Pancreatic Polypeptide Is Recognized by Two Hydrophobic Domains of the Human Y4 Receptor Binding Pocket*

Background: The Y4R is involved in regulation of food intake and gastrointestinal transport. Results: Mutagenesis studies revealed several residues displaying a significant loss of potency for hPP. Conclusion: Tops of TM2, TM6, and TM7 interact with the hY4R native agonist hPP. Significance: Characterizing the structure of the Y4R binding pocket is crucial for the development of new anti-obesity drugs. Structural characterization of the human Y4 receptor (hY4R) interaction with human pancreatic polypeptide (hPP) is crucial, not only for understanding its biological function but also for testing treatment strategies for obesity that target this interaction. Here, the interaction of receptor mutants with pancreatic polypeptide analogs was studied through double-cycle mutagenesis. To guide mutagenesis and interpret results, a three-dimensional comparative model of the hY4R-hPP complex was constructed based on all available class A G protein-coupled receptor crystal structures and refined using experimental data. Our study reveals that residues of the hPP and the hY4R form a complex network consisting of ionic interactions, hydrophobic interactions, and hydrogen binding. Residues Tyr2.64, Asp2.68, Asn6.55, Asn7.32, and Phe7.35 of Y4R are found to be important in receptor activation by hPP. Specifically, Tyr2.64 interacts with Tyr27 of hPP through hydrophobic contacts. Asn7.32 is affected by modifications on position Arg33 of hPP, suggesting a hydrogen bond between these two residues. Likewise, we find that Phe7.35 is affected by modifications of hPP at positions 33 and 36, indicating interactions between these three amino acids. Taken together, we demonstrate that the top of transmembrane helix 2 (TM2) and the top of transmembrane helices 6 and 7 (TM6–TM7) form the core of the peptide binding pocket. These findings will contribute to the rational design of ligands that bind the receptor more effectively to produce an enhanced agonistic or antagonistic effect.

G protein-coupled receptors (GPCRs) 5 are the most prominent group of cell surface proteins. They are formed by seven transmembrane helices (TM) that are connected by intracellular and extracellular loops (ECL). GPCRs can be activated by several stimuli such as hormones, light, or odorant molecules (1). It is estimated that ϳ30% of all prescribed pharmaceuticals modify the activity of GPCRs (2) indicating that these receptors are fundamental drug targets in modern pharmacology.
The Y 4 receptor (Y 4 R) is a member of the NPYR, a class A GPCR family composed of Y 1 R, Y 2 R, Y 4 R, and Y 5 R receptors in humans. The NPYR is closely related to other class A GPCR families such as the neuropeptide FF receptor family and the orexin receptor family (3). NPY receptors are physiologically coupled to the G i or G o proteins; however, other reports show that rabbit Y 2 R and rabbit Y 4 R are also coupled to the G q protein, triggering an increase in inositol phosphate (4). These receptors are activated by the NPY family of peptide hormones, consisting of NPY, peptide YY (PYY), and pancreatic polypeptide (PP). NPY peptides and receptors form a multiligand/multireceptor system that plays a role in several physiological and pathological processes such as obesity and cancer (5). NPY peptides consist of 36 amino acids, are C-terminally amidated, and share high sequence identity. Shared structural features include a C-terminal helix. Despite high sequence homology and common structural features among NPY receptors and peptide hormones, however, there are significant differences in the affinity of these peptide hormones to the different receptor subtypes as well as differences in how the peptides bind their receptor (6).
The Y 4 R, cloned in 1995 (7), has 375 amino acids and was found to be expressed in the colon, small intestine, pancreas, prostate (8), brain, and coronary arteries (7). Physiologically, the Y 4 R is involved in the regulation of food intake (9), colonic anion transport (10), and adipose tissue and bone formation synergistically with Y 2 R (11). The Y 4 R sequence is one of the least conserved members of the NPYR family among different species, making it the fastest evolving functional member of the family (12). This makes it difficult to transfer conclusions from other Y receptor members to this subtype. Its main agonist, hPP, is produced by endocrine cells of the Langerhans islets of the duodenal part of the pancreas. These cells are also found in the gastrointestinal tract (13), albeit in much lower numbers. hPP was the first member of the NPY family of peptides to be identified. It is secreted after food ingestion in proportion to its caloric content (14), and it promotes appetite suppression and inhibition of gastric emptying (15). This ligand was already found in ancient tetrapod evolution and appears to be one of the fastest developing peptides of the family (12). Because of its role in appetite suppression, this system is a very attractive target for the design of new therapeutic compounds for fighting obesity.
