Structural basis of the ligand binding and signaling mechanism of melatonin receptors

Melatonin receptors (MT1 and MT2 in humans) are family A G protein–coupled receptors that respond to the neurohormone melatonin to regulate circadian rhythm and sleep. Numerous efforts have been made to develop drugs targeting melatonin receptors for the treatment of insomnia, circadian rhythm disorder, and cancer. However, designing subtype-selective melatonergic drugs remains challenging. Here, we report the cryo-EM structures of the MT1–Gi signaling complex with 2-iodomelatonin and ramelteon and the MT2–Gi signaling complex with ramelteon. These structures, together with the reported functional data, reveal that although MT1 and MT2 possess highly similar orthosteric ligand-binding pockets, they also display distinctive features that could be targeted to design subtype-selective drugs. The unique structural motifs in MT1 and MT2 mediate structural rearrangements with a particularly wide opening on the cytoplasmic side. Gi is engaged in the receptor core shared by MT1 and MT2 and presents a conformation deviating from those in other Gi complexes. Together, our results provide new clues for designing melatonergic drugs and further insights into understanding the G protein coupling mechanism. Melatonin receptors (MT1 and MT2) are the targets for melatonin, the major neurohormone involved in circadian rhythm and sleep regulation. Here the authors describe the structures of 2-iodomelatonin and ramelteon bound MT1–Gi and MT2-Gi, revealing that MT1 and MT2 possess distinctive features within the ligand-binding pocket.

M elatonin receptors (MTs) are family A G proteincoupled receptors (GPCRs) expressed in the central nervous system (CNS) and peripheral tissues and are targets for melatonin, the major neurohormone involved in circadian rhythm and sleep regulation [1][2][3] . In vertebrates, melatonin (N-acetyl-5-methoxytryptamine) is mainly synthesized in the pineal gland and follows a circadian pattern synchronized to the dark phase of the natural light/dark cycle 4,5 . Modern life, such as worldwide travel, shift work, and artificial lighting, could disturb physiological melatonin production and cause sleep disorders, which was estimated to affect~20% of the population in the US. As a result, melatonin is one of the most popular supplements for the treatment of insomnia and jetlag. In addition to entraining the sleep-wake rhythm 6 , the effects of melatonin via activation of MTs regulate other physiological processes, including modulating the cardiovascular system 7 and buffering the immune system 8 . Emerging roles of MTs have also been confirmed in cancer protection 9,10 , bone formation 11 , glucose maintenance 12 , and neurodegenerative disorders 9 . Therefore, designing therapeutic agents targeting MTs has been continuously pursued in the drug discovery field 9 .
In humans, the MT family consists of two highly conserved members, termed MT 1 and MT 2 , and they share 55% and 70% sequence similarity for the overall region and the transmembrane part, respectively. Both MT 1 and MT 2 predominantly signal through G i/o proteins 13 , leading to inhibition of adenylyl cyclase and decreased intracellular concentrations of adenosine 3',5'-cyclic monophosphate (cAMP). For other G-protein families, the exact coupling profile remains unclear. However, coupling of the MT 1 receptor to G q/11 was consistently observed 14 . In the brain, MT 1 is expressed in the Locus Coeruleus and lateral hypothalamus (REM areas) 15,16 , but MT 2 is mainly located in the reticular thalamus (NREM area) 15,16 . The signaling variation and localization difference probably result in the distinct in vivo functions of MT 1 and MT 2 17 : MT 1 is mainly implicated in the regulation of rapid eye movement (REM) phases of the vigilance state in sleep 18 ; MT2 selectively increases non-REM (NREM) sleep [17][18][19] . In addition, single nucleotide polymorphisms (SNPs) revealed that MT 2 is exclusively functionally relevant to type-2 diabetes (T2D) 12 . Given the different physiological roles of MT 1 and MT 2 , obtaining selective ligands is highly desirable. However, most of the drugs on the market or under clinical evaluation, such as ramelteon 20 , tasimelteon 21 , and agomelatine 22 , are nonselective 23,24 . Currently, developing subtype-selective drugs, especially MT 1 -selective drugs, is very difficult. Moreover, the pharmacophores of the currently available melatonergic agonists are limited to very few types 25 . For example, an indene or naphthalene bioisostere group was repeatedly applied in melatonergic ligands [23][24][25] . Therefore, delineating the detailed configuration of the ligand-binding pocket of MTs is of great value for effective drug design. In this study, we determined the cryo-EM structures of active melatonin receptors engaged with the heterotrimeric G i protein. These structures reveal distinctive features of the orthosteric pocket in active MTs and provide insights into MT activation and G protein-coupling preference.

