Structure of the melatonin–related orphan receptor, GPR50

TL;DR

This study reveals the structure of the GPR50 receptor, offering insights into its signaling activity and potential for drug development.

Contribution

The paper presents the first 3.4 Å cryo-EM structure of ligand-free GPR50 and proposes its activation mechanism.

Findings

GPR50 exhibits moderate constitutive activity through interaction with Gα12.

The structure reveals a distinct ligand-binding pocket and access channels compared to melatonin receptors.

Comparison with AlphaFold3 predictions suggests a possible activation mechanism for GPR50.

Abstract

GPR50 is an orphan G-protein–coupled receptors that belongs to the member of melatonin–related receptor family. GPR50 plays roles in various physiological processes, including cancer progression, Notch signaling, and insulin, leptin, and glucocorticoid signaling. GPR50 forms a complex with melatonin–receptor type 1A or 1B, and regulates signaling activity of melatonin–receptor type 1A. Although endogenous agonists have not been characterized, GPR50 may have its own signaling activity, which is undefined at present. In this study, in an attempt to characterize the orphan activity of GPR50, we determined the 3.4 Å structure of ligand-free GPR50 using cryo-electron microscopy. We showed that GPR50 exhibits moderate constitutive activity through interaction with Gα12. The structure reveals a putative ligand–binding pocket and ligand access channels of GPR50 that differ from those of melatonin receptors. Based on the comparison with the AlphaFold3–predicted active state, we propose an activation mechanism of GPR50. Our findings could serve as a platform for identifying the synthetic or endogenous ligands of GPR50, providing insights into the elusive G-protein–dependent signaling of GPR50.

ga1

Figures

B and C](#fig0005)). While this feature is shared with MT1 and MT2, it is rare among class A GPCRs ([Nicoli et al., 2022](#bib27)) (Supplementary Fig. S5). Because GPR50 was engineered by inserting BRIL into ICL3 to facilitate structure determination, we examined possible BRIL–induced structural artifacts by comparison with the AlphaFold3-predicted model (Supplementary Fig. S6). The AlphaFold3-predicted structure of ligand-free GPR50 closely resembles the cryo-EM structure, with a root-mean-square-deviation of 1.2 Å over all Cα atoms ([Abramson et al., 2024](#bib1)). The AlphaFold3–predicted G

B and Supplementary Fig. S8B). Entrance of this solvent-exposed channel is disclosed by Q194^ECL2^ and Y294^7.39^ in MT2, whereas it is partially sealed off by Q181^ECL2^ and Y281^7.39^ in MT1 ([Johansson et al., 2019](#bib18), Supplementary Fig. S8B-D) Consistent with this, calculations on channel access route reveal that this solvent-exposed route is computed with a sufficiently large entrance in MT2, whereas it is narrower by 1.8 Å in MT1 compared to MT2 because of the difference between the orientations of Y^7.39^ (Supplementary Fig. S8B and C). In GPR50, Y182^ECL2^ is located at the same

A-D](#fig0015), Supplementary Fig. S4, S7E-H). However, several conserved residues adopt distinct orientations in GPR50 compared with the MT receptors. F180^ECL2^ of GPR50 is very close and nearly collide to 2-iodomelatonin in the MT1 in the superimposed structures, while F179^ECL2^ of MT1 is bent at Cβ and faces the indole ring of 2-iodomelatonin ([Fig. 3A and B](#fig0015)). Moreover, F192^ECL2^ in MT2 is oriented in a similar manner as its counterpart in MT1 ([Fig. 3C and D](#fig0015)). Thus, F^ECL2^ in the MT1 and MT2 is critically important for stabilizing ligand binding through aromatic s

D-F](#fig0020), Supplementary Fig. S8E). Residues F188^5.38^, I166^4.63^, and N185^5.35^ at the beginning and end of ECL2 may constitute the entrance of the channel ([Fig. 4I](#fig0020)). The end of ECL2 in GPR50 protrudes by 4.7 and 4.5 Å more than in MT1 and MT2 (comparing the Cα atoms of N185^5.35^ in GPR50, S184^5.35^ in MT1, and S197^5.35^ in MT2), respectively ([Fig. 4I](#fig0020)). Displacement of TM5 and the end of ECL2 from TM4 generates wider space of GPR50 which is absent in both MT1 and MT2. Diameter of the entrance of this channel is 3.8 Å, which is significantly wider than the fi

I](#fig0025), Supplementary Fig. S9E). In the inactive state, R126^3.50^ forms the core of the hydrophilic network of the N^3.49^R^3.50^Y^3.51^ motif by interacting with N125^3.49^. Notably, Y127^3.51^ of the N^3.49^R^3.50^Y^3.51^ motif and Y208^5.58^ are not involved in either the inactive or active structures. In addition to the interaction between R126^3.50^ and N125^3.49^, R126^3.50^ of ligand-free GPR50 also interacts with Y136^ICL2^, E232^6.30^, and N235^6.33^, whereas interactions involving Y136^ICL2^ and E232^6.30^ are not observed in inactive MT1 and MT2. The ionic interaction between

C-G](#fig0030)). In the model, the α5 helix of Gα_12_ engages a pocket formed by ICL2, TM2, TM3, TM5, and TM6 via hydrophobic interactions and H-bonds ([Fig. 6C-E](#fig0030)). In addition, the αN also forms an ion pair with ICL2 ([Fig. 6E](#fig0030)). Interestingly, Q381 at the tip of the α5 helix is surrounded by a polar environment in GPR50, comprising residues N61^2.37^, N64^2.40^, R126^3.50^, N298^8.47^, and N300^8.49^ ([Fig. 6D-F](#fig0030)). Gα_i3_ and Gα_13_ contain tyrosine and glutamine, respectively, at the position corresponding to Q381 ([Fig. 6G](#fig0030)). By contrast, other G-pr

Full text

INTRODUCTION

G-protein–coupled receptors (GPCRs) are a vast group comprising ∼800 family members, playing a pivotal role in receiving stimuli transmitted to the cell and relaying them into the cell interior with positioning on the cell membrane (Yang et al., 2021). Despite their significance in signal transduction, agonists and the physiological functions remain veiled for nearly 100 GPCRs (Watkins and Orlandi, 2021). Although several molecular structures of orphan GPCRs have been reported for the past 2 to 3 years with the help of cryogenic electron microscopy (cryo-EM), a large number of orphan GPCRs still remain uncharacterized (Lin et al., 2020, Lin et al., 2023, Jeong et al., 2021, Shin et al., 2024).

GPR50 is an orphan GPCR and highly expressed in the pituitary gland and tanycytes, which function as important hormone endocrine glands and regulators of the flow of substances via the blood-brain barrier, respectively (Sidibe et al., 2010). GPR50 is related to the protection of cancer, such as breast cancer and hepatocellular carcinoma (Saha et al., 2020). Diminished expression of GPR50 correlates with an unfavorable survival prognosis regardless of the breast cancer subtype (Biswas et al., 2023). Elevated expression of GPR50 induces the proliferation, migration, and autophagy of hepatocellular carcinoma cells (Zhao et al., 2023). These tumor-associating functions of GPR50 are likely related to its interaction with the transforming growth factor β receptor and a disintegrin and metalloproteinase 17 (ADAM17), leading to the activation of SMAD and Notch signaling, respectively (Saha et al., 2020, Wojciech et al., 2018). Additionally, GPR50 has been proposed to participate in several physiologically important signaling pathways, including those related to insulin, leptin, and glucocorticoids (Bechtold et al., 2012, Li et al., 2011). Furthermore, GPR50 has been implicated in the regulation of neuronal development as a mitophagy receptor (Liu et al., 2024).

