Allosteric activation of the glutamate receptor mGlu2 by the serotonin receptor 5-HT_2A

Abstract

G protein-coupled receptor (GPCR) dimers and higher-order oligomers are the subject of intense debates. The complexes formed between two or more different GPCRs enable the cell to integrate several signals. But the mechanisms of the allosteric interaction within these oligomers remain unclear. Here, we use as a model the heteromer formed in the brain between two targets of anti-psychotic drugs, the metabotropic glutamate receptor 2 (mGlu2) and the serotonin receptor 5-HT_2A. Using molecular dynamics, a nanobody-based sensor and resonance energy transfer assays, we demonstrate that the 5-HT_2A receptor behaves as a positive allosteric modulator by stabilizing the active conformation of mGlu2. Using cysteine cross-linking experiments and mGlu2 sensors, we reveal the molecular basis for this allosteric modulation of the mGlu2 receptor by 5-HT_2A, an effect also mediated by other GPCRs. Our results thus provide insights into the allosteric control of a GPCR activity via heteromerization with another receptor.

The interaction between different GPCRs is important in integrating their intracellular signaling. Here, the authors provide insight into the mechanisms of allosteric activation of the mGlu2 receptor induced by 5-HT2A or other GPCRs in their heteromeric complexes.

Introduction

G protein-coupled receptors (GPCRs) are the largest family of membrane proteins and are the target of around 30% of marketed drugs^1,2. Upon activation by extracellular stimuli, they transduce signals by coupling to heterotrimeric G proteins from different families³. Given that individual cells express dozens of distinct GPCR subtypes, it raises the question of how the cells integrate signals from such a diverse receptor repertoire. For over two decades, GPCR dimers and higher-order oligomers were viewed as a possible way of signal integration (Fig. 1a)^4–7, but this subject raised intense debates^8–10. Today, although such complexes are being considered, a more accepted view is that different GPCRs can interact transiently^11,12, but the functional consequence of such interaction remains more than elusive^13–15. It is therefore important to understand how one GPCR subtype regulates the function of another, as such knowledge may enable the development of drugs that precisely tune GPCR signaling¹⁶.

Fig. 1. mGlu2 receptor interacts with co-transfected 5-HT_2A receptor at cell surface.

a Scheme illustrating signal integration mechanism between two GPCRs by heteromerization between the two receptors. The receptor A and B are represented as gray or blue 7TM bundles respectively. b TR-FRET saturation analysis of the indicated pairs of constructs between Snap-tagged mGlu2 (^SmGlu2) and the indicated Halo-tagged receptors (^Hreceptor), when HEK293 cells were transfected with fixed and increasing amounts of cDNA for ^SmGlu2 and ^Hreceptor, respectively. Non-cell permeant fluorophores were used to label the Snap-tag (black circle) with Lumi4-Tb donor (orange star) and Halo-tag (gray circle) with a fluorescent acceptor (blue star). In the same cells, FRET signal and cell surface expression of ^SmGlu2 (Exp_S) and ^Hreceptor (Exp_H) by the emission signal of the donor and acceptor, respectively, were measured. Of note, for clarity a stoichiometry of one mGlu2 homodimer and one 5-HT_2A receptor is shown, but we cannot exclude that two molecules of 5-HT_2A could interact with one mGlu2 homodimer The FRET signals between the co-expressed ^SmGlu2 and ^HmGlu2 and between ^SmGlu2 and the ^HM2 muscarinic receptor 2 class A GPCR are used as positive and negative controls, respectively. Data are shown in means ± SEM from three independent experiments (n = 3).

The molecular mechanisms of allosteric interaction between GPCRs within these oligomers, and in particular within heteromers formed of two different receptors⁶, are still unclear. The molecular basis of direct interactions between GPCRs begins to be understood, thanks to recent structural^17–21 and biochemical^13,22 studies that have revealed possible dimerization interfaces between receptors. Interestingly, homodimers can adopt an asymmetric interface in the active state^19,20, consistent with the asymmetric functioning reported for many class A GPCR dimers^23–25, the most abundant receptors accounting for 90% of mammalian GPCRs. Class C GPCRs^19,20, which classically form obligate dimers²⁶, have also been studied in this regard. This stands in contrast to class A receptors, for which the monomer has been demonstrated to be the necessary and sufficient signaling unit²⁷. In dimers and higher-order oligomers formed by class A GPCRs, the seven transmembrane domains (7TMs) play a key role in stabilizing complexes, with almost all TMs that could be involved in dimerization or oligomerization interfaces²¹. For dimeric class C GPCRs, rearrangement of the dimerization interface during activation is well-characterized. In the inactive state, the dimer interface is generally stabilized by interactions involving TM4 and TM5 from each protomers^22,28,29. A major reorientation of the dimer interface is observed during agonist activation, where the two TM6s come into contact, a conserved conformation that enables G protein coupling³⁰. Recent structures of class C GPCRs also revealed the structural basis of the asymmetry that occurs at the 7TM interface during G protein coupling^19,20. The rearrangement of the 7TM interface for other GPCRs, such as class A receptors, is unclear, and some studies have also proposed a relative repositioning of these transmembrane domains between the protomers³¹.

On the other hand, the molecular basis of allosteric interactions between two different GPCRs in a heteromer remains poorly defined. Indeed, very few structures of complexes between two different 7TMs have been solved, with the exception of the obligatory class C GABA_B heterodimer^28,32, the sweet taste receptor^33–35 and the recently discovered mGlu heterodimers^29,36,37. Heterodimeric complexes formed between a class A GPCR and a class C GPCR are valuable models for studying receptor heteromers, owing to both structural and physiological insights. Firstly, the 7TM interfaces of several class C GPCR dimers in the inactive and active states has recently been well characterized^19,20. Secondly, several heteromers formed between class C and class A GPCRs have been shown to be physiologically relevant, including combinations between several mGlu receptors and receptors of neurotransmitters (serotonin, dopamine or adenosine receptors), including the mGlu2-5-HT_2A³⁸, mGlu5-D1³⁹ and mGlu5-A_2A-D2⁴⁰ heteromers. Among them, the properties of the mGlu2-5-HT_2A heteromer have been extensively studied, as such heterocomplex has been proposed as an important target for antipsychotics^38,41–46. Nevertheless, the structural basis for the allosteric control of the mGlu2 receptor by 5-HT_2A remains elusive.

In this study, we used the mGlu2-5-HT_2A receptor complex as a model to examine the molecular basis of the allosteric interaction between a class A and a class C GPCR. This receptor complex was first reported over fifteen years ago, and subsequent in vivo studies have confirmed the negative synergy of the mGlu2 and 5-HT_2A receptors during schizophrenia and antipsychotic treatment^42,47. For example, the antipsychotic effects of clozapine on the 5-HT_2A receptor require the expression of mGlu2 receptor⁴⁸. Furthermore, the activation of the mGlu2 receptor has been proposed to promote the anti-dyskinetic and anti-psychotic effects of 5-HT_2A in the treatment of Parkinson’s disease⁴⁹. Through molecular modelling, resonance energy transfer assays and cysteine crosslinking, we demonstrate that 5-HT_2A receptor behaves as a positive allosteric modulator, stabilizing the active conformation of mGlu2. Strikingly, we also find that this allosteric activation of mGlu2 is not specifically mediated by 5-HT_2A receptors, but also mediated by many other class A GPCRs, including several dopamine and serotonin receptors. Collectively, our results provide insights into the allosteric control of a GPCR activity through heteromerization.

