Distinct ETA Receptor Binding Mode of Macitentan As Determined by Site Directed Mutagenesis

The competitive endothelin receptor antagonists (ERA) bosentan and ambrisentan, which have long been approved for the treatment of pulmonary arterial hypertension, are characterized by very short (1 min) occupancy half-lives at the ETA receptor. The novel ERA macitentan, displays a 20-fold increased receptor occupancy half-life, causing insurmountable antagonism of ET-1-induced signaling in pulmonary arterial smooth muscle cells. We show here that the slow ETA receptor dissociation rate of macitentan was shared with a set of structural analogs, whereas compounds structurally related to bosentan displayed fast dissociation kinetics. NMR analysis showed that macitentan adopts a compact structure in aqueous solution and molecular modeling suggests that this conformation tightly fits into a well-defined ETA receptor binding pocket. In contrast the structurally different and negatively charged bosentan-type molecules only partially filled this pocket and expanded into an extended endothelin binding site. To further investigate these different ETA receptor-antagonist interaction modes, we performed functional studies using ETA receptor variants harboring amino acid point mutations in the presumed ERA interaction site. Three ETA receptor residues significantly and differentially affected ERA activity: Mutation R326Q did not affect the antagonist activity of macitentan, however the potencies of bosentan and ambrisentan were significantly reduced; mutation L322A rendered macitentan less potent, whereas bosentan and ambrisentan were unaffected; mutation I355A significantly reduced bosentan potency, but not ambrisentan and macitentan potencies. This suggests that – in contrast to bosentan and ambrisentan - macitentan-ETA receptor binding is not dependent on strong charge-charge interactions, but depends predominantly on hydrophobic interactions. This different binding mode could be the reason for macitentan's sustained target occupancy and insurmountable antagonism.


Introduction
Endothelins (endothelin-1, endothelin-2 and endothelin-3) are vasoactive peptides mainly produced by endothelial cells, but also by smooth muscle cells, fibroblasts and macrophages. Endothelin-1 (ET-1), which is known to be a potent and long lasting vasoconstrictor, also acts as a mitogen, angiogenic factor, mediator of fibrosis and inflammation, and has a pathogenic role in a variety of cardiovascular disorders [1]. ET-1 responses are mediated via activation of two homologous G protein-coupled receptor subtypes, endothelin receptor subtype A (ET A ) and endothelin receptor subtype B (ET B ) [2,3]. Both receptor subtypes activate Gq protein-mediated pathways leading to phospholipase Cb and PKC activation and increased intracellular calcium concentrations [4]. In lung tissue of patients suffering from pulmonary arterial hypertension (PAH) ET-1 concentrations are elevated [5,6]. These increases in local ET-1 concentrations cause activation of endothelin receptors in pulmonary arterial smooth muscle cells (PASMC). Consequently, increased intracellular calcium levels promote cytoskeletal contraction and cell proliferation [4,5,7] and thereby mediate persistent constriction and remodeling of pulmonary arteries, two hallmarks of PAH pathology [8][9][10][11].
The central pathogenic role of ET-1 in PAH has been demonstrated in several clinical trials evaluating different endothelin receptor antagonists (ERAs) [12].
The kinetic behavior of bosentan and ambrisentan is typical of high-affinity compounds with diffusion-controlled receptor interaction. Such compounds display fast receptor association and dissociation rates, and a typical receptor occupancy half-life for a diffusion-controlled 1-nM compound is less than 10 minutes (bosentan and ambrisentan have half-lives of 1 min!). In contrast, the kinetic behavior displayed by macitentan is typical of compounds for which factors beyond diffusion are limiting receptor association [20][21][22]. Such compounds are characterized by slower receptor association as well as slower dissociation rates.
Mechanisms potentially affecting ligand-receptor interaction kinetics include 1) the need for conformational changes of ligand and/or receptor during binding, 2) the type of interaction (electrostatic versus hydrophobic), and 3) the ease of release and re-entry of water molecules from/into the binding site.
In this study we characterized affinities and kinetic properties of various macitentan and bosentan analogs and show that the different kinetic properties originally found for macitentan and bosentan are replicated by structurally close analogs. We further demonstrate that macitentan prefers a compact conformation in aqueous media with minimized hydrophobic surface. Supporting molecular modeling studies suggest that this compact conformation optimally occupies a sub-pocket of the ET-1 binding site of the ET A receptor. Finally, functional studies in point-mutated ET A receptor variants revealed that the interaction of macitentan and the ET A receptor does not depend on charge-charge interactions, but is dominated by tight hydrophobic interactions that result from an optimal shape-match between antagonist and binding pocket.
