Single hormone or synthetic agonist induces G_s/G_i coupling selectivity of EP receptors via distinct binding modes and propagating paths

Significance

Nature has diversified functions of a single hormone to accomplish the concerted physiological reactions, but one particularly important question is how a single hormone–GPCR pair generate diverse signals, and dissecting the mechanisms underlying these questions could offer therapeutic routes for refractory diseases. In the present article, we illustrated the important role of the EP4-biased signal in treating AKI, and by resolving the agonist-induced G_s/G_i coupling selectivity of EP receptors, we revealed two hypotheses that account for a single hormone–induced functional diversity: 1) The single ligand–receptor interaction mode induces the different intracellular conformational states; 2) the receptor assumes different conformational states to recognize the same ligand, and further generates the different structural propagation paths.

Abstract

To accomplish concerted physiological reactions, nature has diversified functions of a single hormone at at least two primary levels: 1) Different receptors recognize the same hormone, and 2) different cellular effectors couple to the same hormone–receptor pair [R.P. Xiao, Sci STKE 2001, re15 (2001); L. Hein, J. D. Altman, B.K. Kobilka, Nature 402, 181–184 (1999); Y. Daaka, L. M. Luttrell, R. J. Lefkowitz, Nature 390, 88–91 (1997)]. Not only these questions lie in the heart of hormone actions and receptor signaling but also dissecting mechanisms underlying these questions could offer therapeutic routes for refractory diseases, such as kidney injury (KI) or X-linked nephrogenic diabetes insipidus (NDI). Here, we identified that G_s-biased signaling, but not G_i activation downstream of EP4, showed beneficial effects for both KI and NDI treatments. Notably, by solving Cryo-electron microscope (cryo-EM) structures of EP3-G_i, EP4-G_s, and EP4-G_i in complex with endogenous prostaglandin E₂ (PGE₂)or two synthetic agonists and comparing with PGE₂-EP2-G_s structures, we found that unique primary sequences of prostaglandin E2 receptor (EP) receptors and distinct conformational states of the EP4 ligand pocket govern the G_s/G_i transducer coupling selectivity through different structural propagation paths, especially via TM6 and TM7, to generate selective cytoplasmic structural features. In particular, the orientation of the PGE₂ ω-chain and two distinct pockets encompassing agonist L902688 of EP4 were differentiated by their G_s/G_i coupling ability. Further, we identified common and distinct features of cytoplasmic side of EP receptors for G_s/G_i coupling and provide a structural basis for selective and biased agonist design of EP4 with therapeutic potential.

To accomplish concerted physiological reactions and maintain whole-body homeostasis, single hormone may generate distinct effects at different organ/tissue locations. Nature has evolved a system to diversify hormone function at three primary levels: 1) Different receptors can recognize the same ligands (1–12), probably via variable amino acid composition at the receptor level; and 2) the same receptor–ligand pair can couple to different cellular effectors, in certain cases potentially dependent on different structural states with the same primary amino acid sequence for the particular receptor at the conformational level (additionally regulated by the presence of the effector or local effector concentrations); and 3) different alternative splicing or posttranslational modifications of G protein–coupled receptor (GPCR) may serve as a putative layer for the generations of diverse signals downstream of a single hormone–receptor pair engagement. Notably, examples of sensing the same hormone by different receptors or connection of the same receptor–ligand pair to different downstream effectors are often observed within the largest membrane receptor superfamily GPCRs, which account for more than one-third of all clinically used drug targets (13–16). Well-known instances include adrenal, dopamine, or prostaglandin E₂ (PGE₂) receptor systems (SI Appendix, Fig. S1 A and B) (1, 2, 4, 5, 17).

Mechanistic understanding of hormone function diversified at GPCR level requires the knowledge of how ligand engagement within the receptor pocket propagates to distinct intracellular rearrangements of the receptor for selective G protein/arrestin signaling. Crystal structures of AT1R have provided important mechanistic insights into the structural propagation paths underlying ligand-induced G_q/arrestin bias (18, 19). However, the structural basis for selective downstream signaling via plastic shaping of the hormone-binding pocket at the amino acid or conformational state levels, and mechanistic insights for the path delivering information for ligand binding to intracellular conformational states for GPCRs are still elusive. Moreover, biased ligands with the ability to selectively activate one of several downstream GPCR signaling pathways have the potential to avoid undesirable side effects (15, 20). Therefore, there is tremendous interest in understanding how the same or different ligands interact with their corresponding receptor pockets, and further transduce the binding signals to specific downstream effectors via selective propagating paths.

PGE₂ is an unstable fatty acid hormone produced by cyclooxygenase (COX) and three PGE synthases (mPGESs) and is widely distributed throughout the human body (21, 22). In response to PGE₂ stimulation, differential coupling of each EP receptor (EP1-EP4) to selective G proteins is fundamental to dictating their diverse functional outputs (SI Appendix, Fig. S1A) (17). In particular, EP4 can couple to both the excitatory G_s and the inhibitory G_i proteins to confer diverse functions such as renoprotection, cardioprotection, and antiinflammation, indicating the prevalence of signaling plasticity (23). For example, EP4 activation prevented the acute kidney injury (AKI) by reducing the expression levels of kidney injury molecule-1 (KIM-1) protein and other inflammatory factors such as IL-1β and TNF-α. Systemic administration of EP4 agonist reduced the serum creatinine (SCR) values and increased the survival rate in the rat model of AKI (Fig. 1A) (24). Moreover, the selective agonist ONO-AE1-329 induced EP4-G_s-cAMP signaling to promote the phosphorylation and plasma membrane translocation of aquaporin 2 (AQP2), thus providing an alternative strategy to treat NDI caused due to V2 vasopressin receptor (V2R) defects (25) (Fig. 1A). In addition, EP4-G_s signaling prevented the occurrence of glomerulonephritis, increased vasodilation, maintained vascular integrity, and showed cardioprotective effects (26–28). In contrast, the EP4-G_i pathway not only antagonizes cAMP accumulation, but also activates PI3K to inhibit the infiltration of eosinophils, resulting in an antiallergic inflammatory effect (29–31). Although the regulation of EP4 activity has therapeutic potential for the treatment of many diseases, the specific contribution of each specific G protein signaling, including both G_s and G_i, remains largely unknown (32–34). Therefore, it is important to understand the function and structural basis of the EP4-G_s and EP4-G_i signaling and to develop selective G_s/G_i-biased agonists of EP4, which will not only help dissect the function of biased EP4 signaling and the development of more precise drugs for a spectrum of diseases with EP4 as a promising therapeutic target, but also facilitate the elucidation of how a single hormone–receptor pair to recognize each other and transduce specific interactions via propagating path located in 7TM bundle for selective G protein subtype coupling in a broader context.

Fig. 1.