Results

Direct interaction between mGlu2 and 5-HT_2A receptors at cell surface

We first verified the interaction between mGlu2 and 5-HT_2A receptors by time-resolved fluorescence resonance energy transfer (TR-FRET) saturation assays⁵⁰. HEK-293 cells were co-transfected with Snap-tagged mGlu2 (^SmGlu2) and Halo-tag 5-HT_2A (^H5-HT_2A) for orthogonal labeling of two receptors with non-cell permeant fluorescent substrates specific of these tags attached to the extracellular N-terminal of each receptor. To minimize the difference in size of the N-terminal region between mGlu2 receptor, which has a large extracellular domain (563 residues), and 5-HT_2A, which has a short N-terminal sequence (67 residues), the Halo-tag (33 kDa) and Snap-tag (21 kDa) were fused to the 5-HT_2A and mGlu2 receptors respectively⁵¹. We also optimized the labelling conditions to obtain a strong FRET signal (Supplementary Fig. 1). Saturation curve analysis showed a strong FRET signal between ^SmGlu2 and ^H5-HT_2A (Fig. 1b and Supplementary Fig. 2), albeit lower than the FRET signal measured between the ^SmGlu2 and ^HmGlu2 subunits, which form a constitutive homodimer^26,52. A strong FRET signal was also observed between ^SmGlu2 and the Halo-tagged dopamine D1 receptor ^HD1 (N-terminal region of 20 residues) (Supplementary Fig. 3). In contrast, a weak FRET signal was obtained between ^SmGlu2 and the Halo-tagged M2 muscarinic receptor (^HM2, this negative control containing 18 amino-acids in its N-terminal region; Fig. 1b), as well as sphingosine-1-phosphate receptor 1 (^HS1P1R, 41 amino-acids in its N-terminal region; Supplementary Fig. 3). Notably, the FRET signal intensity did not correlate with the N-terminal length of the class A GPCRs tested. These data demonstrate a strong propensity for mGlu2 to form a complex with 5-HT_2A, consistent with previous studies^38,42,53.

3D model of the mGlu2-5-HT_2A complex

Structures of the full-length mGlu2 receptor in the inactive and active states^20,29 revealed the relative rearrangement of the transmembrane domain interface in the homodimer upon agonist activation. In the inactive state, stabilized by a competitive antagonist bound into the glutamate binding site and a negative modulator bound into a transmembrane domain of the homodimer, the transmembrane interface is composed of TM4 and TM3²⁰ (Fig. 2a, b). In the active state, the two TM6s form the homodimer interface^20,29 (Fig. 2c, d). These structural data are consistent with the model of 7TM rearrangement we previously proposed based on cysteine cross-linking experiments of the mGlu2 receptor expressed at the surface of living cells²². In our previous work, we showed that the mGlu2 receptor in its active state can form a homo-oligomer via the TM4-TM5 interface. This finding suggested that the mGlu2 receptor can form higher-order complexes with other membrane proteins across this interface.

Fig. 2. mGlu2 receptor structures and mGlu2-5-HT_2A heteromer homology model.

Top view of inactive (a), active (c) cryo-EM structure of mGlu2 homodimer and homology model of mGlu2-5HT_2A heteromer (e). Schemes are shown to illustrate the 7TM interfaces of inactive mGlu2 homodimer (b), active mGlu2 homodimer (d) and mGlu2-5-HT_2A heterotrimer (f). mGlu2 and 5-HT_2A receptors are shown in light blue and gray, respectively, except for TM4 and 5 (red) and TM6 (yellow) of mGlu2 promoters and TM4 and 5 (dark red) of 5-HT_2A receptor.

To investigate the heteromeric interface, we generated a realistic and unbiased 3D model of the active mGlu2 homodimer in complex with the 5-HT_2A receptor. As our study was carried out before the structures of the two receptors were available, thus an initial homology model was built for the active mGlu2 homodimer, where the two protomers interact via a symmetrical TM5-TM6 interface, and the inactive 5-HT_2A receptor. In all-atom molecular dynamics (MD) simulations of the isolated mGlu2 homodimer, the TM5-TM6 interface between the mGlu2 protomers progressively decreases its contact surface on the intracellular side and rotates towards a TM6-TM6 interface, which is maintained until the end of the simulation (Fig. 2e, f). Subsequently, a heterodimeric mGlu2-5-HT_2A complex between mGlu2 homodimer and a 5-HT_2A monomer was assembled by protein-protein docking, and an-unbiased molecular dynamics revealed a symmetrical interface TM4-TM5 between the two receptors. During this simulation, we observed that the conformation of mGlu2 promoters was stable in the mGlu2-5-HT_2A complex, with RMSDs similar to those observed in the mGlu2 homodimer alone (average RMSD of 2.0 Å for the protomer interacting with 5-HT_2A and 1.8 Å for the other; Supplementary Fig. 4a). Similarly, the 5-HT_2A maintained initial inactive conformation was also stable throughout our MD simulations, either in the mGlu2-5-HT_2A complex or in monomeric form (average RMSDs of 2.0 Å and 2.1 Å, respectively, Supplementary Fig. 4b). Consequently, this initial model of the mGlu2-5-HT_2A complex was deemed to be an appropriate tool for studying inter- and intra-receptor functional effects and for guiding the subsequent cross-linking experiments. More recently, we have further validated this model through a new complex model built from experimental structures, the cryo-EM structures of the mGlu2 homodimer²⁰ and 5-HT_2A receptor⁵⁴ (see Materials and Methods). To confirm their consistency, we submitted both models, based on homology modeling and on experimental structures, to the same protocol of coarse-grained molecular dynamics simulations (CG MD; see Methods). This level of representation is well suited for comparing the organization and persistence of pre-formed dimer interfaces, as it allows access to longer effective simulation timescales than would be feasible with all-atom models while retaining the key features of protein–protein contacts. Because our analysis focused on the stability and similarity of existing interfaces rather than on association/dissociation processes or detailed kinetics, standard coarse-grained MD simulations were sufficient for this purpose, and no enhanced sampling methods were required. In these CG trajectories, both models displayed the same general orientation of the protomers (Supplementary Figs. 5a and 7e), with several common contacts at the interfaces (Supplementary Figs. 5b and Fig. 7f). Moreover, the stability of the interfaces was preserved across the CG MD trajectories (Supplementary Figs. 6d, 7d, e).

Fig. 7. The mGlu2 receptor active state is stabilized by different class A GPCRs.

a Stabilization of the mGlu2 extracellular domain active state upon co-expression of the indicated class A GPCRs measured by the FRET signal between two nanobodies that selectively bind on the active mGlu2 receptor. Mock indicates the conditions where the mGlu2 cDNA was cotransfected with the empty pRK5 plasmid. The receptors that induced significant change of FRET signal were shown in red. Data were normalized to Mock and shown in means ± SEM of triplicates from at least three independent experiments (n = 3). One-way ANOVA with Dunnett’s multiple comparisons test with following ****P < 0.0001, *P = 0.017, **P = 0.0041, *P = 0.0287, ****P < 0.0001, ****P < 0.0001, ****P < 0.0001, ****P < 0.0001, **P = 0.0048. The other data being not significant. b, d Percentage of active ^HmGlu2 receptor in the presence of increasing amount of co-expressed ^SD1 (b) or ^SM2 (d) receptor revealed by the nanobody-based sensor in (a) in the absence of ligands (red) and in the presence of LY379268 (10 μM, green) or LY341495 (10 μM, blue). The expression (Exp) ratio of the co-transfected of ^SD1 or ^SM2 and ^HmGlu2 were determined by the donor emission after labeling either with Snap-Lumi4-Tb or Halo-Lumi4-Tb. FRET signal was normalized to ^HmGlu2 expressed alone in the presence of LY379268. Data are means ± SEM of triplicates from three independent experiments (n = 3). c, e IP₁ production of HA-tagged mGlu2 co-expressed with increasing amounts of Flag- tagged D1 (c) or M2^R121P (e) and Gqi₉ in basal condition (blue), compared with IP₁ production of increasing amounts of Flag-tagged D1 expressed only with the cotransfected Gqi₉ (black). IP₁ production is proportional to the amount of Flag-tagged D1 (c) at the cell surface as measured by ELISA, in contrast to what observed for Flag-tagged M2^R121P (e). Data are mean ± SEM from a typical experiment performed three times and normalized to mGlu2 expressed alone in response to LY379268 (red, n = 3).

Cys cross-linking to characterize the mGlu2-5-HT_2A complex

We first validated this model of interaction between mGlu2 and 5-HT_2A receptors with receptors expressed at the cell surface of HEK293 cells. First, we show that the mGlu2 agonist LY379268 increases the FRET signal between the two receptors when co-transfected (Fig. 3a and Supplementary Fig. 8). This indicates that additional heteromers between these two receptors are formed, although we cannot exclude that the increase in FRET signal results from a conformational change between the two receptors (Supplementary Fig. 9a). In contrast, we verified that the competitive antagonist LY341495 of mGlu2 did not alter the saturation curve (Supplementary Fig. 9b). This is consistent with our model that only the mGlu2 receptor in the active conformation can form a complex with two 5-HT_2A receptor (Fig. 2e, f).

