Crystal structure of human endothelin ETB receptor in complex with peptide inverse agonist IRL2500

Endothelin receptors (ETA and ETB) are G-protein-coupled receptors activated by endothelin-1 and are involved in blood pressure regulation. IRL2500 is a peptide-mimetic of the C-terminal tripeptide of endothelin-1, and has been characterized as a potent ETB-selective antagonist, which has preventive effects against brain edema. Here, we report the crystal structure of the human ETB receptor in complex with IRL2500 at 2.7 Å-resolution. The structure revealed the different binding modes between IRL2500 and endothelin-1, and provides structural insights into its ETB-selectivity. Notably, the biphenyl group of IRL2500 penetrates into the transmembrane core proximal to D2.50, thus stabilizing the inactive conformation. Using the newly-established constitutively active mutant, we clearly demonstrate that IRL2500 functions as an inverse agonist for the ETB receptor. The current findings will expand the chemical space of ETR antagonists and facilitate the design of inverse agonists for other class A GPCRs.

E ndothelin receptors (ETRs) are G-protein-coupled receptors (GPCR) activated by vasoactive peptide, endothelins 1 . Two ETR subtypes (ET A and ET B ) are widely expressed in the vascular endothelium, brain, and other circulatory organs 2,3 . Endothelin-1 (ET-1) activates the both ETRs with sub-nanomolar affinities. The activation of the ET A receptor leads to potent and long-lasting vasoconstriction, whereas that of the ET B receptor induces nitric oxide-mediated vasorelaxation. Therefore, the upregulation of ET-1 is related to circulatory-system diseases, including pulmonary arterial hypertension (PAH) [4][5][6][7] . Moreover, the autocrine and paracrine signaling functions of ET-1 through the ET A receptor play a critical role in tumor growth and survival 8 . Thus, ETR antagonists have been developed for the treatment of circulatory-system diseases and cancers 6,7 . Bosentan is the first orally-active ETR antagonist 9,10 , and is used to treat PAH. The ET B receptor is the prominent ETR subtype in the brain, with high expression levels in astrocytes 11 . Stimulation of the ET B receptor modulates astrocytic responses, indicating its important roles in regulating astrocytic functions 12 . The upregulation of the astrocytic ET B receptor by ET-1 increases the vascular permeability and reduces the AQP4 levels, thereby aggravating vasogenic brain edema 11 . The application of ET Bselective antagonists may provide preventive effects against brain edema in the acute phase of brain insults [13][14][15][16] .
To date, most ETR antagonists have been developed based on bosentan 17,18 . The ETR antagonists that have been developed till now are mostly N-heterocyclic sulfonamides with similar structures and molecular weights, and non-sulfonamide antagonists (atrasentan, ambrisentan, darusentan, and enrasentan) still retain high similarities with each other and with the sulfonamides 6 . Since the ETR antagonists are chemically very similar 7 , and the expanded chemical space should be exploited. IRL2500 is a peptide-mimetic ETR antagonist developed based on the partial region of ET-1 19 , not on bosentan. IRL2500 has been characterized as an ET B -selective antagonist with an IC 50 value of 1.2 nM 20 , which shows higher affinity than that of bosentan. In an animal model, the intracerebroventricular administration of IRL2500 attenuated the cold injury-mediated brain edema and disruption of the blood-brain barrier, indicating the neuroprotective effect of IRL2500 14,15 . Clarification of the IRL2500 binding mode would facilitate the expansion of the chemical space of ET agents.
We previously reported the crystal structures of the ET B receptor bound to ET-1 21 and bosentan 22 ; however, both the binding mode and ET B -selectivity of IRL2500 remained to be elucidated. Here, we present the crystal structure of the ET B receptor in complex with IRL2500. This structure revealed the unique binding mode of IRL2500, which differs from those of ET-1 and bosentan. Structure-guided functional analyses clearly demonstrate that IRL2500 functions as an inverse agonist for the ET B receptor, and thus will provide the basis for the design of inverse agonists for other class A GPCRs.