Result
The overall structure of the MT 1 and MT 2 signaling complex.
To obtain stable MT 1 -G i protein complexes, the nonselective agonists 2-iodomelatonin 26 and ramelteon 20 exhibiting high potency and affinity for the melatonin receptor were used. The receptor and G protein were co-expressed in insect cells. The assembled complexes were then purified to homogeneity for single-particle cryo-EM studies. The structures of 2-iodomelatonin-and ramelteon-bound MT 1 Table 1). The relatively high-quality density maps of the two complexes together with the previously solved inactive structure of MT 1 enabled us to confidently build atomic models composed of the ligands MT 1 , G i , and scFv16 27 . The majority of the side chains of MT 1 and G i protein residues are well defined in the structures (Supplementary Figs. 1 and 2). In addition to the protein part, an extra density located between the N-terminal portion of TM1 and TM7 was modeled as a cholesterol molecule. The same strategy was then applied to prepare the agonistbound MT 2 -G i complex. Eventually, the reconstituted ramelteonbound MT 2 -G i -scFv16 complex was subjected to cryo-EM data acquisition. After processing 120k particles from 3519 movies, an EM map of only 4.5 Å resolution was acquired. We speculated that improving the stability of the receptor might be helpful for obtaining high-resolution maps. Therefore, the three thermostable mutations, L108 ECL1 F/F129 3.41 W/C140 3.52 L, based on previous findings 28 , were chosen and introduced into MT 2 . These three residues were located far away from the ligandbinding pocket and the G protein-coupling interface 28 ; as a result, their mutations minimally interfered with the function of the receptor in terms of ligand recognition and G protein coupling 28 . With this triple mutant complex of ramelteon-MT 2 -G i -scFv16, we obtained an EM density map at a global nominal resolution of 3.5 Å, which enabled us to model most portions of the receptor, the ramelteon molecule, the G i protein, and the scFv16 (Fig. 1, Supplementary Fig. 3 and Supplementary Table 1).
The determined structures reveal that the overall assembly of MT 1 -G i and MT 2 -G i is similar to that of the most active GPCR-G-protein complexes (Fig. 1b). Both the ligands of 2-iodomelatonin and ramelteon used for complex reconstitution are bound at the orthosteric pockets of MT 1 and MT 2 (Fig. 1b). The 2-iodomelatonin-bound MT 1 -G i complex shares a nearly identical overall structure with that of ramelteon-bound MT 1 -G i , with root-mean-square deviation (RMSD) values of 1 Å for the Cα atoms of the whole complexes and 0.8 Å for the Cα atoms of MT 1 alone. The ramelteon-bound MT 2 -G i complex adopts a conformation similar to that of MT 1 -G i , with an RMSD of 1.4 Å for the Cα atoms of the receptor part, reminiscent of the structural similarity of the two receptors in their inactive states. However, noticeable structural differences are also observed on the extracellular side and in the region involved in G-protein engagement, as discussed below.
The orthosteric site and determinants of the ligand selectivity. In both the MT 1 and MT 2 structures, the ligand of 2-iodomelatonin or ramelteon binds at the orthosteric pocket constituted by TM3, TM5, TM6, TM7, and ECL2 (Fig. 2a, c). The binding pose of the ramelteon in active MT 1 or MT 2 is superimposable with those in inactive structures ( Supplementary  Fig. 4). However, the binding pose of the 2-iodomelatonin in the active structure is slightly different from that in the inactive structure, especially in the region of the alkylamide tail, which moves closer to the toggle switch residue W 6.48 in active MT 1 ( Supplementary Fig. 4). ECL2 sits on the top of the pocket and completely blocks ligand access from the extracellular side in both structures (Fig. 2b, d). The only entrance to the orthostericbinding site in the active state is the previously identified lateral channel between TM4 and TM5 (Fig. 2b, d). Comparison of the orthosteric pocket from active and inactive structures revealed remarkable differences: the pocket in the active structures is more constricted in the central part, but it extends to connect with a wide "longitudinal channel" formed by TM3, TM4, and TM5 (Fig. 2b, d). Inspection of the residues around the pockets reveals that while those surrounding the iodine group and alkylamide tail (named R 3 position in melatonin, Fig. 1) align well, several residues lining the solvent channel present very different conformations (Fig. 2e, f), which results in the varied configuration of the active pocket. Specifically, the aromatic residue Y187 5.38 in MT 1 rotates from the solvent-facing conformation in the inactive structure toward TM4 and forms a hydrogen bond with N162 4.60 in the active structure (Fig. 2e). This hydrogen bond further reduces the diameter of the ligand entrance and could restrict the release of the bound agonist, as the Y187 5.38 A mutation led to a fast ligand dissociation rate 27,28 . In addition, the N162 4.60 A mutation disrupted the function of MT 1 , again highlighting the functional importance of this hydrogen pair in MT 1 activation 27,28 ( Supplementary Fig. 5). In MT 2 , the corresponding pair N175 4.60 and Y200 5.38 also experienced conformational changes during transmission from the inactive to the active state (Fig. 2f). However, they take an opposite manner, and the hydrogen bond formed in the inactive structure disappears in the active structure (Fig. 2f). The absence of this hydrogen bond indicates that MT 2 does not need an MT 1 -like entrance-restricting hydrogen bond for activation, correlating well with the previous finding that the N175 4.60 A mutation showed no functional impairment 27,28 ( Supplementary Fig. 5). Notably, to facilitate the conformational changes of this N 4.60 -Y 5.38 pair, a conserved proline (P 4.59 ) is located close to N 4.60 in MT 1 and MT 2 , and mutation of P 4.59 disrupts the ligand-binding activity of MT 2 29 .