Fig. 3. mGlu2 receptor in active state is more likely to interact with 5-HT_2A receptor.

a TR-FRET saturation analysis of ^SmGlu2 and ^H5-HT_2A with or without treatment with the indicated mGlu2 receptor agonist LY379268 (10 μM). Data are shown in means ± SEM from three independent experiments (n = 3). b–d Schematic representation of ^HmGlu2 and ^S5-HT_2A constructs used in this study. Halo-tagged mGlu2 wild-type (^HmGlu2^WT), active control (^HmGlu2^Ctrl) and active control mutants (^HmGlu2^Ctrl + Cys mutation) (b) together with Snap-tagged 5-HT_2A wild-type (^S5-HT_2A) and mutants (^S5-HT_2A + Cys mutation) (c) were used to form the mGlu2-5-HT_2A hetero-oligomer (d) by cysteine cross-linking. In ^HmGlu2^Ctrl, the amino acid tyrosine located on 6.58 in TM6 was replaced with cysteine to stabilize the active state of mGlu2 receptor by crosslinking. e, f Cysteine cross-linking of the indicated ^S5-HT_2A and ^HmGlu2^Ctrl constructs at cell surface after labeling with fluorescent Snap substrates. The ^HmGlu2^Ctrl constructs containing or not a cysteine at position 5.42, and ^S5-HT_2A^5.41 contains a cysteine at position 5.41, as indicated. The results were obtained after treatment with mGlu2 receptor agonist LY379268 and CuP. MW, molecular weight. Data are representative of a typical experiment performed three times. (f). Change of the percentage of mGlu2-5-HT_2A hetero-oligomers without (open bars) or with (solid bars) LY379268 (10 μM) was quantified by the total amount of 5-HT_2A protein composed by monomer, dimer and mGlu2-5-HT_2A heteromer. Data are mean ± SEM from at least three independent experiments (n = 5). One-way ANOVA with Dunnett’s multiple comparisons test with **P = 0.001 and ns, not significant. g Cross-linking of the indicated ^S5-HT_2A mutant and ^HmGlu2 wild-type or mutant after labeling with fluorescent Snap substrates. The mutants contain a single cysteine at positions mGlu2^5.42 and 5-HT_2A^5.41. The results were obtained after pre-incubation with mGlu2 receptor agonist LY379268 and CuP. h Change of the percentage of cross-linked mGlu2-5-HT_2A hetero-oligomers at the cell surface in the presence or absence of LY379268 (10 μM). It was quantified by the amounts of ^S5-HT_2A subunits labeled at cell surface by the Snap fluorophore and quantified its emission 800 nm. Data are mean ± SEM from at least three independent experiments (n = 6). One-way ANOVA with Dunnett’s multiple comparisons test with ****P < 0.0001 and ns, not significant.

Next, to further validate our model of the mGlu2-5-HT_2A complex, we performed a cysteine disulfide crosslinking analysis. This technique probes proximity (<8 Å) between engineered cysteine residues, enabling the detection of possible protein-protein interactions on the cell surface. This approach has been successfully applied to study the dimeric 7TM interfaces of several class C GPCRs, including the mGlu2²² and calcium-sensing (CaS) receptor homodimers⁵⁵, the mGlu2-7 heterodimer²⁹ and the GABA_B receptor hetero-oligomer⁵⁶. To detect the crosslinked heterodimer in blots, Snap-tagged ^S5-HT_2A and Halo-tagged ^HmGlu2 subunits were used to allow their specific labeling at the cell surface using non-cell permeant fluorescent substrates. This allows direct fluorescence visualization of the receptors on blots, as previously reported²². Prior to probing the heterodimeric interface, we first designed a mGlu2 mutant that stabilizes the active state of mGlu2 in the absence of agonist. Based on our 3D model, the mGlu2^Y6.58C mutation was introduced into Halo-tagged wild-type mGlu2 (Fig. 3b and Supplementary Fig. 9c, d). This construct, in which the active TM6-TM6 interface is stabilized by a disulfide bridge between the two Cys^6.58, was named “Active control” (mGlu2^Ctrl) (Fig. 3b). We have shown that this mGlu2^Ctrl construct has a strong constitutive activity upon CuP treatment, which promotes disulfide crosslinking, by measuring the constitutive accumulation of inositol phosphate-1 (IP₁; Supplementary Fig. 9e). To verify the presence of disulfide bridge crosslinking between the two Cys^6.58 in the mGlu2 homodimer in blot analysis, we used a construct in which the natural intersubunit disulfide bridge through Cys121 was removed (Supplementary Fig. 9f), as it was done previously²². A majority of mGlu2 dimers were observed after CuP treatment, an effect reduced by an mGlu2 antagonist (Supplementary Fig. 9f). Overall, these results demonstrate that Cys6.58 effectively stabilizes the active conformation of mGlu2 homodimer.

Next, to crosslink the ^HmGlu2^Ctrl and ^S5-HT_2A receptors (Fig. 3c, d), we introduced Cys at positions 5.42 and 5.41 in the ^HmGlu2^Ctrl and ^S5-HT_2A receptors, respectively, based on our 3D model (Supplementary Fig. 9c, d). We quantified on blots the ^S5-HT_2A monomer (M), dimer (D) and ^S5-HT_2A cross-linked to mGlu2 (mGlu2-5HT_2A oligomers; O) thanks to the fluorescent Snap substrates labelling. The efficiency of cross-linking between the two receptors induced by CuP was quantified by the change of the ratio of the amounts for O/(M + D + O) before and after CuP treatment. Interestingly, when co-expressing these two mutants, we obtained a CuP-induced high molecular weight band (over 300 kDa) (Fig. 3e, f) that was suppressed by treatment with the reducing agent DTT (Supplementary Fig. 10a, b). A similar band was obtained with wild-type ^HmGlu2 containing a Cys in position 5.42, but the amount of cross-linked oligomer was clearly increased in the presence of the mGlu2 agonist LY379268 (Fig. 3g, h). This shows the advantage of using the ^HmGlu2^Ctrl construct over wild-type ^HmGlu2 to study the heteromeric interface by crosslinking since it does not require agonist treatment. Finally, the heteromeric composition of this complex was confirmed by labeling the ^HmGlu2 and the ^S5-HT_2A constructs with the fluorescent substrate Halo-Tag Alexa Fluor 660 and Snap-Surface Alexa Fluor 782, respectively (Supplementary Fig. 10c–e). The high molecular weight band was visualized with both fluorophores, confirming it contains both receptor subunits.

Our crosslinking analysis revealed that the heteromeric interface between ^HmGlu2 and ^S5HT_2A involves TM5 of both receptors, while a distinct disulfide bond between TM6 helices stabilizes the active conformation of the mGlu2 homodimer (Fig. 3d). Furthermore, our structural model of the heterocomplex suggests a potential additional interaction interface involving the TM4 helices of both receptors (Fig. 2e, f). This possibility is supported by prior evidence implicating TM4 in the oligomerization interface of active mGlu2 homodimers²². In contrast, the dimerization interface of the 5-HT_2A receptor is unclear⁵⁷, and the experiments above indicate that only TM5 is involved (Fig. 3e–h). Therefore, we investigated the ^S5-HT_2A homodimer interface by Cys crosslinking (Supplementary Fig. 11a), before further studying the mGlu2-5-HT_2A interface. Symmetrical Cys crosslinking showed that TM4 of ^S5-HT_2A was extensively crosslinked in the homodimer, in addition to TM5 and TM1 (Supplementary Fig. 11b–e). These results indicate that 5-HT_2A can homodimerize across two possible symmetrical interfaces involving either TM4-TM5 or TM1 (Supplementary Fig. 11d, e).

The above experiments indicate that TM4 of both mGlu2 and 5-HT_2A receptors may be involved in their direct interactions. We then systematically mapped possible crosslinks involving the TM4 and TM5 of these two receptors (Fig. 4a, b and Supplementary Fig. 12). Our data showed that the greatest number of cross-links were obtained between TM5 of ^HmGlu2 and TM4 of ^S5-HT_2A receptor (Fig. 4a–c and Supplementary Fig. 12). In contrast, no crosslinking was detected between mGlu2 and either TM1 or TM6 of the ^S5-HT_2A receptor (Supplementary Fig. 13a, b). These results show that TM4s and TM5s of mGlu2 and 5-HT_2A receptors are primarily responsible for heteromer formation (Fig. 4d). This is highly consistent to previous findings that Ala4.40, Ala4.44 and Ala4.48 located in mGlu2 TM4 are necessary for the mGlu2 to form a GPCR complex with 5-HT_2A receptor^43,46.