Detailed knowledge of the receptor-peptide interaction is essential for rational structure-based drug design. Although several studies characterizing the binding pocket of NPY receptors have been published in the past years (6, 16 -19), little is known about the Y 4 R. To our knowledge, only one study describes a subtype-selective interaction between Y 4 R and its ligand hPP (20) where the conserved residue Asp 6.59 of human Y 1 R/Y 4 R binds to Arg 35 of the peptides pNPY or hPP. In contrast, Asp 6.59 of hY 2 R and hY 5 R interacts via Arg 33 of pNPY and pPYY. The importance of this conserved residue was later identified in other systems closely related to the NPYR, such as the NPFF receptor 1 and 2 systems (3) or the prolactin-releasing peptide receptor (21).
In this study, we characterize in detail for the first time structural determinants of the hPP-hY 4 R interaction. Residues located at extracellular regions of the hY 4 R, chosen according to their location in the receptor sequence and in comparative models of hY 4 R, were mutated to determine their role in hPP binding. Simultaneously, a set of hPP analogs was also developed to pinpoint specific interactions between hPP and hY 4 R. We identified Tyr 2.64 , Asp 2.68 , Asn 6.55 , Asn 7.32 , and Phe 7.35 as members of the hY 4 R binding pocket. Furthermore, hPP analogs with modifications in residues 27, 33, or 36 revealed these positions to be interaction partners with the receptor. These results clearly demonstrate the importance of the top of transmembrane helix 2 (TM2) and the top of transmembrane helices 6 and 7 (TM6 and -7) for hPP binding and illustrate the complexity of the intermolecular interactions within the hY 4 R subtype.
Peptide characterization was achieved by matrix-assisted laser desorption/ionization (Ultraflex III MALDI-ToF/ToF, Bruker Daltonics, Billerica, MA) and by electrospray ionization mass spectrometry. Peptide purities were determined on two analytical reversed-phase HPLC systems using a linear gradient of 0.1% (v/v) TFA in H 2 O (eluent A) and 0.08% (v/v) TFA in acetonitrile (eluent B). The gradient used was 20 -70% of eluent B in eluent A in 40 min. The purity of the synthetic peptides was higher than 92% (data not shown).
Preparation of hY 4 R Mutants-The single mutations were inserted by site-directed mutagenesis into the protein sequence. Pfu Turbo DNA polymerase (Agilent) (2.5 units/l) was used as a reaction enzyme in combination with 10ϫ reaction buffer. The plasmid hY4_EYFP_N1 was used as a template (50 -100 ng), and sense/antisense oligonucleotides were used in 2.5-10 pmol/l concentrations depending on the reaction conditions. The deoxyribonucleotide (dNTP) mix was added in 10 mM concentration, and dimethyl sulfoxide (DMSO) was used occasionally to reduce secondary structures. DpnI (Thermo Fischer Scientific) was used to eliminate the original dsDNA template. Constructs were transformed in semi-competent Escherichia coli DH5␣ or E. coli JM109 cells, and the plasmid DNA was isolated using a Wizard plus Mini or Midi DNA purification system kit (Promega). The desired mutations were confirmed by sequencing of the complete coding sequence. The mutated positions are named after the system of Ballesteros and Weinstein (22).
Fluorescence Microscopy Studies-HEK293 cells were seeded and transfected with cDNA encoding hY 4 R constructs as earlier described (23). The nuclei were stained with Hoechst 33342 (0.5 mg/ml) for 10 min after starving the cells for 20 min in Opti-MEM medium. Fluorescence microscopy pictures were captured using an ApoTome Imaging system with an Axio Observer microscope (Carl Zeiss, Jena, Germany).
Signal Transduction Assays-Signal transduction assays were performed on 24-or 48-well plates as described previously with minor changes (3,23). As transfection reagents, Metafectene and Metafectene Pro (Biontex) were used. The analysis of the data obtained was performed using the GraphPad Prism 5.03 software (GraphPad Software, San Diego). For each hypothesis, the data were processed using a nonlinear regression analysis, obtaining concentration-response curves displaying EC 50 (37), and ␦-opioid receptor (PDB code 4EJ4) (38).
These structures were aligned with MUSTANG (39), and the resulting multiple sequence alignment was aligned with a multiple sequence alignment of hY 1 R, hY 2 R, hY 4 R, and hY 5 R using ClustalW (40). Sequence alignments were adjusted to remove gaps within transmembrane ␣-helices and ensure that highly conserved residues remain aligned (supplemental Fig. S1). hY 4 R residues were threaded onto the three-dimensional coordinates of aligned residues in each of the 14 GPCRs.
Missing Atom Coordinates Were Constructed Using Rosetta Loop Construction Protocols-Missing density and loop regions were reconstructed using Monte Carlo Metropolis fragment replacement and cyclic coordinate descent loop closure algorithms in Rosetta (41). All models underwent repacking and gradient minimization with RosettaMembrane (42). An additional constraint was included to account for the expected disulfide bond between hY 4 R residues Cys 3.25 and Cys 5.25 .
