Key interactions by conserved polar amino acids located at the transmembrane helical boundaries in Class B GPCRs modulate activation, effector specificity and biased signalling in the glucagon-like peptide-1 receptor

Class B GPCRs can activate multiple signalling effectors with the potential to exhibit biased agonism in response to ligand stimulation. Previously, we highlighted key TM domain polar amino acids that were crucial for the function of the GLP-1 receptor, a key therapeutic target for diabetes and obesity. Using a combination of mutagenesis, pharmacological characterisation, mathematical and computational molecular modelling, this study identifies additional highly conserved polar residues located towards the TM helical boundaries of Class B GPCRs that are important for GLP-1 receptor stability and/or controlling signalling specificity and biased agonism. This includes (i) three positively charged residues (R3.30, K4.64, R5.40) located at the extracellular boundaries of TMs 3, 4 and 5 that are predicted in molecular models to stabilise extracellular loop 2, a crucial domain for ligand affinity and receptor activation; (ii) a predicted hydrogen bond network between residues located in TMs 2 (R2.46), 6 (R6.37) and 7 (N7.61 and E7.63) at the cytoplasmic face of the receptor that is important for stabilising the inactive receptor and directing signalling specificity, (iii) residues at the bottom of TM 5 (R5.56) and TM6 (K6.35 and K6.40) that are crucial for receptor activation and downstream signalling; (iv) residues predicted to be involved in stabilisation of TM4 (N2.52 and Y3.52) that also influence cell signalling. Collectively, this work expands our understanding of peptide-mediated signalling by the GLP-1


Introduction
GPCRs mediate signal transduction across cell membranes in response to a wide range of extracellular stimuli [43]. Understand-ing how these receptors function at the molecular level requires knowledge of how agonist binding is converted to receptor activation and consequently stimulation of downstream signalling cascades that can be both G protein-dependent and G proteinindependent [37]. GPCRs are dynamic proteins that can explore multiple conformational states and with the advances in GPCR structural biology, new insights into the structural basis of GPCR activation have revealed the importance of inter-connected networks of residues for conformational transitions that allow agonist bound receptors to activate intracellular signalling cascades [29,40].
Sequence alignments of related membrane proteins suggest that polar residues are under evolutionary pressure for conservation and hence maintain common structural and functional roles [25,26]. In support of this, there are a number of highly conserved http polar residues present in Class A GPCRs that participate in key interactions associated with their activation [4,5,42]. Class B GPCRs typically contain more conserved polar residues in their transmembrane (TM) bundle than Class A GPCRs, which may be reflective of the diversity of receptors/ligands found within the Class A subfamily, however, it may also reflect the mode by which Class B ligands bind and activate their receptors. Peptide ligands associate primarily with the large extracellular N-terminal domain of Class B GPCRs, but also need to interact with the TM bundle to promote receptor activation [6,47,48,41]. Previously, we revealed the importance of networks of conserved polar residues located in the TM bundle of Class B GPCRs for controlling receptor activation and downstream signalling of the glucagonlike peptide-1 receptor (GLP-1R) [64,66,68]. This receptor plays an essential role in nutrient regulated insulin release, and has emerged as a major target for therapeutic treatment of type 2 diabetes and obesity. The GLP-1R is pleiotropically coupled to multiple signalling pathways with evidence for biased agonism by the physiological ligand oxyntomodulin, clinically used peptide mimetics and synthetic non-peptide ligands, relative to the cognate agonist GLP-1 [33,65,67]. In our previous studies, we identified conserved buried polar residues were not only important in receptor activation, but that some of these residues were also important for biased agonism at this receptor. The breakthrough crystal structures of the inactive TM domain of two Class B GPCRs (the glucagon receptor (GCGR) and the corticotrophin releasing factor receptor-1 (CRF1R)) that were subsequently published, largely supported the predictions and conclusions from the molecular modelling in these studies, highlighting that these conserved residues may form conserved hydrogen bond networks that are important for activation transition of all members of this class of GPCRs [22,49].
The high resolution TM domain structures have provided better structural templates for Class B GPCR modelling and enabled the generation of a homology model of the inactive state of the GLP-1R TM bundle [64,68]. In addition to the hydrogen bond networks predicted in our previous model, the new model identified another network of residues in the inactive GLP-1R. This was formed between conserved Class B polar residues located within TMs 2, 6 and 7 at the intracellular face of the receptor and was also evident in the crystal structures of the GCGR and the CRF1R [22,49]. In addition to participation in hydrogen bond networks, polar side chains located within the TM bundle of GPCRs can have other important functions. These include the formation of interactions with ligands or effectors and their ability to snorkel out towards phospholipid head groups, thereby stabilising TM helices within the bilayer [51]. These functions of polar TM residues are often (although not always) limited to residues that reside either towards the extracellular or intracellular TM boundaries. While our earlier studies on the GLP-1R focused on conserved polar residues that our original model predicted to reside in water-mediated hydrogen bond interaction networks, or in the central region of the TM bundle forming helical packing interactions, this current study explores the roles of the remaining conserved Class B GPCR TM polar residues, which are predicted to reside close to the TM boundaries (Fig. 1). This set of residues includes the amino acids located within the additional hydrogen bond network at the intracellular face of Class B GPCRs. We have assessed the role of these residues on GLP-1R function using a combination of mutagenesis, molecular modelling and pharmacological characterisation of multiple ligands for affinity and activation of three signalling endpoints. This identified residues important for ligand affinity, receptor folding and those contributing to biased agonism, expanding the current understanding of the functional role of highly conserved polar residues within Class B GPCRs.

Residue numbering
Throughout, residues were numbered using the numbering system described previously [66], whereby the most conserved residue in each Class B GPCR TM domain was assigned 0.50 with this number preceded by the TM number. Each residue is numbered according to its relative position to the residue at 0.50 in each helix and its absolute residue number is shown in superscript. The relative positions of the residues assessed in this study are shown in Fig. 1B-D.