Fig. 4. mGlu2 receptor interacts with 5-HT_2A receptor through their respective TM4 and TM5 to form hetero-oligomer.

a Change of cross-linked hetero-oligomer percentage induced by CuP treatment for the indicated cysteine substitutions in TM4 and TM5 of both ^S5-HT_2A and ^HmGlu2^Ctrl. Positions with a significant change in ^HmGlu2^Ctrl were highlighted in red. Data are mean ± SEM from at least three independent experiments (n = 3). Data were normalized in the consistent way to Fig. 3f. One-way ANOVA with Dunnett’s multiple comparisons test with following **P = 0.0025,**P = 0.0038, ****P < 0.0001, *P = 0.0435, ****P < 0.0001, **P = 0.0096, ****P < 0.0001, ****P < 0.0001, ****P < 0.0001. The other data being not significant. b Cross-linking of the indicated positive mutants of ^S5-HT_2A and ^HmGlu2^Ctrl labeled with fluorescent Snap substrates, without (−) or with (+) CuP treatment. 5-HT_2A monomer, dimer and mGlu2-5-HT_2A hetero-oligomer were separated via SDS-PAGE under non-reducing conditions. Data are representative of a typical experiment performed three times. c Summary of the number of crosslinked ^S5-HT_2A and ^HmGlu2^Ctrl mutants. d Cartoon to show the heteromerization interface in mGlu2-5-HT_2A complex based on the cross-linking results. TMs that can be cross-linked of mGlu2 and 5-HT_2A are highlighted in red and dark red respectively, and mGlu2 TM6 are highlighted in yellow. The second 5-HT_2A which is also possible to interact with another protomer of mGlu2 is shown in dotted lines.

In summary, our 3D models of the mGlu2-5-HT_2A complex, together with crosslinking evidence, have strongly demonstrated that mGlu2 can interact directly with 5-HT_2A to form a heteromer at the cell surface via their TM4-TM5 interface. Based on cross-linking results for the 5-HT_2A homodimer, we cannot exclude that a 5-HT_2A homodimer stabilized by TM1-TM1 interface is formed in the mGlu2-5-HT_2A complex (Supplementary Fig. 14).

5-HT_2A stabilizes the mGlu2 receptor active state

We used our recently developed nanobody-based FRET sensor⁵⁸ to monitor the active conformation of the mGlu2 receptor by detecting the active state of the extracellular glutamate binding domain (Fig. 5a). This sensor is based on the measurement of a FRET signal between a pair of fluorescent anti-mGlu2 nanobodies in the presence of LY379268 (Fig. 5a, b). The DN10 nanobody specifically binds the active conformation of the mGlu2 receptor favored by the agonist, while the DN1 is not sensitive to the conformational state of the mGlu2 receptor^58,59. Interestingly, the increase in transfected 5-HT_2A receptor in the cell in the presence of a fixed amount of cotransfected mGlu2 receptor corresponded to an increase in FRET signal (Fig. 5b and Supplementary Fig. 15), even in the absence of LY379268. This effect was prevented by the mGlu2 antagonist (Fig. 5b and Supplementary Fig. 15), but was not influenced by 5-HT_2A agonist or antagonist (Fig. 5c). Altogether, these data demonstrate that 5-HT_2A receptor, either in its inactive or active conformation, stabilizes the active conformation of the mGlu2 receptor. To rule out the possibility that the DN10 nanobody is responsible for this increase in FRET by stabilization of the active mGlu2 receptor, we monitored the conformational rearrangement of the mGlu2 extracellular binding domain using our Snap-tagged mGlu2 sensor^59–61. Consistent with the nanobody-based FRET sensor results, the increase in transfected 5-HT_2A receptor in the presence of a fixed amount of cotransfected mGlu2 receptor decreased TR-FRET ratio of ^SmGlu2 dimers (Supplementary Fig. 16a, b) and increased in the proportion of the mGlu2 receptors in an active state (Supplementary Fig. 16a, c). This effect occurs at lower concentration of 5-HT_2A than when measured with the nanobody-based assay, a difference that can be explained by the different timing, the nanobody-based assay being measured after overnight incubation with the nanobodies. Of importance, this effect was not observed with the M2 receptor used as a negative control (Supplementary Fig. 16d, e), since it does not have a tendency to interact directly to the mGlu2 receptor.

Fig. 5. mGlu2 receptor is stabilized in active state by co-expressed 5-HT_2A receptor.

a Schematic representation of the inactive and active states of mGlu2 detected by the nanobody-based TR-FRET sensor and modulated by 5-HT_2A. In presence of mGlu2 antagonist LY341495 (10 μM), no FRET signal is expected since the nanobody DN10-d2 does not bind to the inactive conformation of mGlu2 receptor. In contrast, when ^SmGlu2 receptor is stabilized in the active state by the agonist LY379268 (10 μM) or by the co-expressed ^H5-HT_2A, binding of DN10-d2 induces a strong FRET signal. b Percentage of active ^SmGlu2 receptor in the presence of increasing amount of co-expressed ^H5-HT_2A receptor revealed by the nanobody-based sensor shown in (a), in the absence of ligands (red) and in the presence of LY379268 (10 μM, green) or LY341495 (10 μM, blue). The expression ratio of the co-transfected ^SmGlu2 and ^H5-HT_2A were determined by the donor emission after Snap-Lumi4-Tb and Halo-Lumi4-Tb labeling, respectively. FRET signal was normalized to ^SmGlu2 expressed alone in the presence of LY379268. Data are means ± SEM of triplicates from three independent experiments (n = 3). c The 5-HT_2A receptor agonist 5-HT (10 μM, blue) or antagonist ketanserin (50 μM, purple) have no effect on the activation of mGlu2 in presence of 5-HT_2A. The expression ratio of ^H5-HT_2A:^SmGlu2 was 1.1 as shown in (b). Data are means ± SEM of triplicates from three independent experiments (n = 3). One-way ANOVA with Dunnett’s multiple comparisons test with **P = 0.0089 and ns, not significant. d Percentage of active ^SmGlu2 receptor when ^H5-HT_2A wild-type or ^H5-HT_2A^R173P mutant are co-expressed, in the absence of ligand (red) or in the presence of LY379268 (10 μM, green). The expression ratio of ^H5-HT_2A:^SmGlu2 was 1.4 as shown in panel b. Data are mean ± SEM from at least three independent experiments. Data are means ± SEM of triplicates from more than three independent experiments (n = 5). One-way ANOVA with Dunnett’s multiple comparisons test with following ****P < 0.0001, *P = 0.0492, *P = 0.0161 and ns, not significant.

This conclusion is further supported by the behavior of a 5-HT_2A receptor mutant (R173P) located in the intracellular part of TM3. This mutant is deficient in G protein coupling (Fig. 6a), but it can still stabilize the active state of the mGlu2 receptor, as observed by the increased of FRET signal (Fig. 5d). Of note, this 5-HT_2A R173P mutant corresponds to the mutation of the Arg residue of the DRY motif that is highly conserved in class A GPCRs^62,63. Although mutation R173 into proline has not been reported to our knowledge in the 5-HT_2A receptor, mutation of this residue in other class A GPCRs strongly impairs G protein coupling⁶³. Here, we show that R173P mutant was expressed at the cell surface similarly to the wild-type (Supplementary Fig. 17a). Furthermore, this mutation prevents the transactivation of 5-HT_2A receptor by the LY379268 activated mGlu2 receptor (Supplementary Fig. 17b-e), as measured by the intracellular calcium release induced by mGlu2 in the presence of co-transfected 5-HT_2A receptor⁴⁴.

