Impact of N-terminally substituted glucagon family receptor agonists on signal bias, trafficking and downstream responses

Receptors for the peptide hormones glucagon-like peptide-1 (GLP-1), glucose-dependent insulinotropic polypeptide (GIP) and glucagon (GCG) are important regulators of insulin secretion and energy metabolism. Here we sought to investigate how signal bias between cyclic AMP and βarrestin-2 recruitment can modulate the effects of prolonged agonist stimulation at each of these receptors. We generated analogues of GLP-1, GCG and GIP which in some cases showed selective reduction in β-arrestin-2 recruitment versus cAMP signalling compared to the parent peptide. Despite reduced acute signalling potency and/or efficacy, some biased GLP-1 and GIP analogues increased maximal sustained insulin secretion, although only at high agonist concentrations. Biased GCG analogues did not affect maximal insulin release from beta cells, or glucose output in hepatocytes.


Introduction
The receptors for the glucagon-like peptide-1 (GLP-1R), glucose-dependent insulinotropic polypeptide (GIPR) and glucagon (GCGR) are major pharmacological targets in metabolic diseases such as type 2 diabetes (T2D) and obesity [1]. Each of these receptors is present on pancreatic beta cells, and an important component of their overall metabolic actions when physiologically or pharmacologically activated is augmentation of glucose-stimulated insulin release [2,3]. In hepatocytes, GCGR facilitates glucose output, which may be undesirable in T2D; however, its "energy wasting" effect in peripheral tissues [2] could mitigate hyperglycaemia by weight loss and associated improvements in insulin sensitivity.
GLP-1R, GIPR and GCGR are closely related G protein-coupled receptors (GPCRs) in the class B (secretin) family. When activated, they engage the G protein Gαs, which is coupled to insulin secretion via generation of cyclic adenosine monophosphate (cAMP) [4], and β-arrestins, which are multifunctional scaffold proteins widely reported to initiate non-G protein signalling cascades such as phosphorylation of mitogen-activated protein kinases (MAPKs) [5], and concurrently diminish G protein signalling by steric hindrance [6].
The balance between G protein and β-arrestin recruitment can be ligand-specific even at the same receptor -a pharmacological concept known as "biased signalling" [7]. A number of examples of signal bias at the GLP-1R have been described for naturally occurring [8,9] and pharmacological [10,11] orthosteric agonists. We recently described how biased GLP-1R agonists derived from exendin-4, characterised by reduced recruitment of β-arrestins, lead to increases in sustained insulin secretion through avoidance of GLP-1R desensitisation, reduction of GLP-1R endocytosis and resultant attenuation of GLP-1R downregulation over pharmacologically relevant time periods [12,13].
In view of the current drive to develop incretin analogues jointly targeting GLP-1R, GCGR and GIPR [1], we sought to establish whether signal bias could similarly be achieved at the latter two receptors, and to determine if this is associated with prolonged signalling responses, as seen at the GLP-1R.
In the present study, we report that a number of substitutions close to the N-terminus of the cognate ligand for each receptor result in reductions in both cAMP signalling and β-arrestin-2 recruitment, with bias in favour of cAMP in some cases. However, compared to our previous study with biased exendin-4 analogues at GLP-1R [12], the degree of bias achieved was more modest. Moreover, whilst biasrelated differences were apparent in downstream responses such as insulin secretion, these were only apparent at high agonist concentrations.

Peptides
All peptides were obtained from Insight Biotechnology and were at least 90% pure.

Transfections
Transient transfections of plasmid DNA containing the SNAP-tagged human GLP-1R, GIPR or GCGR (Cisbio) were performed using Lipofectamine 2000 according to the manufacturer's recommendations; unless indicated, cells were assayed 24 hours after DNA transfection.

Cyclic AMP assays
All experiments were performed at 37°C. PathHunter CHO-K1 cells: Cells were resuspended in serum-free Ham's F12 medium and treated with indicated agonist, without phosphodiesterase inhibitors, for 30 minutes, followed by application of detection reagents for determination of cAMP by HTRF (cAMP Dynamic 2, Cisbio). β-arrestin-2 recruitment responses (section 2.6) were measured in parallel. Huh7-GCGR cells were treated with indicated concentration of agonist without phosphodiesterase inhibitors before lysis. Curve fitting was performed using Prism 8.0 (Graphpad Software).