GPR50 shares 45% and 43% sequence identity with those of melatonin–receptor type 1A (MT1) and 1B (MT2), respectively, which are involved in regulating circadian rhythms and promoting sleep (Pévet, 2016). However, melatonin does not bind to GPR50 (Clement et al., 2017), and GPR50 is not directly involved in the melatonin-related pathways. Instead, it indirectly modulates the melatonin-related pathway through direct interaction with MT1. GPR50 antagonizes MT1 by blocking the binding of melatonin and related agonists to the receptor (Levoye et al., 2006). Although it is unclear how GPR50 inhibits MT1, both noncanonical C-terminal region (350-617) and the 7 transmembrane (TM) domain of GPR50 are involved in antagonizing MT1 (Levoye et al., 2006).

Although GPR50 is involved in various signal pathways, it is unclear whether the GPCR controls them via direct G-protein–dependent functions, and if so, which type of G-protein is engaged by GPR50 for signaling remains to be resolved. As a first step toward deorphanizing GPR50, we determined the cryo-EM structure of GPR50 and characterized its association with specific G-proteins.

MATERIALS AND METHODS

Construct, Expression, and Purification of Anti-BRIL Fab

The anti-BRIL Fab heavy chain and light chain sequences (Tsutsumi et al., 2020) were each inserted into separate pAcGP67a vectors. The heavy chain sequence includes an 8×His tag, thrombin and Tobacco Etch Virus cleavage sites, and an ALFA tag at the C-terminus (Supplementary Fig. S1A). These vectors were gifts from Dr Jie-Oh Lee (POSTECH). Each baculovirus carrying the constructs was separately amplified in Spodoptera frugiperda (Sf9; CRL-1711, ATCC) insect cells, which were cultured in ESF921 medium (Expression Systems) using the BestBac baculovirus system (Expression Systems). The recombinant baculoviruses were cotransfected into Trichoplusia ni HiFive (BTI-TN-5B1-4; Invitrogen) at a density 1.3 × 10^6^ cells ml^−1^, using a total volume equal to 8% of the medium of HiFive (with heavy chain: light chain volume ratio of 1:1) at 27°C and the medium was collected 72 hours after infection by centrifugation to separate it from the cell pellets. The pH of the medium was adjusted to 8.0 by adding Tris-HCl, resulting in a final Tris-HCl concentration of 50 mM in the medium. To remove the chelating reagent in the medium, 1 mM NiCl_2_ and 5 mM CaCl_2_ were further added. After 1 hour of incubation, the precipitate was removed by centrifugation. The sample was incubated with Ni Sepharose 6 Fast Flow resin (GE Healthcare) at 4°C. After 3 hours of incubation, the resin was washed with a buffer containing 20 mM HEPES (pH 7.5), 150 mM NaCl, 5% glycerol, and 20 mM imidazole. The anti-BRIL Fab was gradually eluted with 100 to 500 mM imidazole in a buffer that had the same composition as the wash buffer, except for the imidazole concentration. The eluant was further purified by size-exclusion chromatography by loading onto a Superdex 75 HiLoad 16/60 column (GE Healthcare) equilibrated with a buffer containing 20 mM HEPES (pH 7.5), 150 mM NaCl, and 5% glycerol. Fractions containing the anti-BRIL Fab were collected and concentrated to 14 ml^−1^ and then stored at –80°C until further use.

Construct, Expression, and Purification of Anti-Fab Nb

The anti-Fab Nb sequence (Tsutsumi et al., 2020), featuring an N-terminal 6×His tag and a C-terminal Strep tag, was cloned into the pET28a vector, a kind gift from Jie-Oh Lee. Escherichia coli BL21 (DE3) cells transformed with this vector were cultured in LB broth until an OD_600_ of 0.7, followed by induction with 0.5 mM IPTG for 20 hours at 18°C. The cells were harvested by centrifugation and lysed in a buffer containing 20 mM HEPES (pH 7.5), 200 mM NaCl, 5% glycerol, and 0.5 mM PMSF. The lysate was homogenized using a sonicator, and the insoluble material was removed by centrifugation at 37,300g for 1 hour at 4°C. The soluble fraction was applied to a Ni²⁺-NTA affinity column (GE Healthcare) and eluted using a 0 to 500 mM imidazole gradient. Fractions containing the anti-Fab Nb were collected, concentrated, and further purified by size-exclusion chromatography on a Superdex 75 HiLoad 16/60 column (GE Healthcare) pre-equilibrated with a buffer containing 20 mM HEPES (pH 7.5), 200 mM NaCl, and 5% glycerol. The fractions containing the anti-Fab Nb were pooled, concentrated to 19 mg/ml, and stored at −80°C for further use.

Construct, Expression, and Purification of GPR50

The GPR50_ICL3_BRIL construct was designed as previously described (Tsutsumi et al., 2020). Briefly, the BRIL sequence was inserted into the intracellular loop (ICL)3 region of the wild-type human GPR50 sequence (residues 1-219 and 227-355, Uniprot Q13585). Two short linkers derived from the A2A adenosine receptor (RQL between residue 219 and the N-terminus of BRIL, and ARSTL between the C-terminus of BRIL and residue 227) were used to connect BRIL to the GPR50 sequence. The construct included an N-terminal signal sequence from lysosomal α-mannosidase of Trypanosoma cruzi (TCM) and a Twin-Strep II tag at the C-terminal end, and it was cloned into a BacMam expression vector (Goehring et al., 2014). Baculovirus harboring the construct was amplified in Sf9 cells using the Bac-to-Bac baculovirus system (Invitrogen). Amplified baculovirus was then used to infect HEK293S GnTI^−^ cells (CRL-3022; ATCC) at a cell density of 3.0 × 10⁶ cells ml^−1^ with a virus volume equal to 9% of the culture medium. The cells were cultured in FreeStyle 293 medium (Gibco) at 37°C with 8% CO_2_ for 12 hours, after which 10 mM sodium butyrate was added. The culture was continued for an additional ∼48 hours at 30°C before harvesting the cells by centrifugation. The cell pellet was resuspended in lysis buffer containing 20 mM HEPES (pH 7.5), 100 mM NaCl, 5 mM MgCl_2_, 5 mM CaCl_2_, and a protease inhibitor cocktail. Cell lysis was performed using a Dounce homogenizer with 30 strokes. The lysate was then mixed with solubilization buffer comprising 20 mM HEPES (pH 7.5), 100 mM NaCl, 5 mM MgCl_2_, 5 mM CaCl_2_, 40% glycerol, 2% LMNG, 0.4% CHS, 2 mg ml^−1^ iodoacetamide, 0.3 mg ml^−1^ benzamidine, and a protease inhibitor cocktail. The mixture was incubated for 1.5 hours at 4°C. Solubilized membrane fractions were clarified by centrifugation at 186,000g for 30 minutes at 4°C, and the resulting supernatant was incubated with StrepTactin XT resin (IBA) for 3 hours at 4°C. The resin was transferred to an EconoPac column and washed with a buffer containing 20 mM HEPES (pH 7.5), 300 mM NaCl, 5 mM MgCl_2_, 10% glycerol, 0.1% LMNG, and 0.02% CHS. During the washing process, the detergent was gradually exchanged from 0.1% LMNG and 0.02% CHS to 0.1% LMNG, 0.033% glyco-diosgenin (GDN; Anatrace), and 0.02% CHS. Proteins were eluted in a buffer comprising 20 mM HEPES (pH 7.5), 200 mM NaCl, 5 mM MgCl_2_, 10% glycerol, 0.01% LMNG, 0.0033% GDN, 0.002% CHS, and 50 mM D-biotin (IBA). The eluted protein was incubated with a 1.4 molar excess of purified anti-BRIL Fab, followed by the addition of anti-Fab Nb at a molar ratio of 1.8 to the anti-BRIL Fab. After incubation on ice for at least 12 hours, the sample was concentrated and further purified using a Superose 6 10/300 column (GE Healthcare) pre-equilibrated with a buffer containing 20 mM HEPES (pH 7.5), 200 mM NaCl, 1 mM MgCl_2_, 0.0016% LMNG, 0.0002% GDN, and 0.0002% CHS. The final protein concentration was adjusted to ∼1.2 and ∼4.5 mg ml^−1^, respectively, for cryo-EM analysis.

