Distinct but overlapping binding sites of agonist and antagonist at the relaxin family peptide 3 (RXFP3) receptor

The relaxin-3 neuropeptide activates the relaxin family peptide 3 (RXFP3) receptor to modulate stress, appetite, and cognition. RXFP3 shows promise as a target for treating neurological disorders, but realization of its clinical potential requires development of smaller RXFP3-specific drugs that can penetrate the blood–brain barrier. Designing such drugs is challenging and requires structural knowledge of agonist- and antagonist-binding modes. Here, we used structure–activity data for relaxin-3 and a peptide RXFP3 antagonist termed R3 B1–22R to guide receptor mutagenesis and develop models of their binding modes. RXFP3 residues were alanine-substituted individually and in combination and tested in cell-based binding and functional assays to refine models of agonist and antagonist binding to active- and inactive-state homology models of RXFP3, respectively. These models suggested that both agonists and antagonists interact with RXFP3 via similar residues in their B-chain central helix. The models further suggested that the B-chain Trp27 inserts into the binding pocket of RXFP3 and interacts with Trp138 and Lys271, the latter through a salt bridge with the C-terminal carboxyl group of Trp27 in relaxin-3. R3 B1–22R, which does not contain Trp27, used a non-native Arg23 residue to form cation–π and salt-bridge interactions with Trp138 and Glu141 in RXFP3, explaining a key contribution of Arg23 to affinity. Overall, relaxin-3 and R3 B1–22R appear to share similar binding residues but may differ in binding modes, leading to active and inactive RXFP3 conformational states, respectively. These mechanistic insights may assist structure-based drug design of smaller relaxin-3 mimetics to manage neurological disorders.