The final set of models was clustered based on r.m.s.d. using bcl::Cluster (43). The top scoring models from the five largest clusters were used for docking studies.
Docking of Pancreatic Polypeptide (PP) into the Comparative Model of hY 4 R-A set of NMR structure conformations of bovine pancreatic polypeptide (PDB code 1LJV) (44) was docked into the hY 4 R comparative models. Bovine pancreatic polypeptide differs only on positions 6 and 23 with respect to hPP and has similar affinity for the hY 4 R as earlier reported (45,46). The use of ILJV provided a guide for the structural distinc-tion between the peptide's helical region and dynamic tail region. The helical region (residues 14 PEQMAQYAAELRRY-INML 31 ) was first docked into the hY 4 R models. Four distinct helix conformations were docked into 37 hY 4 R comparative models without ECLs, guided by a predicted interaction between hY 4 R Tyr 2.64 and hPP Tyr 27 .
C-terminal Residues of hPP Were Added Using de Novo Folding with Experimental Restraints-The five C-terminal residues of hPP (TRPRY) were constructed using Rosetta's low resolution de novo folding algorithm where residues are represented as "centroids" (47). Three experimentally derived restraints between hY 4 R and PP residues were used to guide this step using an 8-Å distance cutoff between residues Asp 6.59 and Arg 35 , Phe 7.35 and Arg 33 , and Asn 7.32 and Arg 33 (20,48). All restraints are detailed in Table 1.
The ECLs were rebuilt as described for the comparative modeling of hY 4 R, with the addition of these experimental constraints. Additionally, these models were refined to atomic detail, replacing centroids with side chain rotamers based on a backbone-dependent rotamer library and energy minimization with RosettaMembrane (49 -51).
Models Were Relaxed Using Atomic Resolution Experimental Restraints-Models were again clustered based on r.m.s.d. Top scoring models from the largest clusters were visually inspected for binding poses that preserved the experimental restraints. Selected models underwent an additional relaxation step with constraints adjusted to reflect atomic level interactions between residues Asp 6.59 and Arg 35 (3 Å distance between the two ␦-oxygen atoms on Asp 6.59 and the side chain nitrogen atoms on Arg 35 ), and residues Asn 7.32 and Arg 33 (4 Å distance between the ␦-oxygen atom on Asn 7.32 and the two side chain nitrogen atoms on Arg 33 ). These constraint distances allow for possible hydrogen bonding and salt bridge interactions. An additional restraint between hY 4 R Phe 7.35 and PP Tyr 36 was introduced. Final models were clustered and visually inspected, and nine representative models were selected. The overall workflow for receptor modeling and peptide docking is summarized in Fig. 1.

RESULTS
The comparative models presented here reflect an iterative process where multiple rounds of modeling were performed in parallel with in vitro experiments. Early models were generated based on comparative modeling with only seven GPCR templates and limited experimental restraints. The number of tem- plates eventually increased to 14 as more GPCR structures became available. Additionally, predicted interactions seen in earlier models were used to guide some of the mutational assays. These assays provided additional restraints that were included in later models. The final models represent one plausible way that binding between hPP and hY 4 R can occur based on all experimental evidence available. Tyr 2.64 of TM2 Interacts with Tyr 27 -Tyr 2.64 is located on top of TM2 and was found to be important in hY 1 R (16). This amino acid is conserved in all receptor subtypes except for hY 5 R ( Fig.  2A); therefore, we hypothesized that this amino acid would be relevant in hY 4 R. To investigate Tyr 2.64 , this position was mutated to Ala and showed a significant shift of the EC 50 value with 65-fold loss of potency but with 130% of efficacy (Fig. 3A). This high efficacy value matches with excellent membrane localization of the receptors revealed by fluorescence microscopy studies. Furthermore, the modification of Tyr 2.64 to the larger aliphatic amino acid, Leu, was much better tolerated and displayed wild type-like activity with slightly reduced efficacy (83%). Furthermore, to test the importance of the hydroxyl moiety at this position, the mutant Y2.64F was constructed. The native ligand hPP displayed a 4-fold loss of potency and wild type-like efficacy (94%).