Cryo-EM Sample Preparation and Data Collection

Concentrated samples at ∼1.2 and ∼4.5 mg ml^−1^ were each loaded onto glow–discharged holey carbon grids (Quantifoil Au, R1.2/1.3, 400 mesh; SPI) at 3 μl per grid. The grid with the ∼1.2 mg ml^−1^ sample was blotted for 10.0 seconds with a blot force of 7 under conditions of 4°C and 100% humidity. Similarly, the grid with the ∼4.5 mg ml^−1^ sample was blotted for 9.0 seconds with the same blot force under conditions of 4°C and 100% humidity. The grids were then plunge-frozen in liquid ethane using a Vitrobot Mark IV (ThermoFisher Scientific). Cryo-EM data were collected on a Titan Krios G4 (ThermoFisher Scientific) equipped with a K3 direct electron detector, operating at an accelerating voltage of 300 kV, at the Institute of Membrane Proteins in Pohang, Republic of Korea. Data acquisition was conducted at a nominal magnification of 105,000× in counting mode, corresponding to a pixel size of 0.851238 Å at the specimen level. For the ∼1.2 mg ml^−1^ sample (Data 1), a total of 8,858 movies were acquired, each consisting of 50 frames, with a total electron dose of ∼65 electrons/Ų and a defocus range of –1.0 to –2.0 µm. For the ∼4.5 mg ml^−1^ sample (Data 2), a total of 5,725 movies were collected under the same conditions as Data 1.

Data Processing

Dose–fractionated image stacks from Data 1 and Data 2 were imported into CryoSPARC v4.5.3 (Punjani et al., 2017). The imported images underwent patch–based motion correction, followed by contrast transfer function (CTF) estimation for each dataset. Micrographs with a CTF fit resolution better than 3.5 Å were selected, resulting in a total of 12,433 micrographs being retained for further processing, while the rest were excluded. Template-based particle picking was performed using the electron microscopy map of the GPR61_ICL3_BRIL–anti-BRIL Fab–anti-Fab Nb complex (EMD-41145 (Lees et al., 2023)) as a template, yielding 15,552,692 particles. Additionally, particles were picked using a blob picker, resulting in 8,530,839 particles. These 2 particle sets were combined, and duplicates were removed using the Remove Duplicate Particles, resulting in a total of 15,892,998 particles. These particles were extracted with a box size of 400 pixels, followed by multiple rounds of two-dimensional classification, which reduced the dataset to 1,255,607 particles. The selected particles were subjected to 2 rounds of ab-initio reconstruction and heterogeneous refinement. From these processes, a single class producing the best map was identified, comprising 283,982 particles. This particle set was used for TOPAZ Train (Bepler et al., 2018), TOPAZ Extract, and Remove Duplicate Particles. Subsequently, two-dimensional classification was performed on the resulting particle set, yielding 733,282 particles. Through ab-initio reconstruction and heterogeneous refinement, the best map was obtained from 3 classes, corresponding to 397,644 particles. These 397,644 particles were combined with the initial 283,982-particle set that had produced the best map before TOPAZ Train (Bepler et al., 2018). Duplicate particles were removed using the Remove Duplicate Particles, resulting in a final set of 608,641 particles, which were used for postprocessing. The final particle set was subjected to per-particle motion correction via local motion correction, followed by global CTF refinement and nonuniform refinement, producing a map with a global resolution of 2.98 Å at a Fourier Shell Correlation of 0.143. To further improve map quality, 2 local refinements were separately performed. First, to enhance the TM region of GPR50, particle subtraction was applied to exclude densities for the micelle, BRIL, anti-BRIL fragment antigen-binding (Fab), and nanobody. The remaining particles, including residual densities, were refined locally using a mask for the TM region of GPR50. Second, local refinement of the Fab–nanobody–BRIL region was performed after subtracting the micelle and the TM region of GPR50. The resolutions of the locally refined maps were 3.34 Å (TM) and 2.88 Å (Fab–nanobody–BRIL) at a Fourier Shell Correlation of 0.143. Finally, the 2 locally refined maps were separately sharpened using the DeepEMhancer (Sanchez-Garcia et al., 2021) and composited in UCSF ChimeraX (Pettersen et al., 2004) using the “vop maximum” command. The resulting composited map was subsequently used for model building.

Model Building and Refinement

To construct the atomic model of the TM region of GPR50, the AlphaFold3 (Abramson et al., 2024)-predicted structure of GPR50_ICL3_BRIL was docked into the composited map using UCSF ChimeraX (Pettersen et al., 2004). The human GPR50 sequence (Uniprot Q13585) was used as the input for AlphaFold3 (Abramson et al., 2024) to generate the model, which was then manually adjusted in Coot (Emsley and Cowtan, 2004). For the Fab–nanobody region, atomic model building was initiated by docking the coordinates of the Fab–nanobody region from the human frizzled receptor 5 structure (PDB: 6WW2 (Tsutsumi et al., 2020)) into the same composited map. After manual adjustments in Coot (Emsley and Cowtan, 2004), the coordinates for both the TM and Fab–nanobody regions were combined, refined, and validated using Coot (Emsley and Cowtan, 2004), real-space refinement (Afonine et al., 2018), and Molprobity (Chen et al., 2010) in PHENIX (Adams et al., 2010).

AlphaFold3 Modeling of BRIL-Free Inactive and Putative Active hGPR50 Structures

Structural models for the active hGPR50–hGα_12_ and ligand-free hGPR50 were generated using AlphaFold3 (Abramson et al., 2024). Amino acid sequences were retrieved from UniProt (Q13585 for GPR50 and Q03113 for Gα_12_) and submitted to the AlphaFold3 server (http://www.alphafoldserver.com) using default parameters. For each target, the top–ranked predicted model based on pLDDT was selected for downstream analysis, and model confidence was evaluated using predicted LDDT and PAE metrics. AlphaFold3-predicted structures were used solely for qualitative comparison of putative active and ligand-free conformations and were not refined against experimental cryo-EM density or subjected to additional restraints. Predicted structures were visualized and analyzed using ChimeraX (Pettersen et al., 2004).

