Structural Determinants of Binding the Seven-transmembrane Domain of the Glucagon-like Peptide-1 Receptor (GLP-1R)*

The glucagon-like peptide-1 receptor (GLP-1R) belongs to the secretin-like (class B) family of G protein-coupled receptors. Members of the class B family are distinguished by their large extracellular domain, which works cooperatively with the canonical seven-transmembrane (7TM) helical domain to signal in response to binding of various peptide hormones. We have combined structure-based site-specific mutational studies with molecular dynamics simulations of a full-length model of GLP-1R bound to multiple peptide ligand variants. Despite the high sequence similarity between GLP-1R and its closest structural homologue, the glucagon receptor (GCGR), nearly half of the 62 stably expressed mutants affected GLP-1R in a different manner than the corresponding mutants in GCGR. The molecular dynamics simulations of wild-type and mutant GLP-1R·ligand complexes provided molecular insights into GLP-1R-specific recognition mechanisms for the N terminus of GLP-1 by residues in the 7TM pocket and explained how glucagon-mimicking GLP-1 mutants restored binding affinity for (GCGR-mimicking) GLP-1R mutants. Structural analysis of the simulations suggested that peptide ligand binding mode variations in the 7TM binding pocket are facilitated by movement of the extracellular domain relative to the 7TM bundle. These differences in binding modes may account for the pharmacological differences between GLP-1 peptide variants.

cellular signaling processes. Crystal structures of peptide ligand-bound ECDs of 11 of the 15 secretin-like class B GPCRs have been solved, and these structures exhibit similar binding modes with respect to the C-terminal conformation of the peptide ligands (3,5). Peptide ligand binding modes revealed by crystal structures of GLP-1-and exendin-(9 -39)-bound GLP-1R ECD (6, 7) (supplemental Fig. S1) are consistent with GLP-1 and GLP-1R truncation, chimera, site-directed mutagenesis, and cross-linking studies (4,(7)(8)(9)(10). In contrast, structural information on the 7TM domain of class B GPCRs is limited to GCGR (11) and corticotropin-releasing factor 1 receptor (12). So far, no full-length class B receptor structures have been determined resulting in a relative lack of information regarding the bioactive conformation of the N-terminal portion of the peptide hormones, as well as the role of the 7TM domain in peptide ligand recognition. NMR and x-ray crystallography studies indicate that the N-terminal region preceding the conserved ␣-helix of the class B peptide hormones (GLP-1 His 7 -Thr 13 , exendin-4 His 1 -Thr 7 , and glucagon His 7 -Thr 13 ) is flexible in solution as well as in the ligand-bound state (6,7,(13)(14)(15). The receptor-bound conformation of the N-terminal region of class B GPCR peptide ligands is proposed to be stabilized by an amino acid motif that is conserved in the class B GPCR peptide ligands (Thr 11 -Phe 12 -Thr 13 in GLP-1) (16), and it may induce an N-capping conformation similar to that observed in the ECD-bound NMR structure of pituitary adenylate cyclase-activating polypeptide (PDB ID 1GEA) (17). Recently, NMR structures of an 11-mer GLP-1 analogue were solved in alterative conformations containing a C-terminal ␣-helix (PDB ID 2N08) and an N-terminal ␤-turn (PDB ID 2N09), and stabilization of these conformations by cyclization cross-links (PDB IDs 2N0N and 2N0I) differentially influenced GLP-1R binding affinity and agonist potency (18). The accumulated ligand and receptor structure-activity relationships suggest that a flexible conformation of the first seven residues allows the peptide ligands of class B GPCRs to interact with residues deep in the 7TM binding pocket (5) and that the ligand N terminus may adopt a more constrained conformation to activate the receptor as proposed for GLP-1 (18). A central polar hydrogen bond network has been identified in GLP-1R that plays a role in peptide ligand biased signaling (19,20).
Although GLP-1R and GCGR share over 50% sequence identity, they have opposing physiological roles; GCGR is activated by glucagon during fasting and mobilizes glucose into the blood, whereas GLP-1R functions postprandially by stimulating insulin secretion to control circulating glucose levels (1,2). It is thus necessary to understand the ligand selectivity and recognition mechanism of the GLP-1R/GCGR axis to design specific drugs directed at either receptor. Chimeric constructs of GLP-1 in combination with glucagon have been used to identify structural determinants of ligand selectivity for GLP-1R and GCGR (31)(32)(33). Substitution of C-terminal GLP-1 residues with corresponding glucagon residues decreases GLP-1R binding and maintains low binding affinity for GCGR. The affinity of this GLP-1/glucagon chimera is however rescued by substituting the 7TM domain of GCGR with that of GLP-1R, suggesting that the N-terminal region of GLP-1 (His 7 -Leu 20 ) is selectively recognized by the 7TM domain of GLP-1R (31)(32)(33). Chimeric peptide ligands consisting of the N-terminal part of glucagon combined with the C-terminal part of GLP-1 have high affinity for both GCGR and GLP-1R (31). These GLP-1/glucagon chimera experiments demonstrate that the ECD and 7TM domains of GLP-1R both play a role in GLP-1/glucagon selectivity but do not provide detailed structural insights into the molecular mechanism that drives selective recognition of the N-terminal region of GLP-1 by the 7TM domain of GLP-1R. The objectives of this study were therefore to identify the structural determinants of GLP-1 binding in the 7TM domain of GLP-1R and to elucidate the molecular mechanism of selective recognition of GLP-1.