Because alanine-scan studies of pNPY (52) 27 to Leu pursued the aim of introducing a longer aliphatic amino acid, whereas the introduction of Cha was constructed to investigate the effects of a more bulky hydrophobic amino acid. Additionally, the substitution of Tyr 27 to Phe was made to investigate the relevance of the hydroxyl group and to discard a possible hydrogen bond. [Ala 27 ]hPP displayed an 8-fold loss of activity (EC 50 11.78 nM), whereas the Leu and Cha variants showed wild type like potency on hY 4 R (  27 ]hPP displayed only a 9-fold loss of potency on Y2.64F. Taken together, this confirms that the presence of a bulky hydrophobic amino acid is favorable for this interaction site of the binding pocket and suggests that Tyr 2.64 might interact with a second amino acid. This is supported by the comparative models, because seven models have Tyr 2.64 within 8 Å of Tyr 27 and eight models have Tyr 2.64 within 8 Å of Leu 31 , which might be the second interaction point. Nevertheless, experimental data are needed to confirm this second interaction point. Other Positions Highlight the Importance of ECL1-Another amino acid investigated was Asp 2.68 . This residue is conserved in all receptor subtypes except in hY 2 R ( Fig. 2A), which contains Gly at this position. Asp 2.68 was found to be important in mutagenesis studies on the hY 1 R (17), and it is one of the interaction points between receptor and NPY on the hY 5 R system (6). In hY 4 R, the exchange of Asp 2.68 to Ala led to a 94-fold loss in activity and decreased the efficacy dramatically to 39% of the wild type receptor response ( Fig. 3B and Table 3). The loss in efficacy fits with the high intracellular accumulation of receptors demonstrated by fluorescence microscopy (Fig. 4). In contrast, the exchange to Glu or Asn regained the efficacy (93 to 87%, respectively) and displayed a loss of potency for hPP of only 9-and 16-fold, respectively.
Additionally, position Trp 2.70 , which is conserved in all receptor subtypes, was described in some GPCR to belong to the motif WXFG and to be important for receptor activation (53). To prove its relevance, Trp was mutated to Ala, leading to a 107-fold loss in potency. Mutation to Tyr displayed only a 2-fold loss of potency (Table 3). These exchanges reveal that an aromatic or bulky side chain is necessary at this position to keep wild type-like activity. An ensemble of hY 4 R comparative models was constructed through several rounds of loop building and energy minimization followed by selection of the best models. Alongside the flowchart are representative models to illustrate the evolution of the comparative model. hPP was docked through the placement of the PP helix, de novo addition of the C-terminal residues, and finally the addition of the ECLs of hY 4 R. Those steps were guided by experimentally derived restraints and followed by selection of the top models.
Tested Residues in TM3, ECL2, and TM5 Do Not Play a Relevant Role in the Binding Pocket-The single residue tested in TM3 was Gln 3.32 . This position has been shown to participate in the binding pocket of nearly all crystallized class A GPCRs (2). In the hY 4 R, the exchange of Gln 3.32 to Ala displayed a wild type-like potency when tested with hPP.
The ECL2 is the least conserved region between receptor subtypes. To elucidate the role of this ECL, several amino acids were mutated to Ala (Fig. 2B). To investigate polar interactions with the positively charged residues of the peptide Arg 33 Table 4) but wild type-like potency. A further mutation to Ser at this position led to a partial regain of efficacy and displayed a 3-fold loss of potency for hPP, whereas an additional exchange to Phe at this position, to prove the relevance of the hydroxyl moiety, led to wild type-like potency for hPP.
TM6 Plays a Crucial Role in Building the Binding Pocket-Asp 6.59 on top of TM6 has already been shown to form a direct contact to the ligand (20). To further characterize the role of this part of the receptor, residues in close proximity to Asp 6.59 were chosen for further mutagenesis studies. Phe 6.54 was mutated to Ala because of its importance in other GPCRs closely related to the hY 4 R, its aromatic character, and its close proximity to Asp 6.59 . F6.54A displayed a 4-fold loss of hPP potency with reduced efficacy (73%, Table 4) that corresponds to poor membrane localization observed by fluorescence microscopy (Fig. 4). Asn 6.55 was found to be involved in the hY 1 R binding pocket (18), and because initial comparative models oriented the side chain of this amino acid toward the interior of the hY 4 R binding pocket, this indicated its possible involvement in ligand-receptor interactions. Stimulation of N6.55A with hPP resulted in an 8-fold loss in activity and 79% of efficacy ( Fig. 5A and Table 5) suggesting that this position is important in the receptor pocket.
Additionally, His 6.62 was investigated because of its proximity to Asp 6.59 at the beginning of ECL3. H6.62A revealed an EC 50 of 0.4 nM, which is moderately better than wild type.