Functional characterization and biased properties of different EP4 agonists. (A) Schematic diagram showing that EP4 mediates membrane trafficking of AQP2, and restrains the KIM-1, inflammatory factor responses in AKI via the EP4-G_s pathway in renal. (B) Chemical structures of PGE₂, Rivenprost, L902688, and KMN-80. The carboxylic acid ester of Rivenprost’s α chain can be hydrolyzed into carboxylic acid (35). (C) Schematic diagram of the effects of Rivenprost and L902688 on AKI mice. The Cdh16Cre⁺/Gnas^flox/flox (Cko) mice represented the renal tubule-specific Gα_s knockout mice, while the Gnas^flox/flox (flox) mice were used as control. “√”, beneficial to treat AKI; “X”, not conducive to treat AKI; “XX”, accentuated the AKI. (D and E) The effects of Rivenprost on SCR (D) and BUN (E) levels of AKI mice (n = 8). Data are mean ± SEM. ***P < 0.001. Data were analyzed by Student’s t test. ^###P < 0.001, ^&&&P < 0.001. Data were analyzed by one-way ANOVA with Tukey’s test. (F) The Emax of ΔBRET values of AQP2 trafficking stimulated with Rivenprost alone, Rivenprost and PTX (100 ng/mL), Rivenprost and NF449 (100 μM), or Rivenprost and MF498 (1 μM). Data are mean ± SEM, (n = 4).^###P < 0.001. Data were analyzed by Student’s t test. ns, no significant difference. ***P < 0.001; **P < 0.01. Data were analyzed by one-way ANOVA with Tukey’s test. (G) Dextran permeability in response to L902688 (5 μM), Rivenprost (5 μM), Rivenprost (5 μM) and PTX (100 ng/mL), Rivenprost (5 μM) and NF449(100 μM), Rivenprost (5 μM) and MF498 (1 μM) in HUVECs. Data are mean ± SEM, (n = 4). ^###P < 0.001; ns, no significant difference. Data were analyzed by Student’s t test. ***P < 0.001, **P < 0.01. Data were analyzed by one-way ANOVA with Tukey’s test.

Results

The G_s- and G_i-Biased Properties of Different EP4 Agonists.

We evaluated the biased properties of available EP4 agonists using endogenous ligand PGE₂ as a reference. Among EP4 agonists, Rivenprost is a potential treatment for ulcerative colitis (36) and osteoporosis (37) (Fig. 1B and SI Appendix, Table S1). Whereas L902688 is G_i-biased, Rivenprost shows greater G_s-biased properties (SI Appendix, Fig. S1 C–F and Table S2). The separation window of the absolute value of biased factor β-value between Rivenprost and L902688 in human or mouse EP4 is more than 1 to 2, respectively, representing at least 10- to 100-fold difference once comparing the potency or efficacy on selectivity of two pathways between two ligands (38–41). Therefore, the Rivenprost and L902688 compound pairs may be good candidates for analysis of different pharmacological functions mediated by the G_s/G_i-biased properties of EP4.

We next investigated the effects of Rivenprost and L902688 on AKI induced by cisplatin, a widely used chemotherapy drug for various solid malignant tumors (42, 43). However, with high mortality and morbidity, a consistent therapeutic treatment option to prevent cisplatin-induced AKI in patients remains unavailable. Importantly, the administration of 0.1 mg/kg Rivenprost, but not 0.1 mg/kg L902688, was found to significantly reduce the levels of serum creatinine (SCR), blood urea nitrogen (BUN), kidney injury molecule-1 (KIM-1) protein, and the inflammatory factor related to AKI, including IL6, TNF-α, and MCP1 in the kidney induced by intraperitoneal injection of cisplatin (30 mg/kg) (Fig. 1 C–E and SI Appendix, Fig. S3 A–L). The concentrations of these two compounds were administrated according to previous studies (SI Appendix, Table S3) and further examined by our bioactivity measurements (SI Appendix, Fig. S2). Notably, specific deficiency of the Gαs protein in renal tubes by Cdh16Cre⁺/Gnas^flox/flox (SI Appendix, Fig. S3M) abolished the beneficial effects of Rivenprost on AKI induced by cisplatin (Fig. 1 C–E and SI Appendix, Fig. S3 A–E). In addition, L902688 treatments aggravated the SCR, BUN, and inflammatory responses of AKI by G_s deficiency in Cdh16Cre⁺/Gnas^flox/flox mice (SI Appendix, Fig. S3 F–L). We further examined the effects of EP2 antagonist TG6-129, EP3 antagonist DG-041, and EP4 antagonist GW627368X to inspect the contributions of these EP receptors in preventing AKI after Rivenprost administration. The results indicated that the G_s-biased protective effect of Rivenprost in vivo was mainly through EP4 signaling, but not EP2 or EP3 (SI Appendix, Figs. S3 N–P and S4 A–H). These results suggested the essential role of the G_s signaling in renoprotection of EP4 activity, and the G_s-biased EP4 agonist may have better therapeutic potential than G_i-biased EP4 agonist for AKI treatments.

In addition to AKI, activation of EP4 was recently proposed as an alternative therapy for NDI (25, 44). We then screened EP4 agonists to explore their efficacy in regulating AQP2 trafficking to the plasma membrane using BRET assay (SI Appendix, Fig. S4I) (45). Notably, all four EP4 agonists promoted the plasma membrane AQP2 trafficking, and the maximal response of AQP2 trafficking by Rivenprost was significantly higher than with other EP4 ligands (SI Appendix, Fig. S4 J and K and Table S4). The application of either the selective EP4 antagonist MF498 or the G_s inhibitor NF449 significantly blocked AQP2 trafficking, whereas administration of the G_i inhibitor PTX promoted it (Fig. 1F and SI Appendix, Fig. S4 L–N and Table S4). Therefore, the G_s and G_i pathways of EP4 inversely regulate AQP2 trafficking, and a G_s-biased agonist may have greater therapeutic potential to treat NDI by promoting AQP2 trafficking.

Activation of EP4 may have protective effects against mechanically induced lung injury, atherosclerosis, or other inflammatory diseases by enhancing the endothelial barrier function (46–48). Rivenprost, but not L902688, markedly reduced FITC-dextran permeability in a G_s-dependent manner, indicating the potential of a barrier-protective effect mediated by EP4 (Fig. 1G and SI Appendix, Fig. S4 O and P). Taken together, these series of experiments indicated that G_s-biased EP4 agonist may have greater therapeutic potential for AKI, NDI, and inflammation-related diseases.

Overall Structures of Agonist-Bound EP3-G_i, EP4-G_s, and EP4-G_i Complexes.

We determined the structures of EP4-G_s-Nb35 in complex with PGE₂, L902688, or Rivenprost; and EP4-G_i-ScFv16 in complex with PGE₂ or L902688 by cryo-EM at an overall resolution of 3.1 Å, 3.3 Å, 3.2 Å, 3.1 Å, and 3.5 Å, respectively. To dissect how PGE₂ induced differential G protein subtype coupling through engagements with different EP receptors, we additionally solved the PGE₂-EP3-G_i structure at 3.5 Å resolution (Fig. 2 A–C and SI Appendix, Figs. S5–S8 and Tables S5 and S6). These structures, together with our recently solved PGE₂-EP2-G_s-Nb35 structure (49), enabled a paralleling comparison of the interaction modes of the PGE₂ inside of ligand pockets of these receptors (SI Appendix, Fig. S9 A and B). The overall structures of EP4-Gs complexed with agonists were highly similar in the receptor region, with similar X-shaped extracellular loop 2 (ECL2), but shorter ECL1 and different conformations of ECL3 compared with previously solved EP2 structure (49) (SI Appendix, Fig. S9C).

Fig. 2.