Receptor mutagenesis
To study the influence of polar TM amino acids on receptor function, the desired mutations were introduced to an Nterminally double c-myc labelled wildtype human GLP-1R in the pEF5/FRT/V5-DEST destination vector (Invitrogen); this receptor had equivalent pharmacology to the untagged human GLP-1R. Mutagenesis was carried out using oligonucleotides for sitedirected mutagenesis purchased from GeneWorks (Hindmarsh, SA, Australia) and the QuikChange TM site-directed mutagenesis kit (Stratagene). Sequences of receptor clones were confirmed by automated sequencing at the Australian Genome Research Facility. Mutated residues and their conservation across human Class B peptide hormone receptors are illustrated in Fig. 1.

Transfections and cell culture
Wildtype and mutant human GLP-1R were isogenically integrated into FlpIn-Chinese hamster ovary (FlpInCHO) cells (Invitrogen) and selection of receptor-expressing cells was achieved through treatment with 600 lg ml 1 hygromycin-B. Transfected and parental FlpInCHO cells were maintained in DMEM supplemented with 10% heat-inactivated FBS and incubated in a humidified environment at 37 C in 5% CO 2 . For all experiments cells passages 8-20 were used.

Radioligand binding assay
FlpInCHO wildtype and mutant human GLP-1R cells were seeded at a density of 3 10 4 cells/well into 96-well culture plates and incubated overnight at 37 C in 5% CO 2 , and radioligand binding carried out as previously described [32]. Briefly, binding assays were performed on whole cells incubated overnight at 4 C with 0.05 nM 125 I-exendin-4(9-39) tracer and increasing concentrations Conservation and location of polar residues mutated in this study. (A) Conservation of polar residues mutated in this study across the human Class B GPCRs (the secretin-like subclass). Residues absolutely conserved are highlighted in grey. These residues shown are conserved as polar (with the exception of 5.56 and 6.35 where one receptor subtype is not) across all mammalian species of receptor cloned to date. GLP-1R; glucagon-like peptide-1 receptor, GLP-2R; GLP-2 receptor, GIP, gastric inhibitory polypeptide receptor; GluR, glucagon receptor; PTH-1R, parathyroid hormone receptor 1; PTH-2R, PTH receptor 2; SecR, secretin receptor; CTR, calcitonin receptor; CLR, calcitonin-like receptor; CRF1, corticotropin-releasing factor receptor 1; CRF2, corticotropin-releasing factor receptor 2; GHRHR, GH-releasing hormone receptor; VPAC1R, vasoactive intestinal polypeptide type-1 receptor; VPAC2R, vasoactive intestinal polypeptide type-2 receptor, PACR, pituitary adenylate cyclase activating polypeptide 1 receptor. (B) Schematic representation of the TM domain of the human GLP-1R. The most conserved residue in each helix is highlighted as a square with a bold letter and represent residue 0.50 for that helix. Residues mutated in the present study are shown in grey. (C) Three-dimensional molecular homology model of the inactive TM bundle of the GLP-1R. (D) Three-dimensional molecular model of the TM bundle of the active full length model of the GLP-1R. The bound GLP-1 peptide is shown dipping into the bundle (dark red helix) and the G as peptide fragment bound at the intracellular face is shown in dark blue. In (C) and (D), side chains mutated in this study are highlighted in space fill with dark green indicating positively charged residues located towards the extracellular face of the bundle and interact with ECL2; pale green, positively charged residues located towards the intracellular face that may interact with lipid headgroups; red, residues in TMs 2, 6 and 7 that form a hydrogen bond network in the apo receptor; purple, residues in TMs 2 and 3 that stabilise interactions with TM4. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) of unlabelled peptide. Cells were washed, solubilised in 0.1 M NaOH and radioactivity determined by c-counting. For each cell line in all experiments, total binding was defined by 0.05 nM 125 I-exendin-4(9-39) alone, and nonspecific binding was defined by co-incubation with 1 lM exendin-4 . For analysis, data are normalised to the specific binding for each individual experiment.

cAMP accumulation assay
FlpInCHO wildtype and mutant human GLP-1R cells were seeded at a density of 3 10 4 cells/well into 96-well culture plates and incubated overnight at 37 C in 5% CO 2 . cAMP assays were carried out as previously described [33]. Briefly, cells were incubated with increasing concentrations of peptide ligands for 30 min at 37 C in the presence of IBMX. Cells were lysed and cAMP levels were detected using a cAMP AlphaScreen TM detection kit (PerkinElmer). All values were converted to concentration of cAMP using a cAMP standard curve performed in parallel, and data were subsequently normalised to the response of 100 lM forskolin in each cell line.

pERK1/2 assay
FlpInCHO wildtype and mutant human GLP-1R cells were seeded at a density of 3 10 4 cells/well into 96-well culture plates and incubated overnight at 37 C in 5% CO 2 . Ligand-mediated pERK1/2 was determined using the AlphaScreen TM ERK1/2 SureFire TM protocol as previously described [39]. Briefly, cells were serum starved for 6 h prior to assay. Initial pERK1/2 time course experiments were performed over 1 h in the presence of either vehicle or 1 lM peptide to determine the time at which agonist-mediated pERK1/2 was maximal. pERK1/2 was detected using the AlphaScreen TM ERK1/2 SureFire TM kit. Subsequent experiments were then performed with increasing concentrations of peptides at the time required to generate a maximal pERK1/2 response using 1 lM peptide. The kinetics of pERK1/2 response for each mutant receptor was similar to WT, peaking at 6 min. Data were normalised to the maximal response elicited by 10% FBS in each cell line, determined at 6 min (peak FBS response).

i Ca 2+ mobilisation assay
FlpInCHO wildtype and mutant human GLP-1R cells were seeded at a density of 3 10 4 cells/well into 96-well culture plates and incubated overnight at 37 C in 5% CO 2 , and receptor-mediated i Ca 2+ mobilisation determined as previously described [61]. Briefly, cells were incubated for 1 h with the cell-permeant Ca 2+ fluorophore, Fluo-4/AM (10 lM) in the presence of 2 mM probenecid prior to determining peptide-mediated changes in fluorescence in a Molecular Devices FlexStation (Molecular Devices, Palo Alto, CA, USA). Fluorescence was determined immediately after peptide addition, with an excitation wavelength set to 485 nm and an emission wavelength set to 520 nm, and readings taken every 1.36 s for 120 s. Peak magnitude was calculated using five-point smoothing, followed by correction against basal fluorescence. The kinetics for ligand-mediated i Ca 2+ were not altered by any of the mutations. The peak value was used to create concentration-response curves. Data were normalised to the maximal response elicited by 100 lM ATP.