Relaxin-3 is the most recently identified member of the relaxin peptide family and is a neuropeptide highly expressed in the brain. It was discovered in 2002 based on a search of the human genome database (1), and subsequently homologs of the relaxin-3 gene (RLN3) were found in other species, such as mouse, rat, pig, frog, chimpanzee, and macaque, and orthologs of relaxin-3 were found in fish, such as the Fugu (Takifugu rubripes) and zebrafish (Danio rerio) (2). Relaxin family peptides bind to class A G protein-coupled receptors known as the relaxin family peptide (RXFP) 5 receptors (3). The relaxin family peptides are part of the insulin peptide superfamily (4) and share structural homology with two peptide chains, A and B, cross-linked by three disulfide bonds: two interchain disulfide bonds between the A-and B-chain and one intrachain disulfide bond in the A-chain (4,5). The endogenous receptor for relaxin-3 is RXFP3 (6) and was previously known as GPCR135 or somatostatin and angiotensin-like peptide receptor (SALPR) (3,6,7). Relaxin-3 can also bind to RXFP1 (8) and RXFP4 (9,10), the endogenous receptors for relaxin-2 (also known as just "relaxin") and insulin-like peptide 5 (INSL5), respectively.
Relaxin-3 is highly expressed in the brains of all species where the gene has been identified, including zebrafish (11), mouse (1), rat (12), macaque (13,14), and humans (6). Extensive studies in mice and rats have determined high levels of relaxin-3 expression in the ventromedial dorsal tegmental area also known as the nucleus incertus (NI) (1,12,(15)(16)(17). Its high expression and extensive overlap of relaxin-3-containing fibers with the expression of RXFP3 in certain areas of the brain suggests that relaxin-3 has distinct and important functional roles (1,17,18). Indeed, relaxin-3 has been shown to be involved in a wide range of behaviors, which include spatial learning and memory (19); modulation of stress, mood, and arousal (20,21); feeding and appetite (22); and addiction (23). The importance of the relaxin-3/RXFP3 system suggests that its receptor is a highly relevant target for developing receptorspecific drugs to treat neurological diseases.
Numerous studies have been focused on understanding structure activity relationship of relaxins (24). The relaxin-3 B-chain alone is capable of interacting with RXFP3, suggesting that the A-chain of relaxin-3 is not crucial for RXFP3 interactions (6,25). Studies using a recombinant expression system to generate chimeric peptides consisting of A-chains from other relaxin family peptides together with the relaxin-3 B-chain demonstrated that a peptide consisting of the INSL5 A-chain and relaxin-3 B-chain (R3/I5) was a selective agonist for RXFP3 over RXFP1 (26). The NMR solution structures of native relaxin-3 and this chimera revealed that the center of the B-chain adopts a helical conformation (27,28). The helix is important for activity, and recently, Hojo et al. (29) and Jayakody et al. (30) showed that hydrocarbon stapling on the B-chain of relaxin-3 can produce relaxin-3 single-chain analogs that retain potent agonist activity. NMR analysis highlighted that the stapled peptide mimicked the ␣-helical structure of the relaxin-3 B-chain (29) and further demonstrated that the B-chain is solely responsible for binding and activation of RXFP3. These studies demonstrate that the A-chain of relaxin-3 is not involved in binding or activation of RXFP3, and its role is to provide a scaffold for maintaining the helical structure within the B-chain. Mutagenesis studies of the B-chain in relaxin-3 led to the identification of the key amino acid residues for receptor binding on the B-chain central ␣ helix (Arg B12 , Ile B15 , Arg B16 , Ile B19 , and Phe B20 ), whereas the amino acid residues Arg B26 and Trp B27 at the C terminus make only a minor contribution to binding affinity but are essential for receptor activation (Fig. 1, A and B) (31).
Truncation of the C terminus of the B-chain led to the generation of an RXFP3 antagonist. Interestingly, the addition of a non-native arginine after Cys B22 at the C terminus, as a result of an artifact of the recombinant production and processing, in the peptide R3(B⌬23-27)R/I5 increased the affinity for RXFP3 (31). A further study showed that the additional arginine is a major contributor for the high-affinity binding of R3(B⌬23-27)R/I5 to RXFP3 (32). The introduction of this amino acid residue has been suggested to provide an additional binding interaction that compensates for the reduction in affinity from the loss of the C-terminal amino acid residues (32). Subsequently, a single-chain variant of this truncated peptide with the cysteine residues mutated to serine was developed. This variant, R3 B1-22R (Fig. 1C), is also a high-affinity RXFP3 antagonist and, furthermore, does not have any affinity for the RXFP4 or RXFP1 receptors and thus is RXFP3-specific (33). Importantly, this peptide is unstructured in solution, and the Arg 23 residue is important for the high-affinity binding, as the variant lacking this arginine demonstrated no binding to RXFP3 (33).
Knowledge of the precise agonist and antagonist binding sites on RXFP3 would further assist the structure-based design of drug leads. Despite having a good understanding of the important B-chain amino acid residues that are responsible for the activity of relaxin-3, little is known about the amino acid residues in RXFP3 that they are interacting with. Mutagenesis studies on RXFP3 together with molecular modeling and ligand docking demonstrated that Arg B12 and Arg B16 are likely interacting with Glu 244 in extracellular loop 2 (EL2) and Asp 145 in EL1, respectively, whereas Arg B26 , one of the critical amino acid residues for activating the receptor, is interacting with Glu 141 in transmembrane domain 2 (TM2) ( Fig. 2A) (34,35). However, the interaction sites for the hydrophobic amino acid residues, Ile B15 , Ile B19 , Phe B20 , and Trp B27 , with the receptor are still unknown. A further study using RXFP3 mutagenesis hypothesized that Trp B27 interacts with Trp 138 in RXFP3 after a conformational change of the C terminus of the peptide upon binding the receptor (36). Importantly, there is no information as to the amino acid residues that bind the RXFP3-specific antagonist R3 B1-22R, particularly the non-native arginine Arg 23 on its C terminus, which is the largest contributor to its affinity. Given the prominent neurological effects following the activation and inhibition of RXFP3 signaling in the brain, RXFP3 presents immense potential as a drug target to treat neurological disorders. Designing small, receptor-specific drugs that can penetrate the blood-brain barrier is challenging and requires understanding of the molecular details of the ligand-receptor complex and the conformational changes that occur upon activation. In the present study, we investigated the roles of several amino acid residues in RXFP3 that we predicted might be interacting with relaxin-3 and the R3 B1-22R single-chain antagonist. Using site-directed mutagenesis, we studied the effects of mutating these amino acid residues on both agonist and antagonist binding as well as the effects of mutations on the activation of RXFP3 following agonist stimulation. Based on these data, we present detailed models of the agonist and antagonist binding sites of RXFP3 and new mechanistic insights into its function. The key amino acid residues (shown as sticks) involved in binding are shown in cyan, and those involved in activation are shown in red. Relaxin-3 A-chain is colored in wheat, whereas the B-chain is colored in green. The C terminus of the relaxin-3 B-chain folds back to interact with the ␣-helical core residues. B, peptide sequence of human relaxin-3 with the amino acid residues important for binding highlighted in cyan, residues involved in activation highlighted in red, and the cysteine residues forming disulfide bonds highlighted in yellow. C, peptide sequence of R3 B1-22R with the binding residues utilized by relaxin-3 highlighted in cyan, whereas Arg 23 , highlighted in magenta, is a non-native residue shown to be very important for binding to RXFP3.

Predictions of relaxin-3/RXFP3 interactions
In previous studies exploring the role of acidic amino acid residues in RXFP3 for relaxin-3 binding, interaction partners for the key basic amino acid residues in relaxin-3, Arg B12 , Arg B16 , and Arg B26 , were identified (34, 35) ( Fig. 2A). These interactions suggest that the relaxin-3 B-chain C terminus inserts into the RXFP3 binding pocket, whereas the B-chain helical region primarily packs against EL2. To further confirm and refine the model of ligand binding and to also identify hydrophobic amino acid residues in RXFP3 that are interacting with the remaining amino acid residues critical for bioactivity (Ile B15 , Ile B19 , Phe B20 , and Trp B27 ), we designed a new set of rationally predicted RXFP3 mutations based on the proximities of ligand to receptor residues generated from our docking studies of relaxin-3 on the RXFP3 homology model (Fig. 2B) that we created in our previous study (34). Based on the model, Ile B15 and Ile B19 are exposed on a surface of the B-chain helix adjacent to an RXFP3 hydrophobic patch that includes Val 239 , Val 241 , Leu 246 , and Leu 248 in EL2; thus, these were mutated to alanine (Fig. 2B). Phe B20 is exposed on a different face of the helix and is observed to be in close proximity to another bulky residue on TM7, Phe 364 , which was therefore also mutated to alanine (Fig.  2B). Finally, the C terminus of the B-chain is rather flexible, and a conformational change has been suggested (35), meaning there are a number of potential interaction partners for Trp B27 , including Trp 138 in TM2; Thr 162 and Met 166 in TM3; His 268 and Lys 271 in TM5; Trp 339 , Thr 346 , Ser 349 , and Lys 353 in TM6; and Phe 371 in TM7 (Fig. 2B). These amino acid residues were also all individually mutated to alanine. The combination of these mutations represents an extensive exploration of a large portion of the RXFP3 ligand-binding pocket (Fig. 2B).