TM7 Is a Contact Point of hPP in the hY 4 R and Asn 7.32 Interacts with Arg 33 of hPP-The first position investigated in TM7 was Asn 7.32 . This residue was found to be relevant for PYY binding on the Y 1 R (54), and initial comparative models indicated the possible importance of this position as well. Asn 7.32 was mutated to Ala, Arg, and Asp. The effect of these substitutions increased in the following manner: Arg Ͼ Ala Ͼ Asp (Table 5 and Fig. 5B). This suggested that the introduction of a positive charge might cause a repulsion that was eliminated with the introduction of the negative charge Asp. Hence, we supposed that one of the C-terminal positively charged arginines might be the interaction partner. Because Arg 35 of hPP was already identified to interact with hY 4 R Asp 6.59 , Arg 33 was suggested as a possible interaction partner of Asn 7.32 . To clarify the binding hypothesis, the hPP analogs [ADMA 33 ]hPP, [SDMA 33 ]hPP, and [Lys 33 ]hPP were synthesized. Position 33 of hPP was modified to SDMA and ADMA to maintain the positive charge and simultaneously reduce the possibility of hydrogen bond formation. Whereas [Lys 33 ]hPP revealed wild type-like activity, the exchange of Arg to ADMA or SDMA displayed a 7-to 6-fold loss of activity (  Fig. 5B). These experiments demonstrated that the introduction of a shorter amino acid such as Lys was better tolerated than the double methylation of Arg, which was not tolerated at all. Unlike N7.32A, N7.32D showed only a 2-3-fold loss of potency when tested with [Lys 33 ]hPP and [ADMA 33 ]hPP, revealing that the introduction of the double methylation or the reduction of the side chain length did not affect the binding pocket.
Phe 7.35 Interacts with Arg 33 as Well as Tyr 36 -Phe 7.35 on the top of TM7 was also investigated, because this conserved position might be a suitable interaction point because of its aromatic characteristics and location. Furthermore, in the hY 5 R, Tyr 7.35 was found to be relevant for the receptor and was suggested to belong to the receptor binding pocket (6). The initial comparative models also suggested its orientation to the interior of the proposed binding pocket. The amino acid was mutated to Ala and Ile, displaying a moderate loss in potency (7-fold) for the Ala mutant, whereas the Ile mutant revealed 41-fold loss of potency compared with wild type ( Table 6). The higher potency loss caused by the Ile variant might indicate that the distance or the space available between Phe 7.35 and the ligand or other positions in the receptor is important. F7.35I also showed reduced efficacy that corresponds to high intracellular receptor localization as demonstrated by fluorescence microscopy (Fig. 4). Preliminary comparative hY 4 R models suggested that this residue was in close proximity to Arg 33 and Tyr 36 . Furthermore, Arg 33 and Tyr 36 revealed to be critical residues for pNPY binding on the hY 4 R (52). Taken all these facts together, we hypothesized that Phe 7.35 might interact with one or both amino acids, Arg 33 or Tyr 36 , of the peptide. As described above, the exchange of Phe 7.35 to Ile displayed higher impact on receptor activity than the Ala substitution ( plus the loss produced by the analog. Finally, F7.35I leads to a dramatic loss in activity when tested with both analogs. Preliminary models suggested Tyr 36 as a second interaction partner to Phe 7.35 . To characterize this hypothetical interaction, several analogs with modifications on Tyr 36 were synthesized ( Table 6). The introduction of Phe 36 was well tolerated on the hY 4 R, indicating that the hydroxyl group of the side chain of Tyr was not playing a relevant role. The introduction of a nonaromatic amino acid such Ile brought a dramatic loss in potency on the wild type receptor (123-fold). Shortening the length of the side chain to Ala led to an even higher loss in potency (Ͼ2000-fold). Surprisingly, the introduction of unnatural amino acids such as Cha and Nle was better tolerated (EC 50 values 0.6 and 14.40 nM, respectively, see Table 6 and Fig. 6). These hPP analogs were tested on F7.35A to investigate the type of interaction between these two positions. F7.35A showed a 17-fold loss of activity when tested with [Phe 36 ]hPP (EC 50 25.29 nM). [Ile 36 ]hPP was not tolerated at all with F7.35A. This peptide displayed a dramatic loss of activity (EC 50 not determinable), whereas Nle was slightly better tolerated revealing an EC 50 value 679-fold over wild type. Finally, the second unnatural amino acid, [Cha 36 ]hPP, displayed a 138-fold loss of activity over wild type (EC 50 of 211.2 nM) on F7.35A and was thereby substantially better tolerated than Ile and Nle. Taken together, this indicates the need for a bulky hydrophobic amino acid at this position (Table 6 and Fig. 6).
The final comparative models support these results because within the nine best models obtained, 8 out of 9 showed Asn 7.32 within 8 Å of distance to Arg 33 , and 9 out of 9 models showed Phe 7.35 within 8 Å of proximity to Arg 33 (Fig. 8).
In addition to the residues mentioned above, the conserved residue His 7.39 , which was one helix turn deeper in TM7, was also investigated. The Ala mutant did not reveal any detectable activity, and fluorescence microscopy pictures confirmed intracellular localization of the receptor. No further studies have been performed as this receptor variant is stuck in trafficking.