BRET2 Assay

A BRET2 assay was conducted to measure GPCR–mediated G-protein activation using a TRUPATH biosensor, as previously described (Olsen et al., 2020). To identify which G-protein mediates signaling from the basal activity of the GPR50, we measured the basal activity of the wild-type receptor while varying the G-protein subtype. Briefly, HEK293T (CRL-3216; ATCC) cells were seeded in 6-well plates at a density of 1.5 × 10⁵ cells per well. After 24 hours, plasmid DNA encoding WT GPR50 was transfected into the cells along with fixed amounts of 100 ng of Gα (G_i3_, G_s_, G_q_, or G_12/13_)-RLuc8, 100 ng of G_β_, and 100 ng of G_γ_-GFP2, while the amount of WT GPR50 plasmid DNA was varied. Transfections were performed using TransIT-2020 (Mirus Bio) at a 1:3 (w/w) DNA-to-transfection reagent ratio in OptiMEM (Gibco). The following day, the cells were harvested and seeded into poly-L-lysine-coated 96-well plates at a density of 40,000 to 50,000 cells per well in DMEM supplemented with 1% dialyzed FBS (Gibco) and incubated for 12 hours. The cells were then washed once and replaced with 60 μl of assay buffer containing 20 mM HEPES (pH 7.4), 1× HBSS, and 0.1% BSA. Next, 20 μl of 25 μM coelenterazine 400a (Nanolight Technologies) was added, followed by an additional 20 μl of assay buffer. The plates were incubated for 15 minutes before being read on a Synergy HTX Multimode Reader (Bio-Tek) using emission filters of 410 nm (RLuc8) and 515 nm (GFP2), with 1-second integration time. The BRET ratio was calculated as the GFP2/RLuc8 emission value, and ΔBRET was determined by subtracting the G-protein–only BRET signal from the GPR50 and G-protein BRET signals. ΔBRET values were plotted against the amount of transfected GPR50 plasmid on the y- and x-axes, respectively, using GraphPad Prism 9.4.1. All experiments were independently performed at least 3 times.

ELISA Assay

Cell surface expression levels of WT GPR50 were quantified by ELISA with coexpression of different G-protein subtypes at varying receptor expression levels. In brief, HEK293T cells were transfected as described in the “BRET2 assay,” and the cells were plated in poly-l-lysine-coated 96-well plates. The next day, cells were fixed with 4% (v/v) paraformaldehyde for 10 minutes at room temperature. Then, cells were washed with PBS and incubated with 5% (v/v) BSA-blocking solution in PBS for 1 hour. After incubation, cells were incubated with an anti-FLAG-HRP–conjugated antibody (Sigma-Aldrich) diluted 1:10,000 in 5% BSA for 1 hour at room temperature. Cells were washed 5 times with PBS, and Super Signal ELISA chemiluminescent substrate (ThermoFisher Scientific) was added to all wells. The luminescence signal was monitored using a Synergy HTX Multimode reader (Bio-Tek). The cell surface expression levels were quantified in GraphPad Prism 9.4.1. The experiments were performed independently 3 times.

RESULTS

Structure Determination

To aid the structure determination of GPR50 by cryo-EM, we inserted thermostabilized Escherichia coli cytochrome b562 RIL (BRIL) into the intracellular loop (ICL) 3 located between TM5 and TM6 of GPR50 (Fig. 1A and Supplementary Fig. S1A-C). In addition, we used an anti-BRIL Fab along with an anti-Fab nanobody (Nb) to increase the molecular size of the complex (Tsutsumi et al., 2020, Fig. 1A and Supplementary Fig. S1A-C). We successfully obtained an electron density map of the GPR50_ICL3_BRIL complex with anti-BRIL and anti-Fab nanobody at 2.98 Å global resolution (Supplementary Fig. S1D, E, S2A-D). To improve the local electron density of GPR50, we performed the local refinement using a mask fitting with the heptahelical transmembrane bundle (7TM) domain of GPR50, which enhanced a local resolution of the map for the 7TM domain to 3.3 Å (Supplementary Fig. S2A-D, Supplementary Table S1). The resulting map was high quality and allowed us to build a complete GPR50 model (Supplementary Fig. S3).Fig. 1. Overall structure of the ligand-free GPR50. (A) Surface representation of the overall structure of the GPR50 (green) in complex with anti-BRIL Fab (light blue and blue) and an anti-Fab nanobody (gray). The BRIL domain (white) is fused into the ICL3 between TM5 and TM6. (B) Overall structure of the ligand-free GPR50. The putative orthosteric ligand–binding site is highlighted with a circle. (C, D) Two different views of (B). (C) A view from the extracellular side. (D) A view from the intracellular side.Fig. 1

Overall Architecture of the Ligand–Free GPR50 Receptor

GPR50 belongs to the canonical class A GPCRs, featuring an extracellular N-terminus (residues 1-23), extracellular loops (ECLs), intracellular loops (ICLs), seven-transmembrane (7TM) domain, and a short amphipathic helix VIII (H8) that runs parallel to the membrane (Fig. 1B-D). The N- and C-terminal regions (residues 311-355) are notably disordered in the map. Among class A GPCRs, the ligand–free GPR50 structure most resembles those of inactive MT1 (PDB ID: 6ME4 (Stauch et al., 2019)) and MT2 (6ME9 (Johansson et al., 2019)), with the root-mean square-deviation values for Cα atoms of 1.03 and 0.91 Å, respectively. Additionally, the DALI analyses revealed that GPR50 exhibited the highest Z-scores (similarity scores) with inactive MT1 (Stauch et al., 2019) and MT2 (Johansson et al., 2019), with values of 32.0 and 31.8, respectively.

In the intracellular face, the most notable differences of ligand-free GPR50 with inactive MT1 and MT2 are observed in ICL2. Overall, the Cα tracing of the ICL2 regions in the 3 GPCRs is different (Fig. 2A and B). Although the ICL2 sequence of GPR50 exhibited almost identical levels of sequence with MT1 and MT2, its structural trace aligns more closely with that of MT2 than with MT1 (Fig. 2A and B and Supplementary Fig. S4). The ICL2 of MT2 and GPR50 contains a short helical part, whereas that of MT1 is predominantly composed of a loop (Fig. 2A). Additionally, although the extent of ICL2 downward movement is similar between MT2 and GPR50, ICL2 in MT1 is further down-shifted by 5.5 Å (comparison between the Cα atoms of L134^ICL2^ in GPR50 and D136^ICL2^ in MT1, Fig. 2B). According to previously reported studies, ICL2 in class A GPCRs plays a crucial role in G-protein binding and selectivity (Duan et al., 2022, Yano et al., 2017). Therefore, these differences in ICL2 raise a possibility that GPR50 might be more similar to MT2 than to MT1 in terms of G-protein binding and selectivity. On the other hand, in the extracellular face, MT1 and MT2 are very similar, whereas GPR50 exhibits different patterns (Fig. 2C and D). The ECL1 of GPR50 is located farther from ECL2 compared to that of MT1 and MT2 (Fig. 2C). At the ECL2, the loop of GPR50 is elevated by 5 Å at the terminal region compared to MT1 and MT2 (comparing the Cα atoms of Q184^ECL2^ in GPR50 with S184^ECL2^ in MT1) (Fig. 2D). Additionally, ECL3 of GPR50 protrudes outward by 7.5 Å from the center of the extracellular face compared to that of MT1 (comparison between the Cα atoms of E266^ECL3^ in GPR50 and M268^ECL3^ in MT1, Fig. 2C). Considering that the ECLs are closely positioned near the orthosteric ligand–binding site and contribute to ligand specificity, ligand specificity of GPR50 may differ significantly from those of MT1 and MT2 (Peeters et al., 2011).Fig. 2. Comparison of the ligand–free GPR50 structure with MT1 and MT2 structures. (A) Comparison of the intracellular side of the GPR50 (green) with the melatonin receptors (MT1 (PDB: 6ME4) in light gray and MT2 (PDB: 6ME9) in brown). (B) Close-up view of (A) for comparison of the ICL2 and TM3 of the melatonin–receptor family members. (C) Comparison of the extracellular side of the GPR50 with the melatonin receptors, color schemes are same as in (A). (D) Close-up view of the superimposed ECL2 and TM5 of the melatonin–receptor family members. (E) Comparison of ECL2 and ECL3 between ligand-free GPR50 (green) and active MT1 (blue). The agonist is shown at the center in pink. (F) Comparison of ECL2 and ECL3 between apo GPR50 (green) and active MT2 (orange), shown in the same view as in (E). (G) Comparison of ECL2 and TMs between ligand-free GPR50 (green) and active MT1 (blue). (H) Comparison of the ECL2 between ligand-free GPR50 (green) and active MT2 (orange), shown in the same view as in (G).Fig. 2