Cell surface receptor expression
FlpInCHO wildtype and mutant human GLP-1R cells, with receptor DNA previously incorporated with an N-terminal double c-myc epitope label, were seeded at a density of 25 10 4 cells/well into 24-well culture plates and incubated overnight at 37 C in 5% CO 2 , washed three times in 1 PBS and fixed with 3.7% paraformaldehyde (PFA) at 4 C for 15 min. Cell surface receptor detection was then performed using a cell surface ELISA protocol to detect the cMyc epitope tag located at the extracellular Nterminus of the receptor, as previously described [32]. Data were normalised to the basal fluorescence detected in FlpInCHO parental cells. Specific 125 I-exendin-4(9-39) binding at each receptor mutant, as identification of functional receptors at the cell surface, was also determined (corrected for nonspecific binding using 1 lM exendin-4(9-39)) as described in [65].

Molecular dynamics simulations
The GLP-1R model was inserted into a hydrated equilibrated pal mitoyloleoylphosphatidylcholine (POPC) bilayer using the CHARMM-GUI interface [28]. Potassium and chloride ions were added to neutralise the system at an ionic strength of approximately 150 mM. Lipid14 (for POPC), AMBER99SP (for the protein) and TIP3P water model parameters were added using ambertools [7]. The simulations were carried out using ACEMD [19] on a purpose-built metrocubo GPU workstation. The system was energy minimised, heated from 0 K to 300 K in the NVT ensemble for 160 ps then simulated in the NPT ensemble, with 10 kcal mol 1 -A 2 positional harmonic restraints applied to the protein heavy atoms, which were progressively reduced to 0 over the course of 15 ns. Bond lengths to hydrogen atoms were constrained using M-SHAKE [34]. Production simulations were performed in the NPT ensemble at 300 K and 1 atm, using a Langevin thermostat for temperature coupling and a Berendsen barostat for pressure coupling. Non-bonded interactions were cutoff at 10.0 Å, and long-range electrostatic interactions were computed using the particle mesh Ewald method (PME) with dimensions of 86 86 142 using a spacing of 1.00 Å. The unconstrained simulation was run for 500 ns. Quantitative analysis of the trajectory was conducted in VMD.

Data analysis
All data were analysed using Prism 6 (GraphPad Software Inc., San Diego, CA, USA). For all analyses the data are unweighted and each y value (mean of replicates for each individual experiment) is considered an individual point. To calculate IC 50 , EC 50 and E max values, concentration response signalling data were analysed as previously described [30] using a three-parameter logistic equation. IC 50 values obtained from binding studies were then corrected for radioligand occupancy as previously described using the radioligand affinity (K i ) experimentally determined for each mutant.
To quantify efficacy in the system, all data were fitted with an operational model of agonism to calculate estimated s values. s is the operational measure of efficacy in the system, which incorporates signalling efficacy and receptor density. This model has been extensively described previously [30,66,64]. All estimated s values were then corrected to cell surface expression (s c ) as determined by cell surface ELISA and errors propagated from both s and cell surface expression.
Signalling bias was also quantified as previously described by analysis of concentration-response curves with nonlinear regression using an operational model of agonism, but modified to directly estimate the ratio of s c /K A [30,66,64]. All estimated s c /K A ratios included propagation of error for both s c and K A . Changes in s c /K A ratios with respect to wildtype of each mutant were used to quantitate bias between signalling pathways. Accordingly, bias factors included propagation of error from s c /K A ratios of each pathway.

Statistics
Changes in peptide affinity, potency, efficacy, cell surface expression and bias of each mutant receptor in comparison to the wildtype control were statistically analysed with one-way analysis of variance and Dunnett's post test, and significance was accepted at p < 0.05.