In addition to the conformational changes in the N 4.60 -Y 5.38 pair, residue H 5.46 (H195 5.46 in MT 1 and H208 5.46 in MT 2 ) (Fig. 2e, f), which is positioned two helical turns below Y 5.38 , displays the most striking difference in the pockets from the inactive and active structures (Fig. 2e, f). In the inactive structures, H 5.46 packs against TM4 and stays far away from the bound ligand (Fig. 2e, f). However, in the active structures, it undergoes a 2.4 Å inward displacement and adopts a flipped rotamer of the side chain, creating van der Waals interactions with the alkylamide tail of the bound ligand and the toggle switch residue W 6.48 (W251 6.48 in MT 1 and W264 6.48 in MT 2 ) (Fig. 2e, f). This rearranged conformation of H 5.46 explains the previous observation that the H208 5.46 A mutation affected MT 2 function 28,30 . Interestingly, although H 5.46 undergoes similar conformational changes in MT 1 and MT 2 , its functional importance seems to be different in the two receptors, as the mutation of H208 5.46 (MT 2 ) only moderately impaired the function of MT 2 , but the H195 5.46 A (MT 1 ) mutation caused a deleterious effect, as reflected by the low expression level of the mutant 27,28 . Owing to the movement of H 5.46 , additional space adjacent to the methoxy group of 2-iodomelatonin (named R 1 position in melatonin) is produced, resulting in the formation of the abovementioned "longitudinal channel" (Fig. 2b, d). We then docked two representative ligands that possess moderate MT 1 selectivity, 5-hydroxyethoxy-Nacetyltryptamine (5-HEAT) 31 and CTL 01-05-B-A05 27 , to both receptors ( Supplementary Fig. 6). Compared to melatonin, both 5-HEAT and CTL 01-05-B-A05 have substituents at the R1 position. In MT 1 , 5-HEAT occupied a position superimposable with that of the bound 2-iodomelatonin by forming hydrogen bonds with N162 4.60 and Y187 5.38 (Fig. 2g). Thus, the molecular structure of 5-HEAT is probably compatible with the pose of the active pocket required to activate MT 1 , resulting in 5-HEAT being an agonist of MT 1 . In contrast, as an antagonist of MT 2 , 5-HEAT did not take a plausible docking position in the observed active pocket of MT 2 , probably due to the varied conformations of N175 4.60 and Y200 5.38 . 5-HEAT probably utilizes an induced-fit model for binding MT 2, and the resultant configuration of the MT 2 pocket probably loses the ability to induce intracellular structural rearrangement. CTL 01-05-B-A05 adopts an extended conformation in MT 1 with the second naphthalene group packing against F194 5.45 (Fig. 2h). However, CTL 01-05-B-A05 binding to MT 2 was suboptimal, as the stacking interaction was disrupted by the precluding side chain of I207 5.45 at the corresponding position (Fig. 2i). Notably, based on our results, it is worth further optimizing the bitopic ligand to obtain more selective MT 1 agonists. For example, introducing novel substituents into the second unit for optimal fitting with the "longitudinal channel" will be a promising strategy.
