Proposed ligand binding site of the transmembrane receptor for neurotensin(8-13).

We report here the first proposed ligand binding site of the transmembrane receptor for neurotensin(8-13) in human and rat, the corresponding bound conformation of the peptide ligand, and site-directed mutagenesis studies that support the binding site model. These three-dimensional structures were generated by using a heuristic approach in conjunction with experimental data. The proposed neurotensin(8-13) binding site is primarily composed of eight residues (i.e., Phe326, Ile329, Trp334, Phe337, Tyr339, Phe341, Tyr342, and Tyr344 in the human receptor; Phe331, Ile334, Trp339, Phe342, Phe344, Phe346, Tyr347, and Tyr349 in the rat receptor) located in the third extracellular loop. The seven aromatic residues form an aromatic pocket on the extracellular surface of the neurotensin receptor to accommodate its ligands apparently by cation-pi, pi-pi, and hydrogen bonding interactions. The neurotensin(8-13) ligand adopts a compact conformation at the proposed binding site. In the bound conformation of neurotensin(8-13), the backbone of Arg9-Pro10-Tyr11-Ile12 forms the proline type I turn, and the hydroxy group of Tyr11 interacts with the two guanidinium groups of Arg8 and Arg9. These guanidinium groups are curled toward the hydroxy group so that they interact electrostatically with the hydroxy group, and that the guanidinium group of Arg9 forms an intra-hydrogen bond with the hydroxy group. The proposed three-dimensional structure may not only provide a basis for rationalizing mutations of the neurotensin receptor gene but also offer insights into understanding the binding of many neurotensin analogs, biological functions of the neurotensin receptors, and structural elements for species specificity of the neurotensin receptors, and may expedite developing nonpeptidic neurotensin mimetics for the potential treatment of the neuropsychiatric diseases.

We previously reported the development of two partial nonpeptidic neurotensin (8 -13), abbreviated as NT (8 -13), 1 mimetics whose Arg 8 -Arg 9 -Pro 10 portion is replaced with substituted indole-2-carboxylates as the partially flexible non-peptidic equivalents by our multiple template approach (1)(2)(3). NT (8 -13) is a peptide fragment of neurotensin (NT) and is more biologically relevant than the parent peptide (4). While we have been focusing on developing full nonpeptidic NT (8 -13) mimetics for the potential treatment of the neuropsychiatric diseases such as schizophrenia and Parkinson's disease (1,5), we have also engaged in delineating the ligand binding site of the transmembrane receptors for NT (8 -13) whose amino acid sequences in human and rat have been deduced (6 -8). Knowledge of the three-dimensional structure of the NT (8 -13) binding site can expedite the development of such mimetics and non-peptidic antagonists of NT for evaluating the physiological and putative pathological roles of NT, provide a basis for rationalizing mutations of the NT receptor gene, and enhance our understanding of the binding of many NT analogs, biological functions of the NT receptors, and structural elements for species specificity of the NT receptors.
One common approach to model G-protein-coupled transmembrane receptors for biogenic amines (i.e., dopamine, adrenaline, serotonin, and acetylcholine) involves (i) identifying the seven transmembrane helices by sequence similarity analysis based on the sequence of bacteriorhodopsin, whose crystal structure of the seven helices is available (9), and/or by hydropathy analysis; (ii) constructing the tertiary structure of such helices by using bacteriorhodopsin as a scaffold; and (iii) patching the extracellular and intracellular loops to the helical bundle (10,11). The last operation was not important to the modeling of these receptors, since the binding sites of the receptors were assumed to be located in the helical bundle. Unfortunately, this approach is not appropriate for modeling of the NT receptor, because ligands of the NT receptor are generally hexapeptides or large organic molecules, while the ligands of the G-protein-coupled receptors for biogenic amines are relatively small organic molecules. This suggests distinct binding regions for the two different types of ligands on their receptors. Intuitively, the binding site of NT (8 -13) should involve the residues in the extracellular loops. However, the three-dimensional structures of the loops of bacteriorhodopsin are not available, and, furthermore, these loops are not homologous to the loops of the NT receptors.
Another common approach to modeling of transmembrane receptors is de novo tertiary prediction (12). This involves (i) aligning a family of homologous sequences to identify the locations of the ␣-helices, ␤-strands, and ␤-turns, (ii) predicting such secondary structures, (iii) packing such secondary structures into a group of rough tertiary folds, (iv) refining the folds into detailed tertiary structures, and (v) selecting the most reasonable structure according to experimental data. This approach is, in principle, applicable to the modeling of the NT receptors. However, the difficulty of this approach in practice is that there are a myriad of different ways to pack the secondary structural elements in three dimensions given limited geometric constraints. Without further information, the selection of three-dimensional candidates is impractical.