We conclude that the topological and physicochemical properties of macitentan contribute to its slower receptor interaction kinetics, which differentiate is from the other two ERAs.

Kinetic properties of macitentan and a diverse set of structural analogs
Previous experiments in human primary PASMC revealed an almost 20-fold longer ET A receptor occupancy half-life of macitentan compared to ambrisentan and bosentan. This longer ET A receptor occupancy translated into insurmountable antagonism of ET-1-induced ET A receptor signaling by macitentan, but not bosentan or ambrisentan [18]. Here, we extended this analysis and included a set of 12 structural macitentan analogs and two bosentan-like structures, tezosentan [23] and clazosentan [24], to further characterize ET A receptor inhibition kinetics in functional assays (structures in Fig. 1A and Figure 2). To measure target affinity, the inhibitory potency of each compound was assessed using calcium flux assays in human primary PASMC, which predominantly express the ET A receptor subtype. Cells were preincubated with compounds for 120 min and the ET-1-induced calcium release (EC 80 : 4-12 nM; determined for each individual experiment) was measured using the fluorescence imaging plate reader (FLIPR). Fig. 1B shows a representative ET-1 concentration-response-curve (CRC) in calcium flux assays.
Insurmountability was assessed using IP 1 accumulation assays, which were previously used to demonstrate insurmountability of AT 1 blockers [25]. To this end, PASMC were pre-incubated with ERAs at different concentrations and then stimulated with a very high concentration of ET-1 (5 mM, corresponding to EC 100 and being ,1000-fold above the EC 50 ). Intracellular IP 1 levels were determined 20 min after ET-1 addition. Fig. 1D shows a representative ET-1 CRC in IP 1 accumulation assays. Fig. 1E shows representative CRC of macitentan, bosentan and ambrisentan against 5 mM ET-1, demonstrating antagonistic efficiency of macitentan and a lack of antagonistic effect of bosentan and ambrisentan up to 1 mM compound concentration. The ''percent (%) insurmountability'' of the tested ERAs was calculated from the maximal reduction of the ET-1-induced IP 1 signal achieved by the ERA (maximal ERA concentration 1 mM), compared to the IP 1 signal obtained without ERA (as shown Fig. 1E). The mean (n = 7) % insurmountability obtained for macitentan was 64%+/2 15%, whereas bosentan and ambrisentan displayed 0% insurmountability. The mean IC 50 values (in nM and log M) and the mean insurmountability (in %) of all tested ERAs and the peptidic selective ET A receptor antagonist BQ123 are summarized in Figure 2.
When the logIC 50 values of all tested compounds with IC 50 , 5 nM were plotted against the respective % insurmountability (Fig. 1F), a linear relationship (R 2 = 0.88) between the % insurmountability and the logIC 50 values for macitentan and its structural analogs was revealed. This correlation of antagonistic potency and insurmountability -i.e. of affinity and dissociation rate -suggested that the tested macitentan-like structures share a common ET A receptor interaction mode. In contrast, bosentan, bosentan analogs, ambrisentan and BQ123 showeda different relationship between potency and insurmountability, and compounds with comparable potency to macitentan and its analogs, were considerably less insurmountable indicating rapid ET A dissociation kinetics. As previously published, high affinity compounds with very rapid dissociation kinetics are likely to a display diffusion-controlled binding reaction [20][21][22]. This suggests that the ET A binding of bosentan-like structures and ambrisentan is likely diffusion-controlled and macitentan-like structures which display insurmountability and slower receptor dissociation have a distinct ET A receptor binding mode that is not diffusion-controlled.