Results
Overall structure. For crystallization, we used the previously established, thermostabilized ET B receptor (ET B -Y4) 22,23 . The IC 50 value of IRL2500 for ET B -Y4 was similar to that for the wild type receptor in the TGFα shedding assay 24 (Fig. 1a), suggesting that the thermostabilizing mutations minimally affect the IRL2500 binding. In contrast, the IC 50 value of IRL2500 for the ET A receptor is over 3 μM (Fig. 1a), indicating that IRL2500 has over 100-fold ET B -selectivity, consistent with the previous pharmacological analysis 20 . To facilitate crystallization, we replaced the third intracellular loop (ICL3) of the receptor with minimal T4 Lysozyme 25 (ET B -Y4-mT4L). Using in meso crystallization 26 , we obtained crystals of ET B -Y4-mT4L in complex with IRL2500 ( Supplementary Fig. 1a, b). In total, 58 datasets were collected and merged by the data processing system KAMO 27 . Eventually, we determined the ET B structure in complex with IRL2500 at 2.7 Å resolution, by molecular replacement using the antagonist-bound ET B structure (PDB code: 5X93) ( Table 1).
The overall structure consists of the canonical 7 transmembrane helices (TM), the amphipathic helix 8 at the C-terminus (H8), and two antiparallel β-strands in the extracellular loop 2 (ECL2), as in the previously determined ET B structures (Fig. 1b). The IRL2500-bound structure is similar to the bosentan-bound structure, rather than the ET-1-bound structure (R.M.S.D. values for Cα atoms = 1.34 and 1.95 Å, respectively), reflecting the inactive conformation. We observed a remarkable difference in the conformation of ECL2. The β strands are opened up by 9 Å, as compared with those in the ligand-free structure ( Fig. 1c and Supplementary Fig. 2a), and are the widest among the peptideactivated class A GPCRs ( Supplementary Fig. 2b). This conformation is facilitated by the crystal packing ( Supplementary  Fig. 2c), and is not a consequence of IRL2500 binding. This structural feature indicates the ECL2 is highly flexible in the inactive conformation of the ET B receptor, to capture the large peptide ligand endothelin.
IRL2500 binding site. We first describe the IRL2500 binding mode. IRL2500 consists of a tryptophan, a biphenyl group and a 3,5-dimethylbenzoyl group 19 . The biphenyl group forms a peptide bond with the tryptophan, and a peptoid bond with the dimethylbenzoyl group (Fig. 1d). IRL2500 binds to the transmembrane binding cleft exposed to the extracellular side, with a clear electron density (Supplementary Fig. 3a, b). The carboxylate group of the tryptophan moiety in IRL2500 forms salt bridges with R343 6.55 (superscripts indicate Ballesteros-Weinstein numbers 28 ) (Fig. 1e, f). The tryptophan side chain of IRL2500 forms a hydrogen bond with the carbonyl group of the N158 2.61 side chain and a cation-π interaction with the K161 2.64 side chain, The tryptophan also forms extensive van der Waals interactions with V177 3.28 , P178 3.29 , and F240 4.64 . The dimethylbenzoyl group of IRL2500 forms van der Waals interactions with the hydrophobic pocket, and is surrounded by V185 3.36 , L277 5.42 , Y281 5.46 , W336 6.48 and L339 6.51 . The biphenyl group penetrates deeply into the receptor core proximal to D147 2.50 , and forms van der Waals interactions with D147 2.50 , H150 2.53 , W336 6.48 , and S376 7.43 . Overall, the carboxylate of IRL2500 is specifically recognized by the positively charged residues of the ET B receptor, and the other moieties fill the space within the transmembrane binding pocket.