In delineating the NT (8 -13) binding site, we applied the latter approach using constraints that were empirically defined according to the NT receptor specificity and the nature of the observed interactions between the NT receptor and its ligands. A heuristic approach, coupled with experimental data, allowed us to propose a likely three-dimensional structure of the NT (8 -13) binding site. In this article, we report the details of the proposed ligand binding site of the NT(8 -13) receptors in human and rat, the corresponding bound conformation of NT (8 -13), and site-directed mutagenesis studies that support the model of the NT(8 -13) binding site.

MATERIALS AND METHODS
Heuristic Modeling-Our proposed NT (8 -13) binding site was essentially derived by the following three constraints. The first constraint was that the proposed ligand binding site should be rich in aromatic residues. This aromatic binding site can tightly accommodate the positively charged ligands such as NT(8 -13) through cation-pi interaction (13)(14)(15)(16)(17)(18), and the neutral, aromatic ring-rich ligands such as SR48692 (19) through pi-pi interaction (20,21). It seemed plausible that a series of large ligands for the NT receptor, which includes NT (8 -13) and its peptide analogs, as well as the nonpeptidic SR48692, bind with comparable affinity to the same region (the same subsite or distinct and yet partially overlapping subsite) of the NT receptor. Another useful constraint arose from the postulate that, like the bradykinin and neurokinin binding sites (22,23), the NT binding site (i.e. the NT(8 -13) binding site) is located at least partially on the surface of the receptor, namely located in one or more of the three extracellular loops. It seems unlikely that the NT-like ligands bind completely inside the helical bundle, as with the relatively small biogenic amine ligands of the G-protein-coupled receptors as proposed by Trumpp-Kallmeyer and co-workers (10). To accommodate the entire peptide ligand within the helical bundle would require a substantial amount of energy to generate a large cavity by displacing a large number of water molecules and solvent ions bound in a preformed pocket, or by a dramatic structural rearrangement of the receptor. The last useful constraint was that the NT(8 -13) binding site should consist of most of varied residues of a large family of G-proteincoupled receptors. Otherwise, binding of the NT-like ligands will not be specific to the NT receptors.
With the above constraints, we identified the extracellular loops possessing more aromatic residues, generated possible tertiary structures of such loops by homology approach with a geometric constraint that there must also be a cluster of aromatic rings in the three-dimensional space, refined iteratively the structures according to different bound conformations of the NT-like ligands, and finally evaluated the hypothetical models by structure-activity relationship studies and sitedirected mutagenesis studies.
Hydrophacy Analysis-The amino acid sequence of the rat NT receptor was obtained from the literature (8), and the human NT receptor sequence was determined in our laboratory (7). Hydrophacy analysis was performed by using the PEPPLOT module in the GCG sequence analysis software package (version 7.2 for UNIX) (24). In the calculation of Kyte and Doolittle's or (Goldman, Engelman, and Steitz) hydropathy (25,26), we set the window to 9, 11, 13, 15, 17, 19, (27). HindIII and XbaI restriction endonucleases, T4 DNA ligase, Taq DNA polymerase, and the pGEM T-Vector kit were obtained from Promega (Madison, WI). The PCR Optimizer kit and pcDNA3 eukaryotic expression vector were obtained from Invitrogen (San Diego, CA), and M13 RF DNA was obtained from Pharmacia Biotech Inc. Sculptor in vitro mutagenesis system was obtained from Amersham Corp. Oligonucleotides were obtained from the Molecular Biology Core Facility (Mayo Clinic, Rochester, MN). Sequencing was done on an ALF DNA sequencer using the AutoRead Sequencing Kits (Pharmacia). The human neurotensin receptor cDNA was cloned as described previously (6). The template for mutagenesis was a PCR product of the 3Ј end of human neurotensin receptor cDNA that contained a silent mutation to change a HincII site into a HindIII site at the 5Ј end, and an XbaI site at the 3Ј end. This PCR product was subcloned into pGEM T-Vector, sequenced, and subcloned into M13mp18. Oligonucleotide sequences for site-directed mutagenesis were as follows: D340H, 5Ј-d(ACTCCGTTC-CTCTATCACTTCTACCACTACTTC)-3Ј; Y339F, 5Ј-d(TGGACTCCGT-TCCTCTTTGACTTCTACCACTAC)-3Ј.
Mutagenesis was done according to the manufacturer's recommendations. Putative mutants were screened by sequencing. Mutants were sequenced completely and subcloned into pcDNA3 eukaryotic expression vector. A PCR product of the 5Ј end of the human neurotensin receptor cDNA was subcloned into the above clones to assemble the full-length constructs. Transient transfections were done using the calcium phosphate method in HEK-293 cells.