Effect of mutagenesis on agonist binding to RXFP3
A total of 15 mutations were generated based on their proximity to key relaxin-3-binding residues. The mutants were first tested for changes in agonist binding in cell-based binding assays using europium-labeled (Eu-)R3/I5. Receptor mutants were tested in parallel for their cell surface expression, and the final specific binding data were expressed as a ratio of the percentage of specific binding/cell surface expression. Eight of 15 mutations tested (W138A (TM2); T162A (TM3); V241A, L246A, and L248A (EL2); K271A (TM5); and F364A and F371A (TM7)) resulted in a significant loss (ϳ70 -85% compared with WT) in Eu-R3/I5-specific binding without affecting receptor cell surface expression ( Fig. 3 and Table 1). Other mutations, such as V239A (EL2), W339A, T346A, S349A, and K353A (TM6) had no effect on binding or cell surface expression ( Fig.  3 and Table 1). On the other hand, M166A (TM3) and H268A (TM5) showed increased specific agonist binding despite having normal cell surface expression ( Fig. 3 and Table 1). The mutation of the bulky side chains of histidine and methionine to alanine may free up steric constraints in the binding pocket and hence the observed higher specific binding. Additionally, His 268 is positioned one helix turn above Lys 271 , which showed affected agonist binding, and hence the removal of the His 268 side chain may open up the binding pocket.

Activation of mutant RXFP3 receptors
RXFP3, when activated by agonists, couples to G i/o proteins to inhibit cAMP production. To assess the ability of the mutant receptors to signal, we therefore used a pCRE-␤-gal reporter gene assay, which is an indirect readout of cAMP activity and has been extensively utilized to assess agonist activity at RXFP3 (33,37). When the mutant receptors were tested in cAMP activity assays, 9 of 15 mutations demonstrated significantly reduced agonist potency (pEC 50 ) compared with WT RXFP3 ( Fig. 4 and Table 1). Notably, there were no changes in ligandactivated efficacy for all of the mutations where a full-dose response was demonstrated. For the mutants where there was not a full-dose response (W138A, L248A, K271A, W339A, and F371A), efficacy values at 1 M agonist were shown, and potencies were described as Ͻ6 ( Table 1). Seven of the 9 mutations (W138A, T162A, L246A, L248A, K271A, F364A, and F371A) previously showed decreased specific binding ( Fig. 4 (A and B), Fig. S1, and Table 1). Mutations at W138A, K271A, F371A, and L248A were particularly detrimental for activation, resulting in of the homology model of RXFP3-relaxin-3 complex with relaxin-3 removed to provide a clearer view. Amino acid residues rationalized to be interacting with relaxin-3 are shown in sticks and color-coded accordingly. Residues in TM2 and TM3 are shown in blue and residues in EL2, TM5, TM6, and TM7 are shown in yellow, purple, red, and pink, respectively.

Mechanism of agonist and antagonist interaction at RXFP3
more than 1000-fold lower agonist potency compared with WT RXFP3, which suggest that these amino acid residues may be involved in agonist recognition (  Table 1). Interestingly, W339A and T346A, which did not show any significant decrease in specific agonist binding, demonstrated significant losses in cAMP activity, suggesting their roles in stabilizing the active conformation instead of a direct role in ligand recognition (Figs. 3A and 4B, Fig. S1, and Table 1). In particular, W339A demonstrated minimal activity (ϳ30% efficacy; Table 1) to agonist at concentrations up to 1 M, suggesting that this amino acid residue is probably important for stabilizing the active receptor conformation, and mutations to alanine shift the receptor toward the inactive state. Indeed, Trp 339 or W6.48 (according to the Ballesteros-Weinstein numbering scheme (38)), which is located on the CWXP motif that is widely known as the "rotamer toggle switch," is important for activation (39,40). Mutagenesis studies of W6.48 in A 3 adenosine receptor, GPR119, GPR39, NK1, and the ␤ 2 -adrenergic receptor demonstrated that mutation of this amino acid residue significantly affected agonist-induced receptor activation but did not impair agonist affinity (41,42). The other four mutations (T162A, L246A, T346A, and F364A) had significantly lower agonist potency but showed a less dramatic drop (Fig. 4B, Fig. S1, and Table 1). In contrast, V241A, which had a dramatic loss in specific agonist binding, did not show impairment in cAMP signaling ( Fig. 4A and Table 1).