Docking of PP to the hY 4 R Comparative Model-Pancreatic polypeptide was docked into the comparative model of hY 4 R to assist interpretation of experimental results. Because inactive GPCR structures were used for our templates, it was important to consider the effects this may have on docking an agonist to this model. Rosetta's comparative modeling protocol is insensitive to the state of GPCR templates. Templates are used only in the initial transmembrane helix positioning. Several relaxation steps allow for energy-based adjustments to these placements. Additionally, all extracellular loops are rebuilt in accordance with Rosetta's de novo folding algorithm. It is conceivable, however, that the helical conformations of an active template may be altered enough to fall outside of the conformational area explored with inactive templates. We compared the r.m.s.d. of our templates with the latest agonist-bound GPCR crystal structures. Pairwise alignments using the structure-based alignment tool MAMMOTH revealed that the average r.m.s.d. value of our inactive structures (2.9 Ϯ 0.6) is not significantly different from that of the active structures (2.8 Ϯ 0.6). Importantly, the average r.m.s.d. value is unchanged when combining the two groups (3.1 Ϯ 0.5).
In addition to our analysis, Tautermann and Pautsch (55) examined the binding sites of active and inactive ␤ 2 -adrenergic receptors. They show that the binding site is very similar between the inactive and active states. Previous modeling studies with the inactive structure predicted the binding mode of an agonist that overlapped well with that seen in the agonistbound crystal structure (55).
The initial placement of the PP helix was guided specifically by the altered activity of hY 4 R Tyr 2.64 and hPP Tyr 27 mutants ( Table 2). This placement provided a starting position from which the dynamic ECLs and C-terminal tail of hPP might be  folded to simulate additional interactions suggested by the mutational data. These interactions specifically include a predicted salt bridge between hY 4 R Asp 6.59 and PP Arg 35 , a predicted hydrogen bond between hY 4 R Asn 7.32 and PP Arg 33 , a predicted cation-interaction between hY 4 R Phe 7.35 and PP Arg 33 , and an interaction between hY 4 R Phe 7.35 and PP Tyr 36 .
The restraints imposed by these experimental results were included initially as low resolution restraints based on residue proximity. To complete the model, several restraints were adjusted to higher resolution atom level restraints in an attempt to capture the proposed interactions on an atomic level. The specific restraints imposed and their corresponding steps are described in Table 1. When PP was docked using the low resolution restraints, 81% of the generated models did not significantly violate any of the restraints. In the final step, when the high resolution restraints were imposed, 29.8% of the models generated were able to fit these restraints with no significant violations. This was encouraging in that a significant portion of our models were capable of fitting proposed atom-level interactions. A subset of nine top-scoring models that showed no significant violation of high resolution restraints was selected as the final ensemble for discussion. These models fit well with the majority of the experimental results, accurately portraying residues found to affect activity as well as those residues that failed to show any effect on activity. Specifically, the predicted salt bridge between Asp 6.59 and Arg 35 is well represented in eight of the nine models. All models show less than a 4.0 Å distance between both inter-residue oxygen-nitrogen pairs, providing possible salt bridge interactions or hydrogen bonding. Six of the nine models demonstrate a distance of less than 3.2 Å between the oxygen in hY 4 R Asn 7.32 and amine group in PP Arg 33 , providing for the possibility of a hydrogen bond between these residues. hY 4 R Phe 7.35 and PP Arg 33 point toward each other in all nine models, which is conducive to the proposed cationinteraction. Additionally, hY 4 R Phe 7.35 and hPP Tyr 36 were oriented toward each other in four models. Finally, hY 4 R Asp 2.68 is within 8 Å and points toward the PP helix in five models, suggesting an interaction between the hPP helix and hY 4 R Asp 2.68 . One of the nine models is shown in Fig. 7, A and B, highlighting the binding site and residues important for PP-hY 4 R binding.
The importance of hY 4 R Trp 2.70 for hPP binding is the only experimental finding not well reflected in the models. In all but one of the nine models, it is pointing away and/or not in close proximity to hPP. Possible explanations include inaccuracy of the model in this region, increased dynamics of this region as displayed in our models, or an indirect effect that involves a second site on the receptor that interacts with both hPP and Trp 2.70 . It is interesting that the length of TM2 varies in the models, thereby changing the length of the first intracellular loop dramatically from three residues in two of the models, 9 -11 residues in five models, and 12-13 residues in two models. Because the models did not converge on a consistent length of ECL1, and precision is a prerequisite for accuracy, we expect that accuracy in this region might be low. This discrepancy in loop length is shown in Fig. 7C.
The residues that failed to show a significant effect on activity in the mutational assays are generally not contacting PP in the model. The majority of these residues are located in ECL2, which is consistently localized on the edge of the receptor away from hPP. Specifically, Lys 4.72 , Glu 4.79 , Phe 4.80 , Asp 4.83 , His 5.34 , and Phe 6.54 are further than 8 Å away from any PP residue. Gln 3.32 , Glu 4.67 , Trp 5.29 , His 6.62 , and His 7.39 are within 8 Å of an hPP residue in only three of the nine models, and Tyr 5.38 is within 8 Å of an hPP residue in only two of the nine models. ECL2 and the residues not involved in PP binding are shown in Fig. 7D.