Cryo-EM structure of agonist-bound EP4-G_s, EP4-G_i and EP3-G_i complexes. (A) Cryo-EM density of the Rivenprost-EP4-G_s complex (Left), the L902688-EP4-G_i complex (Middle) and the PGE₂-EP3-G_i complex (Right). (B) Vertical positions comparison of the agonists in EP4-G_s complexes, EP4-G_i complexes, and the EP3-G_i complex. The brown dashed line indicates the insertion depth of the E-ring of L902688 in the EP4-G_s complex (lower line) or EP4-G_i complex (upper line). (C) Schematic diagram showing that the nature diversifies hormone function at three primary levels: 1) the yellow box: different receptors can recognize the same ligands, probably via variable amino acid composition at the receptor level; and 2) the green box: the same receptor–ligand pair can couple to different cellular effectors, in certain cases potentially dependent on different structural states with the same primary amino acid sequence for the particular receptor at the conformational level (additionally regulated by the presence of the effector or local effector concentrations); 3) the blue dotted box: different alternative splicing or posttranslational modifications of GPCR may serve as a putative layer for the generations of diverse signals downstream of a single hormone–receptor pair engagement.

A striking difference was the binding modes of the same ligand between G_s- and G_i-coupled EP4 structures. The EM density results indicated that L902688 was completely enveloped and embedded deeply inside the receptor in the EP4-G_s complex, whereas the same ligand was relatively shallow within the EP4-G_i complex (Fig. 2B and SI Appendix, Fig. S8 E and F). Moreover, local configuration of PGE₂ fitted with EM density in the EP4-G_s/G_i complex also exhibited notable differences (SI Appendix, Figs. S8 A and B and S9 A and B). The binding poses of these ligands inside of the EP4 pocket were supported by computational simulations (SI Appendix, Fig. S9 D–I). The E-ring of L902688 was inserted 5.7 Å deeper toward the transmembrane core in the EP4-G_s complex compared to the EP4-G_i structure (Fig. 2B). Engaging with the same endogenous hormone PGE₂, the primary sequence within the orthosteric pocket of EP2, EP3, and EP4 showed marked differences, which may play vital roles for their different downstream effector (G protein subtypes) coupling (Fig. 2C). Moreover, in particular for EP4, the different conformational states of the ligand pocket and corresponding propagating paths connected to the intracellular surface of the receptor, may underlay mechanisms governing the distinct downstream signaling of G_s vs. G_i in response to endogenous PGE₂ or synthetic agonist L902688 stimulation (Fig. 2C).

Primary Sequence Differences in EP Receptor Pocket Contributed to G Protein Subtype Coupling Selectivity.

The four human PGE₂ GPCRs, including EP1-EP4, recognized the same endogenous prostaglandin lipids and were connected to distinct G protein subtypes (50) (Fig. 2C). Because our EM map has better resolution with no artificial truncation or mutations of EP4 compared to the 3.3 Å PGE₂-EP4-G_s structure solved in parallel (SI Appendix, Fig. S10 A–H), we used our 3.1 Å PGE₂-EP4-G_s structural model for further structural analysis.

PGE₂ showed different binding configurations in all solved structures of EP receptors. The positions of the E-ring and ω-chain of PGE₂ in the EP2-G_s complex were lower than those in EP4-G_s/G_i and EP3-G_i complexes (Fig. 3A and SI Appendix, Fig. S9 A and B), though the α-chain of PGE₂ was at similar horizontal heights and recognized by the conserved Y^2.65T^ECL2R^7.40 motif and L/M^3.32L^7.36 motif (Fig. 3A and SI Appendix, Figs. S10 I–M and S11 A and B and Table S7). Totally 16 to 19 amino acids, with only six amino acids identical in primary sequence (Fig. 3A), participated in the interactions between EP receptors and PGE₂. The different amino acids in primary sequence could serve as important determinants for selective interactions of the PGE₂ within the ligand pocket of different EP receptors, and then connecting them to specific downstream G protein subtype couplings. For instance, the carboxy group of the PGE₂ α-chain engaged a hydrogen bond with the S28^1.39 of EP2, which was replaced by the allelic P^1.39 in EP3 or EP4. Compared with the bulged Q103^2.54–Q339^7.46 pair of EP3, the T82^2.54–S308^7.46 pair of EP2 had smaller side chains, providing enough space for penetration of the E-ring and ω-chain of PGE₂ inside the receptor core and establishing stable hydrogen bonds with the hydroxyl of PGE₂. Moreover, whereas the steric packing by S120^3.36-L304^7.42 of EP2 restricted the placement of the ω-chain of PGE₂ at a deep position, the allelic G141^3.36–A335^7.42 pair of EP3 and S103^3.36–A318^7.42 pair of EP4 provided less entrance and the ω-chain assumed a shallower pose (Fig. 3B and SI Appendix, Fig. S9 A and B). Alanine mutation of T82^2.54, T123^3.39, S308^7.46 of EP2 (49); Q103^2.54, Q339^7.46 of EP3 (51), and T69^2.54, P322^7.46 of EP4 (Fig. 3C) significantly affected the G_s/G_i activity of different EP receptors. These results provided important evidence that the difference of the primary amino acid sequence in the ligand pocket was important for the G_s/G_i coupling selectivity of EP receptors by accommodating distinct ligand-binding configurations.

Fig. 3.

Different binding modes of endogenous PGE₂ in distinct EP receptor subtypes. (A) Comparison of PGE₂ binding sites between EP4, EP2, and EP3. The L/M^3.32-L^7.36 hydrophobic interactions and the Y^2.65-T^ECL2-R^7.40 motif are shown as blue- and red-filled circles, respectively. The F102^3.35 engaged with PGE₂ in the EP4-G_s complex is highlighted by orange-filled circles. Residues without interaction are shown as blank circles. Residues circled by the red dashed line constituted the different interaction modes with PGE₂ in EP2 or EP3 compared to those in EP4, and the combined mutation of the allergic positions in EP4 that mutated to those of EP2 (M^3.32T^3.36L^7.42S^7.46) or of EP3 (Q^2.54G^3.36S^3.39Q^7.46) are shown. The F102, S103, and P322 in EP4 are surrounded by a blue dashed box. Ballesteros–Weinstein residue numbers are provided for reference. (B) The ω-chain of PGE₂ showed differences in the ligand interaction pattern in EP receptor structures with magenta in EP2-G_s, blue in EP4-G_s, orange in EP4-G_i, and green in EP3-G_i. The red dashed line represents the same horizontal plane. (C) Mutagenesis studies showing the effects of residues on ligand-binding pocket. Heatmap of the ΔpEC₅₀ (pEC₅₀ of mutant - pEC₅₀ of wild type) and Emax (normalized to 100% wild type) showed differences in the Gα_s-Gγ/Gα_i-Gγ dissociation assay. Data are mean ± SEM, (n = 3). (D) Comparison of the biased properties of the combined mutations of EP4. The bias factor (β value) of EP4-MTLS or EP4-QGSQ was calculated using the EP4-WT as the reference by the method in the Materials and Methods of Gα_s-Gγ/Gα_i–Gγ dissociation assay. Data are mean ± SEM, (n = 3).

Distinct PGE₂ Configurations within the Ligand Pocket in EP4-G_s and EP4-G_i Complex Structures.

