Mapping and testing cholesterol binding sites on GLP-1 receptors in insulin-producing pancreatic cells

To determine the effect of increased cholesterol levels on GLP-1R function, we fed mice a standard chow diet supplemented or not with cholesterol for 5 wk. We then performed intraperitoneal glucose tolerance tests (IPGTTs) post-administration of vehicle or 1 nmol/kg exendin-4 to determine the glucose lowering effect of pharmacologically targeting the receptor under these conditions. Mice on either diet had similar glucose responses under vehicle conditions; however, those on high cholesterol diet presented worse glucose lowering responses to exendin-4 vs chow-fed mice (Figure 1A and B), an effect that was maintained when normalised to the corresponding vehicle control (Figure 1C). Islets isolated from high cholesterol-fed mice displayed a near-significant increase in cholesterol (Figure 1—figure supplement 1A) accompanied by decreased exendin-4-induced cAMP responses ex vivo (Figure 1—figure supplement 1B), highlighting the negative impact of even mild increases in cholesterol on GLP-1R signalling.

To more directly evaluate the effect of increased islet cholesterol content on GLP-1R function, islets isolated from wild-type (WT) mice were loaded ex vivo with MβCD saturated with cholesterol, which led to a significant increase in cholesterol content (Figure 1D). Under these conditions, we measured a decrease in insulin secretion in response to exendin-4 vs vehicle conditions, primarily due to elevated basal secretion levels at 11 mM glucose (Figure 1E and F), confirming that increased cholesterol levels reduce the GLP-1R glucoregulatory potential by decreasing its capacity to potentiate islet insulin secretion. This lack of proportional secretory increase following exendin-4 stimulation of cholesterol-loaded islets was not related to an overall saturation of the secretory response, as the percentage of insulin secretion was further increased in both vehicle and cholesterol-loaded islets by stimulation with a secretagogue cocktail (Figure 1—figure supplement 1C). These effects were accompanied by a decreased capacity for exendin-4 to reduce GLP-1R plasma membrane diffusion (Figure 1—figure supplement 1D, E), a behaviour observed here and previously (Manchanda et al., 2023; Pickford et al., 2020) by Raster Image Correlation Spectroscopy (RICS) in rat insulinoma INS-1 832/3 cells without endogenous GLP-1R [INS-1 832/3 GLP-1R KO cells (Naylor et al., 2016)] stably expressing SNAP/FLAG-tagged human GLP-1R (SNAP/FLAG-hGLP-1R).

We next performed the reverse experiment by decreasing islet cholesterol levels with simvastatin, a long-established hydroxy-methylglutaryl coenzyme A reductase (HMGCR) inhibitor (Pedersen and Tobert, 2004). Filipin staining confirmed decreased islet cholesterol content (Figure 1G), which correlated with increased exendin-4-induced cAMP responses (Figure 1—figure supplement 1F), and exendin-4-induced insulin secretion (Figure 1H, I) in simvastatin-exposed islets. Contrary to cholesterol loading, simvastatin triggered decreased GLP-1R plasma membrane diffusion under vehicle conditions, measured by RICS as above (Figure 1—figure supplement 1G, H), highlighting the importance of cholesterol in determining GLP-1R movement at the plasma membrane. Taken together, these results demonstrate that changes in cholesterol levels have specific effects on beta cell GLP-1R behaviours, indicating the potential existence of direct interactions between the receptor and this plasma membrane-enriched lipid.

To gain insights into the precise location of cholesterol interactions with the GLP-1R, cgMD simulations were employed to study both active (corresponding to agonist bound) and inactive (apo-) states of GLP-1R in a model mammalian plasma membrane using the latest Martini 3 forcefield and cholesterol parameters (Figure 2A). The PyLipID package (Song et al., 2022) was employed to characterise cholesterol interactions for each receptor state. Two key parameters were measured: occupancy, defined as the percentage of frames in the trajectory where any cholesterol molecule is within the cutoff distance from the receptor, and interaction residence time, defined as the time that a cholesterol molecule remains in the binding site, providing information about the strength of the interaction (full data included in Supplementary file 1 and Supplementary file 2). Occupancy values for each receptor residue in both states were calculated and displayed as heatmaps, with the top 10 residues per state indicated in Figure 2B. The receptor active and inactive states exhibited similar top occupancy residues, with hydrophobic residues such as Phe, Leu, Ile, and Val enriched in both states. A generally similar profile was observed for the top 30 highest occupancy residues in active vs inactive states, with some subtle differences noted in the receptor transmembrane domains TM2 and TM6 (Figure 2C).

The top three cholesterol binding sites, displaying the highest cholesterol residence times for both states, were calculated using the PyLipID package (Figure 2D and E). Interestingly, notable differences were observed in these sites between the two receptor states: for the active state, site I (purple) was located in the upper TM1 and TM7 region, while in the inactive state, site I (cyan) was found in the lower TM5 and TM6 region, an area which undergoes key structural changes upon G protein engagement and GLP-1R activation (Zhang et al., 2017) and might, therefore, represent a state-specific cholesterol binding site. Site II (orange) was similar for both states but presented subtle differences, with an additional cholesterol pocket involving the Trp in extracellular loop 2 (ECL2) being more pronounced in the inactive vs the active state. Site III (green) showed some degree of variation between the two states; in the inactive state, two TM5 residues (Val319 and Phe315) formed an additional cholesterol pocket with TM4, presenting the highest average residence time (186 nanoseconds) across all sites, while, in the active state, the cholesterol pocket was located closer to the bottom of TM3 and TM4, involving Leu260 in ECL2.