To elucidate the structural basis for the ET B -selectivity of IRL2500, we compared the residues constituting the IRL2500 binding site between the ET B and ET A receptors ( Fig. 2a and Supplementary Fig. 4). Most of the residues are conserved, while H150 2.53 , V177 3.28 , and S376 7.43 are substituted for the bulkier residues Y129 2.53 , F161 3.28 , and T359 7.43 in the ET A receptor, respectively. These substitutions may cause steric clashes with the aromatic groups of IRL2500 and reduce its affinity. To investigate this hypothesis, we measured the IC 50 values of IRL2500 for the H150Y, V177F, and S376T ET B receptor mutants. These mutants showed similar responses for ET-1 in the TGFα shedding assay ( Supplementary Fig. 5a). The H150Y mutant showed a similar response for IRL2500, and the S376T mutant showed a 3-fold increased potency (that is, 3-fold smaller IC 50 value) of IRL2500. Only V177F showed a 6-fold decreased potency of IRL2500 ( Fig. 2b and Table 2), suggesting that the F161 3.28 in the ET A receptor sterically clashes with the tryptophan moiety of IRL2500. Moreover, the H150Y/V177F double mutant showed the further decreased potency of IRL2500 by about 1.5-fold as compared with that of the V177F mutant only ( Fig. 2b and Table 2), suggesting that the two residues have some cooperativity in the ET A receptor. Notably, these mutations did not reduce (actually increased) the potency of the non-selective antagonist bosentan, showing that the two residues (H150Y and V177F) were selective to the IRL2500 inhibition. Taken together, Y129 2.53 and F161 3.28 in the ET A receptor may specifically clash with IRL2500, partially accounting for its ET B -selectivity. This is consistent with the results of the previous study, in which the replacement of the tryptophan moiety with the smaller moiety valine weakened its ET B -selectivity 29 .
Comparison of the binding modes of IRL2500, ET-1, and bosentan. IRL2500 is designed to mimic the F14, I19, I20, and W21 residues in ET-1, which play critical roles in ligand binding to the ET B receptor 19 . The tryptophan and dimethylbenzoyl groups of IRL2500 seem to be equivalent to W21 and I19 in ET-1, respectively, while the biphenyl group of IRL2500 seems to be equivalent to F14 and I20 of ET-1 (Fig. 3a). However, a comparison between the IRL2500 and ET-1 binding revealed an unexpected difference in their binding interactions (Fig. 3b). The carboxylate of the tryptophan in IRL2500 superimposes well with that of W21 in ET-1, and is coordinated by similar positively charged residues (Fig. 3c). The tryptophan of IRL2500 does not superimpose with the W21 of ET-1, but do with the I20. Instead, the dimethylbenzoyl group of IRL2500 superimposes with the W21. Nevertheless, these hydrophobic moieties of IRL2500 and ET-1 form comparable hydrophobic interactions (Fig. 3c, d). In contrast, the biphenyl group of IRL2500 penetrates into the receptor core, in an opposite manner to the F14 and I20 of ET-1 (Fig. 3b). Overall, while the electrostatic interactions between the carboxylates and the positively charged residues are conserved in IRL2500 and ET-1 binding, the other moieties form distinct interactions with the receptor. The volume of the ligand binding pocket in the ligand-free structure is large, thereby allowing the aromatic moieties of IRL2500 to flip (Fig. 3b).
IRL2500 has distinct chemical moieties as compared with bosentan, because IRL2500 was not developed based on bosentan. To reveal the similarities and differences in their binding modes, we compared the binding modes of IRL2500 and bosentan in detail (Fig. 3e, f). The carboxylate of IRL2500 and the sulfonamide of bosentan are similarly coordinated by the positively charged residue R343 6.55 , suggesting that this electrostatic interaction is a common feature of the antagonist binding      TM1   TM2  TM3   TM4   TM5   TM7   TM1   TM2   TM3   TM4  TM5   TM6   TM1   TM2   TM5   TM6   TM7   TM1   TM3  TM5   TM6   TM4   ECL2 TM1   TM2  TM3   TM4   TM5   TM7   TM1   TM2   TM3   TM4  TM5   TM6   TM1   TM2   TM5   TM6   TM7   TM1   TM3  TM5   TM6   TM4   ECL2   to the ET B receptor. In addition, like bosentan, the aromatic moieties of IRL2500 fit within the local hydrophobic pockets in the ET B receptor. Overall, IRL2500 has moieties that form similar binding interactions to those of bosentan. However, bosentan lacks the moiety corresponding to the biphenyl group of IRL2500, which deeply penetrates into the receptor core (Fig. 3f). Thus, IRL2500 fits into the pocket more tightly as compared with bosentan, contributing to its higher affinity for the ET B receptor.
