Quaternary structure of a G-protein-coupled receptor heterotetramer in complex with Gi and Gs

G-protein-coupled receptors (GPCRs), in the form of monomers or homodimers that bind heterotrimeric G proteins, are fundamental in the transfer of extracellular stimuli to intracellular signaling pathways. Different GPCRs may also interact to form heteromers that are novel signaling units. Despite the exponential growth in the number of solved GPCR crystal structures, the structural properties of heteromers remain unknown. We used single-particle tracking experiments in cells expressing functional adenosine A1-A2A receptors fused to fluorescent proteins to show the loss of Brownian movement of the A1 receptor in the presence of the A2A receptor, and a preponderance of cell surface 2:2 receptor heteromers (dimer of dimers). Using computer modeling, aided by bioluminescence resonance energy transfer assays to monitor receptor homomerization and heteromerization and G-protein coupling, we predict the interacting interfaces and propose a quaternary structure of the GPCR tetramer in complex with two G proteins. The combination of results points to a molecular architecture formed by a rhombus-shaped heterotetramer, which is bound to two different interacting heterotrimeric G proteins (Gi and Gs). These novel results constitute an important advance in understanding the molecular intricacies involved in GPCR function.

Background G-protein-coupled receptor (GPCR) oligomerization is heavily supported by recent biochemical and structural data [1][2][3][4][5][6]. Optical-based techniques are instrumental in studying the dynamics and organization of receptor complexes in living cells [7]. For instance, total internal reflection fluorescence microscopy shows that 30 % of muscarinic M1 receptors exist as dimers (with no evidence of higher oligomers) that undergo interconversion with monomers on a timescale of seconds [8]. Similarly, the β 1 -adrenergic receptors (β 1 -AR) are expressed as a mixture of monomers and dimers whereas β 2 -adrenergic receptors (β 2 -AR) have a tendency to form dimers and higher-order oligomers [9]. Moreover, the monomer-dimer equilibrium of the chemoattractant N-formyl peptide receptor at a physiological level of expression lies within a timescale of milliseconds [10]. Together, these studies in heterologous systems show that a given GPCR is present in a dynamic equilibrium between monomers, dimers, and higher-order oligomers.
Studies in a broad spectrum of GPCRs [11][12][13][14] show that these receptors may form heteromers. GPCR heteromers are defined as novel signaling units with functional properties different from homomers and they represent a completely new field of study [15]. Innovative crystallographic techniques have permitted researchers to obtain crystal structures of GPCR families A, B, C, and F, bound to either agonists, antagonists, inverse agonists or allosteric modulators; in the form of monomers or homo-oligomers; and in complex with a G protein or with a ß-arrestin [16]. However, crystal structures of GPCR heteromers have not yet been obtained. Here, we propose a quaternary structure of a heteromer, taking into account the molecular stoichiometry and the interacting G proteins. Adenosine A 1 -A 2A receptor (A 1 R-A 2A R) complexes constitute a paradigm in the GPCR heteromer field because A 1 R is coupled to G i and A 2A R to G s ; that is, they transduce opposite signals in cyclic adenosine monophosphate (cAMP)dependent intracellular cascades. First described as a concentration-sensing device in striatal glutamatergic neurons [17], the A 1 R-A 2A R heteromer is thought to function as a G s /G i -mediated switching mechanism by which low and high concentrations of adenosine inhibit and stimulate, respectively, glutamate release [17,18]. The structural basis of this switch is key to understanding heteromer function and the biological advantage behind the GPCR heteromerization phenomenon. Here, we have devised the molecular architecture of the adenosine A 1 R-A 2A R heteromer in complex with G proteins using a combination of microscope-based single-particle tracking, molecular modeling, and energy transfer assays in combination with molecular complementation. The results point to A 1 and A 2A receptors organizing into a rhombus-shaped heterotetramer that couples to G i and G s . The overall structure is very compact and provides interacting interfaces for GPCRs and for G proteins.