Cell Culture-Chinese hamster ovary (CHO-K1) cells that had been stably transfected with the hNTR(Leu) (28), and rNTR gene, and HEK-293 cells transiently transfected with chimeric receptors by the CaPO 4 precipitation method (29), were cultured in 150-mm Petri plates containing 20 ml of Dulbecco's modified Eagle's medium containing 100 M minimal essential medium nonessential amino acids (Life Technologies, Inc.) supplemented with 5% (v/v) FetalClone II bovine serum product (HyClone Laboratories, Logan, UT). CHO cells (subculture 9 -19) were harvested at confluence, while HEK-293 cells were harvested 48 h after transfection. Cells were removed by aspiration of the medium, followed by a wash with 50 mM Tris-HCl, pH ϭ 7.4 (6 ml), which was discarded, resuspension in 5-10 ml of Tris-HCl, scraping the cells with a rubber spatula into a centrifuge tube, and collection of cells by centrifugation at 300 ϫ g for 5 min at 4°C, in a GPR centrifuge (Beckman Instruments, Fullerton, CA). The cellular pellet (in 50 mM Tris-HCl, 1 mM EDTA, pH ϭ 7.4) was stored in liquid nitrogen until radioligand binding was performed.
For use in binding assays, crude membranal preparations were prepared by centrifugation of the cellular pellet at 35,600 ϫ g for 10 min. The supernatant was decanted and discarded, and the cellular pellet was resuspended in 2 ml of Tris-HCl ϩ 1 mM EDTA (pH ϭ 7.4) followed by homogenization with a Brinkmann Polytron at setting 6 for 10 s. Centrifugation was repeated as above, the supernatant was decanted and discarded and the final cellular pellet was resuspended in 50 mM Tris-HCl, 1 mM EDTA, 0.1% bovine serum albumin, and 0.2 mM bacitracin. Protein concentration of the membranal preparation of CHO-K1 cells was estimated by the method of Lowry et al. (30) using bovine serum albumin as a standard.
Radioligand Binding Assays-We used a Biomek 1000 robotic workstation for all pipetting steps in the radioligand binding assays as described previously by our group (31). Competition binding assays with [ 3 H]NT (1 nM) and varying concentrations of unlabeled NT, and peptide analogs were carried out with membranal preparations from the appropriate cell lines. Nonspecific binding was determined with 1 uM unlabeled NT in assay tubes with a total volume of 1 ml. Incubation was at 20°C for 30 min. The assay was routinely terminated by addition of cold 0.9% NaCl (5 ϫ 1.5 ml), followed by rapid filtration through a GF/B filter strip that had been pretreated with 0.2% polyethyleneimine. Details of binding assays have been described previously (32). The data were analyzed using the LIGAND program (33).
PI Turnover Assays-Intact CHO-K1 cells were harvested for PI turnover analysis at about 80% confluence. Cells were detached from the Petri plates by removal of culture medium and followed by incubation of the cellular monolayer for 20 min at 37°C with gentle shaking in a modified Puck's D1 solution containing 2 mM EGTA. We have described elsewhere the details of assaying in intact cells the relative changes in PI turnover by using a radioactively labeled precursor (34). Briefly, intact CHO cells were prelabeled with D-myo-[ 3 H]inositol (18.3 Ci/mmol) in the presence of lithium chloride (final concentration, 10 mM). Cells were then stimulated with NT or the appropriate analog. The amount of [ 3 H]inositol 1-phosphate produced by the cells was isolated chromatographically on Dowex 1-X8 (200 -400 mesh). For the experiments described here, the stimulation time was 30 min. The number of CHO cells per assay tube was 8 ϫ 10 4 to 2.25 ϫ 10 5 .
Statistical Analysis-The values presented for K d and EC 50 are expressed as the geometric means Ϯ S.E. (35).
Homology Modeling-Homology modeling involved three steps. First, structures with homologous sequences were identified by searching a protein data base with the Sequence Search module implemented in the Quanta program (36). The protein data base contained 756 crystal structures and was obtained from Molecular Simulation, Inc. (36). The Sequence Search module performed sequence alignment and similarity score calculation. In the sequence search, the cutoff score for short sequence was set to 20. Second, a three-dimensional structure was generated according to the scaffold of the identified homologous structure with the homology modeling modules of the Quanta program. Third, the structure was refined by energy minimization and constrained molecular dynamics (MD) simulations (a subset of the atoms in the system was allowed to move in the dynamics simulation) in vacuum employing the AMBER 4.1 programs with the Cornell et al. force field (37,38).

RESULTS
Hydropathy Analysis-At least seven helices were identified according to the Kyte and Doolittle and the Goldman, En-gelman, and Steitz methods regardless of the window setting. The locations of the rat receptor helices are essentially identical to those found by Tanaka et al. (8), except that the regions of helices II, III, and V are slightly different. The hydropathy profiles of the human and rat NT receptors are essentially identical. We therefore conclude that the regions of the seven helices in the human and rat receptors are the same (Fig. 1).