Refined model of the RXFP3-relaxin 3 complex
In our previous study (34), we generated a homology model of RXFP3 and performed ligand-docking simulations guided by our mutagenesis data based on the inactive structure of the CXCR4 receptor (PDB entry 3OEO (43)). We have utilized this model to predict further interaction sites and perform more RXFP3 alanine mutations in this study. Recently, the active (44) and inactive (45) structures of the angiotensin II receptor type I, which shares higher homology with RXFP3, became available, which allowed us to use these structures as templates for the homology modeling of active and inactive RXFP3 models to refine docking based on our current mutagenesis and peptide SAR data (see companion paper (59)). Based on these results, we further refined our model of how relaxin-3 binds and activates RXFP3 (Fig. 5A). The importance of Leu 246 and Leu 248 strongly supports a binding mode that places Ile B15 and Ile B19 packed against the small ␤-sheet in EL2 with hydrophobic

Mechanism of agonist and antagonist interaction at RXFP3
interactions between Ile B15 and Leu 246 and between Ile B19 and Leu 248 (Fig. 5B). Ile B15 has been shown to be critical for relaxin-3 binding previously (31) with an alanine replacement of this amino acid residue resulting in a dramatic loss of affinity for RXFP3. Replacement of Ile B19 with alanine only shows a moderate drop in affinity, but replacement with a charged amino acid residue (glutamic acid) is detrimental, confirming the importance of a hydrophobic nature at this position (31). The V239A replacement in RXFP3 was well tolerated, suggesting that this amino acid residue is outside the binding site. V241A, despite showing a significant drop in Eu-R3/I5 binding, did not show impairment in signaling, suggesting that it may only make a minor contribution to binding affinity.
Trp B27 was positioned in our preliminary model toward TM5 and TM6, based on the preferred conformation of this amino acid residue in the relaxin-3 solution NMR structure. However, of the amino acid residues mutated in this region (Thr 162 , His 268 , Lys 271 , Trp 339 , Thr 346 , Ser 349 , Lys 353 and Phe 371 ), only Thr 162 , Lys 271 , and Phe 371 were found to be important for RXFP3 binding and activation. NMR studies on relaxin-3 peptide have suggested a degree of flexibility afforded by the Gly-Gly hinge between the binding and activation domains, and such a rearrangement has previously been proposed to allow for receptor activation, as mutation of the flexible Gly-Gly to Ala-

Figure 4. Characterization of mutant receptors in a pCRE-␤-gal reporter gene assay in transiently transfected HEK293T cells.
A, representative dose-response curves for W138A in TM2, a mutation that disrupted signaling, and V241A in EL2, a mutation that did not affect signaling. All individual doseresponse curves of mutant RXFP3 receptors are shown in Fig. S2. B and C, the pEC 50 (B) and efficacy of the inhibition (C) of forskolin-stimulated cAMP accumulation when stimulated with the RXFP3 full agonist, analogue 2, at the HA-RXFP3 and mutant RXFP3 receptors. Data are expressed as mean Ϯ S.D.

Mechanism of agonist and antagonist interaction at RXFP3
Ser resulted in a dramatic drop in relaxin-3 activation potency (46). The strong dependence of Trp 138 on binding instead suggested that the Trp B27 aromatic ring in relaxin-3 is stacked against TM2. This, in turn, places the Trp B27 C terminus in a position to be able to form a salt bridge across the binding pocket to the extended side chain of Lys 271 (Fig. 5B), an interaction that would explain the importance of both Lys 271 and a free acid at the C terminus for maximal activity (47). At the same time this work was undertaken, the Trp B27 interaction with Trp 138 on TM2 was also demonstrated by Hu et al. (36), showing that the mutant W138A was not able to discriminate between relaxin-3 and a variant lacking Trp B27 . Considering that the C terminus is flexible and Trp B27 is a large amino acid residue, the effects of mutations to Thr 162 and Phe 371 in its proximity are also likely a result of either direct interactions or indirect local conformational changes that affect its binding pocket (Fig. 5B).
In the model, Phe B20 is located on the other face of the relaxin-3 B-chain helix and thus does not appear to be able to interact with EL2. Instead, we speculated that through a small rotation, it could contact amino acid residues on the opposite side of the binding site. It was previously shown that the replacement of Phe B20 with alanine showed only a moderate drop in affinity, but a replacement with a charged arginine or a polar group like serine resulted in a significant affinity drop, suggesting that the nature of the Phe B20 interactions is hydrophobic (31). Indeed, when Phe B20 was replaced with a hydrophobic and bulky tyrosine, the affinity of the peptide was rescued (31). A mutation to Phe 364 in TM7 was used to probe this possibility, and the variant F364A showed reduced agonist binding and RXFP3 activation; thus, we propose that it is interacting with relaxin-3 Phe B20 (Fig. 5B).