To evaluate the functional relevance of the GLP-1R cholesterol binding sites identified during the cgMD simulations, 12 residues were selected within or at the periphery of these sites and mutated to alanine by site-directed mutagenesis (Figure 2F). Alanine mutations for four of the selected residues (Ile323, Val160, Gly151, and His363) were predicted to cause changes in the GLP-1R structure in the active state by altering the cavity or causing changes in exposed residues and breakages in salt bridges between residues, while a further three alanine mutated residues (Val229, Gly151, and His363) were predicted to cause changes in the inactive GLP-1R structure (Figure 2G). SNAP/FLAG-tagged human GLP-1Rs (SNAP/FLAG-hGLP-1Rs) with single alanine substitutions for each selected residue were successfully expressed in INS-1 832/3 GLP-1R KO cells with no significant differences detected in cell surface expression levels vs WT GLP-1R (Figure 2H, top graph). Mutant receptors were then screened for their capacity to undergo exendin-4-induced internalisation, as this GLP-1R property is severely inhibited following cholesterol extraction with MβCD (Buenaventura et al., 2019). Four alanine mutated residues, V229A, V160A, I165A, and L401A, caused a significant decrease in GLP-1R internalisation vs WT receptor, with GLP-1R V229A and V160A causing a ~40%, and I165A and L401A a ~30% decrease in internalisation (Figure 2H, bottom graph). GLP-1R V229A was selected for further detailed analysis of the effects of modifying a GLP-1R-cholesterol binding site residue on GLP-1R beta cell behaviours.

We next generated INS-1 832/3 GLP-1R KO sublines stably expressing SNAP/FLAG-hGLP-1R WT or V229A. As expected, no difference in surface expression was detected between WT and V229A receptors, although we observed a non-significant tendency for reduced surface levels with the V229A mutant (Figure 3A and B). To determine the effect of the V229A substitution on the overall GLP-1R binding to cholesterol in active vs inactive conditions, cells from both sublines were labelled with PhotoClick cholesterol and exposed to UV light prior to SNAP/FLAG-hGLP-1R purification by anti-FLAG immunoprecipitation and click chemistry to fluorescently label covalently bound PhotoClick cholesterol (Figure 3C). Resulting samples were separated by SDS-PAGE and gels imaged under fluorescence followed by anti-SNAP Western blotting (Figure 3—figure supplement 1A). Quantification of GLP-1R-bound cholesterol in both WT and V229A receptors following vehicle vs acute (2 min) exendin-4 stimulation showed a significant decrease in cholesterol binding to GLP-1R V229A in vehicle conditions without modifying cholesterol binding after agonist stimulation compared to GLP-1R WT (Figure 3D). Co-localisation of SNAP/FLAG-hGLP-1R with cholesterol in the same cells using the fluorescent D4H*-mCherry probe to label plasma membrane cholesterol at the cytosolic leaflet (Lim et al., 2019) corroborated our previous results as we measured a significant decrease in co-localisation between GLP-1R V229A and cholesterol under vehicle conditions compared to WT GLP-1R (Figure 3E–G).

While we could not measure any changes in receptor conformational shift following exendin-4 binding between GLP-1R WT and V229A (Figure 3—figure supplement 1B, C), or differences in binding affinity to the fluorescent exendin-4 derivative exendin-asp3-TMR (Figure 3—figure supplement 1D), quantification of GLP-1R plasma membrane diffusion using RICS in INS-1 832/3 SNAP/FLAG-hGLP-1R WT vs V229A cells revealed a pronounced reduction in GLP-1R V229A diffusion compared to WT GLP-1R under vehicle conditions (Figure 4A and B), to a similar extent to the reduction caused by exendin-4 binding to GLP-1R WT, recapitulating the effect triggered by reduced cholesterol levels following simvastatin exposure (Figure 1—figure supplement 1G, H). Single particle tracking (SPT) of human GLP-1R fused to monoclonal EGFP (hGLP-1R-mEGFP) WT vs V229A imaged by total internal reflection fluorescence (TIRF) microscopy also showed reduced total displacement and speed (displacement over time) of V229A vs WT receptors under vehicle conditions, with these parameters also decreased following WT GLP-1R stimulation with exendin-4 (Figure 4C). Finally, exendin-4-induced activation of either WT or V229A GLP-1R tended to reduce the diffusion of lipid-ordered nanodomains at the plasma membrane, measured via RICS analysis of generalised polarisation (GP) values obtained from the environmentally sensitive lipid probe Laurdan (Brewer et al., 2010; Figure 4D), with no differences in Laurdan GP lateral diffusion in cells expressing WT vs V229A receptors. However, RICCS cross-correlation analysis of lipid-ordered nanodomain co-diffusion with GLP-1R WT vs V229A showed a reduction in the movement of GLP-1R-lipid nanodomain complexes for WT receptors after exendin-4 stimulation, with a tendency for V229A receptors to exhibit the same behaviour already under vehicle conditions, resulting in no significant change for GLP-1R V229A-lipid nanodomain complex diffusion between vehicle and exendin-4-stimulated conditions (Figure 4E, Figure 4—figure supplement 1).

