Evaluation of efficacy- versus affinity-driven agonism with biased GLP-1R ligands P5 and exendin-F1

The glucagon-like peptide-1 receptor (GLP-1R) is an important regulator of glucose homeostasis and has been successfully targeted for the treatment of type 2 diabetes. Recently described biased GLP-1R agonists with selective reductions in β-arrestin versus G protein coupling show improved metabolic actions in vivo. However, two prototypical G protein-favouring GLP-1R agonists, P5 and exendin-F1, are reported to show divergent effects on insulin secretion. In this study we aimed to resolve this discrepancy by performing a side-by-side characterisation of these two ligands across a variety of in vitro and in vivo assays. Exendin-F1 showed reduced acute efficacy versus P5 for several readouts, including recruitment of mini-G proteins, G protein-coupled receptor kinases (GRKs) and β-arrestin-2. Maximal responses were also lower for both GLP-1R internalisation and the presence of active GLP-1R-mini-Gs complexes in early endosomes with exendin-F1 treatment. In contrast, prolonged insulin secretion in vitro and sustained anti-hyperglycaemic efficacy in mice were both greater with exendin-F1 than with P5. We conclude that the particularly low acute efficacy of exendin-F1 and associated reductions in GLP-1R downregulation appear to be more important than preservation of endosomal signalling to allow sustained insulin secretion responses. This has implications for the ongoing development of affinity- versus efficacy-driven biased GLP-1R agonists as treatments for metabolic disease.


Introduction
With the increasing worldwide prevalence of type 2 diabetes (T2D) [1], there is an urgent need for more effective drugs to treat this condition. T2D results from a combination of relative insulin deficiency and resistance to its action in central and peripheral tissues, and is commonly associated with excess adiposity or obesity [2]. The glucagon-like peptide-1 receptor (GLP-1R), a class B G protein-coupled receptor (GPCR) expressed at high levels in pancreatic beta cells and at lower levels in anorectic centres in the brain, is a wellestablished target for pharmacological T2D treatment [3]. GLP-1R activation augments glucose-stimulated insulin secretion, improves beta cell survival and suppresses appetite, with the latter resulting in weight loss and improvements in insulin sensitivity [4].
Pharmacokinetically optimised GLP-1R agonists based on the amino acid sequence of either the cognate agonist GLP-1(7-36)NH 2 or its homologue exendin-4 [5] are approved for the treatment of T2D. These agents not only improve glycaemic control and induce weight loss but also reduce cardiovascular [6] and all-cause mortality [7] in people with T2D.
Recent preclinical studies have shown that GLP-1R agonists that favour G protein signalling and generation of cyclic adenosine monophosphate (cAMP) over β -arrestin recruitment are particularly effective at reducing blood glucose levels [8][9][10][11][12]. Moreover, it is suggested that the markedly reduced β -arrestin recruitment seen at the GLP-1R with the dual GLP-1R/glucose-dependent insulinotropic peptide receptor (GIPR) agonist Tirzepatide [13][14][15] may contribute to its superior anti-diabetic efficacy in clinical trials [16]. An appealing explanation for the effects of these "biased" GLP-1R agonists is that reductions in β -arrestinmediated desensitisation, as well as a trafficking phenotype that favours preservation of GLP-1R at the plasma membrane, lead to prolonged intracellular signalling and cumulatively greater insulin release over time. However, this has not been demonstrated for all published examples of biased GLP-1R agonists. In particular, "P5", the first bespoke biased GLP-1R agonist to be described, was potently anti-hyperglycaemic but poorly insulinotropic in vivo, with its metabolic effects instead partly attributed to increases in adipogenesis [8]. This contrasts with "exendin-phe1" [9] (referred to here as "exendin-F1"), a peptide that showed marked increases in insulin release compared to exendin-4 in cellular models and in mice.
One possible explanation for this discrepancy is that different approaches were used to evaluate the two ligands, with exendin-F1 tested using prolonged incubations with beta cells and islets, as well as sub-chronic in vivo studies, to specifically seek functional evidence of reduced desensitisation over the course of several hours [9]. However, to date, P5 has not been examined in this way.
Establishing a consensus mechanism of action for biased GLP-1R agonists would help guide their development for the treatment of T2D and related metabolic diseases. In the present work we perform direct pharmacological comparisons of P5 and exendin-F1, with exendin-4 included as the reference peptide. As well as determining relative preferences for G protein and β -arrestin recruitment responses, we focussed in particular on differences in GLP-1R membrane trafficking and activation in different endomembrane compartments [17].
Our study highlights a number of pharmacological properties that diverge between P5 and exendin-F1, suggesting these GLP-1R agonists may in fact possess distinct modes of action.
After 20 min at room temperature, cells were then placed at 4°C, and a range of concentrations of test ligands were applied concurrently with 10 nM Luxendin645 [19] or 5 nM exendin-4-Cy5, with a range of concentrations of Luxendin645 or exendin-4-Cy5 also applied to establish equilibrium binding parameters for the competing labelled GLP-1R probe. After a 24-hour incubation period at 4°C, binding was measured by TR-FRET using a Spectramax i3x plate reader (Molecular Devices) fitted with an HTRF module.