Despite binding to the same EP4 receptor, the binding pose of PGE₂ in the EP4-G_i and EP4-G_s showed marked differences (overall RMSD of ~1.2 Å for average atoms) (SI Appendix, Fig. S11C). Compared to the EP4-G_s complex, the carboxyl end of the α-chain of PGE₂ in the EP4-G_i complex was rotated ~60° toward TM7, losing its hydrogen bond with T168^ECL2. The E-ring of PGE₂ in the EP4-G_i complex was rotated upward by ~60°, forming a new H-bond with T69^2.54, which was not observed in the EP4-G_s complex (SI Appendix, Fig. S11 A–C). Accompanied by E-ring rotation, the ω-chain swing ~18° toward TM7 and the atoms of C18-C20 were moved upward, forming additional contacts with I315^7.39 but lost the hydrophobic interaction with F102^3.35 in EP4-G_i compared with the EP4-G_s complex (SI Appendix, Fig. S11 C and D). Consistent with this observation, whereas T69^2.54A, I315A, and I315S showed greater effects on G_i activity more than G_s activity of EP4 in response to PGE₂ stimulation, F102^3.35contributed to the G_s more than Gi activity of EP4 in response to both PGE₂ and synthetic agonist L902688 stimulation (Fig. 3C and SI Appendix, Fig. S11 E and F and Tables S8 and S9). These structural observations and pharmacological characterizations suggested that specific structural features within the orthosteric ligand-binding pocket of single receptor EP4 determine its downstream G protein subtype coupling in response to endogenous agonist PGE₂ stimulation. Notably, because the current models of the PGE2 are on the edge of the noise of the cryo-EM maps and carboxyl groups tend to be sensitive to radiation damage, the real binding pose of PGE2 to EP4 may be different from the current models, which could be evaluated by other biophysical methods or technologies.

Orientations of the ω-Chain Correlated to G Protein Coupling Selectivity of the EP Receptor.

We noticed that the ω-chain in the EP receptor structures showed marked differences, including both the orientation and depth of insertion (49, 51) (Fig. 3B and SI Appendix, Fig. S11 G and H). In the PGE₂-EP3-G_i complex structure, the ω-chain of PGE₂ extended toward TM6 of the receptor and directly interacted with the “toggle switch” W295^6.48 (Fig. 3B). However, the ω-chain of PGE₂ was inserted deeply inside EP2 and had an orientation toward TM3 and TM7. The insertion depth of the ω-chain in both the PGE₂ bound EP4-G_s and EP4-G_i complexes was shallower than that of PGE₂-bound EP2, but deeper than that of PGE₂-bound EP3 (Fig. 3B). Importantly, the binding orientations of the ω-chain in the EP4-G_s and EP4-G_i complexes were between those of EP2 and EP3. The ω-chain diverged at an angle of 31° or 35° in the EP4-G_s complex in correlation to the ω-chains of the PGE₂-EP3 and PGE₂-EP2 complexes, respectively (SI Appendix, Fig. S11G). Interestingly, the ω-chain in the EP4-G_i complex was oriented closer to the PGE₂-EP3 complex, was separated by angles of 23° and 49°, correlating to the ω-chains of the PGE₂-EP3 and PGE₂-EP2 complexes, respectively (SI Appendix, Fig. S11H). Taken together, the conformational states of the ω-chain were separated by different orientations divided by the G_s/G_i transducer coupling abilities. Therefore, we described the PGE₂-binding modes in different EP receptors as the “goldfish” model, that is, the α-chain and E-ring of PGE₂ form the “trunk part” with relatively conserved interactions among the EP receptor members. The ω-chain serves as the tail part, which wiggles toward different orientations (SI Appendix, Fig. S11I).

It is worth to note that human EP2 specifically interacts with G_s protein, whereas human EP3 selectively couples with G_i protein (Fig. 2C). Combined mutations of EP4 pocket residues with allelic residues in EP2 (EP4-MTLS) rendered EP4 more biased toward G_s signaling, whereas the allelic mutations EP4 to EP3, by combined mutations of EP4-QGSQ enabled EP4 to be more biased toward G_i (Fig. 3D and SI Appendix, Fig. S11 J and K). These results indicate that the binding modes of the ω-chain of the endogenous ligand PGE₂, as well as the different primary sequence between different EP receptor subtypes, are important determinants of EP receptor G protein coupling selectivity.

Structural Basis for Selective EP4 Agonism.

Selective activation of EP4 has therapeutic potential in the treatment of a spectrum of diseases, including acute and chronic kidney failure (24), glomerulonephritis (26), osteoporosis (37) and hypertension (27) etc. The cryo-EM structures of EP4-G_s complexes bound with Rivenprost and L902688 may provide structural basis for their selectivity.

Similar to PGE₂, Rivenprost and L902688 consist of three parts, including the α-chain, E-ring, and ω-chain, which interact with three subpockets within the EP4 orthosteric sites, respectively. Both polar and hydrophobic groups are accommodated in the subpocket A of EP4 because of its plasticity, mostly via interactions with the Y80^2.65-T168^ECL2-R316^7.40 motif (SI Appendix, Fig. S12 A–E). The region B subpocket of EP4 accommodated the E-ring of Rivenprost or L902688. The carbonyl group of the E-ring of Rivenprost or L902688 forms a hydrogen bond with T76^2.61 of EP4 (SI Appendix, Fig. S12 C and D), which was substituted by V^2.61 in all other EP family members (SI Appendix, Fig. S12E). Notably, region C is the tail region of the “goldfish” model, which shows high diversity between members of the EP family (SI Appendix, Fig. S11I). The polar residue S103^3.36 donates H-bonds to both the -COCH₃ group of the ω-chains of Rivenprost, whereas S103^3.36 was substituted by G^3.36 in EP1 and EP3 (Fig. 4 A and B and SI Appendix, Fig. S12 E–G). Moreover, EP4 had a smaller A318^7.42 compared with the allelic L304^7.42 in EP2, which enabled the positioning of the ether group (–COCH₃) of Rivenprost toward TM3 and reached L99^3.32 in the Rivenprost-EP4 complex (Fig. 4 A and B). Importantly, allelic mutations of T76^2.61V or A318^7.42L in EP4 caused a significant decrease in Rivenprost- or L902688-induced EP4 activity. Moreover, the reverse mutations V89^2.61T or L304^7.42A in EP2 also decreased its response to Taprenepag, a selective agonist (Fig. 4C and SI Appendix, Fig. S12 H and I and Table S10). The S103^3.36A mutation of EP4 (related to that of EP1 and EP3) caused significant reduction in EP4 activity in response to both Rivenprost and L902688 stimulation. Consistent with this observation, allelic mutations G141^3.36S of EP3 decreased its activity in response to selective agonist Sulprostone (SI Appendix, Fig. S12 H and I and Table S10), highlighting that these key regions confer agonist selectivity.

Fig. 4.

The selective binding pockets of synthetic agonists in EP4. (A) Cutaway view of Rivenprost in EP4 receptors. The residues T76^2.61 and S103^3.36 form a hydrogen bond with Rivenprost. (B) Sequence alignment of EP family members in ligand pocket. Note that T76^2.61 was only found in EP4; L304^7.42 was only found in EP2; S^3.36 is conserved in EP2 and EP4, but replaced by G^3.36 in EP1 and EP3. (C) The decreasing folds of the equivalent mutations of T/V^2.61, S/G^3.36 and A/L^7.42 between EP subfamily receptors in ligand binding. Data are mean ± sem.

Collectively, key amino acids in region B and region C of the three EP4 subpockets, in particular small and polar residues including T76^2.61, A318^7.42, and S103^3.36 which enabled H-bond formation and accommodated a larger group at region C at A318^7.42 -S103^3.36, contributed to the selective binding of agonists to EP4.

A G-Protein-Subtype-Dependent Agonist-Binding Pocket in EP4.