Results and discussion
Reciprocal restriction of adenosine receptor motion in the plasma membrane To examine the dynamics of A 1 R-A 2A R heteromers in the plasma membrane of a living cell, the motion of the receptors tagged with fluorescent proteins (A 1 R-green fluorescent protein [GFP] or A 2A R-mCherry) was measured by real-time single-particle tracking (SPT) (Fig. 1). Examples of fluorescent images and individual particle trajectories are shown in Additional file 1: Figure S1. Analysis of data corresponding to 500 A 1 R-GFP particles showed a linear relationship between the mean square displacement (MSD) versus time lag in the trajectories of up to 1600 single fluorescent particles (Fig. 1a, c). This is typical for Brownian diffusion, indicating a lack of restrictions in A 1 R-GFP motion. Co-expression of A 2A R-mCherry (Fig. 1b) led to a reduction in the lateral mobility of A 1 R-GFP, which became confined to plasma membrane regions of 0.461 ± 0.004 μm in diameter. Its diffusion coefficient decreased from 0.381 ± 0.002 μm 2 /s to 0.291 ± 0.003 μm 2 /s (p = 0.002, one-tailed t-test). Similarly, A 1 R-GFP also decreased the A 2A R-mCherry diffusion coefficient from 0.317 ± 0.002 μm 2 /s to 0.143 ± 0.005 μm 2 /s (p < 0.0001) (Fig. 1d-f). A 2A R moved within a confinement zone of 0.941 ± 0.007 μm in diameter that was reduced to 0.360 ± 0.001 μm (p < 0.0001) when both receptors were co-expressed. We conclude from these mobility comparisons that reciprocally restricted motion of the individual receptor particles must be due to A 1 R-A 2A R receptor-receptor interactions.

Stoichiometry of A 1 and A 2A receptor heterocomplexes
The stoichiometry of the fluorescent receptors on the cell surface can be calculated from the brightness distribution of the individual particles [19] (see "Methods"). In cells expressing A 1 R-GFP, we found the majority of clusters to consist of either two (~47 %) or four (~34 %) receptors, and clusters with one or three receptors were scarce (~10 % and~9 %, respectively) (Additional file 2: Figure S2A and black bars in Additional file 2: Figure S2C). In the case of A 2A R-mCherry, the stoichiometry analysis showed that the clusters mostly expressed trimers (45 %), with dimers (29 %) and tetramers (12 %) the second and third most common populations (Additional file 2: Figure S2D and black bars in Additional file 2: Figure S2F). Remarkably, this stoichiometry for either A 1 or A 2A receptors was altered when the partner receptor was also expressed. In cells co-expressing A 1 R-GFP and A 2A R-mCherry, the dimer population increased (57 % for A 1 R-GFP and 49 % for A 2A R-mCherry, blue bars in Additional file 2: Figures S2C, F) and became the predominant species (Additional file 2: Figures S2B, C, E, F).
In order to focus the analysis on heteromer complexes, we identified clusters containing both receptors (individual yellow dots in Fig. 1g, displaying both GFP and mCherry fluorescence). In~1000 analyzed colocalized clusters that consisted of a mixture of A 1 -GFP and A 2A -Cherry (yellow dots in Fig. 1g), we found a similar high amount of dimers of A 1 R (75 %, left panel in Fig. 1h and green bar in Fig. 1i) and A 2A R (74 %, right panel in Fig. 1h and red bar in Fig. 1i). Trimers and tetramers of A 1 R, and monomers and tetramers of A 2A R, were in the minority or negligible (see Fig. 1h, i). In summary, given that the percentage of dimers of either A 1 R-GFP or A 2A R-mCherry in the yellow dots (which show co-localization of the two receptors) was similar and high (~75 %), the heterotetramer containing two A 1 Rs and two A 2A Rs must have been the most predominant species. To our knowledge, this is the first stoichiometry data for a GPCR heteromer in living cells.

Arrangement of G proteins interacting with A 1 and A 2A receptors
Monomeric GPCRs are capable of activating G proteins [20]. However, recent findings suggest that one GPCR homodimer bound to a single G protein may be a common functional unit [21]. Thus, an emerging question is how G proteins couple to GPCR heteromers. Because A 1 R selectively couples to G i and A 2A R to G s [22], the working hypothesis was that both G i and G s proteins may couple to the A 1 R-A 2A R heterotetramer. To test this hypothesis, we used bioluminescence resonance energy transfer (BRET) assays [23]. In agreement with the SPT experiments (see above), homodimers and heterodimers were detected by BRET assays in cells expressing A 1 R fused with Renilla luciferase (A 1 R-Rluc) or yellow fluorescent protein (A 1 R-YFP) (Fig. 2a), A 2A R-Rluc and A 2A R-YFP (Fig. 2b), or A 1 R-Rluc and A 2A R-YFP (Fig. 2e). Neither A 1 R-Rluc nor A 2A R-YFP interacted with the ghrelin receptor 1a fused to YFP (GHS1a-YFP), used as a control as a protein unable to directly interact with these adenosine receptors (Fig. 2a, b). In