fig. S8).

BRET Studies.

BRET studies were performed on receptor-expressing Cos-1 cells as previously described (18).

Bimolecular Fluorescence Complementation.

Bimolecular fluorescence complementation assays were carried out in HEK293 cells expressing YN- and YC-tagged receptor constructs, as described previously (19).

cAMP Assays.

cAMP accumulation assays were carried out using the AlphaScreen kit or the LANCE assay as previously described (18, 34).

Radioligand Binding Assays.

Receptor binding assays were carried out either using intact Cos-1 cells expressing the tagged receptor or isolated receptor-bearing membranes, as described previously (25, 34).

ERK1/2 Phosphorylation Assay.

Receptor-mediated ERK1/2 phosphorylation was determined by using the AlphaScreen ERK1/2 SureFire protocol as described previously (34).

Intracellular Ca²⁺ Mobilization Assay.

Intracellular Ca²⁺ mobilization was determined as described previously (34).

Data Analysis.

All data obtained were analyzed in GraphPad Prism 5.0.3 (GraphPad Software). Radioligand inhibition binding data were fitted to a three-site inhibition mass action curve. In whole-cell ligand interaction studies, data were fitted to an operational model of allosterism and agonism to derive functional estimates of modulator affinity and cooperativity. For all other data, concentration–response curves were fitted with a three-parameter logistic equation.

For more detailed methods, see SI Methods.