Molecular modeling studies of the interaction between the ET A receptor and its antagonists
To rationalize this potentially different binding mode a structural level, we built a molecular model of the ET A receptor, based on the recently published orphanin FQ receptor structure [26] (PDB: 4EA3) (see Methods). The 'open' conformation of macitentan that was observed in single crystals grown in organic solvents ( Fig. 3A and [15]) could not be fitted into the expected ligand-binding pocket, which is a sub-pocket of the endothelin binding site in the ET A model (data not shown). However, two dimensional 1 H NMR spectroscopy (ROESY) of macitentan performed in an aqueous environment demonstrated intramolecular proton interactions (Fig. 3B) that suggested a more compact molecular conformation with minimal lipophilic surface exposed to solvent. Our NMR studies also showed that in aqueous solutions containing $10% of CD 3 CN this compact conformation is not detectable indicating a relatively high sensitivity of the macitentan conformation towards solvent polarity. These NMR data and calculations performed with the modeling package MOLOC [27] led us to propose a preferred conformation of macitentan in aqueous solution (Fig. 3C). Fig. 4A, B show macitentan (compact conformation) and bosentan manually docked into the ET A receptor. The neighboring amino acids shaping the binding pocket are Q165, K166, L322, R326, K329, D351 and I355. As an anchor for initial docking, we assumed a charge-charge interaction between the sulfonamide of bosentan and the side chain of R326, as previously published [28]. For macitentan, we assumed an initially similar principle orientation of the sulfamide towards R326. Figure 4C illustrates the structural properties of receptorbound macitentan and bosentan and highlights the different topologies of these two molecules. The structures of macitentan and bosentan are characterized by a central pyrimidine unit bearing three and four substitutents, respectively. The exchange of the 5-ortho-methoxy-phenol unit in bosentan with the 5-parabromo-benzene substituent in macitentan allows a deeper penetration of the latter into the ET A receptor binding pocket (Fig. 4A, B). Unlike bosentan, macitentan carries no substituent in the 2-position of the central pyrimidine resulting in a more complete burial of macitentan in the binding pocket. Replacement of the tert.-butyl benzene sulfonamide in bosentan by the propyl sulfamide in macitentan not only decreases the molecular size of this moiety but also increases the pKa of macitentan by about one unit. Finally, the assumed stacking of the 5-bromo-pyrimidine onto the central pyrimidine appears to stabilize the compact conformation of macitentan in aqueous media (Fig. 3C) and gives rise to a unique globular overall conformation. This conformation of macitentan allows optimal shape complementarity between inhibitor and binding pocket in the ET A receptor. Due to the reduced acidity of the sulfamide moiety, charge-charge interactions are expected to play a minor role in the receptor binding affinity of macitentan. Hence, the proposed binding forces of macitentan differ from those of bosentan-like molecules (and possibly other ERAs like ambrisentan), which all contain an acidic function and whose binding was proposed to depend on chargecharge interactions [28].

Functional characterization of point-mutated ET A receptor variants
To verify the structural model and to distinguish the different molecular interaction modes of macitentan, bosentan and ambrisentan, ten point-mutated ET A receptor variants were designed. They incorporated single mutated amino acids which -according to the model -should differentially affect the interaction of the individual ERAs with the ET A receptor (Fig. 4A, B).
The R326Q mutant was designed to evaluate a potential charge-charge interaction with the sulfamide moiety in macitentan in comparison with the more acidic sulfonamide of bosentan or the carboxylate of ambrisentan. Six additional amino acids lining the predicted binding pocket, i.e. Q165, K166, L322, K329, D351 and I355, were mutated to probe their role as short-distance ERA interaction partners ('nearest neighbors'). Finally, three amino acids, N137, K140 and L141 were mutated to probe interactions of amino acids located in the extended vicinity of the presumed ERA binding pocket ('extended ERA binding pocket'). All mutated amino acids and their suggested role in ERA binding are summarized in Table 1.