Experimental Procedures
Construction of GLP-1R and GCGR Point Mutants-To study the influence of specific residues on receptor function, the desired mutations were introduced to wild-type GCGR (11) and N-terminally FLAG tag-labeled wild-type human GLP-1R in the pcDNA3.1/V5-His-TOPO vector (Invitrogen); this GLP-1R receptor construct had equivalent pharmacology to the untagged human GLP-1R based on radioligand binding and cAMP assays. The single mutants and E364 6.53b N/E387 7.42b Q double mutant were constructed by PCR-based site-directed mutagenesis. Sequences of receptor clones were confirmed by sequencing with the ABI 3730xl DNA analyzer (Applied Biosystems, Foster City, CA).
Cell Culture and Transfection-CHO-K1 or HEK-293T cells were seeded onto 96-well poly-D-lysine or fibronectin-treated cell culture plates (PerkinElmer Life Sciences) at a density of 2.7 ϫ 10 4 per well. After overnight culture, the cells were transiently transfected with wild-type or mutant GLP-1R or GCGR DNA using Lipofectamine 2000 transfection reagent (Invitrogen).
cAMP Assay-cAMP accumulation was measured using HTRF-cAMP dynamic kit (Cisbio International, Gif sur Yvette Cedex, France) according to the manufacturer's instructions. Briefly, HEK 293T cells were transfected with plasmids bearing wild-type and mutant GLP-1R. Twenty four h post-transfection, cells were collected and used to seed white poly-D-lysinecoated 384-well plates at a density of 8000 cells per well. Cells were incubated for a further 24 h at 37°C. They were then incubated for 30 min in assay buffer (DMEM, 1 mM 3-isobutyl-1-methylxanthine) with different concentrations of peptides at 37°C. The reactions were stopped by addition of lysis buffer containing HTRF reagents. Plates were incubated for 60 min at room temperature, and time-resolved FRET signals were measured after excitation at 620 and 650 nm by an EnVision multilabel plate reader (PerkinElmer Life Sciences).
Quantification of Cellular GLP-1R Construct Expression Levels by Flow Cytometry-Approximately 1 ϫ 10 5 transfected cells were blocked with PBS containing 5% BSA at room temperature for 15 min and incubated with 1:1000 diluted primary antibody (anti-FLAG, Invitrogen) at room temperature for 1 h. The cells were then washed three times with PBS containing 1% BSA followed by 1-h incubation with anti-rabbit Alexa-488conjugated secondary antibody (1:300, Invitrogen) at 4°C in the dark. After the washes, the cells were resuspended in 200 l of PBS containing 1% BSA for detection in a flow cytometer (Accuri TM C6, BD Biosciences) utilizing laser excitation and emission wavelengths of 488 and 519 nm, respectively. For each assay point, ϳ20,000 cellular events were collected, and the total fluorescence intensity of positive expression cell population was calculated.
Residue Numbering-Peptide ligand residue numbers are annotated as three-letter amino acid residues with residue number as superscript (e.g. His 7 , histidine at position 7), and the receptor residue numbers are annotated as single letter amino acid, Uniprot number, and Wootten numbering for class B GPCRs as superscript (19), according to IUPHAR guidelines (34) and GPCR residue numbering guidelines (35), respectively. GLP-1 and glucagon peptide ligands start with amino acid residue 7 (His 7 ) due to post-translational processing.
Construction of a Full-length GLP-1R Model-A full-length GLP-1R model was constructed based on an experimentally validated (11,22) full-length GCGR⅐glucagon complex model combining crystal structures of the GCGR 7TM domain (PDB ID 4L6R) (11), the antibody-bound GCGR ECD (PDB ID 4ERS) (21), and the GLP-1-bound GLP-1R ECD (PDB ID 3IOL) (7). The GLP-1-bound full-length GLP-1R model satisfies spatial constraints defined by GLP-1R cross-linking studies (10,23,24) connecting the following: (i) Ala 24  (vii) Glu 9 Gln mutant GLP-1 bound to R190 2.60b K mutant; and (viii) Q234 3.37b E mutant GLP-1R. To set up the simulation systems, the eight starting structures were embedded separately in a 90 ϫ 90 Å 2 palmitoyloleoylphosphatidylcholine bilayer generated with VMD (Version 1.9.2) (36). In each system, lipids located within 1 Å of the complex models were removed. Subsequently, each system was solvated by TIP3P water molecules with 0.15 M NaCl and included ϳ114,160 atoms (90 ϫ 90 ϫ 140 Å 3 ). MD simulations were performed using the GROMACS 4.6.1 package (37) with isothermal-isobaric (NPT) ensemble and periodic boundary condition. The CHARMM36-CAMP force field (38) was applied. For each system, energy minimizations were first performed to relieve unfavorable contacts, followed by equilibration steps of 50 ns in total to equilibrate the lipid bilayer and the solvent, with restraints to the main chain of the protein and the peptide ligand. The temperature of each system was maintained at 310 K using the v-rescale method (39) with a coupling time of 0.1 ps. The pressure was kept at 1 bar using the Berendsen barostat (40) with p ϭ 1.0 ps and a compressibility of 4.5 ϫ 10 Ϫ5 bar Ϫ1 . SETTLE (41) constraints and LINCS (42) constraints were applied on the hydrogen-involved covalent bonds in water molecules and in other molecules, respectively, and the time step was set to 2 fs. The long range electrostatic interactions were computed using the Particle-Mesh Ewald algorithm (43) with a real space cutoff of 1.4 nm.