IRL2500 functions as an inverse agonist. To obtain mechanistic insights into the receptor inactivation by IRL2500, we compared the ET B structures bound to ET-1, bosentan, and IRL2500. Previous structural studies showed that ET-1 binding induces the inward motion of the extracellular portion of TM6 including W336 6.48 , leading to receptor activation on the intracellular side 21 (Fig. 4a). Bosentan binding sterically prevents the inward motion of W336 6.48 with its 2-methoxyphenoxy group, and thus functions as an antagonist 22 (Fig. 4b). The dimethylbenzoyl group of IRL2500 superimposes well with the 2-methoxyphenoxy group of bosentan and similarly prevents the inward motion (Fig. 4b).
Moreover, the dimethylbenzoyl and biphenyl groups of IRL2500 sandwich the W336 6.48 side chain, tightly preventing its inward rotation (Fig. 4c). These observations suggest that IRL2500 tightly prevents the transition to the active state, as compared with bosentan, thereby possibly working as an inverse agonist that reduces the basal activity.
To investigate the inverse agonist activity of IRL2500 for the ET B receptor, we first measured the ligand-induced AP-TGFα release responses. In the wild-type ET B , treatment with IRL2500 or the antagonist bosentan with the cells for 4 h-incubation did not change the receptor activation level (Fig. 4d and Table 3).
These data suggested that IRL2500 either does not have the inverse agonist activity or that the assay is not sensitive enough to detect the inverse agonist activity. Indeed, we observed that the basal activity of the ET B receptor was very low in the assay and thus we could not distinguish whether IRL2500 functions as an antagonist or an inverse agonist by this assay.
Therefore, we tried the same assay by utilizing a mutation to facilitate the constitutive activity of the ET B receptor. Constitutively active mutant GPCRs have been employed in pharmacological characterizations of inverse agonists 30 , because such mutant GPCRs allow the signal measurement in a larger detection window. The substitution of the highly conserved L 3. 43 to glutamine has been identified as a causative activating mutation in the TSHR 31 and CYSLTR2 32 genes, which are related to hyperthyroidism and uveal melanoma, respectively. Therefore, we transferred the L 3.43 Q mutation into the ET B receptor (ET B -L192 3.43 Q) and examined its constitutive activity. We found that ET B -L 3.43 Q induced spontaneous AP-TGFα release (Fig. 4e), indicating that L 3.43 Q functions as a constitutively active mutation in the ET B receptor. We confirmed that this mutation also increased the constitutive activity in the ET A receptor ( Supplementary Fig. 6a). The L 3.43 Q-mutant ET B and ET A increased the potency of both ET-1 (EC 50 ) and the antagonists (IC 50 ) by approximately 5-fold and 2-fold, respectively (Supplementary Fig. 5b-d). We evaluated the inverse agonist activities of these compounds, using the constitutively active mutant, ET B -L 3.43 Q ( Fig. 4f and Table 3). As expected, the antagonist bosentan did not change the receptor activation from the baseline level. Conversely, IRL2500 reduced the basal activity of the ET B -L 3.43 Q mutant (EC 50 = 0.92 nM). Both bosentan and IRL2500 did not change the basal activity of the ET A -L 3.43 Q (Supplementary Fig.  6b, c). These data indicate that IRL2500 functions as a potent inverse agonist for the ET B receptor, consistent with the structural observations. The biphenyl group of IRL2500 prevents the inward motion of W336 6.48 to stabilize the inactive conformation, and thus IRL2500 functions as an inverse agonist.
In addition to IRL2500, K-8794 and BQ-788 have been characterized as potent ET B -selective antagonists. K-8794 is a high-affinity analog of bosentan 22 , whereas BQ-788 is a peptide analog with distinct chemical moieties, as compared with those of IRL2500 and bosentan 33 . We investigated the inverse agonist activities of K-8794 and BQ-788, using the constitutively active mutants of the ETRs (Fig. 4f and Table 3). K-8794 and BQ-788 reduced the basal activity of the ET B -L 3.43 Q with EC 50 values of 0.61 and 0.96 nM, respectively, indicating that they also function as inverse agonists. K-8794 and BQ-788 showed higher and lower efficacies (E max ) of the inverse agonist activity than that of IRL2500, respectively. The K-8794-bound ET B structure has been determined and thus we compare the ET B structures bound to K-8794, bosentan, and IRL2500. The chemical structure of K-8794 is similar to that of bosentan, but K-8794 has a dimethylphenyl group linked to the peptide bond. This modification slightly displaces the overall position of K-8794, thereby moving W336 6.48 outward by the interactions with the 6methoxyphenoxy and the alkyl groups ( Supplementary Fig. 7a). Therefore, K-8794 tightly prevents the inward rotation of W336 6.48 , in a similar manner to IRL2500 ( Supplementary  Fig. 7b). We thus suggest that this structural feature likely contributes to the inverse agonist activity of K-8794.