One striking and unexpected observation is the different binding modes of the same agonist (for both L902688 and PGE₂) in the EP4 receptor in complex with different G protein subtypes (Fig. 5 A and B and SI Appendix, Fig. S9 A and B). Notably, the L902688 could be positioned in the EM map of L902688-EP4-Gs by two orientations and the MD analysis suggested both binding modes are stable (SI Appendix, Figs. S9 D and E and S13 A–C). In the binding mode 1, the tetrazole ring faced extracellular side and form polar contacts with Y80^2.65, T168^ECL2, and R316^7.40 (SI Appendix, Fig. S13A). Because replacing these polar residues with hydrophobic residues, including mutations of Y80^2.65F, T168^ECL2V and R316^7.40M, significantly decreased Gs activity downstream of EP4 in response to L902688 stimulation, we choose mode 1 for further structural analysis (SI Appendix, Fig. S13 D–F and Tables S8 and S9). The exact conformation could be further evaluated by more detailed Cryo-EM structure analysis in the future.

Fig. 5.

A G-protein-subtype-dependent agonist binding mode in EP4. (A) The distinct configurations and positions of L902688 in the EP4-G_s and EP4-G_i complex. The red arrows indicate the movement of corresponding L902688 regions. Hydrogen bonds and polar interactions are shown by red dashed lines. (B) Interactional differences of EP4 in L902688 pocket between G_s and G_i signaling complex. (C) Schematic diagram of the ternary complex model (52–54). The model involves the interaction of the hormone (ligand), the receptor (EP4), and the downstream effector (G protein). The K_Lo represents the ligand interacting with the free receptor to constitute a low-affinity state, which then coupled with G protein, representing as K_C. The K_G represents that the free receptor coupled with G protein, and the K_Hi represents the ligand binding to such a precoupled receptor-G protein complex. The scheme also shows the formation of the high-affinity ternary complex for effective signaling. (D and E) The competition binding curves of L902688 in response to wild-type EP4 receptor, EP4-Gα_i/EP4-Gα_s fusion protein (EP4-G_i/EP4-G_s) and EP4-Gα_i/EP4-Gα_s with A318L mutation (EP4-A318L-G_i/EP4-A318L-G_s). L902688 showed two different binding states to EP4-G_i/EP4-G_s, containing a high-affinity state (K_Hi) and a low-affinity state (K_Lo). Gα fusion induced a fraction of EP4 into the high-affinity state, at 34% and 47% in EP4-G_s and EP4-G_i, and a large shift of K_Hi. In contrast, the K_Lo of both EP4-G_s and EP4-G_i are similar to wild-type EP4. A318L mutant weakened L902688 K_Hi and K_Lo, both in EP4-G_s and EP4-G_i. Importantly, the high-affinity fraction reduced from 34% in EP4-G_s to 9% in EP4-A318L-G_s chimeric mutation, while EP4-A318L-G_i chimeric mutation showed no reduction. Data are mean ± SEM, (n = 3).

Importantly, in contrast to the deep insertion inside of the 7TM bundle with an overall inversion C configuration in the EP4-G_s complex, L902688 assumed a Z shape and was bound to the orthosteric pocket with a much shallower position in the EP4-G_i complex (Fig. 5A and SI Appendix, Fig. S13G). These two binding modes of L902688 shared 11 common interacting residues but showed 10 different contacts (Fig. 5B and SI Appendix, Figs. S12C and S13F).

The α-chain of L902688 moved upward by 6.2 Å in the EP4-G_i complex compared with that in the EP4-G_s complex. The tetrazole ring constituted a polar network with the Y^2.65T^ECL2R^7.40 motif in the L902688-EP4-G_s complex structure. In contrast, the tetrazole ring of L902688 only parallelly packed against T79^2.64 in the EP4-G_i complex (Fig. 5A and SI Appendix, Fig. S13 E and F). In the EP4-G_i complex, the E-ring of L902688 was lifted by 5.7 Å, occupying the position of the tetrazole ring of the α-chain for L902688 in the EP4-G_s complex (Fig. 5A). The hydrogen bond between the carbonyl group of the E-ring and T76^2.61 was substituted by hydrophobic interaction with W169^ECL2 (Fig. 5A and SI Appendix, Fig. S13 H and I). An upward rotation of the indole ring of W169^ECL2 was observed in the EP4-G_i complex to accommodate these different binding poses (SI Appendix, Fig. S13 H and I). Accompanied by the upward movement of the α-chain and the E-ring, the phenyl ring of the ω-chain of L902688 was moved upward by ~2 Å and traverse by 4.4 Å; thus, the ω-chain was removed from the pocket created between TM3 and TM7, losing the interactions with S103^3.36, A318^7.42, and P322^7.46 in EP4-G_i compared to the EP4-G_s complex (SI Appendix, Fig. S13 H and I).

These observations suggest that distinct binding modes of L902688 may determine the coupling difference observed with downstream G protein subtypes. Consistent with this hypothesis, mutations of W169^ECL2L substantially impaired both G_s and G_i activity of EP4 in response to L902688 stimulation, with a larger effect on Gi. Conversely, mutations of S103^3.36A showed greater effects on L902688-induced G_s activity than that for G_i (Fig. 3C and SI Appendix, Tables S8 and S9).

Our structures revealed that the same ligand could bind to receptor-G protein complexes with distinct configurations, which are determinants of G protein subtype selectivity. These different binding modes could occur with free receptors before G protein binding, or were created after G protein association alone. The classical ternary complex model states that the G protein may play an important role in shaping the ligand-binding pocket of the receptor (Fig. 5C) (52, 53, 55); however, direct evidence for distinct ligand-binding poses induced by coupling of different G proteins has not been achieved previously at atomic resolution. Next, we combined the radio-ligand binding assay and the EP4-G_s/G_i chimeras to inspect the agonist-induced high-affinity states for EP4-G_s/G_i protein coupling (53–56). Notably, two ligand-binding sites with different affinities were identified in the competition binding curves of both the EP4-G_s and EP4-G_i chimeras in response to L902688 engagement, competing with [³H]-PGE₂. The goodness of fit of the competition binding curves by one or two sites was evaluated using the residual variance with the ratio of the sum of squares of residuals (RSS) divided by degrees of freedom (df) (Fig. 5D and SI Appendix, Fig. S13 J and K and Table S11) (57). The low affinity (K_Lo) of both EP4-G_s and EP4-G_i is similar to EP4 wild type, suggesting the affinity state of K_Lo represents the binding of the L902688 with the EP4 receptor alone. Conversely, the high-affinity (K_Hi) state should represent the state of binding of the L902688 to the EP4-G_s or EP4-G_i complexes. Importantly, the L906288-induced high-affinity fraction was reduced from 34% in EP4-G_s to 9% in EP4-A318L-G_s chimeric mutation. In contrast, L906288-induced high-affinity fractions in EP4-G_i chimera and EP4-A318L-G_i chimeric mutation are quite similar. These results are consistent with our hypothesis that introducing a larger residue at the A318 position (A318L) could specifically block the EP4-G_s ligand-binding pocket but had not much effect on the EP4-G_i ligand-binding pocket (Fig. 5 D and E). We therefore speculated that precoupling of EP4 to G_s or G_i may have significant effects on their ligand pocket conformations.

Distinct Pocket Configurations and Propagating Paths of EP4 G_s- and G_i-Biased Activation Mechanisms.