For each system, one 1000-ns production run was performed.
Analysis of Protein-Peptide Ligand Interactions-For each snapshot extracted at 100-ps intervals, we calculated the H-bond interactions formed by the main-chain atoms as well as the side-chain atoms of the first three N-terminal residues of peptides with specific receptor residues using the g_hbond program in the GROMACS 4.6.1 package (37), based on a distance cutoff of 3.5 Å and an H-bond interaction angle of 150°-210°. Ionic interactions were monitored using the g_mindist pro-gram, based on a 4-Å distance cutoff between two heavy atoms with opposite formal charge. Apolar contacts were also monitored using the g_mindist tool, based on a 5-Å minimum distance cutoff between carbon atoms.
Analysis of Protein Conformations-To describe the motions between the ECD and 7TM domain of different GLP-1R⅐ligand complexes, we constructed a Cartesian coordinate system in a similar way as described previously for full-length GCGR (22). As the whole ECD and residues preceding I147 1.42b in TM1 are very dynamic in most simulations, we took the C␣ atom of I147 1.42b in TM1 as the origin (designated as point O), the outward membrane normal as its z axis, the plane parallel to the membrane surface as the xy plane, and the plane defined by the z axis and the center of mass (COM) of the 7TM domain as the xz plane. Of all systems, the distance d between the COMs of the ECD and 7TM domain is monitored, as well as the polar angle and the azimuthal angle of vector OC (linking the origin and the COM of ECD) describing the swing and rotation motions of the ECD in the simulations.

Results
Comparative Structure-based Mutagenesis of GLP-1R and GCGR-We constructed a full-length GLP-1-bound GLP-1R model to guide the systematic GLP-1R mutagenesis studies probing the peptide-binding site. The structural model was

Effects of point mutations on human GLP-1 binding to human GLP-1R versus glucagon binding to GCGR
Mutants that show Ͻ4-fold change (blue), 4 -10-fold increase (orange), Ͼ10-fold increase (red), and Ͼ4-fold decrease (cyan) of IC 50 values for GLP-1 binding with GLP-1R or glucagon binding with GCGR are color-coded according to the snake plot in Fig. 1. For data generated in this study and a previous GCGR study (11), mutant receptors with expression Ͻ30% (colored gray) and span Ͻ10% (red) of the wild-type receptor, the IC 50 values are reported as not determined (ND) and no binding (NB), respectively. Overviews of the 66 human GLP-1R and 16 human GCGR mutation data determined in this study and the 101 human GLP-1R mutation data (76 unique, expressing mutants) and 92 human GCGR mutation data (63 unique, expressing mutants) reported in previous studies are presented in supplemental Table S1, and an overview of rat GLP-1R mutation studies is provided in supplemental Table S2.  (19). All other residues on that helix are numbered relative to this conserved position. Structure-based comparison of Wootten residue numbers for class B GPCRs to Ballesteros-Weinstein residue numbers for class A GPCRs are reviewed by Isberg et al. (35).  (11), the antibody-bound GCGR ECD (21), and the GLP-1-bound GLP-1R ECD (7). These two models are consistent with the results from mutation studies of GCGR, GLP-1R, and other class B GPCRs and photo-crosslinking studies connecting residues of GLP-1R and GLP-1, as described under "Experimental Procedures" and summarized in Table 1, Fig. 1, and supplemental Fig. S1 (5,11). Based on this model, a total of 66 mutants were created and tested for expression (complementing the 76 human GLP-1R mutants from supplemental Table S1) (7, 8, 19, 20, 25, 28 -30, 32). Of these mutations, 64 covering 40 different positions had expression levels greater than 30% of wild-type GLP-1R allowing further characterization. Of these expressing mutants, 40 covering 23 positions in GLP-1R had more than 4-fold reduction in GLP-1 binding (IC 50 values) relative to wild-type (Fig. 1). GLP-1R mutants generally had similar effects on GLP-1 binding affinity and agonist potency (supplemental Table S3 (Table 1). In contrast, 28 mutants (containing 21 unique residue positions) exerted differential effects on GLP-1R compared with the corresponding mutants in GCGR. For example, Y152 1.47b H, V194 2.64b A, M233 3.36b F, Q234 3.37b N, E364 6.53b Q, E364 6.53b Y, E387 7.42b N, and L388 7.43b F showed no significant effect on GLP-1 binding, whereas the corresponding mutants in GCGR all had a marked effect (in most cases negative) on glucagon binding (Table 1). In contrast,  (11)). Mutated residues that had Ͻ4-fold change (blue), 4 -10-fold increase (orange), and Ͼ10-fold increase (red) in IC 50 values for GLP-1 binding to GLP-1R or glucagon binding to GCGR are shown. Mutant receptors with expression Ͻ30% of