NT (8 -13) Binding Site-We first identified a region from the third extracellular loop to the first four residues of the seventh helix (E3 loop) as a potential location of the binding site or part of the binding site, since it carries 10 aromatic residues in the human and rat receptors and is located on the surface of the receptors. This region has the highest percentage (48%) of the aromatic residues, compared to only 11% of aromatic residues in the entire NT receptors. Moreover, according to Tanaka's alignment of the rat NT receptor and some representative G-protein-coupled receptors, the E3 loop is the region of greatest structural variation. This is consistent with different ligand binding specificity and strengthens the likelihood of the E3 loop being the location of the ligand binding site.
We then identified a segment from residues 215 to 230 of the crystal complex (2aat.pdb) of aspartate aminotransferase (EC 2.6.1.1) mutant coupled with pyridoxamine phosphate as the most homologous structure of the E3 loop. There was 43% identity in a 14-residue overlap. In the crystal complex of aminotransferase, the segment of residues 215-230 forms a loop structure whose one end smoothly joins an ␣-helix fragment. This structural feature is consistent with our prediction of the E3 loop based on the hydropathy analysis and therefore serves as an ideal scaffold for modeling of the E3 loop. The aspartate aminotransferase-based homology modeling of the E3 loop revealed that the curled backbone of the E3 loop naturally forms an aromatic pocket in the three-dimensional space (Fig. 2). The pocket in the human receptor is primarily composed of Phe 326 , Ile 329 , Trp 334 , Phe 337 , Tyr 339 , Phe 341 , Tyr 342 , and Tyr 344 , while the one in the rat receptor is primarily composed of Phe 331 , Ile 334 , Trp 339 , Phe 342 , Phe 344 , Phe 346 , Tyr 347 , and Tyr 349 . As apparent in Fig. 2, a cluster of aromatic groups in the E3 loop serves as an ideal binding site to accommodate the positively charged ligands and the neutral, aromatic ring-rich ligands. This aromatic pocket was therefore postulated as the ligand binding site of the NT receptor. The rat and human receptors differ in their sequences in the E3 loop: Thr 341 and Phe 344 in rat are mutated to Pro 336 and Tyr 339 in human, respectively. As a consequence, the binding cavity in the human receptor is slightly smaller than that in the rat receptor, which may provide a structural basis for understanding the structure activity relationship differences observed in studies with bulky NT analogs and the site-directed mutagenesis study on Tyr 339 of the human receptor (vide infra).
NT (8 -13) Bound Conformation-In the sequence search of NT(8 -13), we also identified, with more than 60% identity in 4or 5-residue overlap, five crystal structures containing the Arg-Pro-Tyr sequence and similar residues of Ile and Leu. These crystal structures are p-hydroxybenzoate hydroxylase (1phh-.pdb), serine proteinase (1hf1.pdb), cytotoxic t-lymphocyte proteinase I (2cp1.pdb), p-hydroxybenzoate hydroxylase (2phh-.pdb), and rat mast cell protease (3rp2.pdb). The structures of hydroxylases (1phh.pdb and 2phh.pdb) were discarded, since the backbone conformations of Arg-Pro-Tyr in these structures were fully extended, which by visual inspection were obviously too large to bind to the proposed binding site. In the crystal structures of the three serine proteinases, the backbone of Arg-Pro-Tyr adopts a compact conformation with the proline type I turn, and the hydroxy oxygen atom of Tyr forms an intra hydrogen bond with the guanidinium proton of Arg. The compact conformation identified in the three crystal structures thus served as an ideal scaffold for modeling the bound conformation of NT (8 -13).
Manually docking the structure of NT(8 -13) directly derived from the serine proteinase-based homology modeling revealed a fit between the ligand and the proposed binding site. The docking was performed with constraints of avoiding close van der Waals contact and achieving maximal degree of pi-pi interaction of Tyr 11 with the receptor and cation-pi interaction of Arg 8 and Arg 9 with the aromatic pocket. The fit between the two structures essentially cross-validated the modeling of both the ligand and the binding site. In the manual docking-generated NT(8 -13) bound receptor complex, the guanidinium group of Arg 8 of the ligand was, however, pointing to the solvent, rather than interacting with the receptor. Further refining the serine proteinase-based NT (8 -13) conformer by 500 ps of constrained MD simulation at 298 K in vacuo resulted in a ligand conformation as shown in Fig. 3. In the constrained MD simulation, the receptor of the complex was not allowed to move. As depicted in Fig. 3, the guanidinium group of Arg 8 also curls toward the phenol group of Tyr 11 to form the electrostatic interactions with the hydroxy oxygen atom of Tyr 11 and the aromatic groups of the receptor (vide infra). The compact conformation in Fig. 3 requires obviously less desolvation energy to achieve binding than other conformations in which the Arg and Tyr residues form hydrogen bonds with water. It is therefore reasonable to propose the refined compact conformation as a bound conformation of NT (8 -13). This compact conformation is consistent with suggestion that NT(8 -13) adopts a Type I ␤-turn by Garcia-Lopez et al. (39) and other research groups (40).