Although the pockets from inactive MT 1 and MT 2 showed only subtle conformational differences in the region around the R 3 group of the ligand (termed subpocket), they became more discriminative in the active structures. MT 1 has a tyrosine at the position of 7.40 (Y282 7.40 ), while the corresponding residue in MT 2 is a leucine (L295 7.40 ) (Fig. 2j). Packing of Y282 7.40 against TM1 pushes the two adjacent residues Y281 7.39 and Y285 7.43 closer to the core of the pocket than the equivalent residues Y294 7.39 and Y298 7.43 in MT 2 (Fig. 2j). As a result, MT 2 has a larger subpocket, allowing the accommodation of ligands with bulky R 3 substituents, consistent with the molecular structures of most MT 2 selective agonists 23,30 . In summary, the structural and functional data revealed that although the ligand-binding pockets in MT 1 and MT 2 share similarities, they also present distinct  features. The unique configuration of the N 4.60 -Y 5.38 -H 5.46 motif and the "longitudinal channel" in MT 1 and the larger subpocket in MT 2 could be targeted to design MT subtypeselective drugs.
The active conformation of MT 1 and MT 2 . MT 1 from the G icoupled structures adopts an active conformation with the characteristic of TM6 outward displacement, similar to the other active GPCRs. Comparison of the active and inactive structures revealed that MT 1 maintains a similar conformation on the extracellular side, probably because both of them were stabilized by the agonist (Fig. 3a). In contrast, extensive structural rearrangements occur at the intracellular part (Fig. 3a). Among them, the 11 Å outward displacement of TM6 is most striking (Fig. 3a). In general, the outward displacement angle of TM6 is smaller in Gi/o-coupled GPCRs than in Gs-coupled GPCRs. However, MT 1 TM6 has a relatively large displacement, reaching the extent observed in G scoupled GPCRs 32 ( Fig. 3b and Supplementary Fig. 7). This unusual movement of TM6 is coordinated by the unique structural motifs in MTs, as discussed below. Accompanying TM6 movement, TM7 also undergoes conformational changes, and it takes 4.5 Å movement toward TM6 (Fig. 3a). Induced by the reorientation of H195 5.46 , TM5 also shifts 2 Å closer to TM6 (Fig. 3a). Moreover, its cytoplasmic end, which mediates interactions with the Ras-like domain of G αs , is extended by two additional helical turns in the active structure (Fig. 3a). Last, residue C130 3.55 bends the TM3 helix and reshapes the trajectory of ICL2 ( Fig. 3a and Supplementary Fig. 8). This repositioning of TM3 and ICL2 is necessary for G i accommodation; otherwise, they would make a severe steric clash with the αN of G i (Supplementary Fig. 8). Consistently, the MT 1 C130 3.55 mutation led to a decreased G-protein activation ability 33 . As expected, MT 2 also undergoes structural rearrangements in the regions of TM5, TM6, and TM7, resulting in an active conformation almost identical to that of MT 1 (Fig. 3c), consistent with their sequence homology and the structural resemblance between the inactive states. Conformational differences between active MT 1 and MT 2 were only observed in the extracellular end of TM1 and ICL2, where minimal sequence conservation exists (Fig. 3d). As mentioned above, the bulky Y282 7.40 in MT 1 , which is linked to the slight differences between the ligand-binding pockets, pushes TM1 more outward from the TM bundle (Fig. 3d). ICL2 in MT 2 adopts a different trajectory than that in MT 1 (Fig. 3d). The varied ICL2 in terms of sequence composition  TM1   TM2   TM3  TM4   TM5   TM6  TM7   ECL2   H8  TM7  TM6  TM5   ICL2   TM1   TM2  TM4   TM3   TM7  TM6   TM5   extracellular   and spatial localization probably confers the slightly different G protein-coupling profiles of MT 1 and MT 2 , as discussed below.
Comparison of the conserved motifs in MT 1 and MT 2 with those in other class A GPCRs reveals that they possess common as well as unique features. The toggle switch W 6.48 (W251 6.48 in MT 1 and W264 6.48 in MT 2 ) makes direct contact with the alkylamide tail of the ligand (Fig. 3e), accounting for the linkage between ligand binding and receptor activation. Based on the established activation mechanism of GPCRs, the toggle switch residue often undergoes conformational changes upon agonist binding. However, how MT 1 and MT 2 W 6.48 experience conformational changes remain unknown, as the state of W 6.48 from an inactive structure is lacking. W 6.48 was mutated to Phe in the previous inactive structures for stability improvement. The triadmotif P 5.50 -I 3.40 -F 6.44 , which is located close to the toggle switch, also experiences classical rearrangement (Fig. 3f). Interestingly, the repositioned H195 5.46 (H208 5.46 in MT 2 ) in active MT 1 sits right between the ligand and I 3.40 from the triadmotif and simultaneously forms interactions with the ligand and I 3.40 (Fig. 3e, f) (Fig. 3g). Notably, the commonly observed small hydrophobic residue V 6.41 in most class A GPCRs is replaced by F 6.41 in MT 1 and MT 2 . In the inactive structures, F 6.41 points outside and packs against V 5.54 . However, in the active structures, it flips close to Y 7.53 and stacks with Y 5.58 (Fig. 3h). This special hydrophobic core induces the particularly wide opening of TM6. Mutation of F 6.41 into valine or alanine crippled the signaling activity of MT 1 and MT 2 , demonstrating that a large cytoplasmic cavity is required for G i engagement in MT 1 and MT 2 (Fig. 3i). In summary, through structural rearrangements mediated by classical and unique microswitches, MT 1 and MT 2 adopt a similar active conformation with a wide opening in the cytoplasmic core.