wild-type are colored gray. For each position, the results of mutations that show the most distinct differences between GLP-1R and GCGR are reported (see Table 1). The effects of previously reported human GLP-1R mutants are presented in supplemental Table S1 (7, 8, 19, 20, 25, 28 -30, 32) and rat GLP-1R mutation data in supplemental Table S2 (26,27). The most conserved residues in TM helices 1-7 of class B GPCRs (19) Table 1 and supplemental Tables S1 and S3). Y148 1.43b F, M233 3.36b A, Q234 3.37b E, F367 6.56b A, and L388 7.43b I mutations in GLP-1R decreased GLP-1 binding, although the same mutations of the corresponding residues in GCGR had no significant impact on glucagon affinity. The GCGR-mimicking GLP-1R E387 7.42b D mutant had a 3-fold greater negative impact on GLP-1 binding than the inverse GCGR D385 7.42b E mutant on glucagon binding. The GCGR-mimicking GLP-1R mutants R190 2.60b K, M204 ECL1 R, Y220 ECL1 D, and R299 ECL2 S significantly reduced GLP-1 binding affinity (Table 1), whereas the reverse GLP-1R-mimicking GCGR mutants K187 2.60b R, R201 ECL1 M, D218 ECL1 Y, and S297 ECL2 R did not show any marked effect on glucagon binding.
Probing GLP-1R Mutants with Truncated Peptide Ligands-To assess whether the mutational effects in the 7TM-binding site resulted from interactions with the N terminus of peptide ligands, nine GLP-1R mutants were investigated in 125 Iexendin-(9 -39) displacement studies with six different fulllength and truncated peptides ligands (GLP-1, GLP-1-(7-36), GLP-1-(9 -36), GLP-1- (15-36), exendin-4, and exendin-(9 -39)) ( Table 2). The GLP-1R mutant set included three GCGRmimicking GLP-1R mutants (R190 2.60b K, K197 2.67b I, and E387 7.42b D), four mutants that abolished GLP-1 binding, although the corresponding/reverse GCGR mutants did not affect glucagon binding (Y148 1.43b F, M233 3.36b A, Q234 3.37b E, and F367 6.56b A), two mutants that diminished GLP-1 binding in a similar way as corresponding GCGR mutants that affected glucagon binding (D198 2.68b A and L384 7.39b A), and one mutant that increased GLP-1 binding, although the corresponding GCGR mutant decreased glucagon binding (E364 6.53b Q) ( Table  1 and Fig. 2). With the exception of the F367 6.56b A and E364 6.53b Q, all of the other seven mutants led to reduced binding of GLP-1 and GLP-1-(7-36), whereas binding of all the truncated peptide ligands GLP-1-(9 -36), GLP-1-(15-36), exendin-(9 -39) was not affected compared with wild-type GLP-1R (Table 2). These results demonstrate that the effects of these mutations on GLP-1R binding result from interactions between these residues located in the 7TM domain of the receptor and the N-terminal region of the ligand. GLP-1-(15-36) lacks the seven first residues of GLP-1 and GLP-1-  that are proposed to target the 7TM-binding site in our peptide ligand-bound GLP-1R structural models (Fig. 3A). GLP-1-(9 -36) lacks the first two residues that are required to form a tight interaction with the 7TM domain of GLP-1R, reflected by the decreased affinity of this truncated analogue compared with GLP-1 ( Table 2). Most of the mutants affected both GLP-1 and exendin-4 binding ( Table 2), suggesting that the N-terminal regions of GLP-1 and exendin-4 share similar binding modes in general. The M233 3.36b A, Q234 3.37b , and E387 7.42b D mutants, however, only decreased GLP-1 binding but did not significantly affect exendin-4 affinity ( Table 1 and Fig. 3), indicating that interactions and compatibility with these three residues are more important for GLP-1 binding. These observations are in line with previous NMR, x-ray crystallography, and truncated and chimeric GLP-1R and peptide ligand studies showing that GLP-1 binding is largely determined by interactions with the ECD, but also require interactions with the 7TM domain of GLP-1R, whereas exendin-4 binding affinity is mainly deter-

TABLE 2 The effects of GLP-1R mutations on full-length and truncated GLP-1 and exendin-4 ligand binding
The IC 50 values of GLP-1R mutants that show Ͻ4-fold change (blue), 4 -10-fold increase (orange), Ͼ10-fold increase (red), and Ͼ4-fold decrease (cyan) compared with the IC 50 values for wild-type GLP-1R are color-coded according to the snake plot in Fig. 1. Radioligand displacement curves of representative mutants are shown in Fig. 3. Expression and span levels are reported in supplemental Table S3 and  supplemental Table S4,  mined by interactions with the ECD of GLP-1R and is less dependent on interactions with the 7TM domain than GLP-1 (4, 6, 7, 9, 32, 33). The E364 6.53b Q mutant, located deep in the transmembrane bundle, has increased affinity for both fulllength and truncated GLP-1 and exendin-4 ( Table 2), implying that this mutation stabilizes GLP-1R in a conformation with high affinity for peptide ligands.