cAMP assays
Resuspended cells were stimulated at 37°C in their respective serum-free medium in 96-well half area opaque white plates before addition of HTRF lysis buffer and detection reagents (cAMP Dynamic 2 kit, Cisbio). The duration of stimulation and inclusion or not of the phosphodiesterase inhibitor 3-isobutyl-1-methylxanthine (IBMX) is indicated in the relevant figure legend. The assay was read by HTRF.

Homologous desensitisation assay in beta cells
INS-1 832/3 cells were seeded into poly-D-lysine-coated 96-well plates in complete medium at 11 mM glucose in the presence of different concentrations of test agonist. After an overnight incubation, medium was removed and cells were washed 3 times in HBSS, followed by a 1-hour resensitisation period in complete medium. Cells were then stimulated at 37°C with 100 nM GLP-1 + 500 µM IBMX for 10 min followed by lysis and cAMP determination as described in Section 2.4.

β -arrestin-2 GLP-1R recruitment assays
These assays were performed as described previously [24]. HEK293T cells were transiently transfected using Lipofectamine 2000 with 50 ng each of GLP-1R-SmBit and LgBit-βarrestin-2 (Promega) diluted in pcDNA3.1, or with 500 ng each of GLP-1R-SmBiT and LgBiT-mini-G s (a gift from Prof Nevin Lambert, Medical College of Georgia) [25], per well of a 12-well plate, and the assay was performed 24 hours later. For the kinetic mode assay, cells were resuspended in HBSS containing Furimazine (1:50), seeded into half-area opaque white plates, and total luminescent signal at baseline was recorded over 5 min at 37°C using a Flexstation 3 plate reader. Ligands were then added, and signal was serially monitored for up to 30 min. Ligand-induced change was expressed relative to baseline values for each well. For the endpoint mode assay, cells were resuspended in HBSS and seeded into half-area opaque white plates containing prepared ligands. After a 5-minute incubation at 37°C, Furimazine prepared in HBSS was added, and luminescent signal was recorded over 3 min using a Spectramax i3x plate reader. seeded into 96-well half-area opaque white plates, and baseline luminescent signals recorded at 460 nm (Nluc emission peak) and 520 nm (GFP acceptor peak) or 535 nm (Venus acceptor peak) over 5 min at 37°C using a Flexstation 3 plate reader. Ligands were added, and signal was serially monitored for up to 20 min. Signal was expressed ratiometrically for each time-point as GFP or Venus acceptor divided by Nluc donor signal.

NanoBRET GLP-1R recruitment assays
The BRET ratio at each time-point was first expressed relative to the average baseline value for each well, followed by subtraction of the vehicle signal at each time-point to provide net BRET.

DERET assay
The assay was performed as described previously [24]. Cells were labelled with SNAP-Lumi4-Tb (40 nM, 60 min at 37°C, in complete medium). After washing, labelled cells were resuspended in HBSS containing 24 µM fluorescein. TR-FRET signals at baseline and serially after agonist addition were recorded at 37°C using a Flexstation 3 plate reader using the following settings:

GLP-1R clustering assay
The assay was performed as previously described [20]. Cells were dual labelled with SNAP-Lumi4-Tb (40 nM) and SNAP-Surface-649 (500 nM) for 30 min at 37°C, in complete medium. After washing, labelled cells were resuspended in HBSS. TR-FRET signals at baseline and serially after agonist addition were recorded at 37°C using a Spectramax i3x plate reader with HTRF module. GLP-1R clustering was quantified as the ratio of the fluorescence signal at 665 nm to that at 616 nm.