Results
Sequence alignments of the human Class B receptor subtypes reveal 22 conserved polar residues that are predicted to reside either in the TM bundle or at the membrane interface (10 of which are absolutely conserved as the same residue). An additional 2 residues are also very highly conserved in this subfamily (with the exception of 1 receptor subtype for loci 6.35 and 3 receptor subtypes for 5.56). We have previously reported the effects of mutation of 13 of these residues in the GLP-1R [64,66]. In this study we have probed the function of the remaining residues (Fig. 1A). All of these are located at TM helical boundaries/interfaces with loops, with the exception of Q7.65 410 that is located intracellularly within the predicted helix 8 (H8) at the bottom of TM7 (Fig. 1B-D). Each residue was individually mutated to Ala, verified by DNA sequencing and analysed for the effect of mutation on receptor function.
Wildtype and mutant human GLP-1Rs were isogenically integrated into FlpInCHO host cells by recombination that allows for direct comparison of cell surface expression as there should not be variations that arise due to differences in gene transcription. Cell surface expression was assessed by both antibody detection of the N-terminal double c-myc epitope label using ELISA and whole cell binding using [ 125 -I]-exendin-4(9-39) ( Table 1). A number of mutations resulted in significantly altered cell surface expression relative to the wildtype receptor, with consistent expression changes observed using both methods. Whole cell equilibrium competition binding studies were used to assess orthosteric peptide ligand affinities for the wildtype and each of the mutant GLP-1Rs (Table 1). These were performed with the endogenous agonists GLP-1(7-36)NH 2 (GLP-1) and oxyntomodulin, in addition to the exogenous agonist exendin-4 and an antagonist exendin-4(9-39), all in competition with the radiolabelled ligand 125 I-exendin-4(9-39). This revealed a number of mutations that globally altered peptide affinity and those that had selective effects of peptide affinity (Table 1).
Activation/strength of coupling to three cellular signalling cascades (cAMP production, ERK1/2 phosphorylation (pERK1/2) and intracellular calcium mobilisation ( i Ca 2+ )) was evaluated through the generation of concentration-response curves for all receptors with each peptide agonist (Figs. [2][3][4]. In most cases, mutations that resulted in changes in cell surface expression and/or affinity also produced significant changes on EC 50 and/or E max values (Table 2).
A direct measure of efficacy via calculation of Log s c values allows for direct comparison of receptor activation of individual intracellular signalling pathways at the different receptor mutants compared to the wildtype receptor, independently of their ligand affinity and cell surface expression. These were determined by analysing all concentration-response curves using an operational model of agonism to determine relative signalling efficacy estimates (log s values) that were corrected to different receptor expression levels by normalisation to what they would be if the mutant receptor were expressed at the same level as the wildtype (log s c values, Table 3). Cell surface expression data obtained from antibody binding were used for this correction instead of the Bmax from ligand binding studies, as one mutant showed no detectable radioligand binding, however correction with Bmax yielded similar efficacy values (data not shown). In addition, functional affinities (Log K A ) that describe the affinity of the receptor when coupled to a given signalling pathway were also derived from the operational analysis ( Table 4). The assessment of multiple signalling pathways also provided the ability to measure the signal bias of mutant receptors relative to the wildtype to obtain a quantitative measure of the relative bias between two pathways ( Table 5, Fig. 5).
To aid in interpretation of the experimental data, we used our two published GLP-1R models [64]; an inactive apo model of the TM bundle only and a GLP-1R:GLP-1:G as complex that was generated using multiple structural templates ( Fig. 1C-D). The combined results from expression, affinity and efficacy data (derived from the concentration-response curves) are presented in detail in the context of the predicted locations of mutated residues within these molecular models, clustering those located close in 3D space.
3.1. Three conserved positively charged residues located at the extracellular ends of TM helices 3, 4 and 5 are essential for high affnity agonist binding and conformational transitions linked to pleiotropic effector coupling through stabilisation of ECL2 Three highly conserved positively charged residues, R3.30 227 , K4.64 288 and R5.40 310 , located close to the extracellular surface of the GLP-1R are predicted to form direct interactions with residues in ECL2 in the apo and peptide bound models (Fig. 6). R3.30 227 is predicted to interact within the proximal region of Table 1 Effects of mutation on GLP-1R peptide ligand affinities and cell surface expression. Mutant and WT GLP-1Rs were stably expressed in ChoFlpIn cells and agonist affinities determined by equilibrium competition binding using [125-I]-exendin-4(9-39). Ligand affinities were determined using a three-parameter logistic equation and values are expressed as mean ± S.E.M of four to six independent experiments, conducted in duplicate. Cell surface expression was measured by ELISA against the c-myc epitope and by saturation binding, both normalised to the wildtype receptor. All data are expressed as mean ± S.E.M of four to six independent experiments, conducted in duplicate. Differences in affinity or expression were analysed with one-way analysis of variance (compared to the wildtype receptor) and Dunnett's post test ( * p < 0.05). ND means data were unable to be experimentally defined.

Receptor construct
Ligand binding affinity (pKi) Cell surface expression  ECL2 near to the top of TM4 in both the apo model and the GLP-1 In both inactive and active models, R5.40 310 resides close to N300 bound model (Fig. 6). K4.64 288 forms interactions at the opposite that is also predicted to form a direct interaction with GLP-1. end of ECL2, close to the top of TM5 in the apo receptor and forms R5.40 310 also resides close to His 7 of GLP-1 in the active model multiple interactions with ECL2 in the GLP-1 peptide bound model.
After correction for changes in expression, R3.30 227 A showed similar efficacy for generation of cAMP production and pERK1/2 relative to wildtype for the three peptides (Fig. 6, Table 3). However, there was a small, yet significant increase in efficacy for i Ca 2+ for oxyntomodulin that was not observed with the other two peptide agonists. For R5.40 310 A, a small reduction in cAMP efficacy was observed for GLP-1 and exendin-4, but not oxyntomodulin. In addition, pERK1/2 efficacy was also slightly reduced for exendin-4 and GLP-1 (3-5-fold), but not for oxyntomodulin (Fig. 6, Table 3). In contrast, no detectable i Ca 2+ was evident for any peptide at R5.40 310 A. K4.64 288 A impaired cAMP efficacy for all three peptides, but this was greater for GLP-1 and exendin-4 (42-50-fold) compared to oxyntomodulin (18-fold). In addition, there was no detectable calcium response with GLP-1 and exendin-4, although the oxyntomodulin efficacy for this pathway was unaltered. In contrast, all three ligands displayed a similar reduction in pERK1/2 efficacy (7-14-fold) (Fig. 6, Table 3).
Calculation of bias factors revealed that R5.40 310 did not significantly alter the ability of the receptor to sample between distinct conformations for activation of pERK1/2 and cAMP. Bias could not be calculated relative to i Ca 2+ , as there was no detectable response for this pathway (Fig. 5, Table 5). K4.64 288 A biased the receptor towards i Ca 2+ over cAMP and pERK1/2 when activated by oxyntomodulin and for exendin-4 towards pERK1/2 relative to cAMP (Fig. 5, Table 5). R3.30 227 significantly biased GLP-1 towards i Ca 2+ over cAMP, with a similar trend for oxyntomodulin and exendin-4 (Fig. 5, Table 5). This trend may not have been predicted from efficacy values alone as, unlike the majority of mutants assessed in this study, the functional K A values predicted from operational modelling were also altered differentially in the distinct pathways ( Table 4). The functional K A linked to cAMP accumulation tracked with the loss of affinity, however in i Ca 2+ , little reduction in the functional K A was observed compared to the wildtype receptor. Table 3 Effects of mutation on GLP-1R coupling efficiency to downstream effectors, cAMP, pERK1/2 and i Ca 2+ mobilisation. Mutant and WT GLP-1Rs were stably expressed in ChoFlpIn cells and concentration-response curves were generated for each construct in each pathway for the three agonists. All data were analysed with an operational model of agonism to determine log s values that define efficacy. All log s values were corrected to cell surface expression data from the ELISA (log s c ). Values are expressed as mean ± S.E.M of four to six independent experiments, conducted in duplicate. Data were analysed with one-way analysis of variance and Dunnett's post test ( * p < 0.05). ND means data were unable to be experimentally defined.