Comparison of GLP-1R-GLP-1 and GCGR-Glucagon Binding Modes-
The GLP-1-bound GLP-1R model was subjected to MD simulations to evaluate it and to further investigate the molecular details of GLP-1R-ligand interactions aimed at understanding the observed differential mutational effects between GLP-1R and GCGR. GLP-1R residues shown to be important in mutation studies are involved in GLP-1 interactions in the MD simulations (Table 1 and Figs. 1, 2, and 4). MD simulations of wild-type GLP-1R and the high affinity E364 6.53b Q mutant of GLP-1R displayed frequent ionic interactions between the terminal amino group of GLP-1 and E387 7.42b in combination with tight (E364 6.53 Q) and transient (WT) ionic H-bond interactions between K197 2.67b and Glu 9 , consistent with the negative effect of the GCGR-mimicking K197 2.67b I mutant (Table 1, Fig. 4, and supplemental Fig. S2). The simultaneous ionic interactions of GLP-1 with K197 2.67b and E387 7.42b offer an explanation for the positive effect of the E364 6.53 Q mutant on GLP-1 binding ( Table 2). The position of the flexible GLP-1R-specific K197 2.67b residue is stabilized by an H-bond interaction network with D198 2.68b and Y148 1.47b , which may explain the relatively negative effects of D198 2.68b A, D198 2.68b N, and Y148 1.47b F mutants on ligand binding compared with that of corresponding mutants in GCGR (Table 1 and Fig. 4). When Glu 9 forms an ionic H-bond interaction with K197 2.67b , the amine group of His 7 forms ionic H-bond interactions with E364 6.53b exclusively (Fig. 4D), but when GLP-1 adopts a more constrained binding conformation stabilized by intra-molecular ionic interactions between His 7 and Glu 9 , the N terminus of GLP-1 forms ionic H-bond interactions with E387 7.42b exclusively (Fig. 4E). The model is consistent with the observation that mutations of single E364 6.53b N and E387 7.42b Q mutants do not affect GLP-1 binding affinity or signaling potency, although the double E364 6.53b N/E387 7.42b Q mutant completely abolishes GLP-1 binding (span Ͻ10%). The corresponding D385 7.42b N and E362 6.53b Q point mutations of GCGR diminished glucagon binding ( Table 1), suggesting that both negatively charged residues are required for optimally accommodating the N-terminal His 7 residue of glucagon. Analysis of MD simulations of glucagon-bound GCGR, consistent with previously reported hydrogen/deuterium exchange data (15), showed that the N-terminal His 7 residue of glucagon forms a stable ionic H-bond interaction with D385 7.42b and a transient H-bond or ionic interaction with E362 6.53b (Fig. 4,  A-C). In GLP-1-bound GLP-1R, the R190 2.60b residue forms a tight intra-helical H-bond with E364 6.53b at the bottom of the  Table 2. Data are expressed as a percentage of specific binding in the presence of 3.57 pM unlabeled peptide. Each point (ϮS.E.) represents the mean value of at least three independent experiments done in triplicate (IC 50 data presented in Table 2 and supplemental Table S4).
7TM-binding site, whereas in glucagon-bound GCGR the ionic interaction between K187 2.60b and E364 6.53b is more transient (Fig. 4, D-F). This difference between GLP-1R and GCGR may explain why the R190 2.60b K mutant diminishes GLP-1 binding to GLP-1R, while the inverse K187 2.60b R does not affect glucagon binding to GCGR (Table 1). In addition to this tight ionic interaction network deep in the transmembrane helical bundle, our MD simulations support an earlier proposed (29) ionic interaction between Asp 15 and R380 7.35b (Fig. 4) and predict a previously untested electrostatic interaction between Glu 21 and R299 ECL2 . This predicted interaction is supported by the mutation of these residues to Gln and Ser (mimicking GCGR). The GCGR mutation data and glucagon-bound GCGR model suggest that R378 7.35b plays a similar role in binding Asp 15 of glucagon and explain why the reverse S299 ECL2 R mutant in GCGR did not affect glucagon affinity, as the homologous glucagon Asp 21 formed an intra-helical H-bond with Arg 24 (Fig. 4, A-C, and supplemental Fig. S1). In contrast, the previously reported GCGR model (11,22) suggested that D208 ECL1 forms an ionic interaction with the glucagon-specific Arg 23 residue, explaining the negative effect of the GCGR D208 ECL1 Q mutant on glucagon binding as well as the limited effect of the reverse GLP-1R Q211 ECL1 D mutant on GLP-1 binding. Comparison of the GCGR and GLP-1R models indicates that the N-terminal His 7 of GLP-1 interacts with M231 3.36b in GLP-1R, whereas the His 7 residue of glucagon interacts with Y149 1.47b , V194 2.64b , and L388 7.43b in the GCGR-binding site, consistent with the differential effects of GLP-1R and GCGR mutants of the residues at these positions (Table 1 and Figs. 2 and 4).