Fig. 6. 5-HT_2A receptor enhances the basal activity of coexpressed mGlu2 receptor.

a IP₁ accumulation of 5-HT_2A wild-type or 5-HT_2A^R173P mutant without (gray) or with (red) 5-HT (10 μM) treatment in the absence of Gqi₉. Data are mean ± SEM from three independent experiments (n = 3). One-way ANOVA with Dunnett’s multiple comparisons test with ****P < 0.0001 and ns, not significant. b IP₁ production of ^SmGlu2 co-expressed with increasing amounts of Flag-tagged 5-HT_2A^R173P and Gqi₉ in basal condition (blue) in HEK293 cells. IP₁ production is proportional to the amount of Flag-tagged ^H5-HT_2A^R173P at the cell surface, as measured by ELISA. Data are mean ± SEM from a typical experiment performed three times and normalized to mGlu2 expressed alone in response to LY379268 (red; n = 3). c IP₁ accumulation of ^SmGlu2 co-expressed or not with ^H5-HT_2A^R173P, and Gqi9, in absence or presence of the indicated mGlu2 ligands, LY379268 (10 μM, green), LY341495 (10 μM, light blue) and the negative allosteric molecule MNI 137 (50 μM, purple). Mock correspond to cells transfected only with Gqi₉ cDNA. Data are means ± SEM of triplicates from four independent experiments (n = 4). One-way ANOVA with Dunnett’s multiple comparisons test with *P = 0.037, **P = 0.0014, ***P = 0.0008. d Intracellular calcium release by ^SmGlu2 in HEK293 cells co-expressed (red) or not with ^H5-HT_2A^R173P (blue) and Gqi₉ in response to mGlu2 agonist LY379268. Data are mean ± SEM from a typical experiment performed three times and normalized to the Emax value of mGlu2 expressed alone (n = 3).

We then confirmed the stabilization of the active state of mGlu2 receptor by 5-HT_2A by measuring the accumulation of inositol phosphate-1 (IP₁), using a chimeric Gqi₉ protein to couple mGlu2 to phospholipase C (PLC)²². The 5-HT_2A R173P mutant increases the basal activity of the co-expressed mGlu2 receptor (Fig. 6b), which can be blocked by the competitive antagonist LY341495 or a negative allosteric modulator that binds to the mGlu2 transmembrane domain²² (Fig. 6c). This increase in basal activity is rather modest, but it is significant. This finding is consistent with other data above showing that 5-HT_2A receptors stabilize only a fraction of the mGlu2 receptors in an active state (Fig. 5b and Supplementary Fig. 15a). In addition, as expected from a low fraction of the mGlu2 receptors interacting with the 5-HT_2A R173P mutant, no significant increase of the potency of the mGlu2 agonist by co-expression of the 5-HT_2A R173P mutant is observed (Supplementary Fig. 17f). Finally, a lower intracellular calcium mobilization by the LY379268 activated mGlu2 receptor in presence of Gqi₉ and the 5-HT_2A R173P mutant is measured (Fig. 6d), and it is not related to the mGlu2 cell surface expression level (Supplementary Fig. 17g). This observation supports the effect of this 5-HT_2A mutant in increasing the basal activity of mGlu2 receptor by a direct interaction. Indeed, it is recognized that a basal activation of PLC, such as that resulting from a GPCR constitutive activity, can partly empty the intracellular calcium stores, leading to a decrease in the calcium signal induced by the activation of a PLC-coupled GPCR.

Altogether, these data demonstrate that 5-HT_2A acts as a positive allosteric modulator on mGlu2 receptor activation even under basal conditions where only ambient glutamate may be present⁶⁴.

The mGlu2 receptor active state is stabilized by other class A GPCRs

In order to clarify which GPCRs could interact with mGlu2 receptor at the cell surface, we performed a systematic analysis with 44 different class A GPCRs tagged at their N terminus with a Snap-tag, as illustrated in Fig. 7a. By using the mGlu2 nanobody-based FRET sensor, a number of these GPCRs generated a high FRET signal between the two mGlu2 nanobodies that selectively bind on the active mGlu2, when coexpressed with HA-tagged mGlu2 receptor, including 5-HT_2A, the dopamine D1 to D4 receptors, while others generated lower but significant signals, such as the 5-HT_2B, the adrenergic α_1A and the delta opioid receptors. All others did no generate a signal significantly higher than the background, including the histamine H1 to H3 receptors and the muscarinic M2 to M4 receptors. This was not explained by differences in the amount of the receptors at the cell surface (Supplementary Fig. 18a, b) and there is no link with the N-terminal length of these class A GPCRs such as short N-terminus for D1 receptor made of 20 residues, until the long N-terminus of the cannabinoid CB1 receptor which is formed by 111 residues.

We found that the Gs-coupled D1 receptor has a strong tendency to stabilize the active conformation of mGlu2 receptor (Fig. 7a), in an expression level dependent manner (Fig. 7b). Consistently with this stabilization, the D1 receptor increased the basal accumulation of IP₁ by the co-expressed mGlu2 receptor (Fig. 7c), an increase that was not observed when the D1 receptor was expressed alone. This basal activation of mGlu2 receptor by D1 receptor is also supported by the lower calcium mobilization of the agonist-induced mGlu2 when D1 receptor is coexpressed (Supplementary Fig. 18c) despite a similar expression of mGlu2 (Supplementary Fig. 18d). As for 5-HT_2A receptor, no significant increase of the potency of the mGlu2 agonist by co-expression of the D1 receptor is observed (Supplementary Fig. 18e), consistent with a low fraction of the mGlu2 receptors interacting with the D1 receptor. Altogether, these data indicate the effect of D1 receptor increases the constitutive activity of mGlu2 by a direct allosteric interaction.

To further support the specific allosteric effect of both 5-HT_2A and D1 receptors on mGlu2 receptor, we used the Gi/o-coupled M2 receptor, one of the class A GPCRs that generated no increase of FRET signal by the nanobody-based mGlu2 sensor (Fig. 7a). This is consistent with the low FRET signal between the M2 and mGlu2 receptors shown above (Fig. 1b and Supplementary Fig. 3) showing that the M2 receptor is less prone to interact with mGlu2 receptor than 5-HT_2A receptor. Increasing the M2 receptor versus mGlu2 receptor expression ratio did not result in an increase in the nanobody-based FRET sensor for mGlu2 activation (Fig. 7d). Consistent with this absence of effect, a M2 mutant dead for the coupling to G protein (R121P also mutated in the DRY motif; Supplementary Fig. 19a), did not increase the basal accumulation of IP₁ by the co-expressed mGlu2 receptor (Fig. 7e), despite its similar expression at the cell surface compared to that of the wild-type M2 receptor (Supplementary Fig. 19b). Furthermore, this M2 R121P mutant did not alter the potency of the mGlu2 agonist in IP₁ accumulation (Supplementary Fig. 19c) and in the calcium responses (Supplementary Fig. 19d), despite a similar expression of mGlu2 (Supplementary Fig. 19e).

Altogether, this demonstrates that mGlu2 receptor not only can be allosterically activated by 5-HT_2A receptor, but also by other class A GPCRs, even under basal conditions where only ambient glutamate may be present⁶⁴.

Discussion

Allosteric interactions between protomers within GPCR heterocomplexes are known to modulate both agonist binding and G protein coupling⁶⁵. However, the molecular mechanisms underlying these processes remain elusive. The existence of only a few heterodimers having been demonstrated in native tissues^41,66,67. In this study, we indeed choose the mGlu2-5HT_2A complex already well-characterized^38,42,53 to bring evidence of how a 7TM domain can control the conformation of another one. We investigate the molecular basis and functional implication for G protein coupling within this heteromer. Our results demonstrate that the 5-HT_2A receptor can act as a positive allosteric modulator by stabilizing the active conformation of mGlu2 in the complex.

Our results clarify the molecular bases for the direct interaction between mGlu2 and 5-HT_2A receptors although this has been well-investigated previously especially by Gonzalez-Maesoʼs group³⁸. For instance, the TM4 of mGlu2 have been proposed to be necessary for the formation of GPCR heteromers with 5-HT_2A^38,43,45. 5-HT_2A TM4 have also been reported to associate directly with mGlu2 receptor⁴⁴. Furthermore, independent studies have shown that agonist-dependent phosphorylation of mGlu2 at Ser843 is enhanced specifically in the presence of 5-HT_2A receptor, providing additional biochemical evidence for a direct interaction⁶⁸. However, in all these previous studies, it remained unclear whether the direct interaction between these two receptors could only occur in a given conformational state of the mGlu2 receptor, possibly leading to its direct activation.