β-arrestin recruitment assay
β-arrestin-2 recruitment was determined by enzyme fragment complementation using the PathHunter system (DiscoverX). CHO-K1 cells expressing GLP-1R, GIPR or GCGR were treated with indicated concentrations of agonist for 30 minutes at 37°C before addition of detection reagents.

Measurement of receptor internalisation by DERET
Diffusion-enhanced resonance energy transfer (DERET) [18] was used to monitor agonist-induced receptor internalisation in transiently transfected HEK293T cells (24 hours after DNA transfection of N-terminally SNAP-tagged receptors), or in monoclonal stably transfected CHO-K1 cells. Labelling was performed using the time-resolved Förster resonance energy transfer (TR-FRET) SNAP-probe Lumi4-Tb (Cisbio) at 40 nM for 60 minutes at room temperature, either in suspension (for HEK293T) or with adherent cells (for CHO-K1). After washing three times, fluorescein (24 µM in HBSS) was added to cells in opaque bottom white plates, and baseline signal was read for 10 minutes using a Flexstation 3 plate reader (λex 340 nm, λem 520 and 620 nm, delay 400 µs, integration 1500 µs) at 37°C. Agonists, prepared in 24 µM fluorescein, were added, and signal was sequentially monitored.
Receptor endocytosis leads to reduced contact of Lumi4-Tb with extracellular fluorescein, and a reduction in signal at 520 nm with an increase at 620 nm. After first subtracting values from wells containing fluorescein only, internalisation was expressed ratiometrically as signal obtained at 620 nm divided by that obtained at 520 nm.

Measurement of receptor internalisation using a cleavable SNAP-labelling probe
The assay was adapted from a previous description [13]. HEK293T cells were seeded in black, clear bottom, plates coated with 0.1% poly-D-lysine and assayed 24 hours after transfection with plasmid DNA. Cells were labelled with the cleavable SNAP-tag probe BG-S-S-649 (featuring the DY-649 fluorophore, a gift from New England Biolabs) in complete medium for 30 minutes at room temperature. After washing, fresh medium ± agonist was added, with agonists applied in reverse time order in the case of time-course experiments. At the end of the incubation, medium was removed and wells were treated with for 10 minutes at 4°C with Mesna (100 mM, in alkaline TNE buffer, pH 8.6), or with alkaline TNE buffer alone, to remove BG-S-S-649 bound to residual surface receptor without affecting the internalised receptor population. After washing, cells were imaged using an automated Nikon Ti2 widefield microscope with LED light source (CoolLED) and a 0.75 numerical aperture 20X air objective, assisted by custom-written high content analysis software [19] implemented in Micro-Manager [20]. A minimum of 4 epifluorescence and matching transmitted phase contrast images per well were acquired. Average internalised receptor across the imaged cell population was quantified using Fiji as follows: 1) phase contrast images were processed using PHANTAST [21] to segment cell-containing regions from background; 2) illumination correction of fluorescence images was performed using BaSiC [22]; 3) fluorescence intensity was quantified for cell-containing regions.
Agonist-mediated internalisation was determined by comparing 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.

Gene expression analysis
RNA was harvested from INS-1 832/3 and Huh7 cells using the Cells-to-CT kit (Themo Fisher).

Insulin secretion
Insulin secretion from INS-1 832/3 cells [14] was assayed after a prior overnight period of exposure to low glucose (3 mM) complete medium. Cells were added in suspension to plates containing indicated agonists, prepared in RPMI containing 2% FBS and 11 mM glucose, for 16 hours.
Supernatant insulin concentration was determined by HTRF (High Range Insulin kit, Cisbio). Results were normalised to those obtained with 11 mM glucose but no additional agonist. Three-parameter fitting was performed using Prism 8.0.

Glucose production
Overnight serum-starved hepatocytes were washed and glucose production medium (phenol-red free DMEM containing 1mM pyruvate and 20mM lactate) was applied. After 24 hours exposure to agonist, media was assayed for glucose content using a glucose oxidase method (GLUC-PAP, Randox).
Glucose content was normalised to protein content of each well (BCA assay).