To analyze ERA interaction with these ten ET A receptor mutants, recombinant isogenic HEK cell pools were generated via homologous recombination. These cells expressed the receptor variants under a tetracycline-inducible promoter. The cell surface expression of the receptor variants after tetracycline induction was determined by flow cytometry using antibodies against the FLAGtag, which had been added in-frame to the N-terminus. Following tetracycline induction, all receptor variants were present at the cell surface at levels similar to the wild type receptor (Fig. 5A). To determine interaction of ERAs with the different receptor variants we employed ET-1-induced calcium flux assays. To this end, we first performed concentration-response experiments with the wildtype receptor (with and without tetracycline induction) and all mutated receptors to evaluate receptor responsiveness to ET-1. Representative CRC are shown in Fig. 5B, C, D. Mutant R326Q, which was described to play a key role by forming charge interactions with bosentan [28], did not show any major changes in signal amplitude nor EC 50 value for ET-1 (mean EC 50 R326Q: 1.8 nM) compared to the wildtype receptor (mean EC 50 wt: 0.68 nM), indicating that this residue had no major role in interactions with the natural ligand (mean EC 50 values are shown in Table 1). Fig. 5C shows the ET-1-induced responses of the six ET A receptor variants for which the amino acids that are predicted to line parts of the binding pocket were mutated ('nearest neighbors'). Clearly, all mutants except variant K166A retained their functionality and showed signal amplitudes and potencies comparable to wildtype receptors with mean potencies of EC 50 Q165A: 0.73 nM, EC 50 L322A: 0.63 nM, EC 50 K329M: 1.8 nM, EC 50 D351N: 0.84 nM and EC 50 I355A: 1.8 nM. Mutant K166A displayed a pronounced shift in ET-1 potency, with significant calcium flux occurring at $200 nM ET-1. This indicates an important role for K166 in ET-1 binding or signaling. The ET-1-induced responses of three receptor mutants shaping the vicinity of the macitentan binding pocket is shown in Fig. 5D (extended ERA binding pocket). Compared to wildtype receptors, the K140I and L141A variants did not respond to ET-1 at all. The mutant N137A displayed a considerably reduced signal amplitude (,50% residual signal) and a ,5-fold lower potency of ET-1 compared to the wildtype receptor, indicating an important role of these three residues for ET-1 binding or signaling.
In summary, the six ET A receptor mutants Q165A, L322A, R326Q, K329M, D351N and I355A did not show major ET-1 response differences versus the wildtype receptor and were therefore employed to analyze ERA binding in functional assays. Concentration-response experiments were performed in functional assays to quantify ERA potency on cells expressing the six functional ET A mutants or the wildtype receptor. To this end we incubated these cells with macitentan, bosentan or ambrisentan at different concentrations for 120 min followed by addition of 8 nM (EC 90 ) of ET-1.
Calcium traces were recorded and converted into CRC. Fig. 6 shows the CRC for macitentan (A), bosentan (B) and ambrisentan (C) generated in cells expressing the wildtype receptor or the L322A, R326Q or I355A receptor variants. The mean IC 50 values and the IC 50 shift versus the wildtype receptor are summarized in Table 2.
At the wildtype ET A receptor, the three ERAs displayed IC 50 values of 1.2 nM (macitentan), 12 nM (bosentan) and 2.8 nM (ambrisentan), in good agreement with previous measurements demonstrating approximate equipotency of macitentan and ambrisentan and a 10-fold lower potency of bosentan at the ET A receptor [18]. The potency of macitentan was only slightly affected in mutants R326Q (IC 50 shift: 6.3) and I355A (IC 50 shift: 0.9). However, its potency was strongly affected in variant L322A (IC 50 shift: 148). In sharp contrast, the potency of bosentan was not affected in mutant L322A (IC 50 shift: 0.6), but was strongly affected in mutants R326Q (IC 50 shift: 90) and I355A (IC 50 shift: 135). The potency of ambrisentan was moderately affected in receptor mutant L322A (IC 50 shift: 15) and I355A (IC 50 shift: 2.9) and strongly affected in receptor mutant R326Q (IC 50 shift: 317). In mutant Q165A, antagonistic potency was strongly reduced for all three ERAs (macitentan, IC 50 shift: 174; bosentan, IC 50 shift: 264; ambrisentan, IC 50 shift: 1129). The other two receptor mutants, K329M and D351N, did not significantly affect the potency of any of the ERAs, and IC 50 shifts were less than 10 in all cases ( Table 2). Fig. 6D graphically represents the compound-specific IC 50 -shift pattern on the mutants L322A, R326Q and I355A for macitentan, ambrisentan and bosentan.
Confirmation of differential binding mode using structural macitentan and bosentan analogs Macitentan, bosentan and ambrisentan displayed different molecular interactions with the ET A receptor as evidenced by different sensitivity to mutations at positions L322, R326 and I355. To further substantiate these differences in amino acid interactions, the potency of macitentan analogs 1, 2, 7 and 9, and the bosentan analogs, tezosentan and clazosentan, was measured using the described ET A receptor mutants. The antagonistic potencies of these ERAs at the wildtype ET A receptor and the 3 receptor variants and the calculated IC 50 shifts are shown in Table 3. Fig. 7A and 7B represent the IC 50 shift patterns for the tested macitentan and bosentan analogs, respectively. All macitentan analogs showed macitentan-like IC 50 -shift patterns, with strong effects found only in the L322A mutant (IC 50 shifts: 66-228; see Table 3 and Fig. 7A) whereas tezosentan and clazosentan showed a bosentan-like IC 50 -shift pattern, with strong effects observed with mutants R326Q (IC 50 shift: 50/26) and I355A (IC 50 shift: 54/32), but not in mutant L322A (IC 50 shift: 2.0/1.2). Thus, the differential effects of the mutations L322A, R326Q and I355A were confirmed using structurally related molecules of the corresponding compound series.