No high-resolution structure of a GPCR heteromer has been determined to date. Indeed, most GPCRs form highly dynamic complexes^10–12 and it is therefore difficult to solve the structures made by these transient receptor dimers¹⁸. Although 3D models of the mGlu2-5-HT_2A complex have been proposed^38,44, but only a limited validation by site-directed mutagenesis and photocrosslinking^43,45 has been performed. Critically, these models did not address whether the interaction is dependent on a specific conformational state of mGlu2. Our study proposes and validates a 3D model of a membrane receptor heteromeric complex both in the inactive and active states in the absence of its structural analysis by biophysical techniques such as X-ray or cryo-EM. We combined molecular modelling and molecular dynamics to provide a 3D model of the heteromer, a recent approach that has been used for other GPCR heterodimers⁶⁹. Then, we validated this model with a biochemical approach based on engineered Cys crosslinking that has been nicely used to probe transient 7TM interface of several GPCRs^13,70. Indeed, it is a powerful approach to map close contacts (distance below 8 Å between the Cβ of both cysteines) within transient 7TM interfaces, thus revealing possible and precise interfaces between the transmembrane domains, in absence or presence of ligands that stabilize their inactive or active states^22,56. A strong advantage of our analysis is that it is conducted only on receptors that are present at the surface of living cells, due to the labeling of cell surface receptors with not cell permeant fluorophores.

Our results also reveal that Cys cross-linking approach is interesting for producing a realistic 3D model of the transient heteromers formed between a constitutive class C GPCR dimer and a class A GPCR. Indeed, as mGlu subunits form constitutive homodimers²⁶, they have a reduced number of interfaces available to form a complex with another GPCR. Specifically, TM4 and TM5 of mGlu2 participate in the inactive-state homodimer interface, whereas the two TM6 helices form the active-state interface^20,22,71. We have recently shown that we can precisely control the mGlu2 7TM interface through Cys crosslinking. For example, crosslinking between the two TM6 helices stabilizes the active conformation, thereby exposing TM4 and TM5 outside the homodimer interface and rendering them accessible for complex formation with other transmembrane protein²².

Our crosslinking results reveal how the inactive interface within the mGlu2 homodimer is used by 5-HT_2A to interact directly and form a heteromer. There is a major reorientation of the 7TM dimer interface in the class C GPCR, with the receptor rapidly oscillating between an inactive TM4-TM5 interface and an active TM6 interface^61,72. Therefore, another GPCR that could interact with the TM4-TM5 interface of a class C GPCR could stabilize it in the active state, as observed between mGlu2 and 5-HT_2A. This suggests a general mechanism or principle whereby many other GPCRs capable of interacting with TM4-TM5 of mGlu2 may behave like PAMs. This is what we have observed with the systematic analysis of many other class A GPCRs important drug targets, with many of them being able to stabilize the ECD mGlu2 active state. We further validated this principle by demonstrating that the dopamine D1 receptor is also able to induce constitutive activity of mGlu2 receptor, behaving as a PAM similarly to 5-HT_2A. This regulatory mechanism likely extends to other class C GPCRs such as mGlu5 that has been reported to be stabilized in the active state by direct interaction with the dopamine D1 receptor³⁹. This could also be the case for dopamine D2 and A_2A receptors, which have been reported to interact with mGlu5⁴⁰. Overall, understanding the dynamics of GPCR transmembrane interfaces is essential for elucidating the structures of receptor complexes and their allosteric interactions. Moreover, it would be of strong interest to design drugs that target these interfaces, as has recently been illustrated for other class C GPCRs^28,32,73.

Our results clarify the molecular and structural basis for stabilizing the active state of a GPCR through interaction with another transmembrane protein. Similar to other class C GPCRs, mGlu2 receptor possess a large dimeric extracellular domain oscillating between inactive and active states, and the rapid dynamics of this conformational change has been revealed by various FRET sensors in populations⁶⁰ or at the single-molecule scale^61,72. While these sensors require receptor modification, others are compatible with the native mGlu2 receptor, such as those recently from nanobodies⁵⁸. Interestingly, the data from our study show that the presence of the 5-HT_2A receptor is sufficient to stabilize the active state of the mGlu2 receptor, in line with the predominance of mGlu2 signaling in mGlu2-5HT_2A complex interactions observed in vivo⁷⁴. Indeed, this stabilization is similarly observed in the presence or absence of 5-HT_2A ligands, and in the presence of a 5-HT_2A receptor that may or may not couple to a G protein.

Up to now, the allosteric interactions between 7TM domains start to be understood with the structures of the constitutively dimeric class C receptors where an asymmetric interface is observed^19,20,37. In these structures, only one of the two 7TM protomers reach the active state when stabilized either by G protein coupling or a positive allosteric modulator, or both. However, the structural basis of interactions between two different GPCRs in a heterodimer remains unclear due to the lack of direct structural data. To date, structural insights are limited to a few complexes of class A or class B GPCRs with single transmembrane domain proteins, that have been solved^75–77, shedding light on how these membrane protein partners modify the binding and activation of these GPCRs.

In conclusion, our study demonstrates that GPCRs can allosterically control the active conformation and signaling of other GPCRs via their transmembrane domain in dimeric or higher-order oligomeric complexes. This mechanism has important implications for the understanding of many other GPCR heteromers already proposed in the literature, including hetero-oligomeric complexes formed between GPCRs and other transmembrane proteins. Our findings thereby open a path to drug discovery through small molecules (allosteric modulators) that can modulate signaling pathways associated with proteins other than those to which they bind. Finally, our structural and functional study highlights a fundamental question in cell signaling that lies beyond the GPCR landscape, namely the use by living systems of allosteric protein-protein interactions as a key mechanism to fine-tune cellular responses in both health and disease conditions⁷⁸.

Methods

Materials

LY379268, LY341495, MNI137, 5-HT (5-Hydroxytryptamine), Ketanserin tartrate were purchased from Tocris Bioscience (Bristol, UK). Dichloro (1,10-phenanthroline) copper(II) (CuP) were purchased from Sigma-Aldrich (St. Louis, MO, USA). Muscarine were purchased from MedChemExpress (Shanghai, China). Lipofectamine 2000 was obtained from Thermo Fisher Scientific (Carlsbad, CA, USA). Snap-Surface® Alexa Fluor® 782 was from New England Biolabs. Halo-Tag® Alexa Fluor®660 and Coelenterazine-h were from Promega (Beijing) Biotech Co. IP-One Gq kit, Snap-Green, Snap-Lumi4-Tb, Snap-Red, Halo-Green and Halo-Lumi4-Tb were from Revvity Cisbio (Codolet, France), respectively.

Plasmids and transfection

The PRK5 plasmids encoding wild-type rat mGlu2, with HA and Snap or Flag and Halo inserted just after the signal peptide were constructed as previously reported⁵⁶. The plasmids encoding HA and Snap or Flag and Halo tagged wild-type rat 5-HT_2A were obtained in the same way. All the Cysteine mutants for mGlu2 and 5-HT_2A together with G protein activation defective mutant 5-HT_2A^R173P and M2^R121P were generated by site-directed mutagenesis using the QuikChange mutagenesis protocol (Agilent Technologies). All constructs were confirmed by DNA sequencing.

HEK293 cells (ATCC, CRL-1573, lot: 3449904) were cultured in DMEM (Gibco) supplemented with 10% FBS (Gibco) and 1% penicillin/streptomycin (Gibco). For cross-linking assay, HEK293 cells were transfected in 12 well plate with plasmids of interest by Lipofectamine 2000. For IP₁ accumulation measurements, cells were transfected with the chimeric G-protein Gqi9 and the glutamate transporter EAAC1 to enable mGlu2 receptor couple to the phospholipase C pathway²².