Interactions between MT 1 , MT 2, and G i . The interactions between MT 1 , MT 2 , and G i are mainly mediated by the α5 helix of the G αi subunit, and the receptor cores comprise TM3 and TM5-7 from MT 1 and MT 2 (Fig. 4a). Although ICL2 forms hydrophobic or polar interactions with the G protein in most GPCRs, it mediates minimal contact with α5 and αN in G i bound to MT 1 and MT 2 (Fig. 4b). As a result, the G i interfaces in MT 1 and MT 2 are almost indistinguishable (Fig. 4a). Interestingly, even though MT 1 and MT 2 employ a nearly identical set of residues for accommodating G i , the bound G i still displays slightly different conformations (Fig. 4b). α5 and αN in MT 2 -G i have outward displacements of 1 and 2 Å, respectively, compared to MT 1 -G i (Fig. 4b). This conformational difference may be attributed to ICL2. The varied ICL2 in MT 1 and MT 2 might pose different impacts during G i engagement, which results in the observed G i conformations. Another possibility is that these two structures represent different conformational states during G i binding to MT 1 and MT 2 . Although the MT 1 -G i and MT 2 -G i complexes adopt a commonly observed overall assembly mode, they display distinct features as well. When aligned on the receptors, the G i bound to MT 1 and MT 2 show a relative orientation completely different from those presented by the G i in complex with the μ-opioid 34 , dopamine D2 35 , CB1 cannabinoid 36 , and NTSR1 neurotensin receptors 37 (Fig. 4c). The α5 helix in MT 1 -G i and MT 2 -G i undergoes significant rotational and translational movement away from TM3 and forms tight packing interactions with TM5 in an antiparallel manner (Fig. 4c). In contrast, owing to the large outward displacement of TM6, much fewer contacts are formed between TM6 and G αi in the MT 1 -G i and MT 2 -G i complexes (Fig. 4c).
Hydrophobic contacts dominate the part formed between the C-terminus of α5 helix and the receptor core, exemplified by the I344 G.H5.16 , L348 G.H5.20 , L353 G.H5.25 , and C351 G.H5.23 interacting with the residues from TM3, TM5, and TM6 (Fig. 4d, e). Remarkably, the involved Y 3.53 in MT 1 and MT 2 frequently appears as alanine in most class A GPCRs. In MT 1 -G i and MT 2 -G i , the bulky side chain of Y 3.53 probably contributes to establishing the special position of G i by restricting the α5 helix from moving away from TM5 (Fig. 4d, e). Indeed, mutation of Y 3.53 of MT 2 caused impaired G i signaling ability 38 . In MT 1 , V221 5.72 mediates another hydrophobic contact by burying its side chain into a hydrophobic pocket formed by Y320 G.S6.01 , F334 G.H5.06 , and A337 G.H5.09 (Fig. 4f). Polar interactions are also observed in the MT-G i interface and are mainly contributed by the three basic residues located in the cytoplasmic end of TM5: R 5.69 forms a salt bridge with E318 G.h4s6.12 from G αi ; R 5.71 (K 5.71 in MT 2 ) and K 5.73 forms a hydrogen bond with T340 G.H5.12 and the backbone carbonyl of I319 G.S6.01 , respectively (Fig. 4f). Due to the α5 movement away from TM3, most of the previously observed interactions between ICL2 and αN disappear in MT 1 -G i and MT 2 -G i complexes. For example, the well-known interaction mediated by the side chain of residue 34.51 from ICL2 and the hydrophobic cavity formed by the α5, αN, and β2-β3 loop of G αi is not observed in the MT 1 -G i and MT 2 -G i complexes.