In the Rivenprost-EP4-G_s and L902688-EP4-G_s complex structures, the agonists were inserted approximately 6 Å deeper into the transmembrane core compared with that of the EP4-Gi complex structures, by placing the ω-chain between TM3 and TM7 (SI Appendix, Fig. S13L). The ω-chain of Rivenprost and L902688 was packed with P322^7.46 and S103^3.36 in the EP4-G_s complex (Fig. 6 A and B). In contrast, L902688 sat in a much shallower position and did not contact either P322^7.46 or S103^3.36 in the L902688-EP4-G_i complex (Figs. 5A and 6B). Importantly, mutations in S103^3.36A or P322^7.46A mainly impaired G_s but not G_i activity downstream of EP4 activation as observed in G protein dissociation assays in response to both L902688 and PGE₂ (Fig. 3C and SI Appendix, Tables S8 and S9). These results, along with the mutational effect of F102A (Fig. 3C), indicate that specific structural features with in the ligand pocket, including that of F102^3.36, S103^3.36, and P322^7.46, play potential common roles in determination of G_s- vs. G_i-biased property of different EP4 agonists. These specific structural features could serve as potential structural guidance for the development of biased ligands of EP4.

Fig. 6.

Distinct configurations and propagating paths of G_s and G_i-bias activation mechanisms of EP4. (A) Schematic diagram representing the propagating paths that contribute to the G_s/G_i bias decision mechanism. The ligand pocket in L902688-EP4-G_s (magenta) was deeper than that in L902688-EP4-G_i (blue). The propagating paths started with residues constituting two different ligand-binding pockets and extended to the G_s/G_i protein interface region. The residues in TM2, TM3, TM6, and TM7 are shown as brown, blue, magenta, and green circles, respectively. (B and C) The different structural rearrangement of the ω-chain of L902688 in the EP4-G_i and EP4-G_s complex. (D–F) Key residues along the propagating paths connected the ligand-binding pocket to the cytoplasmic side by comparison of L902688-EP4-G_s (green); L902688-EP4-G_i (blue); ONO-AE3-208-EP4 complexes (gray). (G) Effects of the residues’ mutations along the propagating path in response to L902688. The heat map is filled according to the ΔpEC₅₀ and Emax (100% wild type). “X”, no detectable signal. Data are mean ± SEM, (n = 3). (H and I) The competition binding curves of EP4-Gα_s (H) and EP4-Gα_i fusion protein (I) with V320A, N321A, and D325R mutants in response to L902688. Data are mean ± SEM, (n = 3). (J) The FlAsH-BRET responses of EP4 intracellular region between the ICL2 (S2) or TM7 terminus (S4) and the C terminus in response to the G_s/G_i signal. Data are mean ± SEM, (n = 3 or 4). The concentration (M) of Rivenprost and L902688, and the mutants along the propagating path are shown left. The G_s and G_i signal’s effects were surrounded by red and green dotted lines, respectively. **P < 0.01; ns, no significant difference. Data were analyzed by one-way ANOVA with Tukey’s test.

After aligning with TM3, binding of L902688 in the EP4-G_s complex enabled the inward movement of TM7, as observed in a 2.5 Å position shift of P322^7.46 and G106^3.39 compared to the inactive ONO-AE3-208-EP4 structure (Fig. 6 B and C). Moreover, the binding of L902688 in the EP4-G_i complex induced a further inward rotation of TM7, with a 2.3 Å shift of P322^7.46 (Fig. 6C). The interaction between the phenyl ring of L902688 and S103^3.36/P322^7.46 promoted the packing between V320^7.44 and C284^6.47 and the building of the polar network constituted by the side chain of C284^6.47, the main chain carbonyl of V280^6.43, and the side chain of N321^7.45. These rearrangements further led to the rotation of the carboxylate group of D325^7.49, enabling a polar interaction with the main chain carbonyl of N321^7.45 and readjustment of the packing between V281^6.44, I110^3.43, Y329^7.53, and L61^2.46 (Fig. 6 D–F and SI Appendix, Fig. S14 A and B). These structural rearrangements were propagated to the changes in conformational locks, including the D^7.49P^7.50xxY^7.53 and the ion lock E^3.49R^3.50Y^3.51 motifs on the cytoplasmic side. Moreover, the packing patterns of R117^3.50, Y329^7.53, M270^6.33, and L274^6.37showed notable differences in G_s or G_i-coupled EP4 structures (Fig. 6 E and F). These structural alterations enabled a ~1.4 Å outward tilting of TM6 in the L902688-EP4-G_i complex structure compared to that in the Rivenprost/L902688-EP4-G_s complexes (SI Appendix, Fig. S14C), which enabled accommodation of different G protein subtype C-tails (SI Appendix, Fig. S14D).

Notably, introducing a bulky side chain by replacing G106^3.39 with R abolished EP4 activation as the change blocks the structural rearrangement between TM7 and TM3 (58). Moreover, mutations along the propagating path that connected the ligand-binding pocket to the cytoplasmic region and showed greater conformational differences between the G_s- and G_i-coupled EP4 receptors bound with the same agonist, such as C284^6.47A, V320^7.44A, N321^7.45A, P322^7.46A, and D325^7.49A, significantly impaired G_s activity more than influence on G_i activity downstream of EP4 activation. D325^7.49R mutation completely abolished G_s activity, whereas C284^6.47R mutation abolished G_i activity downstream of EP4 activation (Fig. 6G and SI Appendix, Fig. S14E and Tables S8 and S9). We then inspected the contributions of the proposed conformational propagating path on high-affinity states of the agonist-EP4-G protein ternary complex exploiting the radio-ligand assay toward wild-type EP4 and EP4-G_s/G_i chimeras. Importantly, the L906288-induced high-affinity fractions of EP4-V320A-G_s, EP4-N321A-G_s, and EP4-D325R-G_s were reduced from 34 ± 6% in EP4-WT-G_s to 18 ± 4%, 24 ± 2% and 13 ± 3%, respectively, but showed significantly less effects on the high-affinity fraction of EP4-G_i chimeras (from 47 ± 10% to 37 ± 5%, 44 ± 5% and 34 ± 6% respectively (Fig. 6 H and I and SI Appendix, Fig. S14 F–H and Table S11). These results suggest that the conformational differences between key propagating path residues along TM3 and TM6-TM7, such as microswitches including V320^7.44/C284^6.47/N321^7.45 and D325^7.49, played important roles in the G_s/G_i coupling selectivity (Fig. 6A).

Connecting the Structural Propagating Path to Cytoplasmic Configuration of EP4 for G-Subtype Coupling Selectivity.

To provide further insights into the structural rearrangements localized in cytoplasmic side governing selective G protein subtype coupling, we generated a series of FlAsH-BRET sensors by incorporating FlAsH motifs into the three intracellular loops (ICLs, S1-S3) and TM7 terminus (S4), and the luciferase domain at the C terminus (59–63) of EP4 (SI Appendix, Fig. S14 I and J). We noticed that 10⁻⁶ M or 10⁻⁷ M Rivenprost only activate the G_s, but not the G_i downstream of EP4 wild type. Conversely, 10⁻⁶ M L902688 showed only G_i activity, but not G_s, after engagement with D325R mutant of EP4. In addition, 10⁻⁸ M L902688 or 10⁻⁶ M L902688 only activate D325A or F102A mutants of EP4, respectively (SI Appendix, Figs. S1C, S11F, and S14E). We therefore exploited the combined conditions of these agonists’ concentrations and appropriate EP4 mutants vs. EP4 wild type to profile potential structural features located at the cytoplasmic side of EP4 correlating to G_s vs. G_i coupling. Notably, the G_s activation conditions were specifically reported by the S2 sensor, which suggested that the separation between the C terminus and ICL2 (S2) in response to agonists stimulation. On the contrary, the Gi activation conditions were captured by the S4 sensor, which indicated the separation between the TM7 terminus (S4) and C terminus (Fig. 6J and SI Appendix, Fig. S14 K and L). Moreover, two propagating path mutations, the V320A and N321A, which mainly decreased G_s but not G_i signaling of EP4, showed significant reduction in the S2 signal in response to 10⁻⁶ M Rivenprost stimulation. However, they showed no observed changes on conformation reported by S4 compared with D325R mutation in response to 10⁻⁶ M L902688 stimulation (Fig. 6J). These results suggest that ligands of EP4 can induce specific intracellular configuration, which were associated with the residues in the ligand pocket and propagating path, to determine the G_s/G_i coupling selectivity.