Fluorescence resonance energy transfer (FRET)

A total of 30,000 HEK293 cells per well in black non-transparent 96-well plates are transfected with fixed 20 ng of ^SmGlu2 plasmid cDNA and increasing amounts of ^HmGlu2, ^H5-HT_2A or ^HM2R plasmid cDNA with 0, 5, 10, 20, 40, 60, and 80 ng, and completed to a total amount of 100 ng of plasmid DNA per well with the empty vector pRK5. 24 hours after transfection, HEK293 cells were labeled with solution of 100 nM of Snap-Lumi4-Tb and 60 nM of Halo-Green in Tag-lite buffer at 37 °C for 1 h. Afterwards, cells were washed three times with FRET buffer (20 mM HEPES, 146 mM NaCl, 4.2 mM KCl, 1 g/L glucose, 0.1% BSA, pH adjusted to 7.4), and drugs were added to investigate effects of agonist LY379268 and antagonist LY341495. Fluorescence of Lumi4 (excitation at 340 nm, emission at 620 nm, 50 μs delay, and 450-μs integration time), Green (485 nm, 520 nm, 0 μs, and 1000 μs), and TR-FRET (340 nm, 520 nm, 50 μs, and 450 μs) were collected on an Infinite F500 spectrofluorimeter (Tecan, Männedorf, Switzerland). The expression of Snap-tagged and Halo-tagged receptors were determined by the fluorescence intensity of Snap-Lumi4-Tb and Halo-Green, respectively⁵⁰.

Cell surface quantification

The expression of HA-Snap tagged or Flag-Halo tagged constructs in cell surface were detected by ELISA or fluorescent labeling⁵⁶ as previously described.

Nanobody based TR-FRET assay

A total of 30,000 HEK293 cells per well in white non-transparent 96-well plates were transfected with fixed 20 ng of ^SmGlu2 plasmid cDNA and increasing amounts of ^HmGlu2 plasmid cDNA with 0, 5, 10, 20, 40, 60 and 80 ng, and completed to a total amount of 100 ng of plasmid DNA per well with the empty vector pRK5. 24 h after transfection, HEK293 cells were treated with or without drugs and then labeled with solution of 7 nM of DN1- Lumi4-Tb and 15 nM of DN10-d2 in Tag-lite buffer at 25 °C overnight⁵⁸. Afterwards, cells were washed three times with Tag-lite buffer. FRET measurements were performed using the PHERAstar FS (BMG Labtech, German). Calculate the ratio of the acceptor and donor emission signals for each individual well. Ratio=Signal 665 nm/Signal 620 nm * 10⁴.

TR-FRET based mGlu2 conformational assay

30 000 HEK293 cells per well in black non-transparent 96-well plates are transfected with 20 ng of ^SmGlu2 plasmid cDNA and increasing amounts (0, 5, 10, 20, 40, 60 or 80 ng) of ^H5-HT_2A or ^HM2 plasmid cDNA, and completed to a total amount of 100 ng of plasmid DNA per well with the empty vector pRK5. 24 h after transfection, HEK293 cells were labeled with solution of 100 nM of Snap-Lumi4-Tb and 60 nM of Snap-Green in Tag-lite buffer at 37 °C for 1 h. Afterwards, cells were washed three times with Tag-lite buffer, and drugs were added to investigate effects of agonist LY379268 (10 μM) and antagonist LY341495 (10 μM). The TR-FRET measurements were performed on a PHERAstar FS microplate reader with the following setup: after excitation with a laser at 337 nm (40 flashes per well), the fluorescence was collected at 520 nm for a 50 μs reading after a 50 μs delay after excitation (window 1) or for a 400 μs reading after a 1,200 μs delay (window 2). The FRET signal was determined by dividing the signal measured in window 1 by the signal measured in window 2. Besides, the fluorescence of Snap-Lumi4-Tb and Halo-Lumi4-Tb (excitation at 340 nm, emission at 620 nm, 50 μs delay, and 450 μs integration time) were collected on a PHERAstar FS (BMG Labtech) to determine the expression of Snap-tagged and Halo-tagged receptors, respectively, in separate batches of the same transfected cells.

Inositol phosphate measurements

A total of 30,000 HEK293 cells per well in white non-transparent 96-well plates were transfected with 20 ng ^SmGlu2 and 60 ng ^H5-HT_2A^R173P of plasmid cDNA, or with fixed 20 ng of ^HmGlu2 plasmid cDNA and increasing amounts of ^SD1 or ^SM2^R121P plasmid cDNA with 0, 5, 10, 20, 40, 60 and 80 ng, and completed to a total amount of 100 ng of plasmid DNA per well with the empty vector pRK5. IP₁ accumulation in HEK293 cells was measured in the PHERAstar FS (BMG Labtech, German) by using the IP-One HTRF kit (Revvity CisBio) according to the manufacturer’s recommendations. For cross-linking experiments, cells were treated with 1.5 mM CuP in cross-linking buffer for 10 min (as described above for blot experiments) before measuring.

Intracellular calcium release

The intracellular calcium release measurements were performed as previously described⁷⁹. A total of 30,000 HEK293 cells per well were in black transparent 96-well plates were transfected with plasmids encoding the indicated receptors (mGlu2, 5-HT_2A^R173P, D1 and M2^R173P) with chimeric protein Gqi₉, and then incubated for 24 h. The cells were then preincubated for 1 h with Ca²⁺-sensitive Fluo-4 (Life Technologies). The fluorescence signals (excitation at 485 nm and emission at 525 nm) were measured for 60 s and recorded using a Flexstation 3 microplate reader (Molecular Devices, Sunnyvale, CA, USA). After the first 20 s, cells were treated with agonist and the Ca²⁺ signals were indicated as stimulated increase in fluorescence response. Concentration response curves were fitted using Graph Pad Prism.

Cross-linking and fluorescent-labeled blot experiments

200 000 HEK293 cells per well, seeded in 12-well plates, were transfected with wild type or mutant 1.05 μg ^S5-HT_2A and 0.15 μg ^HmGlu2. After 36 h transfection, adherent HEK293 cells plated in 12-well plates were labeled with 100 nM Snap- Alexa Fluor 782 or 100 nM Halo- Alexa Fluor 660 in culture medium at 37 °C for 2 h. Then, cells were incubated with drug (each at 10 μM) or PBS at 37 °C for 30 min. Afterwards, cells were treated with cross-linking buffer (1.5 mM Cu(II)-(o-phenanthroline), 1 mM CaCl₂, 5 mM Mg²⁺, 16.7 mM Tris, pH 8.0, 100 mM NaCl) at room temperature for 20 min. After incubation with 10 mM N-ethylmaleimide at 4 °C for 15 min to stop the cross-linking reaction, cells were lysed with buffer (containing 50 mM Tris (pH 7.4), 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate) supplemented with complete phosphatase inhibitor cocktail (Roche) at 4 °C for 1 h. After centrifugation at 12,000 × g for 30 min at 4 °C, supernatants were mixed with loading buffer at 37 °C for 10 min. In reducing conditions, samples were treated with additional 100 mM DTT in loading buffer for 10 min. Equal amounts of proteins were separated by 6% SDS-PAGE and then transferred to nitrocellulose membranes (Millipore). Membranes were imaged on Odyssey CLx imager (LI-COR Bioscience, Lincoln, NE, USA) at 700 nm and 800 nm⁵⁶. The density of bands was calculated by ImageJ software (Bethesda, MD, USA).

Homology modeling and assembly of mGlu2-5-HT_2A complex model

To generate the 3D model of mGlu2-5-HT_2A complex, wild-type human mGlu2 7TM and full-length 5-HT_2A were homology modeled form the crystal structure of mGlu1 7TM (PDB: 4OR2)⁸⁰ and 5-HT_2B (PDB: 4IB4)⁸¹ using Modeller v9.16, respectively⁸². Prior to modeling, co-crystallized ligand ergotamine and non-native fusion protein between TM5 and TM6 were removed. Secondary structure of non-crystallized intracellular loop 3 (ICL3) of 5-HT_2B (Tyr249 to Gln314) was consensus predicted with PsiPred⁸³, PSSPred⁸⁴ and Jpred⁸⁵ and then added with Modeller⁸². Sequence alignments between mGlu1:mGlu2 and 5-HT_2B:5-HT_2A were generated with PROMALS-3D⁸⁴ and manually curated to ensure all insertions/deletions occurred in loop regions only. For both receptors, from each structural template, 50 different homology models were generated and ranked by positional criteria and zDOPE score⁸⁶. For all systems, complete GPCR homology models were subsequently energy minimized in the AMBER14SB force-field⁸⁷ using Chimera v1.11 in vacuum conditions⁸⁸.