Receptor
Log Tau 346 reside towards the intracellular side of TMs 5 and 6, respectively. In the inactive apo model, both of these residues are predicted to hydrogen bond to regions in ICL2 that may be required to stabilise ground state receptor interactions. Interestingly, alanine mutation of either of these residues increased cell surface expression (Fig. 8, Table 1). For K6.35 346 A, this was detectable by both antibody labelling (175% of wildtype) and whole cell binding (159% of wildtype). While increased expression was detect-able at R5.56 326 A using antibody labelling (112% wildtype), there was significantly enhanced expression when calculating Bmax values from radioligand binding (141% of wildtype) ( Table 1).
In our active, peptide bound molecular model R5.56 326 and K6.35 346 are predicted to undergo a reorientation compared to the apo model, with both residues pointing away from the bundle (Fig. 8). An additional charged residue, K6.40 351 in TM6 is also located in an outward orientation relative to the bundle that is in a distinct orientation in the active model relative to the apo (Fig. 8).
While mutation of R5.56 326 to alanine did not alter affinity of either of the peptide agonists or the antagonist exendin-4(9-39), K6.35 346 A and K6.40 351 A both had small, yet significant selective Table 5 Effects of GLP-1R mutation on signal pathway bias. Data were analysed using an operational model of agonism to estimate log s c /K A ratios. Changes in log s c /K A ratios with respect to WT were calculated to provide a measure of the degree of stimulus bias exhibited by mutant receptors across the three pathways relative to that of the control receptor (WT). Values are expressed as mean ± S.E.M of four to six independent experiments, conducted in duplicate. Data were analysed with one-way analysis of variance and Dunnett's post test ( * p < 0.05). ND indicates data unable to be experimentally defined.  Values greater than 1 denote bias towards pathway 1, and values less than 1 denote bias towards pathway 2 relative to signalling at the wildtype receptor. Left, pERK1/2 (pathway 1) vs cAMP (pathway 2); middle, pERK1/2 (pathway 1) vs i Ca 2+ mobilisation (pathway 2); right, i Ca 2+ mobilisation (pathway 1) vs cAMP (pathway 2). All plots show the bias factors for the mutant receptors relative to the wildtype receptor for GLP-1 (blue), exendin-4 (salmon) and oxyntomodulin (green). Data points plotted as circles indicate statistically significant bias relative to the wildtype receptor (WT highlighted by the black reference line), whereas data plotted as triangles (at a value of 100 or 100) indicate that no significant signal could be detected for a particular pathway and therefor a bias factor could not be calculated. These values at 100 indicate no signalling in pathway 1 (therefore implied bias towards pathway 2), whereas +100 indicates no signalling in pathway 2 (therefore implied bias towards pathway 1). The residues are highlighted in the colour relevant to the clustering (and relevant figure) in which they are discussed in the results section.
effects on ligand affinity (Table 1). K6.35 346 A selectively enhanced the antagonist. In contrast, K6.40 351 A did not alter the affinity of GLP-1 and exendin-4 affinity, with oxyntomodulin displaying a the peptide agonists, but showed reduced affinity for exendin-4 similar trend, however no effect was observed on the affinity of (9-39) compared to the wildtype receptor (Table 1).  Table 1. Statistical significance of changes in affinity or coupling efficacy in comparison with wildtype were determined by one-way analysis of variance and Dunnett's post test, and values are indicated with an asterisk ( ⁄ , p < 0.05). All values are ± S.E.M of four to six independent experiments, conducted in duplicate.
K6.35 346 A enhanced the efficacy of all three agonists for the three signalling pathways, although this did not reach statistical significance for oxyntomodulin in pERK1/2 (Fig. 8, Table 3). While GLP-1 and oxyntomodulin displayed a similar fold increase in efficacy for calcium signalling (5-6-fold), there was a larger enhancement for exendin-4 at this mutant (26-fold) (Fig. 8, Table 2).
The ability of these mutations to selectively alter efficacy of distinct pathways and/or ligands resulted in different bias profiles of these mutant receptors relative to the wildtype (Table 5, Fig. 5). K6.35 346 A altered the coupling preference induced by oxyntomodulin, such that the receptor was even more strongly biased towards cAMP relative to i Ca 2+ than wildtype, with a similar trend also seen for GLP-1 (Table 5, Fig. 5). R5.56 326 A biased GLP-1 signalling towards cAMP relative to i Ca 2+ and pERK1/2. Oxyntomodulin did not signal to i Ca 2+ at this mutant and therefore may be biased towards pERK1/2 and cAMP over i Ca 2+ (Fig. 5, Table 5). Exendin-4 showed no significant change from wildtype at R5.56 326 A. K6.40 351 A was biased away from i Ca 2+ towards both cAMP and pERK1/2 when activated by GLP-1. Exendin-4 signaling also showed a significant bias for cAMP relative to i Ca 2+ . In contrast, oxyntomodulin biased the signaling away from pERK1/2 relative to cAMP and i Ca 2+ at this receptor in comparison to the wildtype (Fig. 5, Table 5).