TABLE 4 Signal transduction results of ECL2 and TM5 mutants of the hY 4 receptor
The immunoprecipitation accumulation assays were performed using increasing concentrations of hPP for 1 h. EC 50 values from dose-response curves were determined. n represents the number of independent experiments, each performed in duplicate.  FEBRUARY 28, 2014 • VOLUME 289 • NUMBER 9 serve as a foundation from which to identify the residues that line the binding pocket. For example, five of nine models show that hY 4 R Ser 5.28 and PP Thr 32 are within 8 Å of each other, suggesting a possible interaction between these two residues.

DISCUSSION
In the NPY receptor family, the ECL1 and TM6 were regions described to form the binding pocket and interact with the peptide (6,20). Additionally, hY 1 R, which shares high sequence homology with hY 4 R, has been extensively characterized in the past. Many amino acids located on ECL1, TM6, and TM7 are crucial for the interaction (16,19,54). Taking all these data into consideration, we expected the hY 4 R binding pocket to be composed of amino acids located in these areas of the receptor. Furthermore, it was also expected that hY 4 R has a second interaction site with the peptide on top of TM2 or beginning of ECL1, as suggested for hY 5 R (6).
We have now identified a binding pocket for the hY 4 R system that is composed of several residues located on TM2, TM6, and TM7. The first identified position in the pocket, Tyr 2.64 , is conserved in hY 1 R, hY 2 R, and hY 4 R. It is also present in the prolactin-releasing peptide receptor from several species, including human, rat, and mouse. Tyr 2.64 was found to be involved in ligand binding on the hY 1 R and was suggested to belong to a hydrophobic pocket (19). In the hY 4 R, Tyr 2.64 demonstrated that bulkiness and not aromaticity is critical for the interaction with hPP.
Y2.64L displayed a small decrease in efficacy for hPP. This might be caused by a small portion of receptors being trapped intracellularly (Fig. 4). The substantial amount of intracellular accumulation could be due to high expression levels of the mutant receptor, although it is not very likely because all constructs share the same promoter. It is more probable that this intracellular increase is due to an impaired folding of the mutant. However, the signal intensities suggest that enough active receptors are present in the cell membrane.
To elucidate a candidate position on the peptide side to interact with position Tyr 2.64 of hY 4 R, earlier Ala-scanning mutagenesis studies on the NPY peptide family were considered (56). Among others, Tyr 27 of NPY and PYY is relevant for binding in all NPY receptor subtypes (52,56). This conserved residue in the three peptide ligands of the NPY family was thought to be a likely candidate to interact with Tyr 2.64 of hY 4 R. Because Leu and Cha at position 27 of hPP have a nonplanar configuration compared with the wild type Tyr, in the presence of Leu or Cha on position 2.64 of hY 4 R the interaction might be slightly impeded, and a lack of space between Tyr 2.64 and Tyr 27 seems to be a limiting factor for the interaction to take place. This fact would support the close distance between these two positions suggested by the comparative models (Fig. 7A). Additionally, a hydrogen bonding interaction could be discarded between hY 4 R Tyr 2.64 and hPP Tyr 27 . Besides this, we could not explain the relevance of the hydroxyl moiety of Tyr 2.64 because an aromatic amino acid lacking the hydroxyl moiety like Phe is not as well tolerated as a hydrophobic amino acid like Leu. One hypothesis could be that Phe would adopt a slightly different orientation than Tyr or Leu, and therefore the interaction with the ligand could be slightly impeded. Overall, our data are most consistent with a hydrophobic interaction between hY 4 R Tyr 2.64 and hPP Tyr 27 . Furthermore, the fact that the activity shift obtained for Y2.64A with hPP was larger than the shift obtained for [Ala 27 ]hPP on hY 4 R would indicate that Tyr 2.64 might interact with another position in the peptide or within the receptor.
Confirming the importance of ECL1, the nearby residues Asp 2.68 and Trp 2.70 proved critical for the hY 4 R-hPP interaction. D2.68A displayed high loss of potency and efficacy for hPP. Further mutations on Asp 2.68 suggest that perhaps a polar or negatively charged amino acid is needed for correct export to or stability in the membrane. The relevance of this position is supported by the fact that on the hY 5 R Asp 2.68 has been proven to interact with Arg 25 of pNPY (6). Additionally this residue was hypothesized to form electrostatic interactions with NPY in the hY 1 R (17,19). On hY 4 R, Asp 2.68 may form hydrogen bonds with the peptide. Also, a polar effect on the structure that stabilizes the receptor binding pocket might be a feasible function for this position. The nearby residue Trp 2.70 needs a bulky hydrophobic amino acid. Although the models show this residue not directly pointing to the peptide, further data are needed to elucidate the role of this residue on the hY 4 R binding pocket. This position could participate on direct ligand binding or have a more structural role affecting the nearby important positions when mutated.