Although MT 1 and MT 2 share almost identical residues in the receptor core, they still possess varied G protein-coupling abilities. MT 2 exclusively couples to G i/o , but MT 1 could also couple to G q/11 . Therefore, MT 1 should possess a special region in addition to the observed G i interface to support G q/11 engagement. As mentioned above, ICL2, which is involved in the G-protein association in most GPCR-G-protein complexes, indeed displays obvious sequence and conformational differences in MT 1 and MT 2 (Fig. 4a). Thus, ICL2 in MT 1 probably participates in the constitution of a unique interface, at which G q/11 binds and produces a resultant conformation different from that of the observed G i . In fact, this phenomenon has been previously observed for the two closely related receptors M1 and M2 muscarinic acetylcholine receptors: G 11 and G i were bound at different regions in the receptors, and their conformations were also different 39 . Altogether, MT 1 and MT 2 share a unique G i interface featuring the antiparallel packing of TM5 against the α5 of G i , and the varied ICL2 probably confers the additional subtype G protein-coupling ability of MT 1 .
Disease-related mutations in MT 2 receptor. MT 2 loss-offunction has been linked to type-2 diabetes (T2D) 12,38 . Of the mutants associated with T2D, nearly half impair G i signaling ability. When we mapped these affected sites onto the MT 2 -G i structure, we observed clustering of residues around the G i interface (Supplementary Fig. 9). Based on their positions, these mutations can be classified into three categories. The first category mutations are in the GPCR key functional motif, such as R138 3.50 H/L/C from the DRY motif and Y308 7.53 S of the NAxxY motif. The second category mutations are positioned in the G i interface, such as Y141 3.53 F and A234 5.72 T. The third category of mutations is located away from the above-identified functional region. They seem to impair G i binding by influencing the conformation of the cytoplasmic core. For example, R222 5.60 , which sits at the cytoplasmic edge of TM5, might contact the phosphate group from the membrane and contribute to shaping the conformation of TM5. Its mutation to histidine probably disrupts this interaction and leads to a varied configuration of TM5 without the G i coupling ability.

Discussion
Melatonin receptors have always been regarded as drug targets for developing therapeutics to treat insomnia, circadian rhythm and mood disorders, and cancer. Efforts to identify novel ligands with exceptional properties have been constantly made in the past four decades. However, only a limited number of chemotypes have been discovered for melatonergic ligands. The reason partially lies in the lack of structures providing information on the orthosteric ligand pocket. Indeed, immediately following the recent structural determination of MT 1 and MT 2 , new melatonergic ligands possessing novel chemotypes were successfully discovered by screening the ZINC database 40,41 . Therefore, uncovering the features of the ligand-binding site of melatonin receptors has valuable pharmacological significance. In this study, we solved the structures of MT 1 and MT 2 coupled to G i protein, the form representing the active state during receptor signaling. Interestingly, we found that the conserved H 5.46 , N 4.60, and Y 5.38 positioned around the ligand pocket adopt conformations different from those captured in the inactive state. These varied conformations result in a quite different ligand-binding site and the associated channel to the site. While our manuscript is in submission, Okamoto et al. report the structure of ramelteon-MT 1 -G i with varied conformations of H 5.46 , N 4.60 , and Y 5.38 42 . However, an inspection of their data revealed that the quality of their density map was poor and that the model was incorrectly built, especially in the regions of TM5, TM6, and ICL2 and for the residues of H 5.46 , N 4.60 , and Y 5.38 (Supplementary Fig. 10). Moreover, their map supports the model built here (Supplementary Fig. 10). Together, the distinctive feature of the MT 1 ligand site revealed here provides an unrevealed clue for designing novel ligands with MT 1 selectivity in the future. To date, the observed ligand-binding pocket in all solved structures is shaped with agonists sharing a melatonin-like chemotype core. However, when binding to the recently identified ligands with a new chemotype, reorganization of the pocket with remarkable differences from the current pocket probably would occur. Such plasticity in the ligand-binding site is also observed in other receptors, such as in the CB1 receptor 36 . Therefore, further delineating the pocket bound with various ligands will potentially drive a more efficient structure-based drug design.
One of the fundamental questions in the GPCR field is the G protein-coupling selectivity. To date, no general rule guiding GPCR-G-protein assembly has been concluded. For example, although muscarinic receptors share high sequence homology, they display different G protein-coupling profiles 39 . Similarly, MT 1 shares high sequence identity with MT 2 , but it couples more G-protein subtypes. Here, we revealed that MT 1 and MT 2 utilize a nearly identical set of residues to engage G i . The combined sequence and conformation of the receptor core result in a unique G i binding mode, consistent with the observation of divergent G i conformations in the reported structures. Although ICL2 participates in the interactions with G i in almost all the reported structures, it does not play a similar role in MT 1 and MT 2 . However, ICL2 is the only cytoplasmic region where clear differences were observed between MT 1 and MT 2 . Therefore, the ICL2 of MT 1 probably confers selective MT 1 coupling to G q/11 by generating a special interface different from that used for G i . Altogether, our study reveals the structural basis of G protein coupling to melatonin receptors and provides new clues for MT 1 and MT 2 selective drug design.