A hydrogen bond network at the intracellular face stabilises the apo-GLP-1R and plays a role in controlling conformational transitions linked to biased signalling
Molecular modelling of the GLP-1R revealed a network of residues residing at the intracellular face of the receptor involving residues in TM2 (R2.46 176 ), TM6 (R6.37 348 ) and TM7 (N7.61 406 and E7.63 408 ). These are predicted to form an extensive hydrogen bond network in the ground state apo model (Fig. 9) that is disrupted in the active state model. We have previously reported the effects of alanine mutation of N7.61 406 that demonstrated little effect on receptor expression, ligand binding, cAMP formation or i Ca 2+ ([66], Fig. 9). However, there were small, yet significant reductions in the ability of this mutant to promote pERK1/2 when activated by GLP-1 and oxyntomodulin, but not exendin-4 (Fig. 9).
Mutation of R2.46 176 , R6.37 348 or E7.63 408 to alanine each resulted in a significant loss of cell surface expression (Fig. 9, Table 1). Interestingly, each mutation reduced this expression to a similar extent (57-66% of wildtype), supporting the role of these residues in a combined network. Despite this, relatively subtle effects were observed on other aspects of receptor function. All three mutants maintained the ability to bind the three agonists and the antagonist, albeit that a small yet significant reduction (4-fold) in exendin-4 affinity was observed for E7.63 408 A (Table 1). In addition, subtle changes to receptor bias occurred that did not always affect all three peptide ligands equally (Fig. 9, Tables 3 and 5). E7.63 408 A reduced cAMP signalling by all peptides, although this did not reach significance for oxyntomodulin (Fig. 9, Table 3). This resulted in E7.63 408 A being biased towards i Ca 2+ relative to cAMP for all ligands, but this only reached significance for GLP-1 (Fig. 5, Table 5). R6.37 348 A selectively altered effector signalling, reducing i Ca 2+ for GLP-1 and exendin-4, but not oxyntomodulin (Fig. 9, Table 3). This resulted in a statistically significant switch in the receptor bias when activated by GLP-1, such that it more readily activated effector coupling linked to pERK1/2 and cAMP compared to i Ca 2+ (Table 5, Fig. 5). R2.46 176 A had no significant effect on efficacy relative to wildtype.