The overall structure of GPR50 is highlighted with ECL2 which acts as a lid for the ligand binding. The entire ECL2 is oriented almost parallel to the direction of the membrane and covers the orthosteric ligand–binding site (Fig. 1B and C). While this feature is shared with MT1 and MT2, it is rare among class A GPCRs (Nicoli et al., 2022) (Supplementary Fig. S5). Because GPR50 was engineered by inserting BRIL into ICL3 to facilitate structure determination, we examined possible BRIL–induced structural artifacts by comparison with the AlphaFold3-predicted model (Supplementary Fig. S6). The AlphaFold3-predicted structure of ligand-free GPR50 closely resembles the cryo-EM structure, with a root-mean-square-deviation of 1.2 Å over all Cα atoms (Abramson et al., 2024). The AlphaFold3–predicted GPR50 structure reveals an additional long helix after the H8 helix (residues 314-353) that is supported by a high pLDDT value (75-90) (Supplementary Fig. S6A-C). However, this region is not observed in our structure, possibly due to the flexibility between the H8 and the additional helix. In addition, the tip of TM6 exhibited noticeable differences between the predicted model and the cryo-EM structure, as evidenced by the distance between the Cα atoms of D227^6.25^ (Supplementary Fig. S6D and E).

Comparison of ECLs Between GPR50 and Melatonin Receptors (MT1/MT2)

Although GPR50 shares significant similarity with MT1 and MT2 in overall structure, melatonin does not bind to GPR50 (Clement et al., 2017). This could be due to the differences in ECLs, orthosteric ligand–binding pockets, or other intrinsic features. To understand the basis for the failure of melatonin binding to GPR50, we compared cryo-EM structure of GPR50 with the 2-iodomelatonin–bound MT1 and the ramelteon–bound MT2 structures (Fig. 2A-H and Supplementary Fig S7). Although the overall structure of GPR50 most closely resembles the inactive MT1 bound to 2-iodomelatonin (PDB: 6ME4, Stauch et al., 2019) and the inactive MT2 bound to ramelteon (PDB: 6ME9, Johansson et al., 2019), both structures contain a mutation at the toggle switch (W251^6.48^F) introduced to facilitate crystallization. Because the toggle switch-mutation may have affected the conformation of MT1 and MT2, we also compared the ligand–free GPR50 structure with the active-state structures of MT1 (PDB: 7VGY, Wang et al., 2022) and MT2 (PDB: 7VH0, Wang et al., 2022) (Fig. 2E-H).

In the extracellular view, the orthosteric ligand–binding pocket is located approximately 14 Å beneath ECL2, embedded within the TM core and surrounded by TM3, TM5, TM6, and TM7 (Fig. 2E and F). The most distinctive feature between GPR50 and the MT receptors is observed in ECL3. ECL3 of GPR50 is shifted 7 Å outward relative to the TM core compared with that in MT1 and MT2 (comparison between the Cα atoms of E266^ECL3^ in GPR50, M268^ECL3^ in MT1, and M281^ECL3^ in MT2). As a result, the extracellular region of the orthosteric ligand–binding site, as defined by residues in ECL2 and ECL3, is most widely open in GPR50 when compared with that of MT1 and MT2 (Fig. 2E and F). In MT1, the entrance to the orthosteric site is formed by R173^ECL2^ and S182^ECL2^ with an inter-residue distance of 8 Å, with a second pair of residues, D171^ECL2^ and V269^ECL3^, located in a perpendicular orientation and separated by 10 Å (Fig. 2E and G). Similarly, in MT2, the corresponding region, R186^ECL2^ and T195^ECL2^, is separated by 6 Å, and a second residue pair, D184^ECL2^ and A282^ECL3^, is separated by 10 Å (Fig. 2F and H). In GPR50, the corresponding pair (R174^ECL2^ and N181^ECL2^) is separated by 9 Å, similar to that observed in MT1. However, A268^ECL3^ (corresponding to V269^ECL3^ in MT1) is located 13 Å away from D172^ECL2^ (corresponding to D171^ECL2^ in MT1). The neighboring residues K265^ECL3^ and E266^ECL3^ are also oriented away from the pocket, resulting in a wider extracellular entrance. Inactive MT receptors also exhibited differences relative to ligand-free GPR50 that are comparable to those observed between active MT receptors and GPR50 (Supplementary Fig. S7A-D).

The ECL2 of MT1 or MT2 contains a short β-hairpin, whereas the ECL2 of GPR50 is comprised of an entire loop (Fig. 2C, G, H). The overall Cα traces are similar among these structures. The most notable difference in ECL2 between GPR50 and the MT receptors involves Y182^ECL2^ in GPR50 and the corresponding residue, Q^ECL2^ in MT1 and MT2. While Q^ECL2^ interacts with TM6 via backbone of G^6.55^ in the MT receptors, sidechain of Y182^ECL2^ forms H-bond with Y280^7.39^ in TM7, blocking the orthosteric ligand–binding pocket in GPR50 (Supplementary Fig. S7C, D). These 2 tyrosine residues form a hydrophobic network with I179^ECL2^ in ECL2 and anchor ECL2 to TM7, which is not observed in the MT receptors. Such structural differences corroborate previous findings demonstrating that substitution of ECL2 and TM6 of GPR50 with their counterparts from MT1 confers melatonin-induced activation upon GPR50 (Clement et al., 2017). Additionally, replacing G258^6.55^ of the MT1 receptor with threonine significantly decreased melatonin binding affinity to MT1 and cAMP signaling (Conway et al., 2000). In GPR50, G258^6.55^ is replaced to T257^6.55^, which could sterically interfere with the interactions involving neighboring residues.

Comparison of the Orthosteric Ligand–Binding Site Between GPR50 and Melatonin Receptors

Structural differences between the active and inactive states of the orthosteric ligand–binding site may provide additional insight into why melatonin does not bind to GPR50. Structural alignment reveals that many residues forming the ligand-binding site are conserved (Fig. 3A-D, Supplementary Fig. S4, S7E-H). However, several conserved residues adopt distinct orientations in GPR50 compared with the MT receptors. F180^ECL2^ of GPR50 is very close and nearly collide to 2-iodomelatonin in the MT1 in the superimposed structures, while F179^ECL2^ of MT1 is bent at Cβ and faces the indole ring of 2-iodomelatonin (Fig. 3A and B). Moreover, F192^ECL2^ in MT2 is oriented in a similar manner as its counterpart in MT1 (Fig. 3C and D). Thus, F^ECL2^ in the MT1 and MT2 is critically important for stabilizing ligand binding through aromatic stacking, which is abolished upon alanine mutation as previously reported (Johansson et al., 2019, Stauch et al., 2019). Structural alignment suggests that the F^ECL2^ rotamer plays a critical role in selecting the ligands at the pocket between GPR50 and MT1/MT2. (Fig. 3A-D) Together with the differences in the ECLs, subtle residue-by-residue differences at the orthosteric ligand–binding site may explain the lack of melatonin response in GPR50.Fig. 3. Comparison of the orthosteric ligand–binding sites between GPR50 and the MT receptors. (A) Close-up view of 2-iodomelatonin and its interacting residues in active MT1 (light blue, PDB: 7VGY). Residues within 4 Å of 2-iodomelatonin are shown. All figures (A-D) are shown in the same view. (B) Close-up view of the putative orthosteric site in GPR50 (green). 2-iodomelatonin was modeled from the active MT1 structure in (A), and residues within 4 Å of 2-iodomelatonin are shown. (C) Close-up view of ramelteon and its interacting residues in active MT2 (orange, PDB: 7VH0). (D) Close-up view of the putative orthosteric site in GPR50 (green). Ramelteon was modeled from the active MT2 structure in (C).Fig. 3