Common Features and Determinants of G_s/G_i Coupling Selectivity of EP4.

Three EP4-G_s complexes have larger buried surfaces than that of the two EP4-G_i complexes (SI Appendix, Table S12). Although 22 to 25 residues of EP4 are involved in direct interaction with the G_s protein, only 15 to 16 residues directly contact G_i (SI Appendix, Fig. S15A). Importantly, 13 residues including F54^2.39, E116^3.49, R117^3.50, Y127^ICL2, and M213^5.68 of EP4 constitute the common structural determinants of EP4 for coupling of both G_s and G_i (SI Appendix, Fig. S15 B–E and Tables S8 and S9).

The different interaction modes between EP4 and biased agonists were propagated to distinct conformational changes located at the cytoplasmic sides of the TM bundles, These structural alterations enabled a larger separation of TM3 and TM6, and created a larger 7TM cavity in EP4-G_i complexes than EP4-G_s complexes (SI Appendix, Fig. S14D), which was distinct from previous results for GCGR-G_s/G_i complex structures (64), where a larger cavity was required to accommodate G_s coupling than G_i (SI Appendix, Fig. S15F). Compared with the EP4-G_s complex, the α5 Helix of Gα_i was rotated ~11° toward ICL2 (SI Appendix, Fig. S14D), enabling a “hook” configuration of the α5 helical end bending toward TM6 in the EP4-G_i complex, similar to previously solved CB1-G_i or GPR97-Go complex structures (63, 65). However, the EP4-G_s complex showed marked structural differences compared to those of classical G_s-coupled receptors, such as β2AR-G_s and DRD1-G_s complexes (64, 66, 67) (SI Appendix, Fig. S15 G and H), with the α5 Helix of the Gα_s inserting into the cavity between ICL1 and Helix 8 by a spiral extension (SI Appendix, Fig. S15 I–K). These structural comparisons indicated that changes in TM6 configuration play critical roles in accommodating different G protein subtypes. In addition, the distinct interactional packing modes between EP4 and Gα_s or Gα_i in G protein interface also contributed to the selective G protein coupling of EP4 (SI Appendix, Fig. S16 A–F). Taken together, different agonists of EP4 could allosterically regulate the conformational states of the cytoplasmic ends of TM3 and TM5-TM7 to selectively regulate its coupling with the G_s or G_i subtype. Although Y130^ICL2 is a determinant for specific G_i contact, E267^6.30 and T335^8.49, V336^8.50 contributes to selective G_s coupling in the EP4-G_s and EP4-G_i complexes.

Determinants for G_s/G_i Selectivity in the EP Family Receptors.

Comparisons of interfaces and sequence alignment results suggested that Y55^2.40 (H68^2.40 in EP2 and L89^2.40 in EP3), F217^5.72 (S229^5.70 in EP2 and K263^5.72 in EP3), and E267^6.30 (T277^6.30 in EP3) are different residues in EP3 structures, indicating that these residues may participate in G_s/G_i coupling selectivity (SI Appendix, Figs. S15A and S16 G–K). Importantly, the substitution of F217^5.72 in EP4 or S229^5.70 in EP2 by K263^5.72 in EP3 caused loss of its favored cation-π or polar interactions with G_s, and introduced repulsion at this equivalent position in EP3 (SI Appendix, Fig. S16 G and H). The replacement of E267^6.30 in EP2 and EP4 by T277^6.30 in EP3 caused loss of the preferred charge interaction with R385^H5.17 (SI Appendix, Fig. S16 G and H). Further, allelic mutations in EP2 and EP4 in corresponding positions significantly decreased G_s activity of EP2 or EP4 (SI Appendix, Fig. S16 L–N and Tables S8 and S9). Therefore, these three residues, and TM2, TM5, and TM6 are determinants of the G_s coupling activity of EP family members.

The EP3 and EP4 coupled with G_i via distinct mechanisms, shared few common structural features. Though the interaction mediated by Y130^ICL2 with G_i is specific for EP4 (SI Appendix, Figs. S16C and S17 A and B), the M270^6.33/T280^6.33 of EP4/EP3 underwent external rotation to create the space to accommodate the insertion of the α5 Helix of G_i, but not for the corresponding allelic residues H262^6.33 of EP2 (SI Appendix, Fig. S17C). These structural observations were further supported by mutagenesis effects (SI Appendix, Figs. S16 M and N and S17D and Tables S8 and S9).

Collectively, our studies identified common structural determinants of EP receptors for G_s coupling, including allelic residues Y/H/L^2.40, S^5.70/F^5.72/K^5.72, E/T^6.30 in TM2 and TM5-TM6. In contrast, the EP3 and EP4 coupled with G_i via distinct mechanisms, whereas the Y^ICL2 and M^6.33 in ICL2 and TM6 of EP4 played important roles in G_i coupling, the extended helix of TM6 in EP3 provided important G_i interface for efficient coupling (SI Appendix, Fig. S17 E and F).

Discussion

In this study, using the PGE₂-EP receptor system as a model (49, 51), we provide preliminary insights into how single hormones are recognized by different GPCR members to enable signal transduction to distinct downstream effectors. Importantly, the residues constituting the ω-chain-binding pockets showed notable diversity, enabling the separation of ω-chains by different orientations related to their G_s/G_i transducer coupling abilities. Among the 15 residues involved in ω-chain recognition in the EP receptors, 11 showed sequence-specific differences. Importantly, replacements of the ω-chain recognition residues of EP4 to the corresponding EP2 or EP3 allelic residues, created a more G_s-biased EP4 mutant or more G_i-biased EP4 mutant, respectively (Fig. 3 D and E). Therefore, the primary sequence differences in the ligand-binding pocket are determinants of PGE₂ recognition by different EP receptors that generate selective G_s/G_i subtype coupling abilities (Fig. 2C).

One particularly important question is how a single hormone–receptor pair generate diverse signals. Our works suggested that two hypotheses may account for this diversity: 1) The single ligand-receptor interaction mode indices two different intracellular conformational states; and 2) the receptor assumes different conformational states to recognize the same ligand, and further generates variable intracellular conformations through different structural propagation paths (Fig. 2C). Recent structural studies of the glucagon-GCGR complexes suggested the first model that the specific coupling of different G protein subtypes was governed by individual conformational states of intracellular loops with the same glucagon interaction mode (64). In contrast, our studies indicated that the EP4 receptor recognized both endogenous PGE₂ and the synthesized agonist L902688 through two different conformations, and correlated their selective coupling to G_s or G_i effectors. The PGE₂ assumes two different poses with distinct ω-chain conformations in the EP4-G_s or EP4-G_i complexes. The different interaction patterns of PGE₂ within the EP4 ligand pocket caused positional shifts and rotamer changes of specific residues in 7TM of EP4, which determined its downstream effector coupling selectivity. This conclusion was particularly supported by the observation that the two distinct pockets of EP4 accommodate L902688 to couple to G_s or G_i downstream effectors. Notably, the E-ring of L902688 was lifted by ~6 Å in the EP4-G_i complex compared with the EP4-G_s complex, and only 11 out of 21 total contact residues alone showed common interactions with L902688 in different binding states. The biochemical data supported these proposed mechanisms. Collectively, the above results demonstrate that the distinct binding modes of the EP4 pocket via a single hormone or synthetic agonist determine the selective coupling between the G_s and G_i effector proteins via a common structural mechanism with defined key motifs.