order to test the presence of the two G proteins in the heterotetramer, we transfected cells with minigenes that code for peptides blocking either G i or G s binding to GPCRs [24]. In addition, cells were treated with pertussis or cholera toxins that catalyze ADP-ribosylation of G i or G s . Clearly, treating cells with pertussis toxin, or expressing the minigene-coded peptide that blocks α i coupling, reduced the value of BRET max for A 1 R-A 1 R homodimers (Fig. 2a) and for A 1 R-A 2A R heterodimers ( Fig. 2e) but not for A 2A R-A 2A R homodimers (Fig. 2b). This indicates that G i is coupled to A 1 R in both the homodimer and the heterodimer. Similarly, blocking G sreceptor interaction using cholera toxin or a minigenecoded peptide that blocks α S coupling reduced BRET max for A 2A R-A 2A R homodimers (Fig. 2b) and for A 1 R-A 2A R heterodimers ( Fig. 2e) but not for A 1 R-A 1 R homodimers (Fig. 2a). Interestingly, BRET curves showed sensitivity to both cholera and pertussis toxins in cells expressing either A 1 R-Rluc-A 1 R-YFP and A 2A R (Fig. 2c) or A 2A R-Rluc-A 2A R-YFP and A 1 R (Fig. 2d). Functionality of constructs and controls in cells expressing minigenes, and in cells expressing the ghrelin GHS1a receptor instead of one of the adenosine receptors, are shown in Additional file 3: Figure S3. To further confirm that G i binds A 2A R in the receptor heteromer, the energy transfer between Rluc fused to the N-terminal domain of the α-subunit of G i (G i -Rluc) and A 2A R-YFP was analyzed in . Data sets were fitted to mathematical models of free and confined diffusion for A 1 R and A 2A R respectively. g Co-localization of A 1 -GFP and A 2A -mCherry is observed (yellow dots). Scale bar: 100 nm. h Distribution of the fluorescence signal of A 1 -GFP (left) and A 2A -mCherry (right) within co-localized receptors (yellow dots in g). Curves approximately delineate the number of monomers, dimers, or trimers within the co-localized complex. i Stoichiometry analysis performed for co-localized A 1 -GFP and A 2A -mCherry receptor particles co-expressed in HEK-293T cells (yellow dots in g). Green corresponds to A 1 -GFP and red to A 2A -mCherry ; and (f) 0.5 μg of cDNA corresponding to A 1 R (except control blue curves that were obtained in cells not expressing A 1 R), 2 μg of cDNA corresponding to G i -Rluc, and increasing amounts of A 2A R-YFP (0.1-0.5 μg cDNA). In panels a, b, and e, cells were also transfected with 0.5 μg of cDNA corresponding to the G i -related (orange curves) or G s -related (blue curves) minigenes. Cells were treated for 16 h with medium (black curves), with 10 ng/ml of pertussis toxin (green curves), or with 100 ng/ml of cholera toxin (red curves) prior to BRET determination. To confirm similar donor expressions (approximately 100,000 bioluminescence units) while monitoring the increase in acceptor expression (1000-40,000 fluorescence units), the fluorescence and luminescence of each sample were measured before energy transfer data acquisition. MiliBRET unit (mBU) values are the mean ± standard error of the mean of four to six different experiments grouped as a function of the amount of BRET acceptor. In each panel (top) a cartoon depicts the proteins to which Rluc and YFP were fused and the presence or not of partner receptors and/or G s or G i proteins [schemes in c to f are not intended to illustrate on stoichiometry because the predominant form in cells expressing the two receptors was the heterotetramer containing two A 1 and two A 2A receptors (see "Results")] cells co-expressing or not co-expressing A 1 R (Fig. 2f ). A hyperbolic BRET curve was observed in the presence of A 1 R, but not in its absence, indicating that G i and G s are bound to their respective receptor homodimers within the A 1 R-A 2A R heteromer.
Further, two complementary BRET experiments were performed to determine the orientation of G i and G s within the A 1 R-A 2A R heterocomplex. First, Rluc and YFP were respectively fused to the N-terminal domains of the α-subunit of G i (α i -Rluc) and G s (α s -YFP) (Fig. 3, bar a); second, they were fused to the N-terminal domain of the γ-subunit (γ-Rluc and γ-YFP) (Fig. 3, bar b). We observed significant energy transfer between γ-Rluc and γ-YFP in cells co-expressing A 1 R and A 2A R (Fig. 3, bar b) but minimal amounts in negative-control cells (Fig. 3, bars c and d). In cells expressing either A 1 R or A 2A R, the energy transfer between γ-Rluc and γ-YFP was also low (Fig. 3, bars e and f ), suggesting that dimers but not tetramers were the most prevalent form of surface receptors in single-transfected cells. These results in co-transfected cells corroborate the 2:2 stoichiometry obtained from analysis of the fluorescence in single particles and are consistent with G i and G s binding to these A 1 R-A 2A R heterotetramers.