Comparison of Ligand Access Channels in GPR50, MT1, and MT2

To assess the differences in ligand specificity between GPR50 and melatonin receptors, we examined the ligand access channels leading to the orthosteric ligand–binding site. Previous studies have revealed a membrane–buried lateral channel for the ligand entry commonly present in both MT1 and MT2. In both receptors, membrane–buried channel entrances are located in proximity to H^5.46^, a key residue that participates in the opening and closure of the channel (Ch1 in Fig. 4B and C; Johansson et al., 2019; Stauch et al., 2019). H^5.46^ is also conserved in GPR50. Using MOLE 2.5 (Pravda et al., 2018), we identified the membrane–buried putative channel in GPR50, which originates from near H^5.46^ and extends toward the putative ligand–binding site, and compared it to those of MT1 and MT2 (Fig. 4A-F).Fig. 4. Comparison of the putative ligand access channels of GPR50 and MT1/MT2. (A-C) A membrane-buried channel in the melatonin-receptor members: (A) A channel in GPR50, (B) in MT1, (C) in MT2. At the bottom of each figure, diagrams indicate length and radius of each channel. (D-F) Three putative ligand access channels of GPR50 in 3 different views. (D) Ch1 (membrane-buried channel). (E) Ch2 (solvent-exposed channel), and (F) Ch3 (unique channel present in GPR50). (G) Comparison of the membrane-buried channels originated from H5.46 residue in GPR50 with those of MT1 and MT2. (H) The solvent–exposed ligand access channel (Ch2) between ECL2 and ECL3 in GPR50. (I) Comparison of the third potential ligand channel (Ch3) of GPR50 with the corresponding regions of MT1 and MT2.Fig. 4

The H^5.46^ residue in GPR50 is tilted by 2.4 Å toward the interior of the 7TM domain relative to melatonin receptors, as determined from the Cγ atoms of H196^5.46^ in GPR50 and H195^5.46^ in MT1 (Fig. 4G). This tilted orientation obstructs the entrance of the ligand accession channel, unlike in melatonin receptors (Fig. 4A-C, G). Moreover, in GPR50, residues P162^4.59^, N163^4.60^, I166^4.63^, and I192^5.42^, which are positioned on TM4 and TM5, are closer to each other compared to those in MT1, creating a narrower space at the entrance of this channel and further obstructing it (Fig. 4G). However, unlike MT1 and MT2, F188^5.38^ of GPR50 is relatively farther from the channel entrance compared to Y^5.38^ in MT1 and MT2 (Supplementary Fig. S8A). Radius of the entrance in GPR50 is 1.1 Å, which is 0.4 and 0.6 Å narrower than those of MT1 and MT2, respectively (Fig. 4A-C). Moreover, Y^5.38^ of MT1 and MT2 pushes the entrance to a lower position (Supplementary Fig. S8A). As a result, the entrances of MT1 and MT2 are positioned lower and increase the radius of entrance compared to that of GPR50. Therefore, the channel in GPR50 is more restricted in accommodating ligands compared to those of MT1 and MT2. Based on these results, the ligand access channel near H196^5.46^ in GPR50 may have a lower chance of actual ligand translocation relative to the corresponding channel of MT1 and MT2.

MT2 possesses an alternative ligand access channel (Ch2) between ECL2 and ECL3 with a wider entrance compared to the primary channel in MT1 (Johansson et al., 2019, Fig. 2B and Supplementary Fig. S8B). Entrance of this solvent-exposed channel is disclosed by Q194^ECL2^ and Y294^7.39^ in MT2, whereas it is partially sealed off by Q181^ECL2^ and Y281^7.39^ in MT1 (Johansson et al., 2019, Supplementary Fig. S8B-D) Consistent with this, calculations on channel access route reveal that this solvent-exposed route is computed with a sufficiently large entrance in MT2, whereas it is narrower by 1.8 Å in MT1 compared to MT2 because of the difference between the orientations of Y^7.39^ (Supplementary Fig. S8B and C). In GPR50, Y182^ECL2^ is located at the same position as Q194^ECL2^ in MT2 and appears to impose greater spatial constraints between ECL2 and ECL3 through its interaction with Y280^7.39^ (Fig. 4H, Supplementary Fig. S8D). However, Y276^7.35^ in GPR50 is tilted by 82° compared to F277^7.35^ in MT1 (Supplementary Fig. S8D). This creates a different spatial allowance between ECL2 and ECL3 compared to MT1, enabling the formation of a new ligand access channel with a 0.4 Å wider entrance compared to that of MT1 (Fig. 4H, Supplementary Fig. S8C).

In addition, GPR50 has a third potential entrance gate (Ch3; Fig. 4D-F, Supplementary Fig. S8E). Residues F188^5.38^, I166^4.63^, and N185^5.35^ at the beginning and end of ECL2 may constitute the entrance of the channel (Fig. 4I). The end of ECL2 in GPR50 protrudes by 4.7 and 4.5 Å more than in MT1 and MT2 (comparing the Cα atoms of N185^5.35^ in GPR50, S184^5.35^ in MT1, and S197^5.35^ in MT2), respectively (Fig. 4I). Displacement of TM5 and the end of ECL2 from TM4 generates wider space of GPR50 which is absent in both MT1 and MT2. Diameter of the entrance of this channel is 3.8 Å, which is significantly wider than the first calculated channel with an entrance near H196^5.46^ for the membrane–buried lateral channel (Fig. 4A, I, and Supplementary Fig. S8E). This suggests that the channel in GPR50, with an entrance formed by I166^4.63^, F188^5.38^, and N185^5.35^, is a good candidate for an alternative ligand access channel (Ch3).

Overall, GPR50 possesses 3 putative ligand access channels—the first one near the H^5.46^, the second, with an entrance near Y276^7.35^, and the third one formed by the beginning and end of ECL2 (Fig. 4D-F). However, as the first cleft is too narrow compared to those of MT1 and MT2, the latter 2 may provide better chances for ligand access.