Visualisation of GLP-1R subcellular localisation
Cells were seeded onto glass coverslips and labelled with SNAP-Surface-549 (1 µM, 30 min, 37°C). After agonist stimulation and paraformaldehyde fixation, coverslips were mounted onto glass slides using Diamond Prolong antifade with DAPI. Slides were imaged using a Nikon Ti2E microscope frame with integrated hardware from Cairn Research incorporating motorised stage (ASI), LED illumination source (CoolLED) and a 100X oil immersion objective. Z-stacks were acquired and deconvolved using Deconvolutionlab2 [28] using the Richardson Lucy algorithm. Images were processed in Fiji.

High content microscopy trafficking assay
The assay was performed as previously described [24]. Cells were seeded into poly-Dlysine-coated, black 96-well plates. On the day of the assay, cells were labelled with BG-S-S-649 (1 µM, a gift from New England Biolabs). After washing, agonists were applied for 30 min at 37°C in complete medium. Agonists were removed, cells washed with cold HBSS and then placed on ice for subsequent steps. Mesna (100 mM in alkaline TNE buffer, pH 8.6) or alkaline TNE without Mesna was applied for 5 min, and then washed with HBSS.
Microplates were then imaged using the microscope system described in Section 2.13 fitted with a 20X phase contrast objective, with data acquisition controlled by the openHCA software written for the MicroManager platform [29]. 9 random images per well were acquired for both epifluorescence and transmitted phase contrast. HBSS was then removed and replaced with fresh complete medium. Receptor was allowed to recycle for 60 min at 37°C, followed by a second Mesna application to remove any receptor that had recycled to the membrane, and the plate was re-imaged as above. Internalised SNAP-GLP-1R was quantified at both time points using Fiji as follows: 1) phase contrast images were processed using PHANTAST [30] to segment cell-containing regions from background; 2) illumination correction of fluorescence images was performed using BaSiC [31]; 3) fluorescence intensity was quantified for cell-containing regions. Agonist-mediated internalisation was determined as the mean signal for each condition normalised to signal from wells not treated with Mesna, after first subtracting non-specific fluorescence determined from wells treated with Mesna but no agonist. The percentage of reduction in residual internalised receptor after the second Mesna treatment was considered to represent recycled receptor. Recycling was then expressed as a percentage relative to the amount of receptor originally internalised in the same well.

Overnight stimulation surface labelling assay
INS-1-SNAP-GLP-1R cells were seeded into poly-D-lysine-coated, black 96-well plates with agonists applied in complete medium at 11 mM glucose. After an overnight incubation, agonists were removed, and cells washed with HBSS. Cells were then labelled with BG-S-S-649 (1 µM in complete medium, 30 min) before washing and imaging as in Section 2.14.
Surface labelling intensity was quantified as in Section 2.14 with subtraction of signal from INS-1 832/3 cells without SNAP-GLP-1R but labelled with BG-S-S-649.

Insulin secretion assay
After a prior 6 hours at 3 mM glucose, INS-1 832/3 cells were seeded overnight in 96-well plates in complete medium at 11 mM glucose ± test agonists. A sample of supernatant was removed and analysed for insulin content by HTRF (Wide Range Insulin HTRF Kit, Cisbio).
Results were expressed relative to insulin released with 11 mM glucose alone.

In vivo study
Animals were maintained in specific pathogen-free facilities, with ad lib access to food (except prior to fasting studies) and water. Studies were regulated by the UK Animals (Scientific Procedures) Act 1986 of the U.K. and approved by Imperial College London (Project License PB7CFFE7A). Male C57Bl/6 mice (8-10 weeks of age, weight 25 -30g, supplied by Charles River, UK) were injected intra-peritoneally (IP) during the light phase with 2.4 nmol/kg agonist prepared in 100 µl 0.9% NaCl, or vehicle. Food was removed from the cages at this point until the end of the study. After an 8-hour delay, mice were injected IP with 20% dextrose (2 g/kg). Blood glucose levels were monitored immediately before glucose administration and at 20-min intervals thereafter from tail vein blood samples using a hand-held glucose meter.