Integration of modeling and mutational data
The ET A receptor inhibition kinetics of macitentan are characterized by slow dissociation and association rates in comparison to other ERAs of similar affinity [18] which indicated a different binding mode for macitentan compared to other ERAs. Here, we used a combination of molecular modeling and functional studies to characterize the macitentan-ET A receptor Figure 2. Affinity (IC 50 ) and insurmountability of bosentan, tezosentan, clazosentan, ambrisentan, BQ123 and a set of macitentan analogs, measured in ET-1 signaling assays using PASMCs. Insurmountability was determined in IP 1 accumulation assays using 5 mM ET-1 (20 min stimulation), arithmetic means, n$3. IC 50 values were determined in calcium flux assays using 4 nM-12 nM ET-1 (EC 80 ), geometric means, n$4. doi:10.1371/journal.pone.0107809.g002 interaction site at a molecular level. NMR studies revealed that the lipophilic and only weakly acidic acitentan molecule assumed a compact conformation in aqueous solutions and molecular modeling suggested that this compact conformation allows macitentan to form particularly tight interactions within a spatially confined ET A receptor sub-pocket, potentially not engaging in charge-charge interactions due to its weak acidity. Bosentan, bosentan analogs and ambrisentan, occupy the same pocket, but with a lower fit and -due to their expanded and rigid conformation -molecular interactions extend beyond this subpocket into more distal areas. Also, charge-charge interactions are likely contributors to binding. Molecular modeling and physico- chemical considerations thus generated a working hypothesis that offered a molecular basis for a different ET A receptor binding mode displayed by macitentan. We next employed functional studies to validate this hypothesis using ten different ET A receptor variants with single amino acid exchanges. Two amino acid changes (K140I and L141A) led to a complete loss of ET-1induced signaling, whereas the K166A mutation led to a significant reduction of ET-1 potency at the ET A receptor. These results are in line with previous findings showing that the natural ligand ET-1 occupies a large binding pocket within the ET A receptor [28]. All other tested ET A receptor mutants retained their functionality towards ET-1 and were subsequently used to study the binding of ERAs and the ET A receptor. Two of our tested ET A receptor mutants (K329M, D351N) did not affect ERA binding at all, demonstrating that the charges in these two residues are not involved in antagonist binding. In contrast, variant Q165A caused a pronounced loss in potency of macitentan, bosentan and ambrisentan indicating that this residue, located in transmembrane helix 3, is a key contributor to small molecule-receptor binding; however, this amino acid is neither required for ET-1 binding nor signaling. This finding is in good accordance with our molecular model, in which Q165 is in close contact with macitentan and bosentan, thereby restricting spatial motion of both antagonists within the receptor. Interestingly, three receptor mutants, L322A, R326Q and I355A, showed pronounced and differential effects on the potency of macitentan, bosentan and ambrisentan. Fig. 8A and 8B show details of this differential interaction of macitentan and bosentan with the three amino acids L322, R326 and I355 in our model of the ET A receptor pocket. R326 was previously identified as an essential component of bosentan -ET A receptor interaction, most likely due to an electrostatic interaction with the acidic sulfonamide group of bosentan (pKa 5.1) [28]. Our studies here confirm and extend this prominent role of R326 to the other charged ERAs, such as ambrisentan (pKa = 3.5), clazosentan (pKa1 = 4.5, pKa2 = 3.3) and tezosentan (pKa1 = 4.4, pKa2 = 4.1). In contrast, the antagonistic potencies of macitentan and its analogs 1, 2, 7 and 9 were not significantly affected in the R326Q variant. These results suggest a significant contribution of the charge-charge interaction involving R326 in the case of bosentan, tezosentan, clazosentan and ambrisentan, and possibly other negatively charged ERAs not tested here. However, for the weakly acidic macitentan (pKa 6.2) charge-charge interaction with R326 does not contribute to the binding affinity. Similarly, the I355A mutant does not affect macitentan but has a major effect on the potency of bosentan, tezosentan and clazosentan. In our model, this amino acid residue forms the outer rim of the hydrophobic binding pocket harboring macitentan (see Fig. 8A). The expanded and rigid molecules bosentan (Fig. 8B), tezosentan and clazosentan (not shown) cannot be buried entirely in the pocket and protrude into the extended endothelin binding pocket where they interact with I355. Finally, mutant L322A had a significant effect on the potency of macitentan and its analogs and displayed little effect on the potency of bosentan, its structural analogs tezosentan, clazosentan, and on ambrisentan. In our model, L322 forms a significant part of the binding pocket and is involved in tight interactions with macitentan (Fig. 8A). In contrast, bosentan-like molecules seem unable to contact L322 as their elongated scaffold forms a contact with I355 (Fig. 8B) spanning over L322 at a distance. Thus, macitentan, optimally fills an ET A receptor sub-pocket and this interaction is driven by hydrophobic rather than charge-charge interactions.
