Overall structure
For the structural study, we used an N-terminally truncated form of QRFP, known as 26Rfa (Fig. 1a). 26Rfa is also found in vivo and possesses receptor activity, as comparable to that of 43Rfa7,8,15. To facilitate expression and purification, we truncated the C-terminal residues after G366 of human GPR103. We also used an engineered mini-Gαq (mini-Gsqi), a mini-Gs protein whose N-terminal and C-terminal residues are replaced by Gi1 and Gq, respectively16. To efficiently purify the stable GPCR-G-protein complex, the receptor and mini-Gsqi were incorporated in a ‘Fusion-G system’ by combining two complex stabilization techniques17 (Supplementary Fig. 1a, b). The modified receptor and G-protein were co-expressed in HEK293 cells and purified by Flag affinity chromatography. After an incubation with Nb35 and scFv16, which binds to mini-Gsqi, the complex was purified by size exclusion chromatography (Supplementary Fig. 1c, d). The structure of the purified complex was determined by single-particle cryo-EM analysis with an overall resolution of 3.37 Å (Table 1, Supplementary Fig. 2, and “Methods”). As the extracellular portion of the receptor was poorly resolved, we performed receptor focused refinement, yielding a density map with a nominal resolution of 3.48 Å, which was combined with the overall refined map. The resulting composite map allowed us to precisely build the atomic model of all the components, including the receptor (residues 3 to 243 and 263 to 346), ligand, G-proteins, and antibodies (Fig. 1b, c).
The receptor consists of the canonical 7 transmembrane helices (TM) connected by three intracellular (ICL1–3) and three extracellular (ECL1–3) loops, the amphipathic helix 8 at the C-terminus (H8), and the N-terminal residues exposed on the extracellular side (Fig. 1d, e). ICL3 was disordered in the cryo-EM map. At the secondary structure level, ICL2 and ECL3 contain short α helices, and ECL2 forms a long β sheet. Two disulfide bonds are observed in the GPR103 structure (Fig. 1e). One is the highly conserved disulfide bond between C1183.25 (superscripts indicate Ballesteros–Weinstein numbers) and C201ECL2, and another unexpected disulfide bond is formed between C2856.47 and C3277.48. Notably, the N-terminal residues extend above ECL2, constituting the extracellular domain (ECD) together with ECL2. An unambiguous density was observed from the interior of the transmembrane domain (TMD) to the ECD, allowing us to model residues 7 to 26 of QRFP (Fig. 1d, e). QRFP adopts an elongated confirmation, consistent with the NMR analysis of QRFP alone18.
On the intracellular side, the GPR103-Gq complex demonstrates the typical Gq binding mode. Specifically, the C-terminal helix of Gαq (α5h) is deeply embedded within the intracellular cavity formed by the outward displacement of TM6, resulting in the formation of an active signaling complex. In the proximity of the widely conserved N7.49P7.50xxY7.53 motif 19, the side chains of Y2345.58 and Y3327.53 are oriented towards each other (Fig. 2a). Around another conserved D3.49R3.50Y3.51 motif (modified to ERH in GPR103), R1433.50 forms a hydrogen bond with the backbone carbonyl of Y243G.H5.23 (superscript indicates the common Gα numbering [CGN] system20) in the α5h of Gαq. Furthermore, numerous residues in TM6 and 7 form hydrogen bonds with the C-terminus of α5h, comprising an electrostatic network (Fig. 2b). Regarding ICL2, the major interface following α5h, two remarkable interactions are observed. One is the hydrophobic interaction between ICL2 and the Gα subunit. Specifically, the bulky hydrophobic residue F151ICL2 is captured within a hydrophobic pocket in the Gα subunit, as in many other GPCRs16,21(Fig. 2c). The other is stabilization through stacking interaction: W155ICL2 forms π-π stacking and π-cation interactions with R32G.hns1.03, contributing to the stability of the active signaling complex. Finally, we compared the GPR103-Gq complex structure with those of other Class A GPCRs bound to various G-proteins. In terms of the binding angles for α5h, GPR103-Gq naturally exhibits similarity to Gq-coupled GPCRs rather than Gs- and Gi-coupled GPCRs13,16,17,22–24(Fig. 2d, e). Overall, GPR103 demonstrates a conserved activation mechanism and G-protein binding mode, consistent with those of many class A GPCRs.