Number and Brightness (N&B) analysis of the RICCS cross-correlation data was then used to determine the effect of the V229A substitution on receptor oligomerisation before and after exendin-4 stimulation (Figure 5A). Both WT and V229A receptors existed predominantly as monomers or dimers at a proportion of ~60 to 80% of total receptors, a percentage that was reduced for the WT receptor at later acquisition times under vehicle conditions, but with otherwise few changes following exendin-4 stimulation. Around 30% of inactive WT or V229A receptors existed as dimers to hexamers or hexamers to decamers, with a tendency for the dimer to hexamer population to reduce and the hexamer to decamer population to increase after exendin-4 stimulation, and for GLP-1R V229A to have a higher proportion of hexamers to decamers under vehicle conditions vs WT receptors. A smaller proportion of receptors existed as higher order oligomers, with a significant increase measured for the WT GLP-1R after stimulation with exendin-4. For the GLP-1R V229A, however, we observed a higher proportion of receptors pre-existing as higher order oligomers under vehicle conditions, with no further increases after exendin-4 stimulation, indicating that the mutant receptor has a higher clustering propensity in the inactive state. Consistent with this finding, GLP-1R V229A also presented with significantly increased levels of receptors segregated to lipid rafts (Lorizate et al., 2021) prior to exendin-4 stimulation vs GLP-1R WT (Figure 5B).

To determine if the V229A substitution affected receptor recruitment to clathrin-coated pits (CCPs) after agonist stimulation, TIRF images were acquired of both unstimulated and exendin-4-stimulated INS-1 832/3 SNAP/FLAG-hGLP-1R WT vs V229A cells (Figure 5C). Before agonist addition, the SNAP/FLAG-hGLP-1R signal was distributed across the plasma membrane and only weakly associated with clathrin-positive regions (ΔF/S=0.03±0.003 and 0.04±0.006 for WT and V229A, respectively), with no differences between WT and V229A receptors. However, after exendin-4 stimulation, the receptor association with clathrin-positive regions increased, an effect that correlates with the known rapid GLP-1R internalisation triggered by exendin-4 (Buenaventura et al., 2019), with this increase, however, significantly less pronounced for the GLP-1R V229A compared to WT (ΔF/S=0.16±0.006 and 0.10±0.004 for WT vs V229A, respectively), in agreement with the reduced internalisation propensity displayed by this mutant receptor (Figure 2H).

To determine the effects of the GLP-1R V229A substitution on GLP-1R trafficking, we assessed GLP-1R internalisation, recycling, and degradation parameters in INS-1 832/3 SNAP/FLAG-hGLP-1R WT vs V229A cells. GLP-1R V229A internalised more slowly than WT receptor after exendin-4 stimulation (Figure 6A–C), with only ~10% of V229A internalised after 5 min stimulation vs ~30% for WT. V229A internalisation eventually recovered after 30 min of exendin-4 exposure, reaching ~50% for both receptor types. In contrast, plasma membrane recycling of internalised GLP-1R V229A was increased vs WT at both 1 hr and 3 hr recycling periods (Figure 6D and E), while no changes in exendin-4-induced GLP-1R degradation were detected between GLP-1R WT and V229A (Figure 6F and G).

Next, we performed NanoBiT complementation assays to determine the effect of the GLP-1R V229A substitution on receptor coupling to G proteins and β-arrestins, assessing mini-G_s, mini-G_q, and β-arrestin 2 recruitment dose response curves to WT vs V229A receptors after stimulation with a range of exendin-4 doses. Results showed a significant increase in both the potency and efficacy of mini-G_s recruitment to V229A vs WT GLP-1R (Figure 7A). Conversely, mini-G_q recruitment was not changed between WT and V229A GLP-1R (Figure 7B). Additionally, GLP-1R V229A caused a small but consistent increase in β-arrestin 2 recruitment compared to GLP-1R WT (Figure 7C). Bias calculations between mini-G_s and β-arrestin 2 recruitment showed that GLP-1R V229A is Gα_s-biased vs WT GLP-1R (Figure 7D). We next performed bystander complementation assays to measure the recruitment of nanobody 37 (Nb37)-SmBiT, which recognises and binds to active Gα_s, to the plasma membrane (labelled with KRAS CAAX motif-LgBiT), or to endosomes (labelled with Endofin FYVE domain-LgBiT) in response to GLP-1R stimulation with exendin-4 (see Figure 7—figure supplement 1 for a schematic of the assay). GLP-1R V229A displayed increased activity at the plasma membrane (Figure 7E), with a tendency for increased activity also from endosomes without reaching statistical significance (Figure 7F).

To determine differences between GLP-1R WT and V229A beta cell downstream signalling, cAMP and insulin secretion responses to exendin-4 were investigated in INS-1 832/3 SNAP/FLAG-hGLP-1R WT and V229A. GLP-1R V229A displayed a tendency for increased exendin-4-induced cAMP AUC, with kinetic responses that appeared delayed but reached significantly higher maximal cAMP peak levels (Figure 8A and B). Compared to WT receptor, GLP-1R V229A caused a significant decrease in basal insulin secretion under vehicle (11 mM glucose alone) conditions (Figure 8C), which resulted in significantly increased insulin secretion fold-responses to exendin-4 (Figure 8D). We then determined the effect of GLP-1R V229A in primary mouse islets. To this end, SNAP/FLAG-hGLP-1R WT or V229A-expressing adenoviruses were transduced into islets extracted from mice lacking endogenous GLP-1R expression (GLP-1R KO, generated in house). Similarly to the stable INS-1 832/3 sublines, there was no difference in surface expression of GLP-1R WT vs V229A in these islets, although again the V229A mutant receptor displayed a tendency for reduced cell surface levels (Figure 8E, top row and F). GLP-1R V229A also caused a significant decrease in receptor internalisation vs WT GLP-1R following exendin-4 stimulation (Figure 8E, bottom row and G). Expression of GLP-1R V229A was again associated with significantly improved insulin secretion fold-increases to exendin-4, but in this case, no changes were detected under 11 mM glucose alone conditions, with the improved fold being a consequence of enhanced insulin secretion vs WT GLP-1R during the exendin-4 stimulation period (Figure 8H, I). Loading of transduced islets with excess cholesterol led to a loss of secretory responses to exendin-4 for both WT and V229A mutant GLP-1R (Figure 8—figure supplement 1A, B).