Effect of mutagenesis of acidic residues on antagonist binding in RXFP3
Little is known about how the single-chain antagonist R3 B1-22R interacts with RXFP3. Given that all features of the helical portion of the B-chain are retained, key amino acid residues in this region might be involved in similar interactions as in relaxin-3. However, unlike relaxin-3, R3 B1-22R lacks the C-terminal Arg B26 and Trp B27 interaction sites but gains an additional high-affinity binding site through the non-native C-terminal Arg 23 . In addition, it is known that R3 B1-22R with a C-terminal acid binds with lower affinity compared with a C-terminal amide (33). On the other hand, for relaxin-3, a B-chain C-terminal acid is required for high affinity and activity (47). Therefore, we would predict that there are some distinct differences in the binding mode compared with relaxin-3 due to the critical new interactions introduced by the non-native Arg 23 and the C-terminal amide (see companion paper (59)). As the unstructured antagonist must still partially rely on the binding amino acid residues of relaxin-3, we hypothesized that R3 B1-22R formed a helical structure upon the first contact with the receptor, which is primarily driven by Arg 23 , subsequently allowing other binding amino acid residues on the central region of R3 B1-22R to interact with RXFP3. Given the involvement of positively charged arginines in the ligand (see companion paper (59)), we first screened binding to acidic mutant receptors to analyze whether these are involved in similar interactions as with relaxin-3.
All of the acidic mutant receptors except E245A have been previously shown not to affect receptor cell surface expression (34). E245A, which is pointing away from the binding site in our model, has impaired cell surface expression (34) (ϳ40% of WT) but importantly has unchanged affinity for R3 B1-22R (Fig. 6A) but reduced maximum binding, as expected ( Fig. S2 and Table 1). Both Figure 5. A, relaxin-3 docked into an active-state homology model of RXFP3 (gray) with amino acid residues predicted to be interacting shown as sticks. B, close-up of the predicted interactions between hydrophobic residues Ile B15 and Ile B19 on relaxin-3 with Leu 246 and Leu 248 in EL2 of RXFP3, Phe B20 with Phe 364 in TM7 of RXFP3 and Trp B27 of the flexible C terminus with several amino acid residues in TM2 (Trp 138 ), TM3 (Thr 162 ), TM5 (Lys 271 ), and TM7 (Phe 371 ). Ligand and receptor amino acid residues involved in interactions are labeled in red and black, respectively.

Mechanism of agonist and antagonist interaction at RXFP3
E244A and D145A, however, demonstrated reduced affinity for R3 B1-22R (Fig. 6 (A and B), Fig. S2, and Table 1), consistent with our hypothesis that they, as in relaxin-3, interact with Arg 12 and Arg 16 . R3 B1-22R lacks the C terminus of relaxin-3, but we predicted that if both the agonist and the antagonist interact with RXFP3 using similar residues, this would mean that R3 B1-22R Arg 23 could bind close to the amino acid residues interacting with Arg B26 and Trp B27 of the agonist. Interestingly, E141A also demonstrated reduced affinity for R3 B1-22R ( Fig.  6B and Table 1), although its interaction partner Arg B26 of relaxin-3 is missing in R3 B1-22R. Additionally, the combination of E141A/D145A further decreased the affinity (Fig. 6B and Table 1). Notably, the double mutant was still able to bind R3 B1-22R, but we were not able to obtain an affinity that was Ͼ100 nM. Thus, Arg 23 could potentially be interacting with Glu 141 . However, as the removal of Arg 23 from the R3 B1-22R completely abolishes binding of the antagonist peptide, we would have anticipated a complete loss of binding if its binding partner was removed from the receptor. Hence, at this point, we hypothesize that Glu 141 is most likely not the only amino acid residue that is interacting with Arg 23 .

Effect of mutagenesis of nonacidic residues on antagonist binding in RXFP3
Given the involvement of Glu 244 and Asp 145 , we predicted that the hydrophobic patch on EL2 of RXFP3 could also retain interactions with Ile 15 and Ile 19 of R3 B1-22R, as for relaxin-3. To test this and further characterize the binding mode, we tested all of the nonacidic mutant receptors for antagonist binding. W138A, V241A, L246A, L248A, K271A, and F364A, which affected agonist binding, also significantly decreased the R3 B1-22R binding, albeit to a lesser extent than agonist binding ( Fig. 6C and Table 1). Because we observed more subtle changes in specific binding, we performed full saturation binding on these mutant receptors. W138A showed a pronounced loss in binding affinity characterized in Eu-R3 B1-22R saturation binding (Fig. 6D). L246A, L248A, K271A, and F364A had a moderate (ϳ2-fold) but significant decrease in K d (Fig. S2 and Table 1), whereas V241A did not show any perturbations in binding R3 B1-22 R as compared with WT ( Fig. S2 and Table  1). H268A, T346A, and S349A resulted in a decrease in antagonist affinity (ϳ2-3-fold) but not agonist binding or potency, suggesting that these amino acid residues are specifically interacting only with the antagonist (Fig. 6C, Fig. S2, and Table 1). On the other hand, K271A, which demonstrated decreased agonist binding and potency, did not significantly impair antagonist binding (Fig. S2 and Table 1). Although K271A showed slightly reduced (ϳ75% of WT binding) antagonist binding, this was not significantly different compared with WT (p Ͼ 0.05). However, saturation binding on K271A indicated a slight reduction in antagonist affinity (Fig. S2 and Table 1). There was no change in B max value for any mutant receptor other than E245A, which also showed lower cell surface binding. Taken together, this suggests that although there are overlapping binding sites common to both antagonist and the agonist, there are also distinct differences in the binding mode.
Despite the mutational data giving clear insights into the antagonist interaction, these mutations so far had not resulted in a complete loss of R3 B1-22R binding, which would be expected if the amino acid residues interacting with Arg 23 on the antagonist had been eliminated. The most plausible explanation is that Arg 23 forms complex interactions with several amino acid residues in the binding pocket, and hence, mutating a single amino acid residue does not result in a complete loss of binding. Analyzing the structural model, we predicted that Arg 23 could be forming both an ionic interaction with Glu 141 and a cation-interaction with Trp 138 . Thus, we made the doubly mutated W138A/E141A. The W138A/ E141A mutant receptor was expressed normally at the cell surface (Fig. 7A), but when tested in saturation binding, it resulted in a complete loss of Eu-R3 B1-22R binding (Fig.  7B), strongly suggesting that our hypothesized interaction with Arg 23 is correct and explaining the absolute requirement of this amino acid residue.