Combined GLP-1R and GLP-1 Mutation Studies-Based on the comparative GLP-1R, GCGR MD simulation and mutation studies, it was hypothesized that glucagon-mimicking GLP-1 mutants Ala 8 Ser and Glu 9 Gln would be able to restore GLP-1 binding to GCGR-mimicking GLP-1R point mutants E387 7.42b D and R190 2.60b K. It was furthermore hypothesized that the glucagon-mimicking Glu 9 Gln mutant of GLP-1 would restore binding to the Q234 3.37b E GLP-1R mutant, based on the differential effect of this mutant (diminished GLP-1 binding) compared with the corresponding Q232 3.37b E GCGR mutant (no effect on glucagon binding) (Figs. 1 and 2 and supplemental Table S1). These predictions were indeed confirmed by our mutation experiments (Table 3). Systematic microsecond MD simulation studies of different wild-type/mutant GLP-1R⅐GLP-1 complexes provided insight into subtle GLP-1R-specific recognition mechanisms of the N terminus of GLP-1 by GLP-1R-specific hot spots in the 7TM domain described below (Fig. 5 and supplemental Fig. S3). The H-bond/ionic interaction between E387 7.42b and the N-terminal amine group of His 7 is disrupted in the GCGR-mimicking E387 7.42b D mutant, resulting in expulsion of the N terminus of GLP-1 from the 7TM binding pocket of GLP-1R in MD simulations and diminished GLP-1 binding in radioligand displacement studies (Fig. 5, A-C, and supplemental Fig. S3). MD simulations suggest that this H-bond interaction network is re-established by the glucagon-mimicking Ala 8 Ser GLP-1 mutant via formation of an additional stabilizing H-bond with the carboxylate group of D387 7.42b with the hydroxyl group of the Ser 8 side chain, leading to restored binding affinity equal to the wild-type GLP-1⅐GLP-1R complex (Fig. 5, A-C, and supplemental Fig. S3). The ionic lock between R190 2.60b and E364 6.53b is broken in the R190 2.60b K mutant, resulting in an increased distance between TM2 and TM7 and destabilization of the ionic interaction network between the N terminus of GLP-1 and the GLP-1R-binding site (Fig. 5, D-F, and supplemental Fig. S3). The GLP-1 Glu 9 Gln mutant lacks strong ionic intra-molecular interaction between Glu 9 and the N-terminal amino group of His 7 and can therefore adopt an extended conformation that allows the N terminus of GLP-1 to form ionic interactions with E364 6.53b and E387 7.42b deep in the GLP-1R binding pocket. In the MD simulations, this binding mode stabilizes an ionic interaction network between K197 2.67b , E364 6.53b , and His 7 , whereas Gln 9 forms a transient H-bond with K197 2.67b , restoring the receptor's binding affinity (Fig. 5, D-F, and supplemental Fig. S3). The Q234 3.37b E mutation causes electrostatic repulsion with Glu 9 leading to a binding orientation of the N terminus of GLP-1 that does not allow the formation of intermolecular ionic interactions with E387 7.42b (Fig. 5G). The MD simulations indicate the neutral Gln 9 residue in the GLP-1 Glu 9 Gln mutant is compatible with the negatively charged E234 3.37b residue in the GLP-1R Q234 3.37b E mutant and allows the N terminus of GLP-1 to penetrate deeper into the 7TM binding pocket of GLP-1R and interact with E387 7.42b , resulting in a restored receptor binding affinity (Figs. 4 and 5, G-I, and supplemental Fig. S3).

Discussion
Molecular Mechanism of Class B GPCR Peptide Ligand Selectivity in the 7TM Domain-By combining systematic receptor and peptide ligand site-directed mutagenesis studies with extensive MD simulations, we have identified conserved and receptor-specific ligand interaction hot spots in the 7TM domains of GLP-1R and GCGR. Mutations of homologous residues in the 7TM domain of other class B GPCRs have been shown to affect peptide ligand binding as well (5), including human gastric inhibitory polypeptide or glucose-dependent insulinotropic peptide (GIP) (R183 2.60b , R190 2.67b , Q224 3.36b , R300 5.40b , and F377 6.53b ) (44,45), human vasoactive intestinal polypeptide (VIP) (1R188 2.60b , K195 2.67b , and L375 7.43b ) (46,47), and rat secretin (Y124 1.43b , Y128 1.47b , R176 2.60b , K173 2.67b , D174 2.68b , F201 3.36b , W274 5.36b , F337 6.56b , and L353 7.43b ) (48 -50) receptors. Our comparative GLP-1R and GCGR studies show differential effects for almost half of the 62 stably expressed mutants of different residue positions in the 7TM domain, demonstrating that binding of homologous peptide ligands of similar size by homologous class B GPCRs is determined by receptor-specific structural features in the 7TM domain (Table 1 and Figs. 1 and 2). The negative effects of GLP-1R mutants located deep in the 7TM pocket were not observed for truncated peptide ligands of GLP-1R, indicating that these differential mutation effects result from GLP-1R specific interactions between the 7TM domain and the N-terminal region of the ligand ( Table 2 and Fig. 3).