In vivo studies were carried out under the approval of the UK Home Office Animals (Scientific Procedures) Act 1986 Project Licence PP7151519 (PPL Holder: Dr A. Martinez-Sanchez, Imperial College London) at the Central Biological Services unit of Imperial College London. Animals were housed in individually ventilated cages in groups up to five adult mice in a controlled environment (temperature, 21 to 23°C; 12 hr light and 12 hr dark cycles) with free access to standard chow (Bitsi et al., 2023) or high-cholesterol diet when relevant.

C57BL/6 J (RRID:MGI:3028467↗) WT mice were fed a standard chow diet or a rodent diet with 10 kcal% fat and 2% cholesterol (high cholesterol diet; D01101902CR, Research Diets, Inc) for 5 wk (8–9 female mice per condition). Animals were fasted for 2 hr prior to intraperitoneal injection of glucose (2 g/kg)+/- exendin-4 (1 nmol/kg) and continued to fast for 6 hr when blood glucose levels were assessed from the tail vein using a Contour glucometer (Bayer) and strips at 0-, 10-, 30-, and 60 min time-points, with glucose response curves generated from these data.

Primary islets from mixed sex backgrounds were isolated from C57BL/6 J WT or GLP-1R KO mice (the latter generated in house by CRISPR/Cas9 deletion). Pancreata were infused post-mortem with RPMI-1640 medium, (21875091, Thermo Fisher Scientific) containing 1 mg/mL collagenase (Clostridium histolyticum) (S1745602, Nordmark Biochemicals), dissected, and incubated in a 37 °C water bath for 10 min to digest the tissue. RPMI-1640 medium with 10% foetal bovine serum (FBS) (F7524, Sigma Aldrich) was added and mixed vigorously for 20 s to stop the reaction. The dissected pancreata were subsequently washed in RPMI-1640 medium prior to islet separation using a gradient with Histopaque-1119 and –1083 (both from Sigma Aldrich). Isolated islets were transferred to non-adherent dishes containing RPMI-1640 medium with 2 mM L-glutamine and 11 mM D-glucose supplemented with 10% FBS and 1% penicillin/streptomycin (p/s) (P4458, Sigma Aldrich) in a humidified incubator at 37 °C and 5% CO₂. Islets were allowed to recover for 24 hr prior to use (Bitsi et al., 2023).

To increase cholesterol, islets isolated from chow-fed C57BL/6 J WT mice were pre-incubated in 20 mM cholesterol loaded onto methyl-β-cyclodextrin (MβCD) (C4951, Sigma Aldrich) diluted in 1 X Krebs-Ringer bicarbonate-Hepes (KRBH) buffer [140 mM NaCl, 3.6 mM KCl, 1.5 mM CaCl₂, 0.5 mM MgSO₄, 0.5 mM NaH₂PO₄, 2 mM NaHCO₃, 10 mM Hepes, saturated with 95% O₂/5% CO₂; pH 7.4 and 0.1% bovine serum albumin (BSA)] for 1 hr at 37 °C prior to assays being carried out. Control islets were pre-incubated in KRBH buffer alone. For reduced cholesterol, islets were pre-incubated in 10 µM simvastatin (S6196, Sigma Aldrich) diluted in RPMI-1640 supplemented with 10% lipoprotein deficient serum (LPDS) (S5394, Sigma Aldrich) and 1% p/s for a minimum of 18 hr prior to the assays. Control islets were pre-incubated in RPMI-1640 supplemented with 10% FBS and 1% p/s for the same time.

INS-1 832/3 cells (RRID:CVCL_ZL55↗), a rat insulinoma subline with preserved incretin responses (Hohmeier et al., 2000), and derivative INS-1 832/3 cells with endogenous rat GLP-1R deleted by CRISPR/Cas9 [INS-1 832/3 GLP-1R KO (Naylor et al., 2016)], were grown in RPMI-1640 medium with 2 mM L-glutamine and 11 mM D-glucose, supplemented with 10% FBS, 1% p/s, 10 mM HEPES solution (H3537, Sigma-Aldrich), 1 mM sodium pyruvate (11360070, Thermo Fisher Scientific) and 0.05 mM β-mercaptoethanol (M3148, Sigma Aldrich) in a humidified incubator at 37 °C and 5% CO₂. INS-1 832/3 SNAP/FLAG-hGLP-1R WT and V229A cells were generated in house from INS-1 832/3 GLP-1R KO cells following transfection with pSNAP/FLAG-hGLP-1R (Cisbio) or pSNAP/FLAG-hGLP-1R-V229A (generated in house by site-directed mutagenesis, see below), followed by selection with 1 mg/mL G418 sulphate (A1720, Sigma Aldrich). Cells were labelled with SNAP-Surface Alexa Fluor 546 (New England Biolabs) and SNAP-expressing fluorescent cells sorted by fluorescence-activated cell sorting (FACS) and maintained as above with media supplemented with 0.5 mg/mL G418. Mycoplasma testing was performed monthly.