Distinct but overlapping binding modes of agonist and antagonist
Based on the mutagenesis data and our refined active-and inactive-state RXFP3 models, we performed docking simulations with the agonist and the antagonist to the active and inactive RXFP3 models, respectively. Our data suggest distinct but overlapping binding modes of relaxin-3 (Fig. 5A) and R3 B1-22R (Fig. 8A). Although R3 B1-22R is unstructured in solution, it likely forms a helix upon the interaction with RXFP3, which is mainly driven by the high-affinity binding of Arg 23 (see companion paper (59)). It is not uncommon for intrinsically disordered peptides that are flexible in solution to switch to a well-ordered structure upon binding to its receptor. From previous structure-activity relationship studies (31,33), it is known that relaxin-3 utilizes binding amino acid residues on the central helix of the B-chain, such as Arg B12 , Ile B15 , Arg B16 , and Ile B19 . In our refined models guided by mutagenesis, we postulate that these amino acid residues in both relaxin-3 and R3 B1-22R interact with amino acid residues Glu 244 , Leu 246 , Asp 145 , and Leu 248 , respectively, forming a binding surface spanning extracellular loops 1 and 2 (Figs. 5B and 8A).
For relaxin-3, the C-terminal tail with the critical Arg B26 -Trp B27 inserts into the binding pocket to activate the receptor. Arg B26 forms a previously identified key ionic interaction with Glu 141 . Trp B27 at the C terminus is a large amino acid residue, and the flexibility imparted by the Gly-Gly hinge makes predictions of its exact position and multiple interaction partners difficult. However, we have refined our models, and based on functional assays, we hypothesized that Trp B27 stacks on Trp 138 of TM2. This is consistent with a previous study that identified Trp 138 as key amino acid residue interacting with Trp B27 of relaxin-3, as the removal of Trp B27 renders relaxin-3 insensitive to the alanine mutation W138A (36). Importantly, this conformational rearrangement allows the backbone carboxyl of Trp B27 to interact with Lys 271 on the other side of the binding site, on TM5 (Fig. 5B). The importance of both Lys 271 and the free acid at the C terminus can thus potentially be explained by a salt bridge between these oppositely charged groups. Finally, we proposed that Phe B20 stretches across the top of RXFP3 to

Mechanism of agonist and antagonist interaction at RXFP3
form an interaction with Phe 364 . Thus, the large relaxin-3 peptide covers the majority of the extracellular parts of the receptor, essentially forming a lid over the binding pocket, which accommodates the activation domain.
For R3 B1-22R, nothing was previously known about how the non-native Arg 23 interacted with RXFP3 to maintain RXFP3 in an inactive state. Given that Arg 23 is positively charged, we expected to easily identify a counterpart negatively Figure 6. A, specific Eu-R3 B1-22R binding to RXFP3 receptors with mutations in residues predicted to be involved in electrostatic interactions with R3 B1-22R. B, representative Eu-R3 B1-22R saturation binding curves fit to one-site specific binding of acidic mutants E141A and D145A and double mutant E141A/D145A. All saturation binding curves for mutant receptors are shown in Fig. S2. C, specific Eu-R3 B1-22R binding to hydrophobic residue mutant RXFP3 receptors predicted to form hydrophobic interactions with R3 B1-22R. D, saturation binding of Eu-R3 B1-22R to hydrophobic W138A mutant receptor fitted to one-site specific binding. Data are expressed as mean Ϯ S.E. (error bars) of triplicate measurements from at least three independent experiments (n ϭ 3-19; *, p Ͻ 0.001; #, p Ͻ 0.05 compared with HA RXFP3). K d values of mutant receptors are summarized in Table 1.