The Ligand-Receptor Complex-The tertiary structures of the helical bundle and the first two extracellular loops were generated by the homology approach in the same way as we did for the E3 loop (Fig. 4). Packing of the extracellular loops to the helical bundle was biased by the conformations of the residues at the ends of the loops determined by the homology approach. It was also biased by our hypothesis that the three loops should stabilize each other. The NT (8 -13) bound receptor complex as shown in Figs. 5 and 6 was derived from the average structure of 100 ps constrained MD simulation at 298 K in vacuo on the manual docking-generated NT(8 -13) bound receptor complex. The first two extracellular loops and the seven helices, except for the first four residues of the seventh helix, were not allowed to move in the simulation. Both ligand and receptor (the E3 loop part) slightly adjusted their conformations to adapt to each other during the MD simulation. The initial structure of the complex consisted of the manual docking-generated NT(8 -13)-E3 loop complex, the helical bundle, and the other two extracellular loops.
In addition to the electrostatic and van der Waals interactions between the ligand and the binding site, other characterized non-bonded interactions (Fig. 7) are: 1) a weak cation-pi interaction of the guanidinium group of Arg 8 with the aromatic ring of Phe 341 in the human receptor, or of Phe 346 in the rat receptor, the weak interaction was judged by the distance between the interacting groups, where the approximate distance between the center of the guanidinium group and the center of the aromatic group of Phe 341 in the human receptor is 6 Å; 2) a cation-pi interaction of the guanidinium group of Arg 9 with the aromatic groups of Phe 326 , Phe 341 , and Tyr 344 in the human receptor, or of Phe 331 , Phe 346 , and Tyr 349 in the rat receptor, where the approximate distance between the center of the guanidinium group and the centers of the aromatic groups of Phe 326 , Phe 341 , and Tyr 344 in the human receptor is 5, 5, and 6 Å, respectively; 3) a pi-pi interaction of the phenol group of Tyr 11 with the aromatic rings of Phe 326 , Trp 334 , and Tyr 339 in the human receptor, or of Phe 331 , Trp 339 , and Phe 344 in the rat receptor (see Fig. 7 for distance); and 4) hydrophobic interactions of the aliphatic side chains of Ile 12 and Leu 13 with the methylene group of Phe 326 and the side chain of Ile 329 in the human receptor, or of Phe 331 and Ile 334 in the rat receptor (see Fig. 7 for distance).
Structure Activity Interpretations-Based on the proposed NT(8 -13) binding site, the binding affinities of various NT analogs can be interpreted in terms of their structural ele-  Tables I and II. As described above, the receptor interaction of the guanidinium group of Arg 8 is weaker than the receptor interactions of the side chains of Arg 9 and Tyr 11 . This may explain why deleting Arg 8 in NT(8 -13) did not abolish the binding of NT(9 -13) ( Table I). In addition, as indicated in Figs. 6 and 7, the side chain of Arg 8 of the ligand is longer than adequate to form electrostatic interaction with the aromatic group of Tyr 11 in the ligand, and is shorter than adequate to form cation-pi interaction with Phe 341 of the human receptor (or Phe 346 in the rat receptor). This is compatible with the biological data, as substitutions of Arg 8 with L-lysine, L-ornithine (L-Orn), and L-2,4diaminobutyric acid (L-Dab) did not significantly change the binding affinity (Table I). The chain lengths of Lys, Orn, and Dab are only one, two, and three methylene groups shorter than that of arginine, respectively. These side chains are still long enough to interact with Tyr 11 keeping the compact conformation to facilitate binding, but not long enough to interact fully with Phe 341 of the human receptor. Slight decreases of the binding affinities of such analogs may result from the weaker cation-pi interaction of the shorter side chain with the residues of the receptor, and the polarity change that resulted from change of guanidinium group to ammonium group. Again, substitutions of Arg 8 with the D-form residues of Arg, Lys, and Orn did not significantly affect the binding affinity (Table I), since the side chains of the D-form residues are also long enough to interact with Tyr 11 , but not with the receptor.
Similarly, the side chain of Arg 9 is adequately long to interact with the binding site. Substitutions of Arg 9 with L-Lys, L-Orn, and L-Dab did not significantly reduce the binding affinity (Table I). However, the cation-pi interaction of Arg 9 with the binding site is stronger than that of Arg 8 , explaining why the affinities of the 9-substituted D-isomers were lower than those of the 8-substituted D-isomers (Table I).