Previous work has shown that macitentan, bosentan and ambrisentan are all competitive orthosteric antagonists of ET-1 binding at ET A [18]. While this finding means that ERA binding sites must at least partially overlap with that of ET-1 (excluding allosteric antagonism), it does not allow conclusions on the location of the ERA binding site(s). The present study now adds such information and suggest that the different ERAs tested in this study occupy a similar region defined by the effects of mutations at the closely spaced amino acids R326, L322, I355 and Q165. The ET A binding pocket for the 21-amino acid peptide ET-1 has been proposed to be composed of several sub-sites [29,30], with the six C-terminal amino acids of ET-1 (His16-Trp21) being accommodated by the most deeply buried sub-site within ET A . The amino acids K140, L322, I355 and K166 are proposed as key ET-1 interaction partners within this site [29] which suggests an overlap of this site with the ERA binding region defined in the present study.

Different binding modes can cause different binding kinetics
Many slowly associating and dissociating drugs operate through a multi-step induced fit mechanism. After rapid formation of an initial loose drug-target complex, subsequent rate-determining slow steps lead to the formation of the final drug-target complex [31]. These subsequent steps (release of water, conformation changes of ligand and/or receptor) can be rather slow and therefore determine association rates. For high-affinity compounds, a slow binding process is accompanied by a slow dissociation (k off = k on 6K i ) and leads to increased receptor occupancy times. The macitentan-ET A interaction is characterized by relatively slow association and dissociation kinetics compared to other similarly potent ERAs ( [18], and this manuscript). Potential explanations might be the lack of guiding charge-charge interactions and the largely hydrophobic nature of the tight interaction which requires time for water to be released from the binding site. In addition, NMR studies in D 2 O containing increasing amounts of CD 3 CN indicated a relatively high sensitivity of the macitentan conformation towards environ- ment polarity suggesting that macitentan may undergo conformational changes during receptor binding. Thus, the hydrophobic binding mode described in this paper might explain why macitentan and its analogs display a slowed association and dissociation kinetics compared to the other ERAs which are characterized by stronger acidity, lower lipophilicity and chargedependent binding.
The features described here for macitentan and its interaction with the ET A receptor are not unprecedented for GPCRantagonist interactions. Recently, an X-ray structure of a GPCR-ligand complex displaying an almost complete burial of the lipophilic ligand vorapaxar in a hydrophobic sub-pocket of the GPCR PAR-1 (protease-activated receptor-1) was described [32]. The authors attributed the experimentally determined sustained binding properties and the insurmountable antagonism of vorapaxar to its precise hydrophobic surface matching of a PAR-1 sub-pocket. The exact opposite was described in an X-ray structure of naltrindole bound to a d-opioid receptor [33]. The crystal structure of this complex reveals binding of this ligand at the bottom of a wide and solvent-exposed chalice-like pocket, with charge-charge interactions and hydrogen bonds being responsible for high binding affinity. When this opioid ligand is bound, it exposes large surface areas towards the aqueous environment, and existing drug-receptor interactions can be easily weakened by intruding water. For this reason opioid ligands, despite their high affinity, display highly reversible binding, i.e. very short receptor occupancy half-lives, explaining why heroin overdosing is rapidly reversible when treated with the competitive antagonist antidote naloxone [34].