The MD simulations of wild-type and mutant GLP-1R⅐ligand complexes provide molecular insights into GLP-1R-specific recognition mechanisms for the N terminus of GLP-1 by residues in the 7TM pocket, and they explain how glucagon-mimicking GLP-1 mutants restore binding affinity for (GCGR mimicking) GLP-1R mutants (Figs. 4 and 5). The combined mutation and molecular dynamics investigations suggest that the negatively charged Glu 9 residue of GLP-1 does not allow the N terminus of GLP-1 to adopt to changes in the 7TM-binding site as follows: (i) by attractive and repulsive electrostatic interactions with the positively charged (K197 2.67b ) and negatively charged (Q234 3.37b E mutant, E387 7.42b ) residues in the receptor; and (ii) by constraining the ligand conformation via intramolecular ionic interactions with the peptide N-terminal His 7 (Figs. 4 and 5). Although the intra-molecular H-bond-stabilized, bend conformation of residues 8 -11 in GLP-1 in the MD simulations (Fig. 4B) is different from the disulfide cross-linkstabilized ␤-turn conformation of homologous residues 2-5 of an 11-mer GLP-1 analogue observed in NMR structures (PDB ID 2N09) (18), both studies suggest that GLP-1 may adopt a constrained conformation in the 7TM-binding site of GLP-1R. The neutral Gln 9 residue in glucagon allows the peptide ligand to adopt a more extended binding mode deeper in the 7TM binding pocket of GCGR in which the N-terminal amine of glucagon (His 7 ) forms an ionic interaction with E 6.53b , observed in our GCGR-glucagon MD simulation runs and mimicked by the binding modes of GLP-1 Glu 9 Gln mutants (Figs. 4, D-F, and 5). Comparison of the GLP-1R and GCGR models explains why mutation of E 6.53b into neutral (Asn, Gln) residues has a negative effect on GCGR-glucagon binding and not on GLP-1R-GLP-1 binding ( Fig. 1 and supplemental Table S1). Comparison of MD simulations of complexes of GLP-1 with GLP-1R (Fig. 4, A-C), glucagon with GCGR (Fig. 4, D-F), and the glucagon-and GCGR-mimicking GLP-1 (Glu 9 Gln) with GLP-1R (R190 2.60b K) (Fig. 5E) indicates that the extended, deeper peptide ligand binding mode of glucagon/GLP-1 (Glu 9 Gln) is facilitated by the weaker ionic interaction network between K 2.60b and E 6.53b in GCGR and the GLP-1R R190 2.60b K mutant compared with the tighter ionic lock between R190 2.60b and E364 6.53b in wild-type GLP-1R. These observations are in

and GLP-1R mutation studies
The IC 50 values of GLP-1R mutants that show Ͻ4-fold change (blue), 4 -10-fold increase (orange), Ͼ10-fold increase (red), and Ͼ4-fold decrease (cyan) compared with IC 50 values for wild-type GLP-1R are color-coded according to the snake plot in Fig. 1. Radioligand displacement curves of representative mutants are shown in Fig. 5. Expression and span levels are reported in supplemental Table S3 and supplemental Table S5,  line with the results of recent mutagenesis and modeling studies, including alanine mutants R190 2.60b A and E364 6.53b A, which suggested that an ionic lock between R190 2.60b and E364 6.53b plays a role in GLP-1R ligand-biased signaling (20), emphasizing the important role of this polar H-bond network in both functional activity and ligand recognition by GLP-1R.  Table 3 and supplemental Table S5).
GIP receptor shares R 2.60b , K 2.67b , and E 6.53b residues with GLP-1R, suggesting that the structurally aligned Glu 3 residue of GIP may play a similar role in determining the receptor-ligand binding mode as Glu 9 in GLP-1. The GLP-1R binding mode model may also serve as a useful template for other class B GPCRs such as vasoactive intestinal peptide receptor type 1 (VPAC 1 ), pituitary adenylate cyclase-activating polypeptide (PACAP) receptor (PAC 1 ), and secretin receptor that all contain R 2.60b , K 2.67b /R 2.67b , and E 7.42b residues (but no E 6.53b ) and that bind natural peptide ligands with conserved Ser and Asp residues structurally aligned to Ser 8 and Glu 9 of GLP-1. Previously reported combined receptor and ligand mutation studies have indeed suggested that homologous residues Gln 9 of glucagon, Asp 3 of secretin, and Asp 3 of VIP may be located within the same vicinity in the 7TM domain of GCGR (K187 2.60b and I194 2.67b ) (51,52), rat secretin receptor (Y128 1.47b , R166 2.60b , Lys173 2.67b , and D174 2.68b ) (48,49), and VPAC 1 (R188 2.60b and K195 2.67b ) (46), respectively, consistent with our mutation studies and structural model.