Data analysis and statistics
All analyses were performed using Prism 8.0. For bias calculations, to reduce contribution of interassay variability, cAMP and β-arrestin-2 assays were performed concurrently, with the same incubation time of 30 minutes to avoid artefactual bias resulting from different activation kinetics of each pathway [23]; bias was determined by calculating transduction coefficients [24,25]; here, due to the matched design of our experiments, we calculated ∆∆log(τ/KA) on a per-assay basis by normalising the log(τ/KA) of each ligand to the relevant endogenous ligand (GLP-1, GIP or GCG, to generate a ∆log[τ/KA] value) and then to the reference pathway (cAMP). All experiments had a matched design, allowing paired, two-tailed t-tests, or randomised block one-way ANOVA to be performed. Specific statistical tests are indicated in the figure legends. Statistical significance was inferred when p<0.05. To determine statistical significance for signal bias, 95% confidence intervals were calculated; bias versus the reference endogenous ligand was considered statistically significant when this confidence interval did not cross zero, as previously recommended [24].

Effects of N-terminal region mutations to GLP-1, GIP and glucagon on signalling
Biased signalling, in which agonists preferentially stabilise certain receptor conformations to engage specific intracellular signalling pathways, may provide a means to selectively increase therapeutic effects [26]. As the ligand N-terminal region plays a key role in activation of class B GPCRs [27] and is linked to GLP-1R biased signalling [12], we introduced single amino acid substitutions at or close to the N-termini of each endogenous ligand (Table 1 for full amino acid sequences) and tested for intracellular cAMP production ( Figure 1A) and β-arrestin-2 recruitment ( Figure 1B). For each receptor target, all N-terminally modified analogues retained full efficacy for cAMP, but with reduced potency (full parameter estimates are given in Table 2). Potency for β-arrestin-2 recruitment was also reduced, and in the majority of cases, a reduction in efficacy was also observed. Transduction ratios [25] were calculated to quantify the relative signalling impact of each N-terminal region substitution in each ligand for each pathway ( Figure 1C, D). Chiral substitution of the first amino acid to dHis1 (GLP-1, GCG) or dTyr1 (GIP), as well as Gly2 and dGln3, were less well tolerated by GIP and GCG than by GLP-1; for example, GIP-dGln3 showed a 100-fold reduction in cAMP potency compared to wild-type GIP, versus a 10-fold reduction seen for GLP-1-dGln3 compared to GLP-1. Moreover, comparison of the relative impact of each substitution on cAMP versus β-arrestin-2 responses showed that all compounds tested exhibited at least a trend for bias in favour of G protein-dependent cAMP signalling, albeit not statistically significant (as indicated by 95% confidence intervals crossing zero) in some cases ( Figure 1E). The large error bars for the bias estimate for GIP-dGln3 reflect the limitations of this method for bias calculation for extremely weak partial agonists [28].
Overall, these results highlight how the N-termini of each ligand play important roles in receptor activation. However, it should be noted that the magnitude of signal bias with the GLP-1 analogues tested here is smaller than for exendin-4-derived biased GLP-1R agonists, a finding that is consistent with our recent exploration of GLP-1/exendin-4 chimeric peptides [13].

Receptor endocytosis induced by biased incretin receptor agonists
β-arrestin recruitment is classically linked to GPCR endocytosis [29], although conflicting evidence exists for a direct role in controlling trafficking of incretin receptors [30][31][32][33][34]. Nevertheless, a clear correlation between β-arrestin recruitment and GLP-1R internalisation with biased GLP-1R agonists has been demonstrated [12]. We therefore used diffusion-enhanced resonance energy transfer (DERET) [35] to monitor agonist-induced loss of surface labelled SNAP-tagged receptors transiently expressed in HEK293T cells. Robust internalisation was noted for GLP-1R stimulated with GLP-1. In contrast, GIPR internalisation with GIP was less extensive, and no change in DERET signal could be detected with GCGR stimulated by GCG (Figure 2A). Endocytic profiles were confirmed using an alternative approach based on reversible SNAP-tag labelling, in which the fluorescent probe BG-S-S-649 is cleaved from residual surface receptors after agonist-induced internalisation using the cellimpermeant reducing agent Mesna [13,36]. A time course study showed rapid and extensive loss of surface SNAP-GLP-1R when cells were treated with GLP-1, whereas SNAP-GIPR and SNAP-GCGR internalisation with their cognate agonists were, respectively, more limited and virtually absent ( Figure   2B). Representative images showing the effect of Mesna cleavage are shown in Figure 2C.
We compared GLP-1R and GIPR internalisation induced by the N-terminally substituted agonists described in Figure 1, finding that in all cases receptor internalisation was reduced when measured by DERET using stably transfected CHO-K1 cells expressing SNAP-GLP-1R or SNAP-GIPR ( Figure   2D, E). The reversible SNAP-tag labelling assay showed a similar pattern ( Figure 2F). The GCG analogues were not studied with this assay as GCG itself was without effect.