Data analysis
All analyses were performed using Prism 8 (GraphPad Software). The average of withinassay replicates was counted as one biological replicate. For equilibrium binding studies, Luxendin645 and exendin-4-Cy5 K d values were fitted using a "one site -specific binding" model. Test ligand K i values were determined using a "one site -fit K i " model. Functional concentration response data were fitted using 3-parameter logistic fitting, with constraints applied where appropriate. Bias analysis was performed by the calculation of transduction coefficients (i.e. ߬ /K A values) as previously described [9,32,33], with subtraction of the Log(߬/K A ) value for one pathway (cAMP or mini-G s ) from the second pathway (β-arrestin-2) to give ∆ Log(߬/K A ). The assays for which bias was calculated were performed with both pathways assessed in parallel using the same ligand stock, allowing statistical comparisons on a per-assay basis. In the instances when a composite parameter was determined from two assay types not performed in parallel (i.e. the pEC 50 -pK i measurement in Figure 1

Contrasting efficacy and affinity with P5 versus exendin-F1.
We first measured the equilibrium binding affinities of exendin-4, exendin-F1 and P5 ( Figure   1A) in HEK293 cells stably expressing SNAP-GLP-1R [34], using a competitive timeresolved FRET assay in which the receptor N-terminus is labelled with the energy donor Lumi4-Tb to detect binding of Cy5-conjugated GLP-1R antagonist exendin  ("Luxendin645") [19] or the equivalent agonist, exendin-4-Cy5 [18], applied in competition with unlabelled test peptides ( Figure 1B and 1C, Table 1). Both G protein-biased ligands displayed significantly lower affinity for receptor binding than exendin-4; of note, the K i for P5 was 22-fold higher than for exendin-F1 when measured using Luxendin645 as the competing probe but only 4-fold higher with exendin-4-Cy5. Measurement of cAMP production in the same cell model showed reduced potency for both exendin-F1 and P5 ( Figure 1D, Table 1), but the differences between agonists were smaller than for equilibrium binding affinity, suggesting ligand-specific differences in coupling of receptor occupancy to cAMP production ( Figure 1E). This observation has been made before for P5 [35] and exendin-F1 [34], but here the ligands are compared directly for the first time.
Balance between G protein pathway engagement and β -arrestin-2 recruitment was assessed using two approaches. Firstly, using the PathHunter system [9], cAMP production and β -arrestin-2 recruitment were measured in parallel, which highlighted how exendin-F1 and P5 show low efficacy as well as reduced potency for β -arrestin-2 ( Figure 1F, Table 1). β -arrestin-2 recruitment efficacy was greater for P5 than for exendin-F1. Comparison of transduction ratios (߬/K A ) determined using the operational model for each pathway [33] indicated a substantial degree of bias in favour of cAMP production for both P5 and exendin-F1, with the effect being most marked for the latter (~60-fold versus 15-fold). Secondly, recruitment of mini-G s and β -arrestin-2 to GLP-1R were measured by nanoBiT complementation [25,36], with response kinetics shown in Figure 1G and concentrationresponses at 5 min in Figure 1H (see also Table 1). These analyses confirmed low efficacy β -arrestin-2 recruitment but also reduced recruitment efficacy for mini-G s with exendin-F1 and P5. Maximum responses with exendin-F1 were reduced in comparison to P5 in both pathways, to the extent that the β -arrestin-2 response was not amenable to logistic curve fitting for the former ligand. Interestingly, no significant bias could be detected for P5 versus exendin-4 from these data; bias for exendin-F1 could not be determined due to the lack of a quantifiable β -arrestin-2 response.
We also investigated the relative propensity for each ligand to recruit G protein-  Figure 1I). Ligand-induced changes were barely detectable for GRK5 and 6.
We also designed a BRET-based sensor strategy to monitor differences in ligandinduced activation (as opposed to recruitment) of endogenous G proteins by co-expressing SNAP-GLP-1R-Nluc with GFP-tagged nanobody-37 (Nb37), a genetically encoded intrabody that recognises the active conformation of Gα S [26]. Interestingly, the Nb37 response to exendin-F1, but not P5, was reduced in comparison to exendin-4 in HEK293 cells ( Figure   1J). The low dynamic range of this sensor configuration precluded concentration-response analyses.
Overall, these data highlight how exendin-F1 has a higher GLP-1R binding affinity than P5 but lower efficacy for recruitment of mini-G s , GRK2 and β -arrestin-2, as well as reduced Gα s activation, which implies greater coupling efficiency of GLP-1R occupancy to intracellular responses with P5 versus exendin-F1. However, the higher affinity of exendin-F1 results in comparable acute cAMP signalling responses between both biased agonists in the heterologous cell system used for these studies.