On TM6, Asn 6.55 participates in the binding pocket of the hY 4 R. Our results are supported by the loss in NPY binding displayed by N6.55A on studies with the hY 1 R (18). The com-  parative models demonstrate this amino acid pointing to the inner side of the proposed binding pocket close to Phe 7.35 (Fig.  7, A and B). Asn 6.55 is a candidate for interaction with Arg 35 of hPP as in six models these residues are within a distance of 8 Å (Fig. 8). Moreover, as reported recently, this position is involved in ligand receptor interactions of many crystallized class A GPCRs (2). This fact strongly supports our data and confirms the role that Asn 6.55 of hY 4 R has in the hPP binding pocket. The results obtained on position Asn 7.32 suggest that this residue is a key player in the binding pocket of hY 4 R. The muta-tion to Ala displays a small loss in potency for hPP. Prior studies on the hY 1 R, PYY and 1229U91 (GR231118), a Y 4 R agonist and a Y 1 R antagonist, displayed a loss in binding for N7.32A (54). It could be shown that Asn 7.32 might be in close proximity with a positively charged residue, probably one of the two Arg of the C-terminal segment of hPP. To characterize the relationship between hY 4 R Asn 7.32 and Arg 33 of hPP, position 33 was modified to Lys to investigate the influence of the side chain length. Also the asymmetric and symmetric side chain dimethylations were tested at this position. Side chain methylations block  hydrogen bond donor positions and increase hydrophobicity and bulkiness of the residue (57). Furthermore, the ability to form polar interactions such as dipole-dipole interactions might be impeded by double side chain methylation. The asymmetric and symmetric double methylation on position Arg 33 of hPP produced a potency loss (6 -7-fold) on hY 4 R, probably by blocking potential hydrogen bonding positions, potential dipoles, or due to steric hindrance. On N7.32A, the double side chain methylation at position Arg 33 of hPP had a more dramatic effect. This might cause conformational changes in doubly methylated Arg 33 of hPP impeding interactions with close by residues such as Phe 7.35 . These data are in agreement with the shortening of the side chain in [Lys 33 ]hPP that resulted in a smaller potency loss on N7.32A, suggesting that Asn 7.32 and Arg 33 of hPP are in very close proximity. The fact that these three peptide analogs displayed potencies similar to wild type on N7.32D also supports an interaction with Arg 33 , because Asp maintains the hydrogen bonding capability and incorporates a negative charge able to form an ionic bond with position Arg 33 . Accordingly, we were able to demonstrate that Asn 7.32 interacts with Arg 33 possibly by hydrogen bonding or polar interactions. This hypothesis is supported by the great relevance of Arg 33 as already demonstrated in the Ala scan (52). Comparative models where Arg 33 of hPP is located between Asn 7.32 and Phe 7.35 of hY 4 R nicely reflect this hypothesis (Fig. 7B).
The last residue of the proposed binding pocket is Phe 7.35 . The exchange of Phe 7.35 to Ile led to a higher potency and efficacy loss than the exchange to Ala, possibly due to steric hindrance. This position has been found to belong to the binding pocket of several class A GPCRs, among them the peptide receptors human CXC chemokine receptor type 4 and the rat neurotensin receptor 1 (2). Furthermore, this position might highlight the singularity of the Y 4 R binding pocket with respect to the Y 1 R. To investigate the role of hydrophobicity and size of Arg 33 33 ]hPP on F7.35A, because the methyl groups can reduce the distance between both positions. A second interaction point of Phe 7.35 was suggested by preliminary models to be Tyr 36 as hypothesized in previous studies on the Y 1 R (58). The fact that an aliphatic amino acid such as Ile with a branched ␤-carbon is not tolerated in contrast to Cha or Phe could suggest a need for space close to the peptide backbone. Moreover, the effect of these ligands on F7.35A indicates that in the absence of Phe at position 7.35 an aromatic amino acid must be present at position 36 of hPP. This may arise for conformational reasons as only an aromatic amino acid with a planar structure might be able to contact position 7.35 in the absence of Phe. So, in the presence of both aromatic groups ainteraction might be established between Phe 7.35 and Tyr 36 . In the absence of the aromatic group at position 36 of hPP, this residue might form hydrophobic interactions instead.
Our data provide the first insights into the complex binding pocket of the hY 4 R system derived from a combination of modeling and mutagenesis. As it may not be possible to solve the structure of all GPCRs, we demonstrate that this iterative method of study is very promising for understanding structurally uncharacterized receptors. As the model is in agreement with experimental data, it can be used to generate further testable hypotheses regarding the receptor-peptide interaction contributing to the development of ligands with enhanced hY 4 R activity.