NT (8 -13) adopts the proline type I turn in its bound conformation, which suggests that Pro 10 plays an important role in keeping the peptide compact or in preorganizing the compact conformation before binding to the relatively small binding site of the receptor. Thus, replacing Pro 10 with glycine resulted in much lower affinity in binding than substitutions at the 8-or 9-position, which do not affect the compact conformation (Table II).

Proposed Binding Site for NT(8 -13) Transmembrane Receptor
As indicated in the bound conformation of NT (8 -13), the phenol group of Tyr 11 also plays a key role in maintaining the critical compact conformation to binding through establishing a hydrogen bond with the guanidinium group of Arg 9 and electrostatic interactions with the guanidinium groups of the two arginine residues. In addition, the aromatic ring of Tyr 11 inter-acts with the aromatic residues of the binding site. Thus, replacement of Tyr 11 with L-alanine, the L-alanine analog whose one methyl hydrogen atom is replaced by a cyclohexyl group (Cha), D-Tyr, or D-Trp resulted in low binding affinities relative to those of the 10-substituted analogs, and much lower binding affinity than those with substitutions at the 8-or 9-position Proposed Binding Site for NT (8 -13) Transmembrane Receptor (Tables I and II). This is because the side chains of alanine and Cha lack the aromatic and hydroxy groups, and D-Tyr and D-Trp move the aromatic groups and the hydrogen bond acceptors away from the binding site.  (Tables  I and II). However, they were less potent in binding than NT (8 -13) and NT because a hydroxy group responsible for the inter-and intra-hydrogen bonds is missing in [L-Phe 11 ]NT or [L-F-Phe 11 ]NT (8 -13).
Since the binding site of the human receptor is slightly smaller than the one in rat according to the proposed models of the binding sites, analogs with large substituents (i.e., L-Trp, L-␣-Nal, and L-␤-Nal) at the 11-position bound to the rat receptor more tightly than to the human receptor (Table I). L-␣-Nal and L-␤-Nal are the L-alanine analogs with an ␣or ␤-naphthyl group in place of one methyl hydrogen atom. This is best shown by [L-␣-Nal 11 ]NT(8 -13), which demonstrated a 127-fold in-crease of binding at the rat receptor in comparison to the binding at the human receptor. Stronger pi-pi interactions of the naphthyl group with the aromatic binding site and the less desolvation energy of the naphthyl group required for binding may well compensate for the loss of the interactions of the hydroxy group of Tyr 11 , and therefore contribute to the increase of the binding of [L-␣-Nal 11 ]NT (8 -13) to the rat receptor.
The smaller binding site of the human receptor also makes the side chain of the valine residue of [L-Val 12 ]NT (8 -13) or [L-Val 13 ]NT (8 -13) easier to approach to the hydrophobic region in the human receptor than in the rat receptor; therefore, both [L-Val 12 ]NT (8 -13) and [L-Val 13 ]NT (8 -13) bound to the human receptor more tightly than the rat receptor (Table II). In both the human and rat receptors, the reduced hydrophobic volumes of the two valine-substituted analogs caused a decrease of their binding affinities relative to NT (8 -13).
The relatively small NT (8 -13) binding sites in both human and rat also shed light on understanding why [ 3 H]NT(8 -13) is more potent in binding than [ 3 H]NT (4). The relatively lower binding affinity of NT is probably because NT has to undergo hydrolysis to release the NT(8 -13) fragment prior to binding, or spend extra energy to arrange the NT(1-7) fragment in such a way that this fragment does not interfere with the binding of NT (8 -13).
Site-directed Mutagenesis Studies-The binding affinities of NT and [L-␤-Nal 11 ]NT (8 -13) to the mutants of the human NT receptor are listed in Table III. The binding affinity of NT was decreased by only 2-fold when the negatively charged Asp 340 was mutated to a neutral or positively charged histidine (Table  III). This suggests that Asp 340 probably interacts indirectly with NT through long range electrostatic interaction, accounting for the 2-fold decrease of the binding affinity of NT to the mutant. In other words, Asp 340 is not directly involved in the binding of NT, although this residue was thought to interact with the positively charged arginine residues of the ligands before the binding site model was proposed. According to the proposed model, Asp 340 is indeed not involved in the binding. The major characterized interactions between the ligand and receptor are cation-pi, pi-pi, and hydrophobic interactions, but not ionic interaction.