GLP-1R trafficking with exendin-F1 and P5
An altered membrane trafficking profile characterised by reduced endocytosis and faster recycling is thought to be an important component of the action of biased GLP-1R agonists such as exendin-F1 [9], but has not been tested for P5. Using high content microscopy [24], we confirmed that both exendin-F1 and P5 showed reduced GLP-1R internalisation propensity but faster recycling than exendin-4 ( Figure 2A, Table 1). The reduction in efficacy was again most pronounced with exendin-F1 but potency was slightly lower with P5; relative potencies for each ligand versus exendin-4 were in line with those for cAMP production in the same cell model. Higher resolution images of GLP-1R subcellular localisation with each ligand are shown in Figure 2B, revealing that a substantial proportion of GLP-1R remains at the plasma membrane after stimulation with both biased GLP-1R agonists, in comparison to exendin-4. Kinetics of GLP-1R internalisation were assessed by diffusion-enhanced resonance energy transfer (DERET) [39], with very little response detected with either exendin-F1 or P5 below 100 nM ( Figure 2C, Table 1).
We also developed a time-resolved FRET (TR-FRET) assay to monitor translocation of lanthanide-labelled SNAP-GLP-1R to the late endolysosomal compartment, which was labelled using the chemical endolysomotropic dye LysoTracker DND99. This highlighted how exendin-4 rapidly targets internalised GLP-1Rs towards this degradative compartment ( Figure 2D, Table 1). In these studies, performed in parallel to DERET measurements, the EC 50 for exendin-4-induced GLP-1R lysosomal localisation was somewhat higher than for internalisation per se, implying that not all internalised GLP-1Rs are targeted to the lysosome.
Agonist-internalised GLP-1Rs continue to generate cAMP signals from the endosomal compartment [40], with this process being ligand specific [41]. To monitor the distribution of GLP-1Rs in their active state between plasma membrane and endosomes with exendin-4, P5 and exendin-phe1, we co-expressed mini-G s tagged with nanoluciferase with Venus-tagged markers of the plasma membrane (KRAS) or early endosome (Rab5) [25]. In this bystander configuration, the location of GLP-1R in its active conformation is inferred when mini-G s is recruited to the vicinity of the relevant compartment marker leading to an increase in BRET signal. In stable HEK293-SNAP-GLP-1R cells, 100 nM exendin-4 induced robust and rapid translocation of mini-G s to the plasma membrane, followed by a gradual decline; a similar pattern, but with a lower peak, was seen with P5 ( Figure 2F). For exendin-F1, the peak response was further reduced, but did not fall below the peak level for the full 30-minute stimulation period. Mini-G s recruitment to Rab5-positive early endosomes was of slower onset than at the plasma membrane, and response magnitude showed a rank order matching that of GLP-1R endocytosis (exendin-4 > P5 > exendin-F1; Figure 2G).
Expressing ligand-specific Rab5 and KRAS signals ratiometrically from their AUC ( Figure   2H), or over time ( Figure 2I), suggested P5 engenders a similar balance between endosomal and plasma membrane activity to exendin-4, but exendin-F1 delivers a predominantly plasma membrane delimited response. We did not attempt to perform full concentration responses with these assays due to the low dynamic range of the Rab5 BRET assay.
Therefore, both P5 and exendin-F1 show reduced GLP-1R endocytosis and accelerated recycling, with this effect being most dramatic for exendin-F1. The differences in GLP-1R internalisation between P5 and exendin-F1 are mirrored by their respective tendencies to elicit GLP-1R activity at early endosomes.
Interestingly, efficacy with P5 was slightly higher than with the other two ligands in GLP-1Rexpressing cells.
Previous work has demonstrated how prolonged stimulation with GLP-1R agonists with different trafficking phenotypes leads to variable levels of receptor downregulation, which can influence their insulinotropic potential [9,10]. In SNAP-GLP-1R stably expressed in INS-1 832/3 cells lacking endogenous GLP-1R, referred to as INS-1-SNAP-GLP-1R cells [42], both exendin-F1 and P5 showed markedly increased preservation of surface GLP-1R levels compared to exendin-4 ( Figure 3B, Table 2). Exendin-F1 led to loss of surface GLP-1R at lower concentrations than P5, as expected from its higher affinity, but the maximum loss of surface receptor effect was less marked.