The observation that the glucagon-mimicking Ala 8 Ser mutant of GLP-1 restores binding of the GCGR-mimicking E387 7.42b D mutant of GLP-1R is consistent with previous mutation studies showing that the affinity of the reciprocal Ser 8 Ala mutant of glucagon restores binding of the GLP-1R- mimicking D385 7.42b E mutant of GCGR (52). Our MD simulations indicate that Ala 8 is compatible with the longer E 7.42b side chain, whereas Ser 8 stabilizes the ionic interaction with D 7.42b by forming an additional H-bond with the shorter D 7.42b side chain (Figs. 4 and 5). These proposed subtle steric requirements for the formation of (alternative) H-bond interaction networks between the 7TM domain of class B GPCRs and the N-terminal region of their peptide ligands motivate detailed combined investigation of both receptor and ligand structure-activity and structure-selectivity relationships.
Motions between the ECD and 7TM Domain of GLP-1R Accommodate Different GLP-1 Binding Modes-In our recent hybrid structural biology studies combining x-ray crystallography, electron microscopy, hydrogen/deuterium exchange, and cross-linking studies, we demonstrated how peptide ligand binding determines the relative orientation of the ECD and 7TM domains of the homologous GCGR (22). In the current study, we investigated how the relative motions between the ECD and 7TM domain of wild-type and mutant GLP-1R can accommodate different binding modes of GLP-1 and GLP-1 mutants (Fig. 6). In 1-s MD simulations, GLP-1-bound wildtype GLP-1R adopts a similar open conformational state as glucagon-bound GCGR, as reported previously (22), with similar distances and swing and rotation motions of the ECD relative to the 7TM domain (supplemental Fig. S4). The relatively smaller swing motion (2-fold smaller polar angle ) and larger rotational motion (2-fold higher Azimuthal angle ) of the ECD of the E387 7.42b D GLP-1R mutant allows the GLP-1 Ala 8 Ser mutant to bind closer to TM7 to form a tight H-bond network with the mutated D387 7.42b residue (Figs. 5A and 6A). The larger swing motion of the ECD of the R190 2.67b K GLP-1R mutant (supplemental Fig. S4) enables the GLP-1 Glu 9 Gln mutant to position its helical region closer to the helix to allow its N-terminal region to bind in an extended conformation in the 7TM pocket and form an ionic interaction network with E364 6.53b and E387 7.42b simultaneously (Figs. 5B and 6B). The relatively smaller swing motion and larger rotational motion of the ECD of the Q234 3.37b E GLP-1R mutant allows the GLP-1 Glu 9 Gln mutant to interact and form an H-bond interaction network with the mutated E234 3.37b residue and simultaneously interact with E364 6.53b and E387 7.42b (Figs. 5A and 6A).
The comparative mutation and modeling studies suggest that binding of GLP-1 to GLP-1R is controlled by a similar conformational selection mechanism as proposed previously for the binding of glucagon to GCGR (22). Although chimeric receptor and peptide/protein ligand studies have indicated that the first binding step with the ECD provides the largest contribution to ligand binding affinity (3,9), previous (5) and current site-directed mutagenesis studies demonstrate that structural and electrostatic compatibility with the 7TM domain plays an equally important role in the molecular recognition of peptide ligands by GPCRs. Integration of the recent hybrid structural biology studies of the conformational states of full-length class B GPCRs (11,22) with previous (3,5) and current structurebased molecular pharmacology investigations suggests that a combination of optimal interaction networks between the peptide ligand and (i) ECD and (ii) 7TM domains and (iii) relative flexibility of the ECD and 7TM domains and (iv) the peptide ligand determines GPCR-peptide ligand selectivity. The importance of these different factors is likely to be receptor⅐ ligand complex-specific and may for example explain pharmacology differences observed between GLP-1 peptide variants. The differential point mutation effects on GLP-1 compared with exendin-4 in the 7TM-binding site, for example ( Fig. 2 and Table 2), are in line with previous truncated and chimeric GLP-1R and peptide ligand studies and NMR and crystal structures showing that exendin-4 has increased binding affinity for the ECD of GLP-1R because it has a more structured ␣-helix than GLP-1 and forms a tighter ionic interaction network with the ECD of GLP-1R (7,14,33). Several N-terminally truncated forms of GLP-1 (53,54), glucagon (55), GIP (56), parathyroid hormone (PTH) (57), and corticotrophin-releasing hormone (CRH) (58) peptides are competitive antagonists for their corresponding receptor, whereas several C-terminally truncated ligands remain active (albeit with lower binding affinity), indicating that interactions with peptide ligands in the 7TM domain of class B GPCRs are required for receptor activation. Differences in the binding modes of the N termini of GLP-1 variants may therefore be important determinants of their functional activity at GLP-1R and control ligand-biased signaling (20).
In conclusion, our combined mutation and MD simulation studies suggest that diminished binding affinity of GLP-1R mutants can be rescued by subtle changes in the peptide ligand, facilitated by a variety of structural mechanisms, determined by ligand peptide conformation, electrostatic side-chain interactions/compatibility, as well as movements of the ECD relative to the 7TM domain. This demonstrates the complexity of peptide ligand recognition by class B GPCRs, and illustrates how full-length receptor models complemented by extensive mutation and simulations of the conformational dynamics of receptor-ligand complexes can be used to investigate class B GPCR structure-activity and structure-selectivity relationships.