The structure of GLP-1R in inactive form was adopted from GPCRdb (Pándy-Szekeres et al., 2023), using refined GLP1R model PDB ID: 6LN2 (Wu et al., 2020; Zhang et al., 2021). The active form was adopted from cryoEM structure PDB ID: 6x18 (Zhang et al., 2020) with the missing loop modelled back in using MODELLER version 10.4 (Eswar et al., 2006). The GLP-1R was coarse-grained using Martinize2 (Kroon et al., 2022) with MARTINI 3 forcefield (Souza et al., 2021). The ElNeDyn elastic network restraint was applied to active and inactive GLP-1R using an elastic bond force constant of 500 kJ/mol/nm² and an upper cut-off of 0.9 nm (Periole et al., 2009). The transmembrane region of GLP-1R was predicted using PPM 3.0 Web Server (Lomize et al., 2022) and embedded into an asymmetric lipid membrane bilayer in a 18 × 18 × 17 nm³ box using insane.py (Wassenaar et al., 2015), with lipid composition as follows: POPC (30%), DOPC (30%), POPE (8%), DOPE (7%) and cholesterol (25%) in the outer leaflet and POPC (5%), DOPC (5%), POPE (20%), DOPE (20%), POPS (8%), DOPS (7%), PIP₂ (10%), and cholesterol (10%) in the inner leaflet. Mammalian plasma membrane composition was adopted from Song et al., 2019. The new MARTINI 3 lipid model for cholesterol was used in this study (Borges-Araújo et al., 2022; Borges-Araújo et al., 2023). Each system was solvated using MARTINI water (Souza et al., 2021) and 150 nM NaCl, followed by the standard minimisation and equilibration protocols from CHARMM-GUI Martini Marker (Qi et al., 2015). The production simulation was 10 microseconds in length and was repeated 3 X. The v-rescale thermostat (tau 1.0 picoseconds) (Bussi et al., 2007) and the Parrinello–Rahman barostat (tau 12.0 picoseconds) (Parrinello and Rahman, 1981) were used to maintain temperature (303.15 K) and pressure (1 bar) on all production runs. All simulations were done using GROMACS 2022.4 (Abraham et al., 2015). Cholesterol interaction profiles were calculated using PyLipID (Song et al., 2022) with a cut-off of 0.7 nm. Trajectory was analysed using gromacs tool and VMD (Humphrey et al., 1996).

Human SNAP/FLAG-tagged GLP-1R (SNAP/FLAG-hGLP-1R) (Cisbio) was used as the template to generate GLP-1R cholesterol binding mutants for screening. Site-directed mutagenesis primers (Table 2) were designed to replace the relevant amino acid residue with alanine and carried out using the PfuUltra II Fusion HS DNA Polymerase kit (600670, Agilent) according to the manufacturer's instructions. PCR products were treated with 1 µL Dpn1 (10 U/µL) for 1 hr at 37 °C to digest parental DNA. Digested DNA samples were transformed into NEB10-beta Competent E. coli High Efficiency cells (C3019H, New England Biolabs) according to the manufacturer's instructions. Mutant plasmids were sent for Sanger sequencing (Azenta/Genewiz) to confirm the amino acid change to alanine.

INS-1 832/3 GLP-1R KO cells were transiently transfected with SNAP/FLAG-hGLP-1R WT or cholesterol-binding mutants; purified GLP-1R KO mouse islets were transduced with SNAP/FLAG-hGLP-1R WT vs V229A adenoviruses (generated by VectorBuilder) at an MOI of 1. Cells or islets were incubated for 24 hr post-transfection/transduction before labelling with 1 µM SNAP-Surface fluorescent probes (New England Biolabs) for 15–20 min for cells, or 7 min for islets, in full media at 37°C. Cells were washed and imaged in RPMI-1640 without phenol red (32404014, Thermo Fisher Scientific), while islets were washed and imaged in 6 mM glucose KRBH buffer using a Nikon spinning disk field scanning confocal microscope with a 60 X oil objective at 37 °C. Cells were imaged by time-lapse microscopy with images taken every 6 s; a baseline recording was taken for 1 min before addition of 100 nM exendin-4 and further imaging for 10 min. Islets were imaged in vehicle conditions or 5 min after stimulation with 100 nM exendin-4 at 37 °C.

Raw images were analysed using Fiji and the full-width half maximum (FWHM) of a line profile macro, developed by Steven Rothery, FILM Facility, Imperial College London (https://www.imperial.ac.uk/medicine/facility-for-imaging-by-light-microscopy/software/fiji/↗), employed to determine the change in SNAP fluorescence intensity at the plasma membrane from individual cells. SNAP/FLAG-hGLP-1R surface expression levels were determined from the average intensity during the baseline period for cells or from the vehicle images for islets. Receptor internalisation was calculated from the loss of fluorescence intensity at the plasma membrane after 10 min of exendin-4 stimulation compared to baseline intensity for cells, or from the loss of fluorescence intensity after 5 min of exendin-4 stimulation vs vehicle for islets.