QRFP binding site in the transmembrane domain
QRFP binds to both the ECD and TMD with its C-terminal amide directed toward the TMD core (Fig. 3a) The C-terminal heptapeptide of QRFP, fits vertically within the TMD and creates an extensive interaction network with TMs 2–7 and ECL2 of the receptor (Fig. 3b, c, Supplementary Fig. 3). This interaction between QRFP and the TMD can be broadly divided into those with RF-amide and the rest (residues 20–24). In the latter, F22 and F24 of QRFF form extensive hydrophobic interactions with the extracellular halves of TM2 and TM3, leading to the upright structure of QRFP inside the TMD (Fig. 3c). In addition, several hydrogen bonds exist between the receptor and the peptide backbone of QRFP. In comparison, RF-amide binds very tightly deep in the pocket (Fig. 3b), consistent with its significance in receptor binding reported in previous studies with mutant peptides. Specifically, the C-terminal amide forms hydrogen bonding interactions with T1022.61, Q1253.32, and Q3187.39. The side chain of F26 is present at the deepest position in the TMD, and surrounded by bulky hydrophobic residues in TM2, TM3, and TM6. The R25 residue forms electrostatic interactions with E203ECL2 and E2976.59, in addition to a hydrogen-bond with T2155.39. These two residues make the ligand-binding pocket of the TMD negatively charged (Fig. 3a). Thus, the C-terminal amidation functions not only in interactions with the receptor but also in enhancing the charge complementarity with the pocket, by neutralizing the negative charge of the C-terminal carboxylate.
Previous studies with mutant peptides have shown that the C-terminal heptapeptide is sufficient to activate GPR10325,26, despite its reduced affinity. The heptapeptide is conserved from fish to mammals, except for S23(Supplementary Fig. 4a), consistent with the fact that the S23 side chain poorly interacts with the receptor. The receptor residues interacting with the heptapeptide are highly conserved among the homologs (Supplementary Fig. 4b, c). Accordingly, this observed heptapeptide-TMD interaction plays a key role in evolutionarily-conserved GPR103 activation.
This RF amide is the only conserved part of the RF amide peptide (Supplementary Fig. 5a). To examine the conservation of its recognition mechanism, we compared the residues interacting with the RF amide with the corresponding RF amide receptors (GPR10, GPR54, GPR74, and GPR147) (Supplementary Fig. 5b, c). T1022.61 and Q1253.32, which recognize the C-terminal amide of QRFP, are completely conserved. Although Q3187.39 is replaced by histidine in all the other RF-amide receptors, it would form hydrogen bonds with the oxygen atom of the C-terminal amide of QRFP. These considerations suggest that amide recognition by hydrogen bonding interactions via these three residues is a common mechanism in RF-amide receptors. The two negative charges near R25 are also conserved in RF-amide receptors, although E2976.59 is replaced by alanine in GPR54. Instead, Q2115.35, which forms a polar contact with R25 in GPR103, is replaced by E2015.35 in GPR54, suggesting the conserved recognition of R25 by the two negative charges. In contrast, the recognition of the F26 side chain is less stringent but is shared by a bulky hydrophobic amino acid. Thus, the RF amide recognition mechanism observed in QRFP-GPR103 is highly conserved in RF amide receptors.
Unique architecture of the N-terminal region
We observed an unambiguous density above ECL2, despite the suboptimal local resolution ranging from 4 to 6 Å (Fig. 4a and Supplementary Fig. 2). The density aligned with the N-terminal structure was predicted by AlphaFold27–29. To enhance the fidelity of the map, we performed a 3D flexible refinement implemented in cryoSPARC30(Fig. 4b and Supplementary Fig. 6a-f). This improved map facilitated the accurate model building of the predicted N-terminal structure onto the density, employing both rigid body and all-atom refinement implemented in COOT. Furthermore, we successfully modeled QRFP26 up to N7. Thus, the current model allowed us to discuss the secondary structures of the N-terminal residues.
The N-terminal residues, numbering 2 through 40, of GPR103 extend from TM1 to above ECL2 and form a helix-loop-helix (HLH) (Fig. 4c). This N-terminal HLH motif is conserved among GPR103 homologs (Supplementary Fig. 4c), and our homology searches have failed to identify any similar sequences in other proteins, indicating unique feature of GPR103. QRFP extends vertically from the G20, interacting predominantly with the ECD and becoming sandwiched by the N-terminal HLH of the receptor. While QRFP20–26 are sufficient for receptor activation, their EC50 value is 75-fold lower compared to QRFP2625,26. This fact suggests that the ECD functions as an affinity trap, achieving high compatibility through interactions with the N-terminal residues of QRFP26.
The 3D flexible refinement also uncovered a significant conformational alteration on the extracellular side (Supplementary Fig. 6g, h and Supplementary Movie 1). We then built the models of this alteration on the two most significantly changed maps among the output (Fig. 4d, e and Table 1). The results revealed the upright and tilted states of the ECD. A structural comparison of the two states elucidated the dynamic movement of the N-terminal HLH by about 10 Å (Fig. 4f). Moreover, QRFP26, the ECD, and the extracellular half of the TMD oscillate like a pendulum with the C-terminus of QRFP26 as the base point. In the original map, the ECD of the refined structure is positioned in between the upright and tilted states, whereas the transmembrane helices are more closely aligned with the upright, implying the fundamental stability of the upright state.
To our knowledge, this N-terminal ECD configuration has never been observed in the class A GPCR structures or the AlphaFold database29, including the other RF-amide receptors. Moreover, among the RF amides other than QRFP, three consist of fewer than 12 residues, and the longer PrRP-31 lacks sequence homology except for the C-terminal RF-amide (Supplementary Fig. 5a). Consequently, the structure and dynamics of the ECD and QRFP observed in this study are unique to GPR103, thus playing a critical role in the GPR103 selectivity of QRFP.