Responses in the beta cell context
As pancreatic beta cells are a target for GLP-1, GIP and GCG [37], we investigated whether the signalling and trafficking characteristics of N-terminally modified agonists described above could enhance insulin secretion, as previously demonstrated for biased exendin-4-derived GLP-1RAs [12].
We used incretin-responsive rat insulinoma-derived INS-1 832/3 cells [14], in which we first confirmed expression of GLP-1R, GIPR and GCGR by qPCR ( Figure 3A). Using N-terminally substituted GLP-1 and GIP analogues, we found subtly increased maximal sustained insulin secretion with a number of analogues compared to their respective parent ligand, but this was only apparent at concentrations above 1 µM ( Figure 3B, C, Table 3). The majority of analogues displayed reduced potency for insulin secretion, as they had for acute cAMP production in the signalling assays presented in Figure 1, except for GLP-1-gly2, which is likely to reflect protection against DPP-4-mediated degradation with this ligand. For GIP analogues, maximal insulin secretion was correlated with maximal β-arrestin-2 recruitment; for GLP-1 analogues the relationship was less clear ( Figure 3D). As GCG can crossreact with GLP-1R in beta cells [38], we tested each N-terminally modified GCG analogue in both wild-type INS-1 832/3 cells and a sub-clone in which GLP-1R had been knocked out by CRISPR/Cas9 [15]. This showed that the overall response was dominated by GLP-1R-dependent high dose effects absent in GLP-1R knockout cells, with no clear GCGR-dependent advantageous effect for any analogue ( Figure 3E).
Overall, these results indicate that analogues of GLP-1 and GIP with reduced β-arrestin-2 recruitment can augment glucose-stimulated insulin secretion, but with the peptides evaluated in this study, this effect was only apparent at high agonist concentrations.