Methods
Constructs and complex expression. The coding sequence of wild-type human MTNR 1 A (MT 1 ) or modified human MTNR 1 B (MT 2 ) was cloned into the pFastbac A vector (Thermo Fisher Cat# 10360014) with a hemagglutinin (HA) signal peptide followed by a FLAG tag at the N-terminus and a hexa-histidine tag at the C-terminus. Three mutations (L108F, F129 W, and C140 L) were introduced into MT 2 to improve the protein behavior. Dominant-negative Gαi1 (DNGαi1) with mutations (G203A and A326S) 43 was cloned into the pFastbac A vector, and Gβ 1 γ 2 was cloned into the pFastbac Dual vector. MT 1, MT 2 , DNGαi1, and Gβ 1 γ 2 were coexpressed in Spodoptera frugiperda Sf9 cells (Invitrogen Cat# A35243) using the Bac-to-Bac baculovirus expression system (Thermo Fisher Cat# 10360014). Cells were infected at a density of 4 × 10 6 cells per ml with baculoviruses expressing MT 1 or MT 2 , DNGαi1, and Gβ 1 γ 2 at a ratio of 30:5:1. The infected cells were cultured at 27°C for 48 h, collected by centrifugation at 1000 × g for 10 min, and then stored at −80°C for future use. Image processing and 3D reconstructions. Image processing of the 2iodomelatonin-MT 1 -G i -scFv16 complex was conducted as previously described 46,47 . Briefly, data binned four times were used for micrograph screening and particle picking, and data binned two times were used for particle screening and classification. The particles after initial cleaning were subjected to particle extraction from the originally cleaned micrograph, and the resultant dataset was used for final cleaning and reconstruction. Specifically, raw movie frames were aligned with MotionCor2 using a 9 × 7 patch, and the contrast transfer function (CTF) parameters were estimated using Gctf and ctf in JSPR 48 . Only the micrographs with consistent CTF values, including defocus and astigmatism parameters, were kept for subsequent image processing. This process kept 3513 micrographs from 3762 raw movies. Templates for particle selection were generated by projecting the 3D volume of the V2R-Gs complex 49 . The 3,578,225 particles picked from template picking were subjected to 2 rounds of 2D classification, reducing their size to 1,131,706 and then reducing to 773,176 by 3D classification. After several rounds of ab initio refinement, the particles kept at 379,512 were subjected to nonuniform refinement for a 3.05 angstrom reconstruction. The image parameters were converted back and to Relion 50 and cryoSPARC 51 by use of the pyem package 52 . For the ramelteon-MT 1 -G i -scFV16 complex, dose-fractionated image stacks of the collected movies were subjected to beam-induced motion correction using MotionCor2.1 53 , and the CTF parameters for each micrograph were determined by Gctf 54 . The 2-iodomelatonin-MT 1 -G i -scFv16 complex was set as the 3D template for autopicking, and 2,074,844 particles were picked in RELION 3.0 55 . Four cycles of 2D classification were performed. After 2D classification, the particles with good features were further subjected to initial model building and 3D classification. Finally, 4,741,844 particles were further subjected to Bayesian particle polishing and CTF refinement. During postprocessing, different masks were applied on the global refinement map, and a map with a global resolution of 3.33 Å (FSC = 0.143) was obtained. To improve the map quality, the particles were transferred to cryoSPARC for homogeneous and nonuniform refinement. Finally, a density map was obtained with a global resolution of 3.29 Å (FSC = 0.143). Estimation of the local resolution was performed in cryoSPARC. For the ramelteon-MT 2 -G i -scFv16 complex, 4609 movies were collected and processed by cryoS-PARC v.3.2.0 56 . Beam-induced motion correction was performed using patch motion correction, and contrast transfer function (CTF) parameters were estimated by patch CTF estimation in cryoSPARC. Templates for particle selection were generated by projecting the 3D volume of the ramelteon-MT 1 -G i -scFV16 complex. A total of 5,234,455 particles were selected by autopicking (particle diameter set as 150 Å and Lowpass filter as 20 Å) as well as inspection picks, and they were then subjected to 2D classification. Three cycles of 2D classification yielded 847,573 particles with clear features. Six initial models were built by ab initio reconstruction. After that, 639,905 particles based on three 3D classes were selected for homogenous refinement in cryoSPARC. To improve the map quality, local refinement and sharpening were performed using the local refinement calculated B-factor. Finally, a density map was obtained with a global resolution of 3.5 Å (determined by gold standard FSC using the 0.143 criteria). Estimation of the local resolution was performed in cryoSPARC using the volume and mask of the local refinement output.