Other classes of GPCRs commonly possess N-terminal domains with various lengths and functional roles, depending on the class. For example, class B GPCRs, akin to GPR103 as peptide receptors, feature an ECD with a ~ 150 amino acid 'hotdog-like' domain that acts as an affinity trap by encasing the agonist peptide31–33(Fig. 4g). Despite significant differences in sequence homology and length, a functional analogy exists between the ECDs of GPR103 and class B1 GPCRs. Additionally, parathyroid hormone receptor 1 (PTH1R) in class B1 GPCRs exhibits structural polymorphism in its ECD32 (Fig. 4g), similar to the fluctuating actions observed in GPR103 (Fig. 4f). However, only the N-terminal ECD of PTH1R moves independently of the TMD, in stark contrast to GPR103 where the entire ligand-binding pocket, iincluding the ECD, undergoes structural changes. This distinction may be attributed to the ligand conformation: class B1 GPCR ligands form a rigid, straight helical structure, whereas QRFP does not adopt any secondary structure.
Structural comparison with related peptide receptors
GPR103 exhibits a high degree of sequence homology with CCK receptors (CCKRs), orexin receptors (OXRs), and RY-amide neuropeptide Y receptors (YRs), which uniformly recognize peptides with amidated C-termini (Supplementary Fig. 5a) via their TMDs. To elucidate the characteristics of the structure and peptide recognition mechanisms of GPR103, we compared the structures of Y1R, CCK1R, and OX2R in complex with Gq12–14 (Fig. 5a). The lengths and secondary structures of the peptide ligands are diverse, and correspondingly, the lengths of the β-sheet in ECL2 are different. This comparison further highlights the distinctiveness of the N-terminal structure of GPR103 for peptide recognition. Despite the differences on the extracellular side, the C-termini of the agonist peptides as well as the intracellular sides of the receptors, superimposed well, suggesting an evolutionary linkage among these receptors.
We also examined the interactions of the C-terminal residues of the ligand in GPR103 and Y1R (Fig. 5b, c), representing the RF and RY amide receptors, respectively. In both instances, the C-terminal arginine of the ligand forms a hydrogen bond with T5.39 and ionic interactions with either aspartate or glutamate. The carbonyl oxygen of the C-terminal amide hydrogen bonds with T2.61 and Q/H7.39. The nitrogen of the amide hydrogen bonds with Q3.32, with C2.57 in close vicinity. In Y1R, the nitrogen and C932.57 are uniquely positioned to form a hydrogen bond (Fig. 5c), a feature not found in GPR103 (Fig. 5b), although this does not preclude the potential for such interactions in vivo. These highly conserved residues within the RF and RY amide receptors imply a shared recognition mechanism for the C-termini of the peptide ligands.
Next, we focused on the difference between the recognition of the RY-amide and RF-amide. In Y1R, Q2195.46 forms a hydrogen bond with the hydroxyl group of the C-terminal Y36 of the peptide ligand (Fig. 5c), which is highly conserved in the NPY receptors (Supplementary Fig. 5b). Exceptionally, Q2195.46 is replaced by L5.46 in Y2R34(Supplementary Fig. 5b), but its cryo-EM structure revealed that S5.42, situated one turn above, alternatively forms a hydrogen bond with Y36. Contrarily, in the QRFP-GPR103 complex, the hydrogen bond between Y36 and Q2195.46 is replaced by a hydrophobic interaction between F26 and L2225.46. L5.46 is highly conserved in the RF-amide receptors (Supplementary Fig. 5b). Exceptionally, in GPR10 L5.46 is replaced by the polar residue T5.46, which is not expected to interact with the C-terminal phenylalanine, due to its shorter side chain. The presence or absence of a residue capable of hydrogen bonding with the hydroxyl group of tyrosine at the C-terminus may distinguish the RF and RY amide receptors.
We also compared the interactions of the C-terminal two residues of GPR103 with CCK1R and OX2R (Fig. 5d, e). In CCK1R, the amidated end of the peptide forms a hydrogen bond with N982.61, similar to T1022.61 in GPR103. Furthermore, in CCK1R and OX2R, the amide group hydrogen bonds with Y7.43. Thus, while the amide recognition mechanism is conserved in CCK1R and GPR103 and in CCK1R and OX2R, GPR103 and OX2R are markedly different. In CCK8, the C-terminal F8 of the ligand is encased in a hydrophobic pocket similar to QRFP, whereas D7 forms a salt bridge with H2105.39, instead of the hydrogen bond between T2155.39 and R25 of QRFP (Fig. 5b, d). Meanwhile, the C-terminal M28 of OxB only contacts a few hydrophobic residues, and T27 forms a hydrogen bond with N3246.55, instead of T2155.39 in GPR103 (Fig. 5b, e). Thus, in addition to the marked differences in the extracellular region, two important residues characterized by the ligands are recognized by different mechanisms in the TMD and are crucial for the selective acceptance of the respective ligand peptides.