To determine the functional impact of differential GLP-1R downregulation, we measured cAMP responses to a fixed concentration of GLP-1 after a prior 16-hour pretreatment phase with each exendin analogue. Both biased ligands produced substantially less homologous desensitisation than exendin-4 ( Figure 3C). Whilst concentration responses could not be accurately quantified by logistic curve fitting of data from assay repeats, P5 showed greater desensitisation than exendin-F1 at pre-treatment concentrations upwards of 10 nM.
We also measured cumulative insulin secretion with each ligand after an overnight stimulation as a therapeutically relevant readout of sustained pharmacological agonism in beta cells. Here, there was a stark difference in insulinotropic efficacy, with exendin-F1 treatment yielding approximately twice as much insulin secretion as exendin-4 and P5 ( Figure 3D, Table 2). Moreover, assessment of anti-hyperglycaemic efficacy in mice by intraperitoneal glucose tolerance testing, performed 8 hours after agonist administration to allow desensitisation-related effects to emerge as previously performed [9,10], showed a pattern compatible with apparent greater sustained action with exendin-F1 ( Figure 3E). Note that both P5 and exendin-F1 have been demonstrated to show identical pharmacokinetics to exendin-4 [8,9].
In this study we have compared two exendin-4 analogues, P5 and exendin-F1, that were the first two synthetic, orthosteric GLP-1RAs reported to show biased agonism favouring G protein-dependent signalling over β -arrestin recruitment and tested in vivo for their metabolic effects. Biased GLP-1R agonism has emerged as a promising therapeutic strategy for T2D on the basis of preclinical evaluations [8,9,11] and the recent recognition that some of the beneficial effects of the dual GLP-1R/GIPR agonist tirzepatide may be due to biased agonism at the GLP-1R [14]. Whilst in our study we confirmed the exendin-F1 and P5 do It is notable that both exendin-F1 and P5, considered G protein-biased from previous reports [8,9], are in fact low efficacy agonists for mini-G s engagement compared to the full agonist exendin-4. We verified bias in favour of cAMP production over β -arrestin-2 recruitment in the PathHunter system, but interestingly, the same operational analysis did not indicate bias between mini-G s and β -arrestin-2 recruitment for P5 (while for exendin-F1 the β -arrestin-2 response was undetectable, precluding formal analysis). There is increasing recognition that bias comparisons between readouts encompassing significant signal amplification (e.g. cAMP) and those without amplification (e.g. β -arrestin recruitment) are susceptible to system non-linearity that may confound current models [43,44]. However, if seen from the pragmatic point of view that efficacy is a key driver of the manifestations of biased agonism [43], considering P5 and exendin-F1 as G protein-biased appear appropriate, as in both cases β -arrestin-2 recruitment is even more markedly reduced than mini-G s recruitment.
Our whole cell binding assays indicated lower affinity for P5 than for exendin-F1.
Indeed, P5 was previously identified as a low affinity agonist, with a reported ~100-fold lower affinity than GLP-1 in competition with iodinated exendin(9-39) [35]. Our P5 pK i measurements were dependent on the choice of competing fluorescent probe, with apparently lower affinity measured when the antagonist LUXendin645 was used rather than the equivalent agonist exendin-4-Cy5. In a previous report, the absence of G proteins, or expression of a dominant negative Gα s , led to smaller differences between P5 and exendin-4 affinity in insect cells [35]. Thus, probe-dependency of measured pK i with P5 could represent a higher affinity GLP-1R active-state complex triggered by exendin-4-Cy5-induced coupling to Gα s that is not seen with LUXendin645 [45]. Additionally, whilst our whole cell binding assays were performed at low temperature and with metabolic inhibitors to avoid the effects of ATP-dependent GLP-1R redistribution and endocytosis on binding phenomena, we cannot absolutely exclude these as ligand-specific confounders. Affinity measurements in cell-free reconstituted systems could be performed to address this issue.
Contrasting with its higher affinity, maximum responses for multiple readouts were further reduced for exendin-F1 than for P5. Thus, recruitment of mini-G s , GRK2 and intermediates and downstream effectors [46]. Specifically, in the presence of adequate GLP-1R expression, the low acute efficacy for both G protein and β -arrestin engagement of exendin-F1 is still sufficient to fully activate cAMP/PKA signalling either due to either signal amplification or differences in capacity to induce activation of recruited G proteins, whilst at the same time avoiding β -arrestin-mediated desensitisation and downregulation phenomena that otherwise limit response duration. In contrast, the same ligand might fail to induce high amplitude responses in tissues with lower GLP-1R expression. This tissue selectivity could potentially influence the risk of side effects when used therapeutically.