PhotoClick cholesterol (37.5 nmoles, 700147 P, Avanti Polar Lipids) was agitated in 2 mM MβCD in Hanks Balanced Salt Solution (HBSS) (14025092, Thermo Fisher Scientific) overnight at room temperature to facilitate intracellular delivery. INS-1 832/3 SNAP/FLAG-hGLP-1R WT or V229A cells were seeded onto 6-well plates and incubated in 18.75 nmoles of PhotoClick cholesterol loaded onto MβCD in HBSS for 1 hr at 37 °C protected from light. Cells were then stimulated with 100 nM exendin-4 or vehicle for 2 min, washed with ice cold PBS and UV irradiated at 365 nm with a UV Crosslinker for 5 min in ice cold PBS to activate the photoreactive diazirine group in the PhotoClick cholesterol. Cells were then lysed in 1 X TBS (50 mM Tris-HCl, 150 mM NaCl, pH 7.4) supplemented with 1 mM EDTA, 1% Triton X-100, phosphatase inhibitor cocktail (P5726, Sigma Aldrich), and protease inhibitor cocktail (11873580001, Roche Diagnostics). Lysates were sonicated 3X 10 s, centrifuged at 10,000 rpm for 10 min, and incubated with anti-FLAG M2 affinity gel (A2220, Sigma Aldrich) rotating overnight at 4 °C to immunoprecipitate the SNAP/FLAG-hGLP-1R. Following immunoprecipitation, the bead-bound SNAP/FLAG-hGLP-1R was incubated with a click chemistry mix (20 μM rhodamine-azide, 1 mM TCEP, 100 μM TBTA, 1 mM CuSO₄) for 1 hr at room temperature gently agitating to fluorescently label the PhotoClick cholesterol. Urea buffer (2 X) was added in a 1:1 ratio and samples incubated for 10 min at 37 °C prior to 10% SDS-PAGE gel separation and imaging using a ChemiDoc MP imaging system to detect fluorescently labelled cholesterol. The proteins were then transferred onto a PVDF membrane and corresponding SNAP/FLAG-hGLP-1R levels detected by Western blotting with an anti-SNAP antibody (P9310S, New England Biolabs, RRID:AB_10631145↗). Data was represented as amount of bound Photo-Click cholesterol relative to SNAP/FLAG-hGLP-1R.

INS-1 832/3 GLP-1R KO, SNAP/FLAG-hGLP-1R WT, and V229A cells were seeded at 50,000 cells per well in a clear-bottom 96-well black plate coated with poly-d-Lysine. 24 hr after seeding, growth media was removed and cells were stimulated with fluorescent GLP-1R agonist exendin-asp3-TMR (0.1–1000 nM) in complete DMEM for 10 min at 37 °C. Cells were washed three times with PBS, then imaged in Krebs-Henseleit buffer using an EVOS M7000 cell imaging system with a Texas Red 2.0 light cube with λex = 585 nm and λem = 628 nm.

pGEX-KG-D4H*-mCherry plasmid (Addgene #134604) was used to purify D4H*-mCherry as previously described (Lim et al., 2019). Briefly, a plasmid colony was grown in 5 mL LB broth overnight at 37 °C. Culture was scaled up into a 1 L flask and grown for a further 4 hr, and protein production induced with 0.5 mM IPTG for 4 hr. Bacterial culture was pelleted at 2500 rpm for 30 min and resuspended in lysis buffer (0.1 M NaCl, 20 mM Tris-HCl, pH 8.0, 1 mM DTT, 1 X protease inhibitor cocktail) and 0.35 mg/mL lysozyme added for 30 min on ice. Samples were sonicated at 50% Amp for 10 s and 10 s break for a total of 4 min. 1% Triton X-100 was added, and samples rocked at 4 °C for 30 min before centrifugation and incubation of supernatants with glutathione beads (GE17-0756-01, Sigma Aldrich) for 2 hr rocking at room temperature to capture the GST-D4H*-mCherry protein. D4H*-mCherry protein was eluted in elution buffer (25 mM L-glutathione, 50 mM Tris-HCl pH 8.8, 200 mM NaCl), filtered through a 30 kDa centrifugal filter unit (UFC203024, Thermo Fisher Scientific) and stored at –80 °C.

INS-1 832/3 SNAP/FLAG-hGLP-1R WT or V229A cells were seeded onto 13 mm coverslips and labelled with SNAP-Surface 488 probe in full media for 20 min before stimulation with 100 nM exendin-4 for 2 min at 37 °C. Cells were washed with PBS and fixed with 4% paraformaldehyde (PFA) for 10 min at 4 °C, permeabilised for 15 s in a liquid nitrogen bath, blocked in 1% BSA/PBS for 1 hr at room temperature, incubated with D4H*-mCherry (1:25 in 1% BSA / PBS) for 2 hr, washed with PBS and post-fixed in 4% PFA for 10 min at room temperature. Coverslips were washed, mounted onto microscope slides using ProLong Diamond Antifade Mountant with DAPI (P36966, Thermo Fisher Scientific), and imaged with a 63 X oil objective Leica Stellaris 8 inverted confocal microscope from the FILM facility at Imperial College London. Raw images were analysed in Fiji using the Coloc 2 plugin to determine SNAP/FLAG-hGLP-1R co-localisation with D4H*-mCherry.

INS-1 832/3 SNAP/FLAG-hGLP-1R WT or V229A cells were seeded onto glass bottom MatTek dishes and labelled with SNAP-Surface 647 (S9136S, New England Biolabs) for 15 min at 37 °C in full media, washed, and imaged in RMPI-1640 media without phenol red using a 100 X oil objective on a Leica Stellaris 8 STED FALCON microscope from the FILM Facility at Imperial College London using confocal settings as previously described (Bernabé-Rubio et al., 2021a). Images were analysed as previously described (Garcia and Bernardino de la Serna, 2018Pickford et al., 2020). Briefly, cells were imaged at the basal plasma membrane under vehicle conditions and following stimulation with 100 nM exendin-4 with a format size of 256 × 256 pixels and 80 nm pixel size for 200 consecutive frames. RICS analysis was carried out to determine the diffusion coefficient of WT vs V229A GLP-1R in vehicle and stimulated conditions using the SimFCS 4 Software (Global Software, G-SOFT Inc). Three different regions of interest of size 32 × 32 pixels for each image and a moving average of 10 was applied to avoid any artefacts due to cellular motion or very slow-moving particles. Average intensity, intensity plots, 2D autocorrelation maps, and 3D autocorrelation fits were generated for each condition and cell line investigated (Garcia and Bernardino de la Serna, 2018; Bernabé-Rubio et al., 2021b).