Putative Activation Mechanism of GPR50

Superposition of GPR50 with MT1 and MT2 revealed no outward displacement of TM6 in GPR50, a hallmark of class A GPCR activation. Thus, we concluded that the present structure represents the inactive state (Hauser et al., 2021). This is consistent with our experimental condition in which no agonist was added during the purification steps, as confirmed in the cryo-EM density in the putative orthosteric pocket of GPR50. We attempted to make a stable complex between GPR50 and the heterotrimeric Gαβγ, which were unsuccessful. For this reason, we compared the ligand–free GPR50 structure with the active GPR50 model predicted by AlphaFold3 (Abramson et al., 2024) focusing on changes in the TMs, ECLs, ICLs, and the putative ligand–binding site during the activation of GPR50 (Fig. 5A-C, Supplementary Fig. S9A-E).Fig. 5. Putative activation mechanism of GPR50. (A-C) Superposition of the ligand-free GPR50 (green) structure with the active GPR50 model predicted by AlphaFold3 (pink) in 3 different orientations. (D) Superimposed overall structures of the ligand-free and active GPR50 to illustrate the conformational changes upon activation. Regions marked with different-colored hexagons are shown in detail in 5(E-I). (E) Superimposition of the C^6.47^W^6.48^xP^6.50^ motif. (F) The P^5.50^I^3.40^F^6.44^ motif. (G) Electron density for the sodium ion near D74^2.50^ and S115^3.39^. (H) The N^7.49^A^7.50^xxY^7.53^ motif. (I) Structural rearrangement of R126^3.50^ disrupts the hydrophilic networking of the N^3.49^R^3.50^Y^3.51^ motif and destabilizes the inactive state.Fig. 5

On the intracellular side, ICL1, ICL2, and the H8 helix do not exhibit significant structural differences between the 2 states, except for the cytoplasmic end of TM6. In the active GPR50 model, the Cα atom of D227^6.25^ is displaced outward by 11.3 Å relative to the ligand-free structure (Fig. 5B). Similarly, only modest conformational changes are observed on the extracellular side. In the active model, TM1 shifts toward TM7 and ECL3 exhibit a modest inward movement, marked by a 2.8 Å displacement of the Cα atom of G266^ECL3^ (Fig. 5C). The largest conformational change is observed in ECL2, where the Cα atom of N184^ECL2^ near TM5 moves outward by up to 6.2 Å, resulting in a widened extracellular entrance to the orthosteric binding site.

GPR50 preserves canonical motifs, including the C^6.47^W^6.48^xP^6.50^ and P^5.50^I^3.40^F^6.44^ motifs (Fig. 5D-F, Supplementary Fig. S9B and C). The C^6.47^W^6.48^xP^6.50^ motif is located directly below the orthosteric ligand–binding pocket in class A GPCRs (Zhou et al., 2019). This motif, due to its specific positional characteristics, effectively serves as a trigger for the state transition to the activation state, primarily through the rotamer change of W250^6.48^, which is usually triggered by the spatial occupancy of the agonist (Zhou et al., 2019). In the superimposition between the inactive and active model of GPR50, a hallmark shift of 3 Å (Nε1 of W250^6.48^) is consistently observed (Fig. 5D and E, Supplementary Fig. S9B). This shift serves as a crucial factor that distinguishes the inactive and active states of GPR50. The rotamer shift of W250^6.48^ leads to a steric clash with F246^6.44^ of the P^5.50^I^3.40^F^6.44^ motif, prompting F246^6.44^ to undergo a conformational change and shift the CE2 by 3.6 Å to avoid spatial restraint (Fig. 5F, Supplementary Fig. S9B).

This structural alteration significantly affects the putative sodium binding site, similar to the case of MT1, which involves the adjacent D74^2.50^, S115^3.39^, S287^7.46^, and N286^7.45^ residues of GPR50—typically implicated in the activation mechanism of class A GPCRs (Filipek, 2019, Hauser et al., 2021, Zhou et al., 2019). We identified electron density consistent with a presumed sodium ion near D74^2.50^ and S115^3.39^ (Fig. 5G). This finding further supports our conclusion that the structure is likely to represent an inactive state. However, we did not model the sodium ion into this density due to the insufficient resolution (3.3 Å) to reliably distinguish between noise and true sodium ion density, as the electron density was only observable at a very low contour level. Meanwhile, the movement of N286^7.45^ of sodium binding site toward F246^6.44^ of the P^5.50^I^3.40^F^6.44^ motif is critical, acting as a middle linker in the structural propagation to the activated state of GPR50 (Fig. 5D and G). Such movement draws TM3, TM5, and TM7 closer, facilitating an interaction between N286^7.45^ and the shifted W250^6.48^ through a hydrogen bond, likely stabilizing the active state of GPR50 by securing W250^6.48^ (Fig. 5D).

Unlike typical class A GPCRs, GPR50 lacks the N^7.49^P^7.50^xxY^7.53^ and D^3.49^R^3.50^Y^3.51^ motifs. Instead, GPR50 possesses the N^7.49^A^7.50^xxY^7.53^ and N^3.49^R^3.50^Y^3.51^ motifs (Fig. 5H and I, Supplementary Fig. S9D and E). The structural changes from inactive to active states for these motifs resemble those of N^7.49^P^7.50^xxY^7.53^ and D^3.49^R^3.50^Y^3.51^. The N^7.49^A^7.50^xxY^7.53^ motif exhibits structural changes due to the attraction of N286^7.45^ to F246^6.44^ at the putative sodium binding site (Fig. 5G and H). At the end of TM7, comprising the N^7.49^A^7.50^xxY^7.53^ motif, I293^7.52^ is directed toward TM5 to TM6, with Y294^7.53^ showing the most dynamic shift toward TM6 (Fig. 5H, Supplementary Fig. S9D). This structural difference implicates 2 significant features in the transition to the active state of GPR50. First, various residues in TM7 encroach upon the occupancy of TM6, leading to spatial displacement and an outward movement of TM6. Second, it disrupts the stabilization of the inactive state resulting from the hydrophilic networking of the N^3.49^R^3.50^Y^3.51^ motif, initiated by the structural rearrangement of R126^3.50^ (Fig. 5I, Supplementary Fig. S9E).

The N^3.49^R^3.50^Y^3.51^ motif may play important roles in the activation of GPR50 (Fig. 5I, Supplementary Fig. S9E). In the inactive state, R126^3.50^ forms the core of the hydrophilic network of the N^3.49^R^3.50^Y^3.51^ motif by interacting with N125^3.49^. Notably, Y127^3.51^ of the N^3.49^R^3.50^Y^3.51^ motif and Y208^5.58^ are not involved in either the inactive or active structures. In addition to the interaction between R126^3.50^ and N125^3.49^, R126^3.50^ of ligand-free GPR50 also interacts with Y136^ICL2^, E232^6.30^, and N235^6.33^, whereas interactions involving Y136^ICL2^ and E232^6.30^ are not observed in inactive MT1 and MT2. The ionic interaction between R126^3.50^ and E232^6.30^ in ligand-free GPR50 likely contributes to the inhibition of the outward movement of TM6, stabilizing the inactive state. During the transition to the active state, the CZ of R126^3.50^ likely shifts toward Y294^7.53^ of the N^7.49^A^7.50^xxY^7.53^ motif by 6.8 Å, disrupting the hydrophilic orchestration of the N^3.49^R^3.50^Y^3.51^ motif and resulting in the outward movement of TM6, thereby contributing to the structural formation of the active state of GPR50 (Fig. 5D and I).