Model building, refinement, and computational docking. To build the 2iodomelatonin-MT 1 -G i -scFv16 complex, the crystal structure of 2-iodomelatonin bound MT 1 (PDB 6ME4) was used as the starting model. To build the ramelteon-MT 1 -G i -scFv16 complex, the built 2-iodomelatonin-MT 1 -G i cryo-EM structure and ramelteon-MT 1 crystal structure (PDB 6ME2) were used as the guiding models. For the ramelteon-MT 2 -G i -scFv16 complex, the 2-iodomelatonin-MT 1 -G i and ramelteon-MT 2 structures (PDB 6ME9) were used as the guiding models. For the G protein in the three complexes, the Gi heterotrimer from the FPR2-Gi cryo-EM structure (PDB 6OMM) was used as the initial model. All the models were first docked into the EM density map using Chimera 57 , followed by iterative manual adjustment and rebuilding in COOT 58 and phenix.real_space_refine in Phenix 59 .
The final model statistics were validated using MolProbity 60 . The molecular graphic figures were prepared with UCSF Chimera, UCSF ChimeraX 61 , and PyMOL. Docking of 5-HEAT and CTL 01-05-B-A05 to melatonin receptors was performed using AutoDock Vina 62 . Receptor or ligand structures were optimized using AutoDockTools-1.5.6. We set a 25×25×35 Å 3 box as the search region, as such a size was large enough for ligand binding. The outputs of the autodock were analyzed with PyMOL.
Pocket diameter determination. Channel dimensions were obtained using CAVER analyst v3.0.3 63 . Before performing the calculation, hydrogens were added to the corresponding coordinates using PyMOL. The region around the perceived channel entrance was manually set as the starting point of the tunnel. Using default program parameters, channel dimensions were extracted and trimmed to the segment between the channel entrance and ligand centroid. For visual comparison and comprehension, the section view prepared with PyMOL is shown.
cAMP assay. G-protein activation was measured using the cAMP-Gi KIT (cisbio, 62AM9PEB). The day before transfection, HEK293 cells (ATCC CRL-1573) were seeded at a density of 0.8 × 10 6 cells per well in 6-well culture plates and incubated at 37°C with 5% CO2. Cells were transiently transfected with plasmids coding the wild-type melatonin receptor or mutant receptor using Lipo8000 (Beyotime, C0533). Twenty-four hours posttransfection, the culture media was removed, and the cells were washed with PBS buffer. The transfected HEK293 cells were then plated into white 384-well plates (4000/well) in stimulation buffer and treated with 20 μΜ forskolin, 500 μM IBMX, and test agonist for 30 min. After that, 5 μL cAMP Eu-cryptate reagent and 5 μl anti-cAMP-d2 working solution were added 64 . After incubation at room temperature for 1 h, fluorescence was acquired at 620/665 nm using a FRET reader (PCLARIOstar Plus) 65 and analyzed with OriginPro 2021. The expression level of each construct was determined by flow cytometry analysis and was used to normalize the cAMP level. Specifically, transfected cells were labeled with DYKDDDDK Tag (D6W5B) Rabbit mAb (cell signaling technology, 15009S), and then washed with PBS. Cells were resuspended in ice-cold PBS, and the fluorescent intensity of single cells was quantified by CytoFLEX (Beckman) flow cytometer equipped with a 488 nm laser. The flow cytometry data were recorded and analyzed with CytExpert 2.3.0.84. Values of mean fluorescence intensity from~20,000 cells per sample were used for analysis. Statistical analysis was performed with either one-way analysis of variance and a Dunnetts post-test or a paired t-test, and significance accepted at P < 0.05.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support this study are available from the corresponding authors upon reasonable request. The density maps and structure coordinates have been deposited to the Electron Microscopy Database (EMDB) and the protein data bank (PDB) with the following accession codes: EMD-31980 and 7VGY for the 2-iodomelatonin-MT 1 -G i -scFv16 complex; EMD-31981 and 7VGZ for the ramelteon-MT 1 -G i -scFV16 complex; and EMD-31982 and 7VH0 for the ramelteon-MT 2 -G i -scFV16 complex. The PDB datasets used for analysis in this study include 6ME2, 6ME4, 6ME9, 7DB6, 6OMM, 7JVR, 6OSA, 6N4B, 7JJO, 7JV5, and 7AUE).