INS-1 832/3 GLP-1R KO cells were transiently transfected with WT or V229A human GLP-1R-monomeric EGFP fusion constructs, generated in house by mutating EGFP Ala207 to Lys in hGLP-1R-EGFP [kind gift from Dr Alessandro Bisello (Syme et al., 2006)] using the following primers: Forward (5' to 3'): CCTGAGCACCCAGTCCAAGCTGAGCAAAGACCCCA and Reverse (5' to 3'): TGGGGTCTTTGCTCAGCTTGGACTGGGTGCTCAGG, and WT hGLP-1R-mEGFP to V229A with the primers from Table 2. Transfected cells were seeded onto glass bottom MatTek dishes, imaged live at 37 °C in RPMI-1640 without phenol red under vehicle conditions or following stimulation with 1 µM exendin-4 and images analysed as previously described (Salavessa and Sauvonnet, 2021) with some changes. Cells were imaged in a Nikon spinning disk field scanning confocal microscope using a 100 X oil objective under TIRF illumination with the following microscope settings using the MetaMorph software: 488 nm excitation laser, exposure time of 150 ms, 10% laser power at a rate of 1 frame per second for 150 frames. Images were loaded onto the Icy software; an ROI was drawn around each cell and Spot Detector was used to determine the spots in each image. A scale of 2 with a sensitivity of 70 was used and kept constant for each experiment. Spot Tracking followed by Track Manager were used to determine the speed and displacement for each particle detected in the plasma membrane of cells.

INS-1 832/3 SNAP/FLAG-hGLP-1R WT or V229A cells were seeded onto µ-Slide 8-well glass bottom microscope plates (Ibidi) and labelled with SNAP-Surface Alexa Fluor 647 for 10 min followed by 40 μM Laurdan (6-dodecanoyl-2-(dimethylamino) naphthalene, D250, Thermo Fisher Scientific), an environmentally sensitive plasma membrane lipid dye (Brewer et al., 2010), for 5 min prior to imaging. Cells were washed with PBS and imaged in RMPI 1640 medium without phenol red using the confocal settings of a Leica Stellaris 8 STED FALCON in an 86 X objective and a 10% power 405 nm excitation laser and emission ranges of 420–460 nm (blue emission, IB) and 470–510 nm (red emission, IR) for Laurdan, plus a 5% power 647 nm excitation laser for SNAP-Surface Alexa Fluor 647. The general polarisation (GP) formula (GP = IB-IR/IB +IR) was used to retrieve the relative lateral packing order of lipids at the plasma membrane: values of GP vary from 1 to −1, with higher values reflecting lower fluidity (or higher lateral lipid order), whereas lower values indicate increased fluidity. Two hundred consecutive frames were captured for 2.45 min at a pixel size of 80 nm and a format size of 256 × 256 pixels. Cells were imaged under vehicle conditions and after stimulation with 100 nM exendin-4. Further analysis was carried out using the SimFCS 4 Software to determine the RICCS cross-correlation between Laurdan GP values and SNAP/FLAG-hGLP-1R. Specifically, Ch2-Ch1 (B1-B2 map) RICS analysis of brightness (intensity of fluorescence signal) in one channel cross-correlates with brightness in the second channel, allowing for the calculation of cross-correlation. N&B analysis was performed from the SNAP signal to determine the oligomerisation of SNAP/FLAG-hGLP-1R after exendin-4 stimulation by assessing the apparent brightness and number of pixels involved using SimFCS 4 software. Three different regions of interest of size 64 × 64 pixels per image and a moving average of 10 was applied to avoid any artefacts due to cellular motion or very slow-moving particles.

INS-1 832/3 SNAP/FLAG-hGLP-1R WT or V229A cells were labelled with 40 nM SNAP-Lumi4-Tb (SSNPTBC, Cisbio), a lanthanide fluorophore probe for TR-FRET, for 1 hr at 37 °C in full media. Cells were washed with HBSS and resuspended in 60 nM NR12A, a solvatochromic photostable plasma membrane lipid probe (Klymchenko, 2023) in HBSS, incubated for 5 min, and seeded onto a white opaque 96-well half-area plate. TR-FRET was carried out as described (Lucey et al., 2021). Briefly, a baseline reading was taken for 5 min using a Flexstation 3 plate reader with excitation wavelength of 335 nm, and emissions of 490 nm and 590 nm. The cells were then stimulated with 100 nM exendin-4, or vehicle, and the plate read for a further 20 min. Receptor conformational shift was calculated as the 590/490 ratio of both wavelength emission signals after normalising to basal signal under vehicle conditions. AUC was calculated and used for statistical analysis.