According to the proposed NT(8 -13) binding sites and the sequence difference in the E3 loop between the human and rat receptors as described above, Tyr 339 is one of the key residues directly involving the binding of the human receptor, and the counterpart of this residue in rat is phenylalanine. This suggests that the Y339F mutant of a human receptor is a rat-like NT receptor. Therefore, the binding affinity of [L-␤-Nal 11 ]NT- (8 -13) to the Y339F mutant should be similar to that of the rat NT receptor, and higher than that of the human receptor, while the binding affinity of NT to the same mutant should not demonstrate significant change (vide supra). As apparent from Table  III, the binding results of the Y339F mutant well supported the predictions made from the proposed NT(8 -13) binding sites.  (8)(9)(10)(11)(12)(13), and its 8-, 9-, 11-substituted analogs to the human and rat NT receptor (28,44) Peptide   (8)(9)(10)(11)(12)(13), and its 10-, 11-, 12-, 13-substituted analogs to the human and mouse NT receptor (41)(42)(43) Peptide a Data assumes the structure of mouse NTR is close to that of rat receptor. Tanaka et al. first reported the putative regions of the transmembrane helices and the loops in the rat NT receptor based on their hydropathicity profile analysis and sequence comparison with some representative G-protein-coupled receptors (8). Watson et al. later reported the putative regions of the helices and loops in the human NT receptor based on sequence similarity between the human and rat receptors and on the rat helical model reported by Tanaka and co-workers (7). However, identification of transmembrane helices should rely more on the hydropathy profile than sequence similarity. This prompted us to study the hydropathy profile of the human NT receptor. Interestingly, our identification of the regions of the helices and loops in the human receptor is slightly different from the report for the rat NT receptor by Tanaka and coworkers, although the two sequences of the receptors share 84% homology. Therefore, we investigated the hydropathy profile of the rat receptor, and identified slightly different regions of the helices and loops in the rat NT receptor. Since the hydropathicity profile of the rat NT receptor was not shown in Tanaka's report, we could not evaluate which one is more appropriate. Although the deletion mutation work on the rat NT receptor by Yamada et al. was not affected by the two different helix models (45), it is noteworthy that two such models exist. One of these models may turn out to be critical for some mutagenesis studies.
We have shown here the three-dimensional structure of the extracellular loops and the helical bundle of the NT receptors. However, this three-dimensional structure, except for the E3 loop, should be viewed as one snapshot of many different possible structures. Our patching of the extracellular loops to the helical bundle and the modeling of the helical bundle per se were highly biased as stated above. Exhaustive exploration of other possible packing modes, conformations of the loops, and alignments of the seven helices was not pursued. This was because we were only interested in the topology of the binding site, which we thought was likely located in the E3 loop, and because it is still impossible to explore all possible conformations of the loops and the alignments of the seven helices due to the limitation of current computing power and to the lack of rigorous studies on the MD simulations of transmembrane proteins in lipid.
Nonetheless, we think the proposed structure of the E3 loop should be considered as a low-resolution and empirical model of the NT binding site for the following reasons. (i) We determined the side chain conformations of the E3 loop by homology method and further refined such conformations by limited MD simulation. However, we did not compare the dihedral angles of those side chains with their most frequently found torsion angle values (46), nor did we exhaustively study the other possible conformations of those side chains. (ii) The hypothesis of a common region (without excluding the distinct and yet partially overlapping subsites) of the NT receptor for different ligand binding is reasonable. Based on this hypothesis, it is therefore reasonable to define the constraint that the binding site is rich in aromatic residues. This constraint is common in other biologically important proteins such as acetylcholinesterase and FKBP12-binding protein, which all possess an aromatic binding site (47,48). The NT binding site model derived by this constraint is further supported by our structure-activity relationship studies on the analogs with different aromatic groups or without an aromatic group (28) and by the binding of the neutral, aromatic ring-rich ligand UK-73,093 (49) and the compounds developed by Johnson et al. (50). (iii) Selection of the E3 loop as a candidate for the location of the binding site or part of the binding site is not arbitrary. Although the first extracellular loop is also rich in aromatic residues, it carries more conserved residues of a family of G-protein-coupled receptors than the E3 loop. It is therefore not likely to carry the pharmacophores of the receptors responsible for the specific binding of NT and its analogs. Consistently, computational and experimental studies on the receptor for bradykinin, a peptide bioregulator that competes successfully with NT in binding to rat mast cell receptors (51), also suggested the bradykinin binding site located in the third extracellular loop (22). (iv) The match of the two independently generated three-dimensional structures of the NT(8 -13) binding site and NT(8 -13) cross validated the NT(8 -13) binding site model. (v) Most importantly, the proposed model agrees surprisingly well with the structure-activity relationship studies on all of our reported NT analogs and with our site-directed mutagenesis studies.
It should also be noted that the ligand-bound receptor complex that we present here was a selection from many different possible complexes. Other complexes may agree as well with the experimental data. However, systematic docking of ligands to the NT receptors employing the SYSDOC docking program appeared unnecessary in this case, since the predicted receptor structure itself is unjustified (18). We had to use experimental data to determine empirically the ligand-bound receptor complex.