Discussion
We have determined the crystal structure of the ET B receptor in complex with the peptide-mimetic drug IRL2500, and thus elucidated the detailed receptor interactions and the structural basis for its ET B selectivity. Although IRL2500 is designed to mimic the partial region of ET-1, its binding mode is quite different. Moreover, using the constitutively active mutant ET B -L 3.43 Q, we revealed that IRL2500 together with K-8794 and BQ-788, but not bosentan, function as a potent inverse agonist for the ET B receptor, and provided the structural basis for their inverse agonist activities. Our study sheds light on the new aspects of the ET B -selective antagonists, and deepens our understanding of ETR pharmacology.
Although small-molecule ETR antagonists have been developed over the years; however, most ETR antagonists have been designed based on bosentan. Thus, the presently available ETR antagonists are chemically very similar. IRL2500 was developed based on ET-1 and has totally distinct chemical moieties, as compared with those of bosentan. However, the comparison of the IRL2500 and bosentan binding modes revealed the unexpected similarity in their binding interactions. This observation suggests that the charge-complementary interactions in the center of the pocket form the core of the receptor-antagonist interactions, and the other aromatic moieties fit the local hydrophobic pocket. The ligand binding pocket in the inactive ET B structures is larger than those in other GPCR structures, and thus aromatic moieties may be necessary to fit well within the pocket.
Our study revealed that the biphenyl group of IRL2500 penetrates deeply into the receptor core proximal to D147 2.50 , preventing the inward motion of W336 6.48 in TM6, and thus IRL2500 functions as an inverse agonist (Fig. 5a, b). The deep binding modes of the inverse agonists are also observed in the 5-HT 2C R and BLT1 structures (Fig. 5c-f). In the 5-HT 2C R structure, the 4-fluorophenyl group of the inverse agonist ritanserin interacts with F320 6.44 and W324 6.48 , the purported "toggle switch" important for GPCR activation 34 (Fig. 5c, d). In the BLT1 structure, the benzamidine group of the inverse agonist   (Fig. 5e, f), which is highly conserved among the class A GPCRs 35 . Sodium selectively competes with agonist binding in most class A GPCRs by stabilizing the inactive conformations, and the sodium binding site is thus an important pocket targeted in the design of negative allosteric modulators and inverse agonists. Instead of the sodium, the benzamidine group of BIIL260 directly hydrogen bonds with D 2.50 , similarly stabilizing the inactive conformation 36 (Fig. 5f). The binding mode of IRL2500 is similar to that of BIIL260 in BLT1, rather than that of ritanserin in the 5-HT 2C R (Fig. 5g, h). Although the biphenyl group of IRL2500 does not form any hydrogen-bonding interactions with the receptor, it prevents the conformational change around the D2.50 in a similar manner to the benzamidine moiety of BIIL260 (Fig. 5b, f). For the design of effective inverse agonists, the biphenyl moiety would also be useful as a modulation part along with another moiety that exerts specific and tight binding to the orthosteric site, as well as a benzamidine group.