Mechanistically, different ligand–receptor pocket-binding modes dictate conformational states of the intracellular region of the receptor to enable selective downstream effector coupling, which lies at the heart of receptor biology and biased drug development targeting GPCRs. Results from recent studies have indicated that the potentially different interactions with L112^3.36 and Y292^7.43, and the subsequent conformational change in the side chains of N111^3.35, N295^7.46, and N298^7.49 serve as the key microswitches that determine the preferential recruitment of G_q/β-arrestin to AT1R (18, 19). However, the mechanism by which the structural propagation path enables a receptor to differentiate G_s/G_i signaling remains unclear. Our study shows that the shallower ligand interaction mode of the ligand in the EP4-G_i complex compared to the EP4-G_s complex caused the overall inward movement of TM7 and a larger separation of the cytoplasmic end of TM3 and TM6 to accommodate the C-terminal end of the α5 Helix of G_i. Consistent with this observation, our FlAsH-BRET biosensor indicated that separation between TM7 end and C terminus of EP4 is a structural feature reporting G_i coupling conformation, strengthened the key role of TM7 movement in G_i/G_s selectivity. In particular, direct interactions of the agonist with S103^3.36 and P322^7.46 in the pocket, as well as the rotamer changes for R117^3.50, V320^7.44/C284^6.47/N321^7.45/D325^7.49, and Y329^7.53/M270^6.33/L274^6.37 were identified as key microswitches for the conformational propagating path that connects ligand binding to the coupling of different G protein subtypes. Therefore, our results delineate how nature has delicately designed the conformational propagation path of a GPCR for translating ligand binding to selective cellular functions, and this property may be widely used in other ligand-GPCR systems.

In addition, it has been reported that the alternative C-terminal tail splicing isoforms of Bos EP3 determined its G-protein subtype coupling specificity (68, 69). For human EP3s, we can only detect Gi activation downstream of human EP3 isoform 4 in response to PGE₂ stimulation. In contrast, human EP3 isoform 6 could activate the G_s, whereas the isoform 7 could activate both the G_s and G_i signaling (SI Appendix, Fig. S17 G–J). In adrenergic receptor system, the PKA-mediated phosphorylation of β2-AR was reported to decrease the affinity of the receptor for G_s and increase its affinity for G_i (1, 3, 70). Furthermore, the PKA-mediated phosphorylation of β1-AR was reported to enable the transition between the G_s to G_i signaling (71). Therefore, the different alternative splicing or posttranslational modifications of GPCR may serve as another hypothesis for generating diverse signals downstream of a single hormone–receptor pair engagement (Fig. 2C).

In addition to insights into the structural basis underlying the biased preference for signaling via hormones and GPCRs, our studies showed that 1) notably, the functional analysis of G_s vs. G_i-biased signaling suggests that an EP4 agonist with greater G_s bias showed better therapeutic potential to treat AKI, NDI, and vascular inflammation. Though with potential side effects for abdominal discomfort and increased risk of cardiovascular disease, the therapeutic approaches using EP4-biased agonists may be developed as precise medicine for specific patients, such as patients with NDI and obesity (72–74); 2) specific residue combinations were only found in EP4 receptor, but not in other EP family members, providing H-bond formation potentials and enabling accommodation of larger chemical groups that could be used to structurally design selective agonists for EP4-targeted therapy. 3) Common and distinct residues contacting with endogenous or synthetic agonists of EP4, such as PGE₂ and L902688, served as determinants for G_s/G_i coupling specificity. 4) We identified the common EP4 activation mechanism and the different activation paths of the EP4 receptor for selective G_s/G_i coupling. 5) Although 13 residues of EP4 function as common structural features for EP4 for coupling to both G_s and G_i proteins, Y130^ICL2 was identified as a determinant for specific G_i contact, and E267^6.30 and T335^8.49-V336^8.50 contributed to the selective G_s coupling of EP4. 6) Finally, different residue components on the cytoplasmic side served as determinants of the G_s/G_i selectivity among EP4 family members.

Materials and Methods

Cisplatin-Induced AKI Model.

The cisplatin-induced AKI model was performed as previously described (75, 76) with details shown in SI Appendix files.

SCR and BUN Assays.

High-performance liquid chromatography (HPLC) was performed to measure SCR and BUN levels using the Agilent 1100 HPLC system. The details are shown in SI Appendix files.

Protein Expression and Complexes’ Formation.

Recombinant baculovirus of the protein for cryo-EM was generated using the Bac-to-Bac Baculovirus Expression System, and expressed in sf9 cell. The complexes for cryo-EM were reorganization in vitro. The details of the method are shown in SI Appendix files.

Cryo-EM Data Collection and Processing.

The detailed methods of cryo-EM data collection and processing are shown in SI Appendix files.

Gα_s- Gγ/Gα_i–Gγ Dissociation Assay.

The Gα_s-Gγ/Gα_i-Gγ dissociation performed in HEK293T or Cos-7 cells are shown in SI Appendix files.

Radioligand-Binding Assay.

Radioligand-binding assays were performed by using cell membrane from COS-7 transfected with EPs receptors. The details of the method are shown in SI Appendix files.

Structure Codes.

The codes in Protein Data Bank (PDB) and Electron Microscopy Data Bank (EMDB) databases of different structures are shown as follows: PDB ID 8GDB, EMD-29945 for the PGE₂-EP4-G_s complex; PDB ID 8GDA, EMD-29944 for the L902688-EP4-G_s complex; PDB ID 8GCM, EMD-29935 for the L902688-EP4-G_i complex, and PDB ID 8GD9, EMD-29943 for the Rivenprost-EP4-G_s complex; PDB ID 8GCP, EMD-29940 for the PGE₂-EP4-G_i complex; PDB ID 8GDC, EMD-29946 for the PGE₂-EP3-G_i.

Statistical Analysis.

All data are representative of at least three independent experiments. The data are presented as mean ± SEM, and the Student’s t test or one-way ANOVA with Tukey’s test was used for comparisons among groups by GraphPad Prism 7.0.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

The cryo-EM density maps and corresponding atomic coordinates of the different G protein-bound EP receptors complexes data have been deposited in Electron Microscopy Data Bank and PDB [The atomic coordinates and the cryo-EM density maps have been validated in Protein Data Bank (PDB) and Electron Microscopy Data Bank (EMDB) databases under accession numbers PDB ID 8GDB (77), EMD-29945 (78) for the PGE2-EP4-Gs complex; PDB ID 8GDA (79), EMD-29944 (80) for the L902688-EP4-Gs complex; PDB ID 8GCM (81), EMD-29935 (82) for the L902688-EP4-Gi complex, and PDB ID 8GD9 (83), EMD-29943 (84) for the Rivenprost-EP4-Gs complex; PDB ID 8GCP (85), EMD-29940 (86) for the PGE2-EP4-Gi complex; PDB ID 8GDC (87), EMD-29946 (88) for the PGE2-EP3-Gi. All the other data, associated protocols, code, and materials are detailed in SI Appendix.