Molecular model of G i and G s bound to the A 1 R-A 2A R heterotetramer
To identify the orientation of the G protein in the receptor homodimer, we combined energy transfer assays between α s -Rluc (Rluc at the N-terminus of the G protein αsubunit) and A 2A R-YFP ( Fig. 4a) with information on transmembrane (TM) interfaces based on crystal structures of GPCRs [3,4], which have been recently summarized [25]. The observed high-energy transfer using α s -Rluc and A 2A R-YFP indicated close proximity between the N-tail of the α-subunit of G s and the C-tail of A 2A R. Interestingly, Rluc and YFP in the "monomeric" A 2A R-G s complex (see "Methods") point toward distant positions in space (Fig. 4b). Therefore, the observed BRET should occur between Rluc in the G protein α-subunit and a second A 2A R-YFP protomer. Among all described TM interfaces for receptor homodimerization (see Additional file 4: Figure S4), we propose the TM4/5 interface, which is observed in the oligomeric structure of β 1 -AR [4] and in structures derived from coarse-grained molecular dynamics (MD) simulations [26]. In fact, this is the only interface that favors BRET between α s -Rluc and a second A 2A R-YFP protomer in a homodimer (Fig. 4c). The homologous A 1 R homodimer was built using the same TM4/5 interface as for A 2A R (see Additional file 4: Figure S4 and its legend).
The remaining possible TMs able to form heteromeric interfaces are TM1 and TM5/6 ( Fig. 5). Both are possible inter-GPCR interfaces as observed in the structure of the μ-opioid receptor (μ-OR) [3]. To discern between these two possibilities, a bimolecular fluorescence complementation strategy was undertaken. For this purpose, the N-terminal fragment of Rluc8 was fused to A 1 R (A 1 R-nRluc8) and its C-terminal domain to A 2A R (A 2A R-cRluc8), which only upon complementation can act as a BRET donor (Rluc8). The BRET acceptor protein was obtained upon complementation of the N-terminal fragment of YFP Venus protein fused to A 1 R (A 1 R-nVenus) and its C-terminal domain fused to A 2A R (A 2A R-cVenus). When all four receptor constructs were transfected, we obtained a positive and saturable BRET signal (BRET max of 35 ± 2 mBU and BRET 50 of 16 ± 3 mBU) that was not obtained for negative controls (Additional file 5: Figure S5). Figure 5a, b shows that the hemi-donor (A 1 R-nRluc8 and A 2A R-cRluc8) and the hemi-acceptor (A 1 R-nVenus and A 2A R-cVenus) moieties, placed at the Cterminus of the receptors, can only complement if A 1 R-A 2A R heterodimerization occurs via the TM5/6 interface. The TM4/5 interface for homodimerization and the TM5/6 interface for heterodimerization give a rhombus-shaped tetramer organization (Fig. 5a). Remarkably, cell pre-incubation with either pertussis or cholera toxins decreased the BRET max by 35 % (Fig. 5c), further suggesting that both G s and G i proteins bind to the A 1 R-A 2A R heterotetramer.
We next evaluated, using computational tools, whether the proposed A 1 R-A 2A R heterotetramer could couple to both G i and G s proteins. Clearly, the external protomers of the proposed A 1 R-A 2A R heterotetramer can bind to G i and G s proteins (Fig. 5d). This model positions the α-subunits of G i and G s in close contact, facing the interior of the tetrameric complex, while the N-terminal α-helices of α i and α s point outside the complex. The N-terminal α-helices of the γ-subunits are in close proximity, facing the inside (Additional file 6: Figure S6), which explains the significant energy transfer observed between γ-Rluc and γ-YFP (Fig. 3, bar b). The model provides experimental insights into the structural arrangement of heteromers consisting of two GPCRs and coupled to two G proteins, the possibility of which has recently been discussed [25]. We used MD simulations to study the stability of this complex. Additional file 7:  Figure S7 shows root-mean-square deviations (rmsd) on protein α-carbons throughout the MD simulation, as well as key intermolecular distances among protomers and G proteins. Clearly, both the A 1 R protomer bound to G i and the A 1 R protomer that does not interact with it maintained a close structural similarity (rmsd ≈ 0.3 nm) relative to the initial structures. Similar results were obtained for the A 2A R protomers (bound and unbound to G s ) (Additional file 7: Figure S7A). The fact that rmsd values of the whole system, formed by the A 1 R-A 2A R heterotetramer bound to G i and G s , are of the order of 0.6 nm indicates that the initial structural model is maintained during the MD simulation (Additional file 7: Figure S7A). As a consequence, selected intermolecular distances among protomers and G proteins remain constant during the MD simulation (Additional file 7: Figure S7B). A key aspect in the assembly of the heterotetramer is the TM interfaces for homodimerization (TM4/5) and heterodimerization (TM5/6). Additional file 8: Figure S8B shows rmsd values of the four-helix bundle forming the TM4/5 and TM5/6 interfaces, the initial and final snapshots of these bundles, and the evolution of the A 1 R-A 2A R heterotetramer during the MD simulation. Clearly, the rather