The activated mGlu2 homodimer model was superimposed onto the respective monomers of the homodimeric μ-opioid receptor crystal structure (PDB: 4DKL)⁴³. This homodimer model was optimized through protein-protein docking with ROSIE web-server⁸⁵, with the best fitting homodimer selected based on highest ROSETTA I_sc score⁸⁹ and lowest RMSD from 1000 docking solutions. After independent all-atom molecular dynamics simulations, conformations of 5-HT_2A monomer and active mGlu2 homodimer were used to model the mGlu2-5-HT_2A heteromeric complex by superimposing individual 5-HT_2A and one protomer of mGlu2 dimer onto the homodimeric adenosine A_2A receptor crystal structure (PDB: 4EIY)⁹⁰, which was deemed the most appropriate crystal structure available at the time. Protein-protein docking between the two receptors was then performed with the ROSIE web-server⁸⁵ for identification of the best possible arrangement. The final oligomeric model was energy minimized without restraints in Chimera v1.11 with the AMBER14SB force-field to optimize protein–protein interactions.

To further validate our proposed model above, we built another model for the complex by using cryo-EM structures of mGlu2 homodimer and 5-HT_2A receptor published after the beginning of this study. We used the inactive conformation of 5-HT_2A (PDB: 7WC8)⁵⁴ and the active conformation of mGlu2 (PDB: 7MTR)²⁰ where dimeric 7TM interface is TM6-TM6. For both receptors, ligands and any non-protein atoms were removed, residues or loops missing from the structures were added with Modeller, and mutations were reverted to wild-type in Chimera⁸⁸. The Venus flytrap domains (VFTs) of the mGlu2 protomers were removed, so that the model of the complex involved only the transmembrane domains of both receptors. Indeed, the VFTs control the relative position of the 7TMs from each subunit, but this is already established the 7TM dimer used as template, and the VFTs are not expected to be in direct contact of the 5-HT_2A receptor. Also, it is worth noting that G proteins were not present in the mGlu2 active structure used for the modeling (PDB: 7MTR). Following the same strategy used to build the previous model, 5-HT_2A and a protomer of mGlu2 were superimposed on a homodimer of the A_2A receptor (PDB: 4EIY) to form a heteromer with a TM4/5-TM4/5 interface between these two receptors. This model of the mGlu2-5-HT_2A complex was then submitted to the “Refinement” tool on the Haddock webserver⁹¹, following the “Simulated Annealing with centroid restraints” protocol, using the default parameters. The structure with the most favorable Haddock score was selected.

One difference between the two models is the position of TM4 of the 5-HT_2A receptor. In homology model, this helix is located between TM5 of one mGlu2 protomer and TM6 of the other protomer. While in the new model, 5-HT_2A TM4 moves towards to the TM5 of mGlu2 protomer, which explains the higher number of frequent contacts between mGlu2 TM5 and 5-HT_2A TM4 observed in this case (Supplementary Fig. 7-8). This difference probably results from its conformation in the homology model, which predicted a shorter helix than the one observed experimentally. However, despite these differences and the inherent limitations of homology modeling, there were still several frequent contacts between mGlu2 and 5-HT_2A which were common between the two models, and among them many involved TM5 of mGlu2 and TM4 of 5-HT_2A. Therefore, the first model adequately represented the interfaces of the complex, and captured contacts between the protomers that are in agreement with the experimental results.

All-atom molecular dynamics (MD) simulations and coarse-grained (CG) analysis

All-atom MD simulations of 5-HT_2A, mGlu2 homodimer alone and mGlu2-5-HT_2A complex model were performed using the CHARMM36 force-field⁹² with ACEMD⁹³ on specialized GPU-computer hardware. All systems were equilibrated for 28 ns each (except mGlu2-5-HT_2A hetero-complex model, which was equilibrated for 56 ns) at 300 K and 1 atm, with positional harmonic restraints on protein heavy atoms progressively released over the first 8 ns (first 16 ns for mGlu2-5-HT_2A complex model) of equilibration and then continued without constraints. After equilibration, 5-HT_2A monomer, mGlu2 homodimer alone and mGlu2-5-HT_2A heteromer model were subjected to unbiased continuous production runs under the same conditions for 5 µs each. Analysis of 5-HT_2A monomer, mGlu2 homodimer alone and mGlu2-5-HT_2A heteromer model MD simulations was performed using VMD software 1.9.2⁹⁴. The protomer conformational change of each system was respectively measured by means of root mean square deviation (RMSD).

The two hetero-complex models, which are based on homology modeling and cryo-EM structures respectively, were further analyzed through CG MD simulations set up in the CHARMM-GUI web server⁹⁵ and performed with GROMACS v.2019.6⁹⁶. The systems were described by Martini 2.2 representation^97,98, with the elnedyn elastic network⁹⁹ and inserted into a POPC membrane. The elastic network connects the beads representing the backbone of the protein, preventing conformational changes; therefore, the initial conformations of both receptors were preserved throughout the trajectories. However, it is important to note that the network only connects beads inside a same protein chain and does not prevent the relative motions between protomers. The final composition of each system comprised around 20000 atoms, in a simulation box with dimensions of approximately 15*15*10 nm³. For each system, four independent replicas were simulated. In each replica, after 10000 energy minimization steps, a 1 ns simulation in the NVT ensemble was performed, with a time step of 2 fs and position restraints on the protein backbone beads (force constant of 1000 kJ*mol⁻¹*nm⁻²) and on the lipid heads (force constant of 200 kJ *mol⁻¹*nm⁻²). Next, in three consecutive equilibration steps in the NPT ensemble totaling 20 ns, position restraints were maintained but the time step was increased to 4, 10 and finally 20 fs, which was kept in all subsequent simulations. Another 20 ns of equilibration in the NPT ensemble were performed in 4 more steps, with progressive reduction of the position restraints, until their complete removal in the last step. Next, the production trajectory, also in the NPT ensemble, was 10 ms long (considering the 4 replicas, the total production dynamics time was 40 µs per system). The temperature was maintained at 310 K by the v-rescale thermostat¹⁰⁰ and the pressure was maintained at 1 bar by the Berendsen algorithm¹⁰¹. Electrostatic interactions were treated by reaction-field method, with a cut-off of 1.1 nm and a dielectric constant of 15, and van der Waals interactions were cut-off at 1.1 nm. The centers of mass of the receptors and the distances between them were determined with the open-source, community-developed PLUMED library¹⁰², version 2.8. The torsional angle describing the relative orientation between mGlu2 and 5-HT_2A was determined according to a previously described strategy⁶⁹ and calculated with PLUMED.

Curve fitting and data analysis

Data in all figures were shown as mean ± SEM from three independent experiments at least and performed in triplicates. Curves are fitted in Y = Bottom + (Top-Bottom)/(1 + 10^(LogEC50-X)) using nonlinear regression in GraphPad Prism 8 software (GraphPad Software, San Diego, CA, USA). Statistical differences were determined using one-way ANOVA with Dunnett’s multiple comparisons test or two-way ANOVA with Tukey’s multiple comparisons test in GraphPad Prism8, considering to be significantly different when P-value < 0.05.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

S.G., L.L., X.L., J.M., Y.W., and C.X. performed nanobodies based conformational sensor experiment, IP₁ assay. L.L. performed FRET saturation assay and crosslinking experiments, and S.Q. crosslinking experiments. S.G. and L.L. performed data analysis. A.R., J.D., P. Renault and J.G. built the two 3D models and MD analysis for mGlu2-5-HT_2A hetero-complex and proposed the cysteine crosslinking mutations. L.L., P. Renault, J.G., J.-P.P., P. Rondard, and J.L. conceived experiments, supervised the work and wrote the manuscript.

Peer review

Peer review information

Nature Communications thanks Javier González-Maeso, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

Data supporting the findings of this manuscript are available from the corresponding authors upon request. The source data underlying Figs. 1b, 3a, 3e–h, 4a, b, 5b–h, 6 and Supplementary Figs. 1–3, 8, 9a, b, 9e, f, 10a, b, 10e, 11b, c, 12, 13, 15, 16b, c, 16d, 17–19 are provided as a Source Data file. PDB: 4EIY, 7WC8, 7MTQ, 7MTR, 4DKL, 4OR2, 4IB4. Source data are provided with this paper.

Competing interests

Philippe Rondard and Jean-Philippe Pin are involved in a collaborative team between the CNRS and Cisbio Bioassays, Revvity (Eidos team, IGF, Montpellier). The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-69939-3.