Conclusions
In summary, macitentan is clearly different from other ERAs, such as bosentan and ambrisentan, in terms of topology and physicochemical properties, and in the present study we show that the unique pharmacological behavior of macitentan -i.e. its increased receptor occupancy time and insurmountability -is paralleled by a class-specific mode of ET A receptor engagement, which is characterized by dominating hydrophobic interactions and minimal electrostatic binding forces.

NMR Studies
Spatial proximity of the various moieties of macitentan's structure (Figure 9) was assessed by measuring spatial transfer of magnetization between the corresponding protons by means of two-dimensional Rotating Frame Overhauser Enhancement Spectroscopy (ROESY). Spectra were acquired on a Bruker Avance II, 400 MHz UltraShield, 1 H (400 MHz), chemical shifts are reported in parts per million (ppm) relative to tetramethylsilane (TMS), and multiplicities are given as s (singlet), d (doublet), t (triplet), q (quartet), quint (quintuplet), h (hextet) or m (multiplet). Approx. 15 mM solutions of macitentan in either 0.1 M Na 2 CO 3 in D 2 O (pD,10) or CDCl 3 were used. Spectral data for macitentan (for proton numbering see below): 1

Molecular Modeling
An initial C-alpha model of the ET A -receptor was built on the structure of the orphanin FQ-receptor (PDB: 4EA3) [26] since this receptor is also agonized by a peptide and -of all to date crystallized GPCRs -showed best sequence identity with ET A (BLAST search [www.pdb.org]: 27% identity, 56% similarity). This identity was considered sufficient to obtain an acceptable model for the ET A receptor [35]. This raw model comprising amino acids T76 to C385 was subsequently optimized using the program MOLOC [27] according to standard procedures [36]. In a next step a full atom model of the ET A -receptor was generated. Phi and psi angles were obtained for aligned amino acids from the template receptor. Chi angles were also adopted from the target structure where possible or in case of non-identical amino acids generated by using the most probable value avoiding steric conflicts. In a first optimization step only amino acid side chains were allowed to move while all backbone atoms were kept in fixed positions. This step largely removed repulsive interactions between side chains, further improved chi angles of non-conserved amino acids, and revealed regions with unfavorable interactions. In a following optimization step only C-alpha atoms were kept fixed while all other atoms were allowed to move. In a third round of optimization no atoms were kept stationary but positional constraints were applied to C-alpha atoms. Optimization was continued until MOLOC converged and no further steric conflicts could be detected. Inhibitors were docked manually into the ligand binding site in multiple start positions and conformations followed by a round of optimization with surrounding amino acids allowed to move in each case.

Generation of recombinant cell pools expressing ET A receptor variants
Isogenic cell pools expressing the different N-terminally FLAGtagged ET A receptor variants were generated using the Jump-In T-Rex HEK293 cell system (Invitrogen, USA) according to the manufacturer's protocol. This system allows the tetracycline- inducible expression of a gene of interest from a single defined insertion site, so that similar expression levels are obtained from all constructs. In brief, both the retargeting vector containing the gene of interest and the pJTI.R4.Integrase vector were transfected into the Jump-In T-Rex HEK293 parental cell line and selection of recombined isogenic cell pools was performed for 2 weeks in medium containing geneticin (1 mg/ml) and blasticidin (5 mg/ml).

IP 1 measurements
All measurements were performed using the IP-One HTRF kit (Cisbio, France) following the manufacturer's protocol. PASMC were seeded at 10,000 cells/20 mmL/well into 384-well, white, small-volume tissue culture plates (Greiner, Germany,) and incubated overnight at 37uC in 5% CO 2 . Then, medium was removed and 5 mL/well of assay buffer (HBSS containing 0.1% BSA, 20 mM HEPES, 0.375 g/L NaHCO 3 , 50 mM LiCl, pH 7.4) were added. Then, cells were supplemented with 5 mL/ well of a 26 concentrated dilution series (3-fold) of antagonists in assay buffer, covering a final concentration range of 0 nM to 370 nM, and incubated for 120 min at 37uC in 5% CO 2 . Then, 2.5 mL/well of 56 concentrated solutions of ET-1 in assay buffer giving a final concentration of 5 mM were added followed by an incubation at 37uC/5% CO 2 for 20 min. Cells were lysed by the addition of 2.5 mL/well of conjugate-lysis buffer. Then, 2 mL/well of IP 1 -d2 conjugate and 3 mL/well of anti-IP 1 -cryptate terbium were added. After 60 min incubation at room temperature, the assay plates were excited at 337 nm and the ratio of emitted light at 665 nm and 620 nm was recorded using a PHERAstar microplate reader (BMG Labtech, Germany). 665 nm/620 nm emission ratios were translated into IP 1 concentrations via an onplate calibration curve generated with known concentrations of IP 1 . IP 1 concentration values were exported into GraphPadPrism software and the antagonistic concentration-response curves (CRC) were fitted by three-parameter non-linear regression, giving compound specific values for the upper plateau (no inhibition) and lower plateau (maximal inhibition). The degree of insurmountability (%) was calculated as (upper plateau2lower plateau)/upper plateau6100%. To compare results from independent experiments, insurmountability values were normalized against the fully insurmountable compound macitentan analogs 1 or 8 (100% insurmountable). Compounds having an insurmountability of less than 15% in an individual experiment were classified as surmountable (0% insurmountable).