small structural variations of these four-helix bundles, also reflected by rmsd <0.3 nm, suggest a stable complex. Notably, the TM5/6 four-helix bundle seems more stable than the TM4/5 bundles, as shown by its lower rmsd value. Additional file 8: Figure S8B, C depicts contact maps of the TM4/5 and TM5/6 interfaces, as well as the evolution of the network of hydrophobic interactions within these interfaces during the MD simulation. inter-receptor interfaces modeled as in the structure of the μ opioid receptor [3]. TM helices 1, 4, and 5, involved in receptor dimerization, are highlighted in dark blue, light blue, and gray, respectively. nRluc8 and cRluc8 are shown in blue and nVenus and cVenus in dark yellow. c BRET and bimolecular fluorescence complementation experiments were performed in HEK-293T cells transfected with 1.5 μg of cDNA corresponding to A 1 R-cRluc8 and A 2A R-nRluc8, and 1.5 μg of cDNA corresponding to A 1 R-nVenus and A 2A R-cVenus. As the negative control, cells were transfected with 1 μg of cDNA corresponding to nRluc8 and 1.5 μg of cDNA corresponding to A 2A R-nRluc8, A 1 R-nVenus, and A 2A R-cVenus. Cells were treated for 16 h with medium (-toxins), 10 ng/ml of pertussis toxin (+ pertussis), or 100 ng/ml of cholera toxin (+ cholera) prior to BRET determination. The relative amount of BRET is given as in Fig. 4 and values are the mean ± standard error of the mean of three different experiments. Student's t-test showed statistically significant differences with respect to the control ( # p < 0.05, ## p < 0.01) and a significant effect in the presence of either toxin over BRET in the absence of toxins (*p < 0.05). A schematic representation at the top shows the protein to which the hemi luminescent or fluorescent proteins were fused. d Molecular model of the A 1 R-A 2A R tetramer in complex with G i and G s . A 1 R bound to G i is shown in red, G i -unbound A 1 R is shown in orange, A 2A R bound to G s is shown in dark green, G s -unbound A 2A R is shown in light green, and the α, β-, and γ-subunits of G i and G s are shown in dark gray, light gray, and purple, respectively. Transmembrane helices 4 and 5 are highlighted in light blue and gray, respectively Conclusions For more than a decade, experimental evidence has supported the occurrence of homo-oligomers and heterooligomers of GPCRs [21]. However, our basic understanding of what makes heteromers different from homomers remains unknown. Our results, studying adenosine receptors as a model heteromer, point to three important new findings. First, the predominant stoichiometry in cells expressing A 1 R-A 2A R heteromers is 2:2; that is, a dimer of dimers (tetramer). Second, two different heterotrimeric G proteins can couple to heteromers, the overall complex constituting a functional unit. Third, the molecular orientation within the heteromer complex affords various qualitatively different interfaces; the two more relevant are the inter-protomer heteromeric interface and the inter-Gprotein interface. Presumably, the two interfaces provide the key characteristic of heteromers: the ability of one protomer/G-protein complex to influence the signaling of the other. Surely, allosteric effects occurring between heteroreceptors and between G s and G i proteins are due to conformational changes transmitted along the intimately interacting molecules in the complex. In our controlled cell transfection system, which expressed a low density of receptors, minor species formed by monomers and trimers were found in addition to a predominance of tetramers in the plasma membrane, strongly supporting the occurrence of an in vivo dynamic distribution of receptors.
Adenosine was, from an evolutionary point of view, one of the first extracellular regulators given that it is involved in energy and nucleic acid metabolisms. Adenosine A 1 and A 2A receptors are expressed in almost every mammalian organ and tissue. In the heart, where adenosine plays a key role in both inotropic and chronotropic regulation, A 1 R-mediated cardioprotection did not occur in A 2A R knockout mice, suggesting an interaction between A 1 and A 2A receptors. In neurons, A 1 and A 2A receptors show co-localization, leading to interreceptor interactions unveiled by pharmacological treatments. For instance, Okada et al. [27] showed that cAMP-dependent protein kinase A plays a role in the regulation of hippocampal serotonin release mediated by both A 1 and A 2A receptors. Similarly, the control of γamino butyric acid transport in astrocytes was attributed to the expression of A 1 R-A 2A R heteromers and to a specific mechanism by which the heteromer signals via G i or G s depending on the concentration of adenosine [28]. The structural basis of the differential signaling by the heteromer/G-protein macromolecular complex likely implies communication at the receptor-receptor level but also between G s and G i . Because the binding of two G proteins to a heterodimer is not feasible due to steric clashes [25], our finding that the A 1 R-A 2A R heterotetramer may bind to both G s and G i provides a structural framework to interpret experimental data.