A conserved polar residue in H8 is selectively important for GLP-1 mediated signalling, with little impact on exendin-4 and oxyntomodulin
Q7.65 410 A was assessed as part of this study as it is highly conserved in class B GPCRs, but it is not located with the TM bundle, rather at the start of the predicted helix 8 (H8) at the bottom of TM7. In our apo model Q7.65 410 is predicted to form a direct hydrogen bond with the backbone of TM7 (F7.59 404 ) and with the side chain of N7.62 407 and therefore may stabilise the hinge region between TM7 and H8 (Fig. 9). In the active model the interaction with the backbone of TM7 is maintained, but the interaction with N7.62 407 is lost due to a reorientation of the bottom of TM7 upon activation where N7.62 407 then resides close to the G as fragment (Fig. 9). While mutation of Q7.65 410 slightly reduced cell surface expression, it had selective effects on GLP-1R efficacy, with no significant effect on affinity of any ligand (Table 1). GLP-1 and exendin-4 mediated cAMP formation and pERK1/2 were also unaffected, however no i Ca 2+ could be detected when activated by GLP-1 and there was also reduced exendin-4 efficacy for this pathway (Fig. 9, Tables 1 and 3). This resulted in a significant bias of this mutant receptor relative to the wildtype towards cAMP formation compared to i Ca 2+ for exendin-4, and implies a similar bias for GLP-1 (Fig. 5, Table 5). For oxyntomodulin a different profile was observed; this ligand displayed reduced efficacy for pERK1/2 with  Table 1. Statistical significance of changes in cell surface expression or coupling efficacy in comparison with wildtype were determined by one-way analysis of variance and Dunnett's post test, and values are indicated with an asterisk ( ⁄ , p < 0.05). All values are ± S.E. M of four to six independent experiments, conducted in duplicate. no effect on i Ca 2+ or cAMP resulting in a significant bias of Q7.65 410 A towards i Ca 2+ relative to pERK1/2 compared to the wildtype receptor (Figs. 5 and 9, Tables 3 and 5). In the active state, while this residue remains close to TM4, it also interacts with Y2.48 178 in TM2 and W3.46 243 in TM3 (Fig. 10).
While the TM3-TM4 interaction does not appear to be important for receptor stability (as mutation of Y3.53 250 had no effect on receptor expression), the interaction of N2.52 182 in TM2 with TM4 residues may be important for receptor integrity as its muta-wildtype through antibody detection) ( Table 1, Fig. 10). Due to this heavily impaired expression, radioligand binding could not be detected and therefore ligand affinities could not be assessed (Table 1). Following correction for the loss in cell surface expression, pERK1/2 efficacy was not significantly altered at this mutation, however cAMP production was impaired for GLP-1 and exendin-4 (5-6-fold) and no i Ca 2+ could be detected for any of the three peptides (Fig. 10, Tables 2 and 3). N2.52 182 A significantly enhanced the coupling preference to pERK1/2 relative to cAMP for exendin-4 only, although a similar trend was observed with oxyntomodulin (Fig. 5, Table 5). The inability to detect an i Ca 2+ signal for N2.52 182 A indicates that this receptor is likely biased towards cAMP and pERK relative to i Ca 2+ for all ligands (Fig. 5, Table 5).
While mutation of Y3.53 250 had little effect on receptor expression, agonist affinity or cAMP formation, pERK1/2 was impaired (around 10-fold) and there was no detectable i Ca 2+ when activated by all three agonist peptides (Fig. 10, Tables 1-3). Despite this, only oxyntomodulin displayed significantly altered bias with bias towards cAMP production relative to pERK1/2, but as there was no detectable i Ca 2+ response for any peptide, it could be speculated that this mutation may also alter the bias of the GLP-1R away from i Ca 2+ , towards cAMP and pERK1/2 for all peptide agonists (Fig. 5, tion to alanine heavily impaired cell surface expression (39% of Table 5).  4. Discussion dramatic effects on pERK1/2 [30,31,65]. In addition, these mutations within ECL2 altered the efficacy of the pERK1/2 biased agonist Class B GPCRs are activated through interaction of the oxyntomodulin differentially to GLP-1 and exendin-4 highlighting a N-terminal region of their peptide agonists with the TM bundle of key role of this domain in biased agonism. Here, we reveal ligandthe receptor [47,48,6,41]. ECL2 plays an important role in this acti-dependent roles in peptide affinity and activation of the GLP-1R vation process [23,30,63] and mutations within this domain in the of three highly conserved positively charged residues (R3.30 227 , GLP-1R result in impaired cAMP production and i Ca 2+ with less K4.64 288 and R5.40 310 ) that have previously been implicated in GLP-1-mediated function (Table 6), and are predicted in our current molecular models to form stabilising interactions with ECL2. The conservation of positively charged residues at positions 3.30 and 4.64 in all Class B GPCRs and the negative effect on receptor function that is observed following mutation in multiple Class B GPCRs (Table 6) implies there may be a common role in stabilisation of ECL2 by these residues for this class of receptors. The distinct effects of mutation of R3.30 227 and K4.64 288 on affinity and efficacy of GLP-1 and exendin-4 relative to oxyntomodulin are particularly interesting as oxyntomodulin is a biased agonist relative to GLP-1 and exendin-4. These observations were more prominent for K4.64 288 and mutation of the proposed interacting residues in ECL2 (E292A and N304A) also resulted in similar ligand-dependent changes [30,31]. These data support a role for K4.64 288 in controlling activation transition leading to biased agonism by influencing the conformation of ECL2 and its interaction with distinct agonists. A recent study also predicted a similar interaction of K4.64 288 with ECL2, further supporting this theory [15]. Interestingly, for the calcitonin-like receptor (CLR) where a receptor activity modifying protein (RAMP) is required for function, mutation of R4.64 altered adrenomedullin function at CLR-RAMP2 or CLR-RAMP3 complexes, but not CGRP function at CLR-RAMP1 [60,63]. This suggests that in Class B receptor-RAMP complexes, stabilisation of ECL2 by R/K4.64 may have distinct functional consequences, in addition to controlling biased agonism of ligands acting at the same receptor. R5.40 310 , also conserved as a positive charge in many Class B GPCRs, interacts with ECL2 in our modelling, residing close to N300 that is predicted to form a direct interaction with GLP-1 (Fig. 6). R5.40 310 and N300 are both required for high affinity binding of GLP-1, exendin-4 and oxyntomodulin, with mutations of each having similar effects on affinity and both affecting efficacy of all three peptide agonists [30,31], therefore their proposed interaction may be important for peptide recognition. A polar residue at 5.40 is also required for function in other Class B GPCRs, particularly those in the glucagon subfamily (Table 6). In contrast to this proposed interaction of R5.40 310 with N300, a recently published study predicted a direct interaction of R5.40 310 with His 7 of GLP-1 [15]. Although absent in our static active state model, these side chains are in close proximity and in MD simulations (500 ns), R5.40 310 forms transient interactions with His 7 of GLP-1 (Fig. 7). Interestingly, for the GLP-1R, R5.40 310 also plays a role in controlling biased agonism, with distinct negative effects upon mutation Table 6 Published information for Class B GPCRs following mutation of the conserved polar residues assessed in this study. h, human; o, opossum; r, rat. GLP-1(R); glucagon-like peptide-1 (receptor); CLR, calcitonin-like receptor; RAMP, receptor activity modifying protein; CGRP, calcitonin gene related peptide; SecR, secretin receptor; PTH-(R), parathyroid hormone (receptor); GCGR, glucagon receptor; VPAC-(R), vasoactive intestinal polypeptide (receptor); GIP(R), glucose-dependent insulinotropic peptide (receptor). CRE; cAMP response element. hVPAC1R Decreased VIP mediated cAMP production [13] for GLP-1 and exendin-4 relative to the biased ligand oxyntomod-as His 7 of GLP-1 form stable interactions with E6.53 364 (Fig. 6) [64,65]; and suggests distinct functional requirements of R5.40 310 , in combination with the central hydrogen bond network for controlling peptidemediated GLP-1R activation leading to biased agonism. These MD simulations with GLP-1 also suggest R5.40 310 and N300 are key residues in guiding the N-terminus of these peptide agonists into the TM cavity for receptor activation (Fig. 7).