It is possible that SR48692 and the series of our NT analogs with different aromatic groups or without an aromatic group bind to distinct regions in the NT receptors rather than to the same region as NT (8 -13). In that case, we would not assume the NT(8 -13) binding site is rich in aromatic residues. However, searching for a region possessing the most varied residues of a family of the G-protein-coupled receptors would still lead us to identify the E3 loop as the location of the binding site or part of the binding site, although we would have fewer constraints. However, the chances of correct modeling of the patching of the extracellular loops to the helical bundle and modeling of the helical bundle per se are too small to evaluate, since there are a myriad of different ways to model this system. We have, therefore, more confidence on the modeling of the E3 loop binding site in comparison with the modeling of the rest of the NT receptor.
There is a recent report suggesting that SR48692 binds to a distinct region of the NT receptor based on the finding that 20% of the SR48692 binding sites in the NT receptors were inaccessible to NT, and on the finding that a rat NT receptor mutant, which had a deletion of residues 45-60 in the amino-terminal extracellular domain, retained the same affinity for [ 3 H] SR48692, but decreased the affinity for [ 125 I]NT by 3000-fold (52). However, according to the same experimental findings, we think it is also possible that SR48692 and NT bind to the same region. The observation of the so-called "NT-insensitive site" (we term it "NT-inaccessible SR48692 binding site") was probably because 20% of the NT receptors at the membrane were artificially oriented in such a way that the binding site for both SR48692 and NT gave access only to the lipophilic SR48692, but not to the hydrophilic NT peptide. Mutation on the aminoterminal extracellular domain, which is probably involved in anchoring the NT receptor at the membrane, can result in steering artificially the NT receptor at the membrane in an orientation that no or few NT molecules can access to the binding region, while the lipophilic SR48692 can freely access the same region. This may explain why the same affinity for SR48692 and the 3000-fold decreased affinity for NT were detected in the mutant with a deletion of a fragment in the NH 2 terminus, as opposed to the wild-type NT receptor. The NH 2terminal mutation result does not necessarily imply that the NH 2 -terminal extracellular domain constitutes part of the binding site for NT, but not for SR48692.
Cusack et al. have recently identified a region from the sixth helix to the COOH terminus as a species-specific domain sensitive to binding of the newly developed [L-␣-Nal 11 ]NT (8 -13) ligand (44). This result implies that the E3 loop is sensitive to the binding of the [L-␣-Nal 11 ]NT(8 -13) ligand, and is consistent with our proposed binding site, assuming [L-␣-Nal 11 ]NT (8 -13) binds to the same region of the receptor as NT (8 -13). However, we emphasize the following: with chimerism study, we cannot determine whether (case I) the identified speciesspecific domain directly interacts with the ligands or (case II) this domain just stabilizes the other putative extracellular loops, which may constitute the binding site, so that it indirectly affects the binding of the ligands, nor can we conclude that the residues, which were found important to binding according to site-directed mutagenesis studies by us and others (53), are directly interacting with the ligands. As illustrated in Fig. 4, the second extracellular loop interacts with the first extracellular loop which interacts with the E3 loop. Conformational change of the E3 loop would affect the other extracellular loops. If case I is correct, we would propose that the E3 loop is directly involved in binding. We would then be able to explain all the experimental data including the identification of the species-specific domain for binding. If case II is true, we would propose that the binding site is not in the E3 loop. We would then find the NT(8 -13) binding site model contradictory to the experimental data that are clearly associated with the E3 loop binding site. Therefore, of the two possibilities, we empirically ruled out the second possibility.
Our proposed NT(8 -13) binding site reported here provides the first testable three-dimensional model. This empirical and low resolution model may not only provide a basis for rationalizing mutations of the neurotensin receptor gene, but also offer insights into understanding the binding of many neurotensin analogs, biological functions of the neurotensin receptors, and structural elements for species specificity of the neurotensin receptors, and may expedite developing nonpeptidic neurotensin mimetics for the potential treatment of neuropsychiatric diseases. The model remains to be verified and refined by further experimental studies. Site-directed mutagenesis studies assisted by our three-dimensional model are under way and will be reported in due course. Even if this model is viewed as an upgrade from a twodimensional cartoon to a three-dimensional sketch, it was this upgraded model that provided us with insights into developing [L-␣or ␤-Nal 11 ]NT (8 -13) and [L-Cha 11 ]NT (8 -13) analogs, and has culminated in [L-␣-Nal 11 ]NT(8 -13) as the first useful, picomolar affinity agonist capable of discriminating the rat NT receptor from the human NT receptor. Again, it was this threedimensional sketch that provided us with a broader vision to interpret the results of the mutagenesis studies, which can sometimes be misinterpreted and abused without a structural basis. Finally, we hope our NT(8 -13) binding site model serves as an incentive to others for proposing alternative three-dimension models to understand better and to utilize the experimental findings of the neurotensin receptor functions.