After endocytosis, a number of GPCRs are known to continue to signal from the endosomal compartment [47]. Pharmacological and genetic inhibition of GLP-1R endocytosis attenuates cAMP production [48,49], and Gα s was found to colocalise with GLP-1R in early endosomes [40]. In our study, we have monitored ligand-specific patterns of redistribution of GLP-1R in its Gα s -preferring active conformation from the plasma membrane to early endosomes. Here, P5 led to more active GLP-1Rs at Rab5-positive early endosomes than exendin-F1, in line with the greater total GLP-1R internalisation recorded with this ligand. This could be important given that "location bias" can modulate agonist effects, as the same intracellular signalling events originating at the endosome versus plasma membrane may result in distinct downstream actions [50]. Endosomal signalling is often referred to as being responsible for sustained responses [51]. However, our results call into question the relative importance of this phenomenon in controlling the duration of action of GLP-1RAs in the therapeutic setting, as exendin-F1 showed the least tendency to promote GLP-1R activation at the endosomal compartment but the greatest maximal insulin secretion after prolonged incubation with beta cells, and a greater glucose-lowering effect than P5 in a delayed glucose tolerance test in mice. This is presumably attributable to a sufficiently enhanced avoidance of receptor desensitisation and downregulation, both of which ultimately limit the global capacity for signalling for the GLP-1R, to the point that any potential advantages of endosomal signalling no longer dominate. Ultimately, the balance between the positive and negative contributions of all the above-mentioned factors is likely to determine the overall capacity of each pharmacological agonist for sustained action.
Despite its reduced importance in a pharmacological setting, endosomal signalling is likely to play a more prominent role in mediating physiological GLP-1R effects, as the lower plasma concentrations and short circulatory half-life of GLP-1 and other proglucagon-derived peptides means that the fraction of downregulated GLP-1R is likely to be minimal. Moreover, it should be emphasised that our mini-G s bystander BRET assay does not measure endosomal signalling per se, but the presence of activated GLP-1R at particular locations.
Targeted biosensors to monitor cAMP generation [41] or PKA activation [52] are available, and Nb37 redistribution can be observed by microscopy [26], although these tend to be lower throughput methods.
At the outset of the present study, we anticipated that P5 would in fact be highly insulinotropic when assessed using the same methodology [9] used to reveal the impact of GLP-1R desensitisation and time-dependent advantages of exendin-F1, but this proved not to be the case. It appears that P5 achieves a less favourable balance between "beneficial" signalling and "non-beneficial" desensitisation which ultimately limits the amplitude of the sustained insulin release response, which was in fact similar to that of exendin-4. The potent effects of P5 on blood glucose lowering reported by Zhang et al [8] were observed in the "hyper-acute" setting, i.e. with a different approach to the delayed IPGTT used in our study, as well as being corroborated by improvements in HbA1c after chronic administration. Our results do not shed any light on how P5 obtains its pronounced anti-hyperglycaemic effects, which remain hard to explain as P5 was less insulinotropic than exendin-4 and had no differential effect on insulin sensitivity. An insulin-independent mechanism remains a possibility but was not explored in the study of Zhang et al [8]. Chronic treatment with P5 led to various metabolic changes that exceed those of the same dose of exendin-4, including increased adipose tissue hyperplasia and reduced inflammation, along with increased circulating GIP and reduced circulating resistin [8]. These may be relevant in the chronic setting but unlikely to contribute to the hyperacute effects of P5 on glucose levels.
In summary, whilst P5 and exendin-F1 superficially resemble each other at the pharmacological level, our study highlights that emphasising lower acute efficacy for transducer coupling, but higher affinity or potency, may be a more viable strategy to achieve therapeutically optimised GLP-1R biased agonism.