Responses to biased GCG analogues in the hepatocyte context
GCG stimulates glycogenolysis and gluconeogenesis in hepatocytes. Recently, ablation of β-arrestin-2 in hepatocytes was found to increase hepatic glucose output in response to GCG [33]. We used Huh7 cells stably expressing GCGR ("Huh7-GCGR") [16] to assess responses to prolonged stimulation with biased GCG analogues, to determine if differences in β-arrestin recruitment could affect sustained GCGR signalling in a hepatocyte context. Comparisons of maximum cAMP accumulation after 16-hour stimulation with each ligand revealed that greater efficacy for cAMP production was achieved by ligands with reduced β-arrestin-2 recruitment ( Figure 4A, Table 3); and in the case of the -Phe1 and -dGln3 compounds, cAMP potency was also greater than for GCG, despite being at least 10-fold less potent acutely (see Figure 1). However, this did not translate to differential changes in GCGR-induced upregulation of the gluconeogenic enzyme glucose-6-phosphatase (G6P; Figure 4B); neither were any significant differences seen for production of glucose in primary mouse hepatocytes ( Figure 4C).
This study builds on our earlier work using biased GLP-1R agonists derived from exendin-4 and GLP-1 bearing amino acid substitutions close to the N-terminus [12,13]. In the former study [12], we found that a number of changes, including dHis1, dTyr1, Phe1 and dGln3 (as included in the present investigation) led to marked reductions in recruitment of β-arrestin-1 and -2, substantially reduced GLP-1R endocytosis, increased insulin secretion, and improved anti-hyperglycaemic effect in mice.
In the latter study [13] we observed that the Phe1 substitutions led to less marked improvements in insulin secretion in a GLP-1 versus exendin-4 context. We therefore sought to expand this concept to the related class B GPCR targets in the present study, which are structurally and functionally related to the GLP-1R and of interest as therapeutic targets in metabolic disease [39]. Overall, our new data indicate that amino acid substitutions at or close to the N-termini of GLP-1, GIP and GCG can indeed diminish β-arrestin-2 recruitment efficacy, with a somewhat lesser effect on cAMP signalling, but that the degree of pathway selectivity is reduced compared to what we have previously observed with exendin-4 analogues at the GLP-1R [12]. Moreover, the impact on prolonged insulin secretion in pancreatic beta cells was more limited.
GLP-1 is more dependent on its N-terminus for binding to the GLP-1R than is exendin-4 [40].
Sequential truncation of the first nine amino acids of exendin-4 results in only a modest reduction in binding affinity, but virtually abolishes binding of GLP-1 [41,42]. Thus, the reduction in signalling potency resulting from an N-terminal amino acid substitution within the GLP-1 backbone may be secondary to reduced affinity, whereas within exendin-4, the same change might have little impact on occupancy, thereby allowing the modified ligand to achieve biased signalling at higher potency.
Verifying this concept, acute cAMP signalling potencies for exendin-dHis1 and exendin-Phe1 in our earlier study were, respectively, no different to and 2.5-fold lower than for exendin-4 [12]. In contrast, in the present work, the same substitutions to the GLP-1 N-terminus reduced cAMP potency by, respectively, a factor of 6 and 10. This might limit the potential for these modified GLP-1 analogues to improve downstream signalling outputs during prolonged stimulation, except at maximal doses when receptor occupancy is high. As for GLP-1, the N-termini of both GCG and GIP are also known to play a major role in affinity for their cognate receptors, with truncation of the terminal amino acid residue resulting in a >10-fold loss of affinity in both cases [43,44].
A number of the analogues tested in this report have previously been described, due in part to the interest in reducing ligand sensitivity to degradation by the N-terminal targeting exopeptidase DPP-4 [45]. Published potency or affinity measures for GLP-1-dHis1 [46] and -Gly2 [47] were broadly in agreement with our results, although GLP-1-Phe1 was found to be well tolerated for cAMP signalling potency in RIN-T3 cells [48], contrasting with the deleterious effect we observed. Differences in cell type, receptor species, incubation times and other factors may influence responses to agonists, complicating direct comparisons with reported values in the literature [23]. The affinity of GIP-dTyr1 was reduced 10-fold compared to unmodified GIP [49], similar to our results, whereas the -Gly2 substitution was well tolerated [50]. GIP-Phe1 has been used as a GIPR I 125 -radioligand [51]. These datasets are complemented here by our measures of bias between cAMP and β-arrestin recruitment, and endocytosis, which have not previously been reported for these ligands, or indeed for any putative biased GCGR or GIPR agonists to our knowledge.
A further factor which might contribute to the relative lack of effect on downstream responses to biased GLP-1, GIP and GCG analogues during prolonged incubations is enzymatic peptide degradation, for example by neutral endopeptidase 24.11 (NEP 24.11), found on pancreatic beta cell membranes and capable of hydrolysing GLP-1, GCG, and to a lesser extent, GIP [52], or endothelin converting enzyme-1 [53] situated predominantly in endosomal compartments. DPP-4, also expressed by beta cells [54], is also likely to contribute, although the modified N-termini of the ligands tested may confer some resistance to its action. Sequence optimisation to increase proteolytic stability during our extended in vitro studies may be required to maintain adequate ligand concentration to fully manifest consequences of signal bias. In the in vivo setting, fatty acid conjugation e.g. liraglutide [55] protects against NEP 24.11 and DPP-4 degradation, presumably as the resultant albumin-bound form of the ligand is inaccessible to the enzymes. One possible future approach would be to test acylated forms of the ligands described herein to determine if sustained exposure to the N-terminally substituted forms led to enhanced metabolic effects.
In summary, we demonstrate in this study that GLP-1, GIP and GCG analogues with a variety of Nterminal substitutions typically show reduced β-arrestin-2 recruitment. In the case of GLP-1 and GIP, this is associated with reduced receptor endocytosis, and this effect can be exploited to increase maximal insulin release in vitro. Generation of long-lasting biased incretin mimetics will be required to determine whether this applies in vivo.

Table 3. Responses to N-terminally substituted ligands in INS-1 832/3 and Huh7-GCGR cells.
Parameter estimates ± SEM from insulin secretory responses depicted in Figure 3F-H and 16-hour cAMP accumulation in Figure 4A. Note that the Hill slopes for GCG analogues in Huh7-GCGR cells are derived from the curves plotted from the pooled data, so no statistical comparisons are shown. * p<0.05, by one-way randomised block ANOVA with Dunnett's test vs. GLP-1, GIP and glucagon, as appropriate. "n.c." indicates not calculated.    Table 2.    Table   3. Parameter estimates for concentration responses are provided in Table 3.