Methods
Total internal reflection microscopy and single-particle data analysis Single-particle imaging and tracking were performed on a Nikon Total Internal Reflection Fluorescence (TIRF) system, as detailed in Additional file 11: Supplementary Methods. Typically, 500 readouts of a 512 × 512-pixel region, the full array of the CCD chip, were acquired. For single-particle data analysis, parameters were calculated by applying the equations described in Additional file 11: Supplementary Methods. Plasmids DNA sequences encoding amino acid residues 1-155 and 155-238 of YFP Venus protein, and amino acids residues 1-229 and 230-311 of RLuc8 protein were subcloned in the pcDNA3.1 vector to obtain the YFP Venus and RLuc8 hemi-truncated proteins. The human cDNAs for adenosine receptors, A 2A R and A 1 R, cloned into pcDNA3.1, were amplified without their stop codons using sense and antisense primers harboring unique EcoRI and BamHI sites to clone receptors into the pcDNA3.1RLuc vector (pRLuc-N1; PerkinElmer, Wellesley, MA, USA), and EcoRI and KpnI to clone A 2A R, A 1 R, or GHS1a into the pEYFP-N1 vector (enhanced yellow variant of GFP; Clontech, Heidelberg, Germany). G αs cloned into the SFV1 vector, G αi cloned into the pcDNA3.1 vector, or G γ cloned into the pEYFP-C1 vector were amplified without their stop codons using sense and antisense primers harboring unique HindIII and BamHI sites to clone them into the pcDNA3.1-Rluc vector, or EcoRI and KpnI to clone G αs into the pEYFP-N1 vector. The amplified fragments were subcloned to be in-frame with restriction sites of the pcDNA3.1RLuc or pEYFP-N1 vectors to give plasmids that expressed proteins fused to RLuc or YFP on the N-terminal end (G αs -RLuc, G αi -RLuc, G γ -RLuc, G αs -YFP, and G γ -YFP) or the C-terminal end (A 1 R-RLuc, A 2A R-RLuc, A 1 R-YFP, A 2A R-YFP, and GHS1a-YFP). The human cDNAs for A 1 R or GHS1a were subcloned into pcDNA3.1-nRLuc8 or pcDNA3.1-nVenus to give plasmids that expressed A 1 R or GHS1a fused to either nRLuc8 or nYFP Venus on the C-terminal end of the receptor (A 1 R-nRLuc8 and A 1 R-nVenus or GHS1a-nRLuc8 and GHS1a-nVenus). The cDNAs for human A 2A or GHS1a receptors were subcloned into pcDNA3.1-cRLuc8 or pcDNA3.1-cVenus to give plasmids that expressed receptors fused to either cRLuc8 or cYFP Venus on the Cterminal end of the receptor (A 2A R-cRLuc8 and A 2A R-cVenus or GHS1a-cRLuc8 and GHS1a-cVenus). Expression of constructs was tested by confocal microscopy and the receptor-fusion protein functionality by measuring ERK1/2 phosphorylation and cAMP production, as described previously [13,14,17,29].
"Minigene" plasmid vectors are constructs designed to express relatively short polypeptide sequences following their transfection into mammalian cells. Here, we used minigene constructs encoding the carboxylterminal 11-amino acid residues from G α subunits of G i1/2 (G i minigene) or G s (G s minigene) G proteins; the resulting peptides inhibit G-protein coupling to the receptor and consequently inhibit the receptormediated cellular responses as previously described [24]. The cDNA encoding the last 11 amino acids of human G α subunit corresponding to G i1/2 (I K N N L K D C G L F) or G s (Q R M H L R Q Y E L L), inserted in a pcDNA3.1 plasmid vector, were generously provided by Dr Heidi Hamm.

Energy transfer assays
For BRET and complementation BRET assays, HEK-293T cells were transiently cotransfected with a constant amount of cDNA encoding for proteins fused to RLuc, nRLuc8, or cRLuc8, and with increasing amounts of the cDNA corresponding to proteins fused to YFP, nYFP Venus, or cYFP Venus (see figure legends). To quantify protein-YFP expression or protein-reconstituted YFP Venus expression, cells (20 μg protein) were distributed in 96-well microplates (black plates with a transparent bottom) and fluorescence was read in a FLUOstar OP-TIMA Fluorimeter (BMG Labtechnologies, Offenburg, Germany) equipped with a high-energy xenon flash lamp, using a 10 nm bandwidth excitation filter at 400 nm reading. Protein fluorescence expression was determined as the fluorescence of the sample minus the fluorescence of cells expressing the BRET donor alone. For BRET measurements, the equivalent of 20 μg of cell suspension were distributed in 96-well microplates (Corning 3600, white plates; Sigma) and 5 μM coelenterazine h (Molecular Probes, Eugene, OR, USA) was added. After 1 min for BRET or after 5 min for BRET with bimolecular fluorescence complementation, the readings were collected using a Mithras LB 940 that allows the integration of the signals detected in the short-wavelength filter at 485 nm (440-500 nm) and the long-wavelength filter at 530 nm (510-590 nm). To quantify protein-RLuc or protein-reconstituted RLuc8 expression, luminescence readings were also performed 10 min after adding 5 μM coelenterazine h. The net BRET was defined as [(long-wavelength emission)/ (short-wavelength emission)] -Cf, where Cf corresponds to [(long-wavelength emission)/(short-wavelength emission)] for the donor construct expressed alone in the same experiment. BRET is expressed as miliBRET units (mBU; net BRET × 1000).