Methods
Expression and purification. The haemagglutinin signal peptide, followed by the Flag epitope tag (DYKDDDDK) and a nine-amino-acid linker, was added to the Nterminus of the receptor, and a tobacco etch virus (TEV) protease recognition  sequence was introduced between G57 and L66, to remove the disordered Nterminus during the purification process. The C-terminus was truncated after S407, and three cysteine residues were mutated to alanine (C396A, C400A, and C405A) to avoid heterogeneous palmitoylation.  In the IRL2500bound ET B structure, the biphenyl group of IRL2500 forms van der Waals interactions with the receptor, and does not form any hydrogen-binding interactions. In the BIIL260-bound BLT1 structure, BIIL260 forms hydrogen bonds with D66 2.50 , S106 3.39 and S276 7.45 . g, h Superimposition of the IRL2500-bound ET B structure with the BLT1 (g) and 5-HT 2C R (h) structures using the Bac-to-Bac baculovirus expression system (Invitrogen). Sf9 insect cells were infected with the virus at a cell density of 4.0 × 10 6 cells per ml in Sf900 II medium, and grown for 48 h at 27°C. The harvested cells were disrupted by sonication, in buffer containing 20 mM Tris-HCl, pH 7.5, and 20% glycerol. The crude membrane fraction was collected by ultracentrifugation at 180,000 g for 1 h. The membrane fraction was solubilized in buffer, containing 20 mM Tris-HCl, pH 7.5, 200 mM NaCl, 1% DDM, 0.2% cholesterol hemisuccinate, 10 μM IRL2500, and 2 mg ml −1 iodoacetamide, for 1 h at 4°C. The supernatant was separated from the insoluble material by ultracentrifugation at 180,000 g for 20 min, and incubated with TALON resin (Clontech) for 30 min. The resin was washed with ten column volumes of buffer, containing 20 mM Tris-HCl, pH 7.5, 500 mM NaCl, 0.1% LMNG, 0.01% CHS, 10 μM IRL2500, and 15 mM imidazole. The receptor was eluted in buffer, containing 20 mM Tris-HCl, pH 7.5, 500 mM NaCl, 0.01% LMNG, 0.001% CHS, 10 μM IRL2500, and 200 mM imidazole. The eluate was treated with TEV protease and dialyzed against buffer (20 mM Tris-HCl, pH 7.5, 500 mM NaCl, and 10 μM IRL2500). The cleaved GFP-His 10 tag and the TEV protease were removed with Co 2+ -NTA resin. The receptor was concentrated and loaded onto a Superdex200 10/300 Increase size-exclusion column, equilibrated in buffer containing 20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.01% LMNG, 0.001% CHS, and 10 μM IRL2500. Peak fractions were pooled, concentrated to 40 mg ml −1 using a centrifugal filter device (Millipore 50 kDa MW cutoff), and frozen until crystallization. During the concentration, IRL2500 was added to a final concentration of 100 μM.
Crystallization. The purified receptor was reconstituted into molten lipid (monoolein and cholesterol 10:1 by mass) at a weight ratio of 1:1.5 (protein:lipid) 37 . The proteinladen mesophase was dispensed into 96-well glass plates in 30 nl drops and overlaid with 800 nl precipitant solution by a Gryphon LCP robot (Art Robbins Instruments) 26 . Crystals of ET B -Y4-mT4L bound to IRL2500 were grown at 20°C in precipitant conditions containing 30% PEG300, 100 mM Bis-tris, pH 7.5, 150 mM sodium phosphate monobasic, and 10 mM TCEP hydrochloride. The crystals were harvested directly from the LCP using micromounts (MiTeGen) or LithoLoops (Protein Wave) and frozen in liquid nitrogen, without adding any extra cryoprotectant.