We have also previously reported on a key hydrogen bond network located at the cytoplasmic side of the TM bundle, between TMs 2, 3 and 6 that is essential for receptor integrity and for global activation of the GLP-1R [64,66]. The current study reveals the importance of an additional hydrogen bond network, also at the intracellular face, formed by residues in TM2 (R2.46 176 ), TM6 (R6.37 348 ) and TM7 (N7.61 408 and E7.63 408 ) that is evident in the crystal structures of the GCGR and CRF 1 R [22,49]. Differences in our apo models vs GLP-1 peptide bound models suggest a reorganisation of these intracellular networks involving a disruption of crucial contacts between TMs 3 and 6, and TMs 2 and 7 result in the TM bundle opening at the intracellular face, allowing for effector coupling. Mutation of these residues in both networks (with the exception of N7.61 406 ) significantly reduced cell surface expression highlighting a role for both networks in receptor stability ( [66], Fig. 9). The role of these networks are also consistent with experimental data from other Class B GPCRs where mutation of residues either induced constitutive cAMP activity, enhanced potency for cAMP production or result in poor receptor expression at the cell surface, observations that are all consistent with destabilisation of the inactive state [59], (Table 6). These combined data across Class B GPCRs, in addition to the conservation of these interactions in the two solved inactive state Class B GPCR TM crystal structures support a common role for hydrogen bond networks at the cytoplasmic face in stabilisation of the apo receptor [22,49].
Residues within the newly reported TM2-6-7 network in the GLP-1R also have independent roles for signal transduction after being released from their ground state constraints. While we did not identify a role for R2.46 176 in transmission of efficacy, it may play a minor role, as observed in a mutational study at the rat GLP-1R (Table 6). In contrast, we revealed distinct roles for R6.37 348 and E7.63 408 in directing signalling specificity. Consistent with other Class B GPCRs (Table 6), E7.63 408 selectively couples the GLP-1R to cAMP (G as ). In contrast, R6.37 348 plays a role in coupling the GLP-1R to i Ca 2+ that is non-G as -mediated [65], but only when the receptor was activated by GLP-1 and exendin-4. Along with R6.37 348 , K6.40 351 forms part of a basic-X-X-basic motif (BxxB) that is highly conserved in both Class A and B GPCRs, but the effects of mutation are variable depending on the receptor being studied. Evidence suggest residues in this motif play only minor roles in G as /cAMP efficacy for Class B GPCRs, but are more important for IP 3 /calcium mobilisation (Table 6). This is consistent with this current study on the GLP-1R, where mutation of both basic residues had little effect on cAMP production by any peptide, but reduced the efficacy of GLP-1 and exendin-4 for i Ca 2+ . However, there was no alteration in oxyntomodulin efficacy, consistent with distinct receptor conformational propagation achieved by the ligand that exposes distinct side chains for effector interaction. Therefore, the BxxB motif may have distinct roles in controlling receptor conformation and effector coupling between ligands acting at the same receptor. The observed effects of mutation of R6.37 348 , K6.40 351 and E7.63 408 for signalling specificity could arise due to direct contacts with effector proteins or indirectly through forming interactions (either within the receptor or with lipids) that stabilise active receptor conformations required for coupling to distinct pathways. Indeed, R6.37 348 and E7.63 408 are in the vicinity of G as in the GLP-1 bound molecular model and therefore relatively small differences in conformational rearrangement upon binding of distinct agonists could subtly alter interactions with effector proteins giving rise to the observed changes in signal bias.
Lys and Arg residues found near the polar/a-polar interfaces can hydrogen bond to phosphate head groups and esterified oxygens of the lipid backbone, anchoring TMs in the bilayer in the optimal orientation in the membrane for receptor function [51]. From our GLP-1R models, three residues R5.56 326 , K6.35 346 and K6.40 351 may play such a role as our active state model places these residues pointing out towards lipid. The reorientation of these three side chains between the two models suggests that these residues may be important for controlling TM movements during activation transition. Mutation of R5.56 326 and K6.35 346 also increased cell surface expression, an effect that is often associated with stabilisation of the ground state conformation. Indeed, Ala mutation of an equivalent residue, Y6.35, in the CRF1R TM domain crystal structure was used to increase the thermostability of the inactive receptor protein and to aid in crystallisation [22]. R5.56 326 A also selectively impaired pERK1/2 by GLP-1 and exendin-4 and heavily impaired i Ca 2+ by all ligands, consistent with stabilisation of an inactive receptor. In contrast, K6.35 346 A enhanced affinity and signalling efficacy by all ligands to all three pathways. This residue is only positively charged in the glucagon subfamily of Class B GPCRs (being a polar Tyr in most others (Fig. 1)), and therefore may play a different role in this glucagon subclass compared to the other Class B members.
TM4 is the most peripherally located TM and forms the interface for GLP-1R homodimerisation in Class B GPCRs that is important for GLP-1R signalling [18]. N2.52 182 and Y3.53 250 pack up against TM4 and play global roles in GLP-1R activation by peptide agonists, with both residues being crucial for i Ca 2+ mobilisation, but selectively involved in cAMP (N2.52 182 ) or pERK signalling (Y3.53 250 ), effects that may arise due to stabilisation of the important dimerisation interface. Consistent with this, mutation of either residue had the largest impact on calcium signalling, which parallels with the greater loss of calcium signalling relative to cAMP and pERK1/2 following mutation of the TM4 dimerisation interface within the GLP-1R [18]. Molecular modelling also predicts a reordering of TM2 relative to TM3 and TM4 that may stabilise residues within TM3 in the activated receptor, a key domain for signal transduction that may also contribute to the altered signalling at these mutant receptors compared to the wildtype.
Collectively, this work expands our understanding of how peptides activate the GLP-1R receptor to promote signalling, highlighting additional key conserved Class B GPCR polar side chains within the TM domain beyond those already reported. There is now a large body of evidence from multiple Class B GPCRs that shed light on how these complex receptors are activated with conserved polar residues playing a crucial role in this process (Table 6 [68,66,64,65,59,9]. Despite their distinct mode of ligand interaction relative to Class A GPCRs, there are some parallels in how these two classes of receptors are activated. There is now substantial evidence that ECL2 plays a major role in the binding and activation of both classes of receptors [30,63,11,62]. However, conformational differences within ECL2 have been identified, even within the Class A subfamily [62], suggesting different networks of interactions are involved in stabilisation of this important domain. In addition, despite different conserved amino acids in the two subclasses, polar interactions are crucial for signal propagation, facilitating conformational TM rearrangements through the reorganisation of hydrogen bond networks in Class A and Class B GPCRs [2,3,42,66,64,9,59]. For Class A GPCRs, there is substantial evidence that this results in a large-scale conformational transition of TM6 relative to TM3 that requires the disruption of key polar networks at the intracellular face [46,45]. Limited evidence supports a similar movement of TM6 relative to TM3 in Class B GPCRs [50]. This study, taken together with our previous studies [66,64], suggest that breaking of key polar networks at the intracellular face of Class B GPCRs (TM2-TM3-TM6 and TM2-TM6-TM7), like Class A GPCRs, are crucial in this subfamily of receptors to facilitate movements within TM6 allowing for effector interaction.
Additionally, there is an increasing body of evidence from mutational studies supporting distinct modes of receptor activation by biased peptides at the GLP-1R, with this study providing additional evidence for the role of polar interaction networks in influencing how these differences may be achieved. There is also evidence that the ability of individual ligands to influence polar interactions within Class A GPCRs contributes to biased agonism [53,70]. While our mutagenesis studies combined with GLP-1R models can be used to facilitate understanding of mechanisms for activation of Class B GPCRs and propagation of biased signalling, additional and more complex structural and biophysical analysis of this receptor, (or any Class B GPCR) are required to gain an in depth understanding of the large scale conformational movements that allow these very complex receptor-ligand systems to transmit signals from the ligand binding pocket at the extracellular face to cytoplasmic signalling molecules.
Wrote or contributed to writing of the manuscript: Wootten, Sexton, Reynolds, Furness, Miller, Christopoulos.