GPR50 Exhibited High Basal Activity Toward Gα12 Signaling

Both MT1 and MT2 signal through the Gα_i/o_ protein in their signaling pathways, although recent findings revealed that MT1 can interact with Gα_s_, whereas MT2 binds to Gα_q_ in some systems (Liu et al., 2016). To date, the specific G-protein and signaling pathways associated with GPR50 have not been clearly identified through cell-based assays. To investigate if GPR50 exhibits any selectivity toward the G-protein signaling, we performed the BRET2 assay of GPR50 with Gα_i3_, Gα_s,_ Gα_q_, and Gα_12/13_ (Fig. 6A and B). GPR50 displayed the highest basal activity toward Gα_12_. GPR50 exhibited moderate activity toward Gα_i3_, which is consistent with a previous study that GPR50 was pulled down with endogenous Gα_i_ protein (Levoye et al., 2006). By contrast, basal activities toward Gα_sS_ and Gα_q_ were considerably lower compared to other subtypes. These results collectively suggest that GPR50 may preferentially induce Gα_12_-mediated signaling but also participates in multiple signaling pathways.Fig. 6. Basal activity of GPR50 across different G-protein subtypes. (A) The activity measured by BRET2 assays. WT GPR50 was expressed in a gradient by adjusting the amounts of plasmid DNA encoding the receptor. (B) Surface expression levels of the proteins for the BRET2 assay. The FLAG-tagged WT GPR50 (400 ng) transfection condition with Gαi3 (100 ng), Gβ3 (100 ng), and Gγ9 (100 ng) was set to 100% on the y-axis. For (A and B), data are mean ± SEM from at least 3 independent experiments, performed in technical triplicate. Statistical differences were determined by 2-way ANOVA with Dunnett’s posthoc test, compared to WT (NS, not significant; *P < .0001). (C) Overall structure of the AF3-predicted GPR50-Gα12 complex model. (D) Close-up view of the interface between GPR50 and the α5 helix of Gα12. (E) Close-up view of GPR50 and Gα12 shown in a different view. (F) Cartoon representation of the GPR50 and Gα12 interaction. (G) Sequence alignment of the G-proteins. The GPR50-contact residues are shown with their numbers on top.Fig. 6

To understand the basis for the preference of Gα_12_ (and Gα_i3_ and Gα_13_) over other Gα proteins, we analyzed the AlphaFold3–predicted GPR50-Gα_12_ complex (Fig. 6C-G). In the model, the α5 helix of Gα_12_ engages a pocket formed by ICL2, TM2, TM3, TM5, and TM6 via hydrophobic interactions and H-bonds (Fig. 6C-E). In addition, the αN also forms an ion pair with ICL2 (Fig. 6E). Interestingly, Q381 at the tip of the α5 helix is surrounded by a polar environment in GPR50, comprising residues N61^2.37^, N64^2.40^, R126^3.50^, N298^8.47^, and N300^8.49^ (Fig. 6D-F). Gα_i3_ and Gα_13_ contain tyrosine and glutamine, respectively, at the position corresponding to Q381 (Fig. 6G). By contrast, other G-proteins contain hydrophobic residues at the corresponding position. Therefore, despite limitations of AlphaFold3-predicted structures in fully explaining G-protein specificity, differences in the interacting residues between GPR50 and G-protein may provide a structural basis for GPR50’s preference for specific G-proteins.

DISCUSSION

GPR50 is associated with various diseases, including cancer (Biswas et al., 2023, Saha et al., 2020, Wojciech et al., 2018), insulin regulation, adaptive thermogenesis, and neuropsychiatric conditions such as bipolar disorder and major depression (Khan et al., 2016), Alzheimer’s disease, and schizophrenia (Chen et al., 2019, Delavest et al., 2012). However, GPR50 remains a highly enigmatic protein with unknown activation mechanism. While GPR50 forms a complex with MT1 and MT2, and regulates their activities, GPR50 is classified as an orphan receptor, with no agonist identified to date. Moreover, no G-protein–dependent functions have been identified for GPR50. As a first step to understand the G-protein–dependent function of GPR50, we determined the structure of ligand-free GPR50 using cryo-EM and investigated G-protein–bound active structure with AlphaFold3 (Abramson et al., 2024). We also analyzed the specific G-protein through which GPR50 exhibits high basal activity.

Although inactive structures of MT1 and MT2 have been determined via X-ray crystallography (Johansson et al., 2019, Stauch et al., 2019), these structures contain mutations introduced to enhance protein stability at the toggle switch. As a result, despite being complexed with agonist, these receptors remain in an inactive state. Therefore, it remains uncertain whether such an inactive state is due to mutation-induced artifacts, insufficient agonist concentration to trigger a conformational change, or an intrinsic property of the protein. Unlike MT1 and MT2, our structure preserves the original sequences and accurately reflects a genuinely inactive state, as evidenced by the orientations of TM6 and the toggle switch, along with the presence of sodium ion electron density. Thus, GPR50 represents the first inactive structure in the melatonin-receptor family.

Although ligand-dependent activation of GPR50 is unknown, mutation analysis revealed that GPR50 can activate G-protein–dependent signaling pathway (Pévet, 2016, Clement et al., 2017). A chimeric GPR50 with the ECL2 and TM6 of MT1 activates the G-protein by melatonin, which suggests that ligand–dependent G-protein signaling mediated by GPR50 is a possible signal transduction mechanism. Moreover, this raises the possibility that GPR50 can interact with unidentified ligands and trigger the downstream signaling. Although we do not know the precise ligands at present, the orthosteric binding site contains overall similarity with those of MT1 and MT2 with some clear differences (Fig. 3A-D, Supplementary Fig. S7E-H), which requires further investigation in the future.

One important finding in this study is the putative ligand access channel of GPR50. Previous studies suggest that subtype selectivity of MT1 and MT2 for the ligands is determined by the differences in ligand access channel (Johansson et al., 2019, Okamoto et al., 2021, Stauch et al., 2019, Wang et al., 2022). While MT1 has membrane-buried channel (Ch1), MT2 possesses solvent-accessible channel (Ch2), which can be exploited to facilitate the subtype selectivity. In this study, we found that 3 potential channels with 2 channels similar in MT1 and MT2; while the first channel near H196^5.46^ is also present in MT1 and MT2, its diameter at entry site of GPR50 is much narrower owing to the conformation of H196^5.46^ compared to the corresponding residues in MT1 and MT2, and this makes the first channel candidate is unlikely at GPR50. By contrast, the solvent–exposed access entry is as wide as MT2, raising the possibility of the access channel. Because the ligand access route in MT2 could accommodate more polar compounds compared to the membrane-buried channel in MT1, it has been proposed that the difference in ligand entry is an important factor in determining the melatonin–receptor subtype selectivity (Johansson et al., 2019, Stauch et al., 2019). Moreover, GPR50 has a third potential site (Ch3) comprised of I166^4.63^, F188^5.38^, N185^5.35^, which is not present in MT1 and MT2. Thus, distinct entry gate of GPR50 in the melatonin-receptor members suggests that GPR50 is likely to have selectivity different from those of MT1 and MT2 (Johansson et al., 2019, Okamoto et al., 2021, Stauch et al., 2019, Wang et al., 2022).

By contrast to MT1 and MT2, which primarily interact with Gα_i/o_ and inhibits cAMP pathway, GPR50 activates Gα_12_ but also moderately activates Gα_i/o_, raising the possibility that it is involved in the multisignaling pathways. The G_12_ – Rho A signaling pathway plays critical roles in cell migration and invasion, cell growth and division, as well as neuronal functions and embryonic development, with direct control over the cellular shape, movement and growth (Suzuki et al., 2003). Future analysis on the link between the GPR50 and biological function may help to elucidate the unknown function of GPR50.

The ligand-free structure of GPR50 may serve as an initial foothold in identifying the endogenous ligand and provides a foundational reference for developing selective modulators (Yang et al., 2021). Although the physiological roles of GPR50–dependent G-protein signaling remain unknown, identifying newly revealed endogenous ligands through structure–based drug discovery studies may help determine the specific pathways in which the G-protein–dependent functions of GPR50 are involved.

Author Contributions

Jinwoo Shin: Writing – original draft, Visualization, Validation, Methodology, Formal analysis, Data curation. Yunje Cho: Writing – review & editing, Writing – original draft, Supervision, Conceptualization. Junhyeon Park: Validation, Methodology. Eunyoung Jeong: Methodology. Dongyoung Baek: Validation, Methodology. Jihan Kim: Writing – review & editing, Visualization, Validation, Methodology, Investigation. Yoojoong Kim: Data curation. Young Jin Kim: Supervision.

Declaration of Competing Interests

The authors declare no competing interests.