INS-1 832/3 SNAP/FLAG-hGLP-1R WT or V229A cells were seeded onto 6 cm dishes and stimulated with 100 nM exendin-4 or vehicle for 2 min at 37 °C, placed on ice, and washed with ice cold PBS before being osmotically lysed with 20 mM Tris-HCl (pH 7.0) supplemented with protease and phosphatase inhibitor cocktails. The cell suspension was homogenized by passing through a 21-gauge needle and ultracentrifuged for 1 hr at 41,000 rpm at 4 °C. The supernatant was discarded, and the pellet resuspended in cold PBS +1% Triton X-100 plus protease and phosphatase inhibitor cocktails, transferred to an Eppendorf tube, and allowed to rotate for 30 min at 4 °C. The suspension was then ultracentrifuged for another hour at 41,000 rpm at 4 °C to separate detergent-resistant (DRM) from detergent-soluble (DSM) membrane fractions: the 'disordered' DSM fraction in the supernatant was retained for analysis and the 'ordered' DRM (lipid raft) fraction was resuspended in 1% SDS plus protease and phosphatase inhibitor cocktails, sonicated, centrifuged for 10 min at 13,500 rpm at 4 °C, and resolved via SDS-PAGE and Western blotting as in Buenaventura et al., 2019. Blotted membranes were analysed using Fiji and SNAP levels normalised to flotillin (detected post-stripping with anti-flotillin-1 antibody, sc-74566, Santa Cruz Biotechnology, RRID:AB_2106563↗), used as DRM loading control.

INS-1 832/3 SNAP/FLAG-hGLP-1R WT or V229A cells were transfected with clathrin light chain (CLC)-GFP and plated onto poly-L-lysine-coated glass coverslips. Prior to imaging, cells were labelled at 37 °C with 5 μM SNAP-Surface 549 probe for 30 min in full media and washed. Microscopy was performed on a custom-built TIRF microscope based on a Zeiss Axiovert 135 with DPSS lasers at 405, 491, and 561 nm, 100 X/1.45 NA objective. Colour channels were split using a Dualview (Optical Insights) and projected side by side onto an EMCCD camera (QuantEM:512SC, Photometrics). Still images were acquired by sequential excitation with the 473 and 561 nm lasers. During the experiments, cells were incubated in an imaging solution containing 138 mM NaCl, 5.6 mM KCl, 2.6 mM CaCl₂, 1.2 mM MgCl₂, 3 mM glucose, and 5 mM HEPES, pH 7.4 at ~35 °C. Exendin-4 (100 nM) was added for at least 10 min to trigger receptor endocytosis. Calibration and alignment were performed using 100 nm polystyrene beads (Molecular Probes). To quantify SNAP/FLAG-hGLP-1R association with clathrin, circular ROIs surrounding individual well-defined puncta of fluorescence (5–50 per cell) were selected on CLC-GFP images and transferred to the corresponding SNAP ones using a Fiji macro and average fluorescence per pixel was measured in a central circle (c) of diameter 5 pixels (0.55 µm) at each puncta location, in a surrounding annulus (a) with an outer diameter of 7 pixels (0.77 µm), and in an area not containing any cell as background (bg). The specific cluster-associated fluorescence ∆F was calculated by subtracting the annulus value from that of the circle (∆F=c a). Local fluorescence from CLC-GFP molecules not associated with the cluster site was calculated by subtracting the background from the annulus value (S=a bg) and used to normalize ∆F to the expression level of the tested protein (Barg et al., 2010). The reported parameter ΔF/S can be interpreted as the GLP-1R affinity for clathrin structures.

GLP-1R-SmBiT WT was previously cloned in house as described in Manchanda et al., 2023. GLP-1R-SmBiT V229A was generated from the WT template by site-directed mutagenesis using the following primers: Forward (5' to 3'): TCTTTCCTGTCGACTCGCGTTTCTGCTTATGCAGT, Reverse (5' to 3'): ACTGCATAAGCAGAAACGCGAGTCGACAGGAAAG. LgBiT-mini-G_s and -mini-G_q were a gift from Prof. Nevin Lambert, Medical College of Georgia, USA, and LgBiT-β-arrestin 2 was purchased from Promega (plasmid no. CS1603B118). For recruitment assays, INS-1 832/3 GLP-1R KO cells were seeded onto 6-well plates and transiently transfected with a combination of 1 µg GLP-1R-SmBiT WT or V229A and either 1 µg of LgBiT-mini-G_s, -mini-G_q or -β-arrestin 2.

Nanobody37 (Nb37) bystander NanoBiT assays (Figure 7—figure supplement 1) were carried out as previously described (Manchanda et al., 2023). The constructs required to determine plasma membrane (KRAS CAAX motif), or endosomal (Endofin FYVE domain) GLP-1R activity were a gift from Prof. Asuka Inoue, Tohoku University, Japan. INS-1 832/3 GLP-1R KO cells were seeded onto 12-well plates and transfected as detailed in Table 3.

Briefly, 24 hr after transfection, cells were incubated in 3 mM glucose for a minimum of 2 hr, detached and resuspended in NanoGlo Live cell substrate (N2012, Promega), 1:20 dilution in HBSS, and seeded onto an opaque white 96-well half-area plate. Using a Flexstation 3 plate reader, a baseline luminescence reading was acquired for 8 min at 37°C, followed by stimulation with vehicle or 100 nM exendin-4 for Nb37 bystander assay, or with serial dilutions of exendin-4 starting from 1 µM, 100 nM, 10 nM to 1 nM for 30 min for the mini-G_s/mini-G_q/β-arrestin 2 recruitment assays. Results were normalised to individual basal signal followed by average vehicle responses. Non-linear fit curves of the AUCs were used to generate dose responses using GraphPad Prism 10.

All data analyses and graph generation were performed using GraphPad Prism 10 and presented as mean ± standard error of the mean (SEM); n numbers represent biological replicates. Statistical analyses were performed with the tests indicated in the figure legend. Data was deemed significant at p≤0.05.