Data collection and structure determination. X-ray diffraction data were collected at the SPring-8 beamline BL32XU, with 10 × 15 μm 2 (width × height) microfocused beams and an EIGER X 9M detector (Dectris). Various wedge data sets (10°) per crystal were mainly collected with the ZOO system 38 , an automatic datacollection system developed at SPring-8. The loop-harvested microcrystals were identified by raster scanning and subsequently analyzed by SHIKA 39 . The collected images were automatically processed with KAMO 40 (https://github.com/ keitaroyam/yamtbx). Each data set was indexed and integrated with XDS 41 , and the datasets were hierarchically clustered by using the correlation coefficients of the intensities between datasets 41 . After the rejection of outliers, 58 data sets were finally merged with XSCALE 41 . The IRL2500-bound structure was determined by molecular replacement with PHASER 42 , using the K-8794-bound ET B structure (PDB code: 5X93). Subsequently, the model was rebuilt and refined using COOT 43 and PHENIX 44  TGFα shedding assay. The TGFα shedding assay, which measures the activation of G q/11 and G 12/13 signaling 24 , was performed as described previously 22 . Briefly, a plasmid encoding an ET B construct with an internal FLAG epitope tag or an ET A construct was transfected, together with a plasmid encoding alkaline phosphatase (AP)-tagged TGFα (AP-TGFα), into HEK293A cells by using a polyethylenimine (PEI) transfection reagent (1 µg ETR plasmid, 2.5 µg AP-TGFα plasmid, and 25 µl of 1 mg per ml PEI solution per 10-cm culture dish). After a 1-day culture, the transfected cells were harvested by trypsinization, washed, and resuspended in 30 ml of Hank's Balanced Salt Solution (HBSS) containing 5 mM HEPES (pH 7.4). The cell suspension was seeded in a 96-well plate (cell plate) at a volume of 80 μl per well and incubated for 30 min in a CO 2 incubator. For the measurement of antagonist activity, IRL2500 was diluted in 0.01% bovine serum albumin (BSA) and HEPES-containing HBSS (assay buffer) and added to the cell plate at a volume of 10 µl per well. After 5 min, ET-1, at a final concentration of 0.5 nM, was added to the cell plate at a volume of 10 µl per well. For the measurement of agonistic activity, after adding 10 µl of the assay buffer, serially diluted ET-1 was mixed with the cells at a volume of 10 µl per well. After a 1 h incubation in the CO 2 incubator, aliquots of the conditioned media (80 μl) were transferred to an empty 96-well plate (conditioned media (CM) plate). Similarly, for the measurement of inverse agonist activity, the cells were mixed with 10 µl of the assay buffer, followed by the addition of serially diluted IRL2500, and incubated for 4 h before the transfer of the conditioned media. The AP reaction solution (10 mM p-nitrophenylphosphate (p-NPP), 120 mM Tris-HCl (pH 9.5), 40 mM NaCl, and 10 mM MgCl 2 ) was dispensed into the cell plates and the CM plates (80 µl per well). The absorbance at 405 nm (Abs 405 ) of the plates was measured, using a microplate reader (Spec-traMax 340 PC384, Molecular Devices), before and after a 1 h incubation at room temperature. AP-TGFα release was calculated as described previously 22 . The AP-TGFα release signals were fitted to a four-parameter sigmoidal concentrationresponse curve, using the Prism 7 software (GraphPad Prism), and the pEC 50 (equal to −log 10 EC 50 ) and E max values were obtained. To obtain equilibrium dissociation constant (K B ), for each experiment performed in parallel, IC 50 values (IRL2500, K-8794, BQ-788), an EC 50 value (ET-1), a Hill slope (K, ET-1), and tested concentration of ET-1 (A; 0.5 nM) were processed as follows 46 : The resulting K B values were logarithmically transformed and their negative values (pK B ) were used to calculate a difference in pK B value (ΔpK B ) for a mutant (MT) receptor as follows by using pK B value for WT receptor performed in parallel: The pK B and the ΔpK B values were used to calculate mean and SEM.
To measure the constitutive activity in a plasmid volume-dependent manner, HEK293 cells were seeded in a 96-well plate at a concentration of 4 × 10 5 cells per ml in Opti-MEM I Reduced Serum Media (Thermo Fisher Scientific), in a volume of 80 μl per well. A transfection mixture was prepared by mixing the PEI transfection reagent (0.2 µl per well) and plasmids (20 ng AP-TGFα plasmid, titrated ETR plasmid, and an empty vector to balance the total plasmid volume) in Opti-MEM I Reduced Serum Media (20 µl). The mixture was added to the cells, which were then incubated for 24 h before the transfer of the conditioned media. After adding the AP reaction solution, the absorbances of the cells and the CM plates were measured at 20 min intervals. The AP-TGFα release signals were calculated as described above, and the signal in the mock-transfected conditions was set at the baseline. A glutamine mutation was introduced into the internal FLAG epitope-tagged ET B and the ET A constructs at L192 3.43 and L176 3.43 , respectively.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Coordinates and structure factors have been deposited in the Protein Data Bank, under the accession number 6K1Q for the IRL2500-bound structure. The raw X-ray diffraction images are also available at Zenodo (https://zenodo.org/record/2803553). All other data are available from the authors upon reasonable request.