Mechanism of agonist and antagonist interaction at RXFP3
charged amino acid residue in RXFP3 as the key binding partner. However, no single acid residue mutation resulted in a total binding loss. Thus, we predicted that Arg 23 is making interactions with several amino acid residues, and with the guide of the inactive-state RXFP3 model and ligand docking, we postulate the binding site of Arg 23 to be sandwiched between Trp 138 and Glu 141 in TM2, forming both ionic and cation-type interactions (Fig. 8B).
In relaxin-3, the B-chain helix ending at Cys 22 and the activation domain Arg B26 -Trp B27 are separated by a short linker sequence, Gly-Gly-Ser. In contrast, in the antagonist, Arg 23 follows directly on the putative helical segment, which has several consequences for the binding. First, for the Arg 23 to reach into the binding pocket, the helix likely sits deeper in the transmembrane domains of RXFP3, which is possible for the smaller peptide lacking the A-chain. A few amino acid residues in TM5 (His 268 ) and TM6 (Thr 346 and Ser 349 ) affected R3 B1-22R binding but not relaxin-3 binding, which may be a result of this difference in positioning (Fig. 8). R3 B1-22R is a more potent antagonist as a peptide amide, with R3 B1-22R acid having a 10-fold lower affinity for RXFP3 (33). This is in contrast to relaxin-3, where the acid is favored. This can be explained by the fact that in relaxin-3, the longer tail allows the acid to potentially interact with Lys 271 ; however, in the shorter tail of the antagonist, this interaction is not possible. Thus, the amide form, which does not require burying a charge that's not involved in an ionic interaction, is favored.
In summary, our current mutagenesis studies have led to refined models of relaxin-3 bound to an active-state model and R3 B1-22R bound to an inactive-state model of RXFP3 with improved accuracy and reliability. These data together with extensive mutagenesis data on R3 B1-22R in the companion article (59) have significantly improved our understanding of how relaxin-3 and R3 B1-22R interact with RXFP3. Such information is extremely valuable for structure-based drug design efforts to design next-generation RXFP3 agonists and antagonists.

Site-directed mutagenesis
The forward and reverse primers for all mutants were designed with partially overlapping ends as described (49) and are listed in Table S1. Mutagenesis reactions on pcDNA3.1zeo (ϩ) HA RXFP3 were performed using the method described previously (34). Individual mutants were screened and confirmed by DNA sequencing on both strands (Centre of Translational Pathology, University of Melbourne) to ensure that there were no introduced random mutations.

Homology modeling and docking
A model of RXFP3 with relaxin-3 docked into the binding site was prepared as described earlier (34) but using a different template for homology modeling. Briefly, the NMR solution structure of relaxin-3 (PDB entry 2FHW (28)) was used together with a homology model of RXFP3 based on the crystal structure of angiotensin II receptor in an activated state (PDB entry 5XJM (44)). Distance restraints corresponding to mutational data were introduced during the homology modeling procedure in Modeler 9v18 (50) to create an initial docking pose of relaxin-3 on RXFP3. The model was then embedded in a POPC/POPE/cholesterol membrane (60:20:20 relative composition) of dimensions 1.7 ϫ 1.7 nm using the insane.py script (51) and the coarse-grained force field Martini version 2.0 (52). The system was then simulated for 500 ns to equilibrate the membrane using the Gromacs 2018.2 molecular dynamics engine (53). The system was then converted back to all-atom representation using an in-house modified version of the Backward algorithm (54). The system was then minimized and then equilibrated over 5 ns using the amber lipid17 and ff14SB force fields (55) and the molecular dynamics engine Amber18 (56). During the equilibration, the system was progressively heated up to 300 K with the protein heavy atoms restrained to their initial coordinates during the first 2.5 ns, and these restraints were then progressively removed over a 2.5-ns simulation time. The system was then simulated for 20 ns without restraints. The Langevin theromostat with a damping coefficient of 2.0 ps Ϫ1 and the Berendsen barostat with a coupling time constant of 1.0 ps were used to maintain pressure and temperature at 300 K and 1.0 bar, respectively. The root mean square deviation of the protein heavy atom coordinates to their initial position stabilized at 2.2 Å after a 4-ns simulation, suggesting that the system reached equilibrium.
A model of the antagonist bound to RXFP3 was generated similarly but using as a template for the receptor the X-ray structure of angiotensin I in an inactive state (PDB entry 4YAY (45)). The initial position of the antagonist was modeled manually with PyMOL version 2.0 (Schrödinger, LLC, New York) by considering the RXFP3/relaxin-3 binding mode, deleting the A-chain of relaxin-3, and repositioning the B-chain deeper in the binding pocket to satisfy the experimental data on the Arg 23 positioning. This model was then embedded in a membrane and refined using molecular dynamics simulation similarly as the RXFP3/relaxin-3 model. The root mean square deviation of the protein heavy atom coordinates to their initial position stabilized at 2.9 Å after a 12-ns simulation, suggesting that the system reached equilibrium.
The final frames of the RXFP3/relaxin-3 and RXFP3/antagonist simulations were energy-minimized and used for interpreting mutational data. The resulting molecular models of the complexes have MolProbity scores (57) of 1.5 and 1.3, corresponding to the 96th and 98th percentile (100th percentile is the best), suggesting that the geometry of the models is excellent.