Computational model of the A 1 R-A 2A R tetramer in complex with G i and G s
The crystal structure of inactive A 2A R [PDB:4EIY] [30] was used for the construction of human A 2A R [UniProt:P29274] and A 1 R [UniProt:P30542] homology models using Modeller 9.12 [31]. These receptors share 51 % of sequence identity and 62 % of sequence similarity, excluding the C-terminal after helix 8. Intracellular loop 3 (ICL3) of A 2A R (Lys209-Gly218) and A 1 R (Asn212-Ser219) were modeled using Modeller 9.12 [31] using ICL3 of squid rhodopsin [PDB:2Z73] as a template. The C-terminus tails of A 1 R, containing 16 amino acids (Pro311-Asp326), and of A 2A R, containing 102 amino acids (Gln311-Ser412), were modeled as suggested for the oxoeicosanoid receptor (OXER) [32] (see Additional file 9: Figure S9 for details). The N-terminus of A 1 R and A 2A R were not included in the model. The "active" conformations of A 1 R bound to G i and A 2A R bound to G s were modeled using the crystal structure of β 2 -AR in complex with G s [PDB:3SN6] [33]. The globular α-helical domain of the α-subunit was modeled in the "closed" conformation [34] Figure S9 for details). Rluc [PDB:2PSD] and YFP were fused to the to the N-terminus of the α-subunits and γ-subunits of G i and G s by a covalent bond. The structures of adenosine receptor oligomers were modeled via the TM4/5 interface for homodimerization, using the oligomeric structure of the β 1 -AR [PDB:4GPO] [4], or via the TM5/6 interface for heterodimerization, using the structure of the μ-OR [PDB:4DKL] [3]. The G i -bound A 1 R and G s -bound A 2A R protomers were rotated 10°to avoid the steric clash of the N-terminal helix of G i and G s with the C-terminal helix (Hx8) of G s -unbound A 2A R and G i -unbound A 1 R, respectively. This computational model, without Rluc and YFP, was placed in a rectangular box containing a lipid bilayer (814 molecules of 1-palmitoyl-2-oleoyl-sn-glycero-3phosphocholine -POPC -) with explicit solvent (102,973 water molecules) and a 0.15 M concentration of Na + and Cl − (1762 ions). This initial complex was energyminimized and subsequently subjected to a 10 ns MD equilibration, with positional restraints on protein coordinates. These restraints were released and 500 ns of MD trajectory were produced at constant pressure and temperature (see Additional file 10: Movie M1). Computer simulations were performed with the GROMACS 4.6.3 simulation package [35], using the AMBER99SB force field as implemented in GROMACS and Berger parameters for POPC lipids. This procedure has been previously validated [36].

Availability of data and materials
The crystal structures 4EIY, 2Z73, 3SN6, 1AGR, 2RH7, 2PSD, 4GPO, and 4DKL are available from PDB (http:// www.rcsb.org). All other relevant data are within the paper and its Additional files.