Intracellular calcium flux measurements using recombinant HEK-TRex cells expressing different ET A receptor variants
Recombinant HEK cells expressing the human ET A variants were seeded in growth medium (in presence or absence of 100 ng/ mL tetracycline) at 12,000 cells/well into poly-l-lysine 384-well black clear-bottom plates (Greiner, Germany) and incubated overnight at 37uC in 5% CO 2 . Then, the growth medium was exchanged with 50 ml/well of dye buffer (HBSS with 20 mM HEPES, 0.375 g/l NaHCO 3 , and 3 mM fluo-4 AM). Cells were incubated for 1 h at 37uC in 5% CO 2 , then medium was exchanged with 50 ml/well assay buffer (HBSS with 0.1% BSA, 20 mM Hepes, 0.375 g/l NaHCO 3 ) followed by equilibration at room temperature for at least 30 min. For the determination of the inhibitory potency (IC 50 ) of the endothelin receptor antagonists, within the Fluorescent Imaging Plate Reader (FLIPR Tetra, Molecular Devices, USA), cells were supplemented with 10 ml of 66concentrated antagonist dilution series prepared in assay buffer covering the final concentration range from 0 nM to 10 mM (0.1% DMSO final assay concentration). After a 120-min incubation at room temperature, cells were stimulated by the addition of 10 ml of 76 concentrated ET-1 in assay buffer to obtain the final assay concentration of 8 nM (EC 90 ). Calcium flux was monitored for 3 min. For the IC 50 calculations, FLIPR traces were normalized by trace alignment at the last time point before agonist addition (ScreenWorks software, Molecular Devices, USA). Then, relative fluorescence units (RFU) of the maximum signal per well were exported to the proprietary IC 50 Witch software (Actelion, Switzerland) and used to calculate IC 50 values [settings: fixed minimum (0 nM ET-1) and curve-intrinsic maximum]. For the determination of ET-1 potency and efficacy at the different receptor variants, cells stained with fluo-4 were stimulated with 10 ml of a 76 concentrated dilutions series of ET-1 in assay buffer and calcium responses were recorded and processed as described above. To optimally compare response amplitude and ET-1 potency for the different cell pools, CRCs were displayed as response over baseline. Potencies were calculated using the IC 50 Witch software (settings: intrinsic minimum and intrinsic maximum). Table 3. The right half of the

Flow cytometry
Recombinant HEK cells expressing the FLAG-tagged human ET A variants were seeded at 10 6 cells per well (in presence or absence of 100 ng/ml tetracycline) into 6-well plates in growth medium and incubated overnight at 37uC in 5% CO 2 . Then, the cells were washed with PBS and detached with cell dissociation buffer (Life Technologies, USA). FLAG-hET A receptors were stained with mouse anti-FLAG M2 antibodies (Sigma, USA), diluted in PBS/2 mM EDTA/0.5% fatty acid free BSA (Calbiochem, Germany). As secondary antibody Alexa Fluor 488 goat anti-mouse IgG (Life Technologies, USA) was used. Then, mean fluorescence intensities (MFI) of the propidium iodidenegative cells were recorded using a FACSAria IIu flow cytometer (BD Biosciences, USA). Surface receptor expression was calculated by subtracting the MFI of the non-induced cell pools from the MFI of the tetracycline-induced cells pools and then comparing this value to the value obtained for the wildtype-expressing cells (100%).