Agonist and antagonist whole-cell binding assays
HEK293T cells were plated out at 25,000 cells/well and were transfected in a 96-well ViewPlate (PerkinElmer Life Sciences) with plasmids encoding the constructs of interest (250 ng/well) and Eu-H3/I5 and Eu-R3 B1-22R binding assays conducted as described previously (34,37,48). Specific binding screening assays were performed using 5 nM Eu-R3/I5 in agonist-binding assays or 5 nM Eu-R3 B1-22R in antagonist-binding assays in the absence (total binding) and presence (nonspecific binding) of 1 M unlabeled R3 B1-22R as described previously (34). Saturation binding assays were performed by preparing different concentrations of Eu-R3 B1-22R with (nonspecific binding) and without (total binding) competition of 1 M unlabeled R3 B1-22R in 1% BSA in relaxin receptor binding buffer (20 mM HEPES, pH 7.5, 1.5 mM CaCl 2 , 50 mM NaCl). Ligands were incubated for 90 min at room temperature and were then aspirated and washed with 200 l/well cold PBS. 100 l of Delfia Enhancement Solution was then added to each well, and the plate was shaken at 500 rpm for 15 min. The time-resolved fluorescence of europium was read on an Omega POLARstar (BMG Labtech) with the excitation/emission settings of 340 nm/615 nm. Specific binding was determined by subtracting the nonspecific binding from the total binding, and the graph was fitted to a one-site specific binding nonlinear regression. Each concentration point was performed in triplicate, and data were expressed as mean Ϯ S.D. from at least three independent experiments. Significance was assessed using one-way analysis of variance (ANOVA) with Fisher's least significant difference (LSD) test on GraphPad Prism version 7.0 (GraphPad Software, La Jolla, CA).

Cell surface expression assays
All receptor constructs were expressed with a hemagglutinin (HA) epitope tag on the N terminus. HEK293T cells were plated at 200,000 cells/ml of DMEM in a 24-well plate (Costar) precoated with poly-L-lysine and transfected with 1 g of plasmid DNA/well for 24 h as described previously (34). 24 h later, the media were aspirated, and cells were washed with Tris-buffered saline (TBS) solution (50 mM Tris, pH 7.4, 150 mM NaCl). Cells were then fixed with 3.7% formaldehyde in TBS for 20 min at room temperature, followed by two washes with TBS and subsequently blocked with 1% BSA/TBS for 30 min. The HA epitope tag was detected using a purified mouse monoclonal anti-HA, HA.11(16B12) (Covance), made up in a 1:1000 dilution in 1% BSA/TBS (400 l/well) for 2 h at room temperature. The cells were washed and incubated with a 1:1000 dilution of

Mechanism of agonist and antagonist interaction at RXFP3
Alexa 488 -conjugated goat anti-mouse IgG made up in 1% BSA/TBS (400 l/well). Cells were then washed two times; 200 l of lysis buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 1 mM EDTA, 0.25% Triton X-100) was added to each well; and the plate was shaken for 15 min at 300 r.p.m. at room temperature. Cell lysates were then scraped, and 180 l of cell lysates were transferred to a 96-well black Optiplate (Costar) to be read on Omega POLARstar (BMG Labtech) with excitation/emission wavelength of 490 nm/520 nm. Cell surface expression was determined by subtracting nonspecific binding of the antibody in cells transfected with pcDNA3.1 (ϩ) zeo and divided by the HA-RXFP3 WT expression to get a percentage HA-RXFP3 expression. Significance was assessed using one-way ANOVA with Fisher's LSD test on GraphPad Prism version 7.0.

cAMP activity assays
HEK293T cells were plated at 25,000 cells/well in a 96-well CellBind plate (Corning) and co-transfected the next day with plasmid encoding various RXFP3 receptor constructs and pCRE ␤-gal as a cAMP-response element (CRE)-controlled reporter of cAMP activity (58). 24 h post-transfection, medium was removed, and increasing concentrations of the RXFP3 full agonist, analogue 2 (37), made up in DMEM containing 500 nM forskolin were added to the respective wells. The stimulation was carried out for 6 h at 37°C, 5% CO 2 , after which medium was aspirated and plates were kept frozen at Ϫ80°C until further analysis. Development of the plates required the plates and assay buffers to be first thawed to room temperature. 25 l of assay buffer 1 (100 mM Na 2 HPO 4 , pH 8.0, 2 mM MgSO 4 , and 0.1 mM MnCl 2 ) was added into each well and shaken at room temperature for 10 min before adding 100 l of assay buffer 2 (similar to buffer 1 with the addition of 0.5% Triton X-100 and 40 mM ␤-mercaptoethanol) into each well and further shaken for 10 min at room temperature. 25 l of substrate for ␤-gal, chlorophenol red ␤-D-galactopyranoside (Roche Applied Science), was added into each well with shaking until color change was observed. The plate was then read and absorbance was measured at 570 nm on a Bio-Rad plate reader. Readings were normalized to medium containing 500 nM forskolin (positive control) as 100% and to medium only (negative control) as 0%. Normalized data were fitted to a nonlinear regression sigmoidal dose-response curve on GraphPad Prism version 6.0. Each concentration point was performed in triplicates, and the data were expressed as mean Ϯ S.E. from at least three independent experiments. pEC 50 values of each mutant construct were subjected to one-way ANOVA with Fisher's LSD test on GraphPad Prism version 7.0.