Additional files
Additional file 1: Figure S1. Examples of receptor trajectories in HEK-293T cells. Images of cells expressing A 1 R-GFP (A) and of particular trajectories of A 1 R-GFP-containing (B) or A 2A R-mCherry-containing (C) particles. (TIF 1164 kb) Additional file 2: Figure S2. Graphical description of the stoichiometry of A 1 R-GFP, A 2A R-mCherry or both A 1 -GFP and A 2A -mCherry. The fluorescence intensity signal distribution (gray area) detected for more than 7000 independent observations is given for HEK-293T cells expressing given as mean ± SD (n = 4-8). One-way ANOVA followed by a Bonferroni post -hoc test showed a significant effect of CPA when compared with that of forskolin (red bars, ***p < 0.001) or of CGS 21680 when compared to basal cAMP levels (blue bars, ## p < 0.01, ### p < 0.001). (C, D) BRET saturation curves were performed in HEK-293T cells transfected with (C) 0.3 μg cDNA coding for A 1 R-Rluc, increasing amounts of cDNA coding for A 1 R-YFP (0.1-1.5 μg cDNA), and 0.4 μg cDNA coding for GHS1a, or (D) with 0.2 μg of cDNA coding for A 2A R-Rluc, increasing amounts of cDNA coding for A 2A R-YFP (0.1-1.0 μg cDNA), and 0.5 μg cDNA coding for to GHS1a. Prior to BRET determination, cells were treated for 16 h with medium (black curves), with 10 ng/ml of pertussis toxin (green curves), or with 100 ng/ml of cholera toxin (red curves). mili BRET units (mBU) are given as the mean ± SD (n = 4-6 different experiments grouped as a function of the amount of BRET acceptor). (TIF 1418 kb) Additional file 4: Figure S4. Possible interfaces in A 2A R homodimers in complex with G s . In A-E, the A 2A R homodimer was modeled through TM4 using the H 1 -receptor structure as template (A), through TM5 using the structure of squid rhodopsin (B), through TM4/5 using the β 1receptor structure (C), and via TM5/6 (D) and TM1 (E) using the μ-OR structure. TM helices 1, 4, and 5 involved in receptor dimerization are highlighted in dark blue, light blue, and gray, respectively. A 2A R protomers bound to G s (in gray) are shown in light green, whereas G s -unbound A 2A R protomers are shown in dark green. Rluc (blue) is attached to the Nterminal αN helix of G s , and YFP (yellow) is attached to the C-terminal domain of the G s -unbound A 2A R protomer (light green). It is important to note that the position of YFP is highly dependent on the orientation of the long and highly flexible C-tail of A 2A R (102 amino acids, Gln311-Ser412), which was modeled as described for the OXER [32] (see Additional file 9: Figure S9 for details). Despite these limitations, we can crudely estimate the approximate distances between the center of mass of Rluc and YFP as 4.6, 10.1, 6.5, 11.6, and 8.3 nm for panels A-E, respectively. Thus, among all these possible dimeric interfaces, only the molecular models depicted in panels A (TM4 interface) and C (TM4/5 interface) would favor the observed highenergy transfer between G s -Rluc and A 2A R-YFP (Fig. 4a in main paper). However, there is a steric clash between the N-terminal helix of Gs and the dark-green protomer in the TM4 interface. Accordingly, we have modeled A 2A R homodimerization via the TM4/5 interface. Unfortunately, similar experiments with cells transfected with G i -Rluc and A 1 R-YFP could not be accomplished because of a lack of receptor expression (not shown); it is likely that the shorter C-tail of A 1 R (16 amino acids, Pro311-Asp326) could not accommodate YFP in the presence of G i in the right three-dimensional structure. The A 1 R homodimer was built using the same TM4/5 interface as for A 2A R. (TIF 3135 kb) Additional file 5: Figure S5. BRET assays in cells expressing fusion proteins containing hemi-Rluc8 and hemi-Venus moieties fused to adenosine receptors or containing the ghrelin GHS1a receptor instead of one of the adenosine receptors. (A) Saturation BRET curve in HEK-293T co-transfected with 1.5 μg of the two cDNAs corresponding to A 1 R-cRLuc8 and A 2A R-nRLuc8 and with increasing amounts of cDNAs corresponding to A 1 R-nVenus and A 2A R-cVenus (equal amounts of the two cDNAs). BRET max was 35 ± 2 mBU and BRET 50 was 16 ± 3 mBU. BRET in cells expressing cRluc8 instead of A 1 R-cRluc8 gave a linear, non-saturable signal. (B) Comparison of BRET responses using complementary and non-complementary pairs, or replacing one adenosine receptor with the ghrelin GHS1a (gn) receptor. Data are mean ± SD of three different experiments grouped as a function of the amount of BRET acceptor. ***p < 0.001 with respect to BRET in cells expressing adenosine receptors and hemi-Rluc8 and hemi-Venus proteins. (TIF 398 kb) Additional file 6: Figure S6. Details of the relative position of Rluc and YFP in a receptor heterotetramer interacting with two G proteins. Computational-based model of G s and G i bound to the adenosine A 1 R-A 2A R heterotetramer. Rluc and YFP fused to the N-terminal domain of the G α -subunits point toward different positions in space (A), whereas Rluc and YFP fused to G γ -subunits are close (B). The color code of the proteins is depicted in the adjacent schematic representations (TM4 and TM5 of GPCR protomers are in light blue and gray, respectively). (TIF 6445 kb) Additional file 7: Figure S7. Molecular dynamics (MD) simulation of the adenosine A 1 R-A 2A R heterotetramer in complex with G i and G s . (A) Root-mean-square deviations (rmsd) on protein α-carbons of the whole system (black solid line), of the two A 1 Rs (orange and red