Author Correction: Super-resolution microscopy compatible fluorescent probes reveal endogenous glucagon-like peptide-1 receptor distribution and dynamics

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

T he glucagon-like peptide-1 receptor (GLP1R) is a secretin family class B G protein-coupled receptor (GPCR) characterized by hormone regulation 1 . Due to its involvement in glucose homeostasis, the GLP1R has become a blockbuster target for the treatment of type 2 diabetes mellitus 2 . The endogenous ligand, glucagon-like peptide-1 (GLP-1) is released from enteroendocrine L-cells in the gut in response to food intake 3 , from where it travels to the pancreas before binding to its cognate receptor expressed in β-cells. Following activation, the GLP1R engages a cascade of signaling pathways including Ca 2+ , cAMP, ERK and β-arrestin, which ultimately converge on β-cell survival and the glucose-dependent amplification of insulin release 4,5 . GLP1R is also expressed in the brain 6,7 , where it further contributes to metabolism via effects on food intake, energy expenditure, locomotion, and insulin sensitivity. Despite this, GLP1R localization remains a challenge and is impeding functional characterization of GLP-1 and drug action.
Such methods have a number of shortcomings. Antibodies possess variable specificity 17 and tissue penetration, and GLP1R epitopes might be hidden or preferentially affected by fixation in different cell types and tissues. Enzyme self-labels allow GLP1R to be visualized in living cells without affecting ligand binding, but require heterologous expression and have therefore not yet been able to address endogenous receptor. Moreover, fluorescent analogues of Exendin4(1-39) and Liraglutide activate and internalize the receptor, which could confound results in live cells, particularly when used as a tool to sort purified populations (i.e. β-cells) 26,27 . Antagonist-linked fluorophores circumvent this issue, but the majority lack thorough pharmacological validation, or possess near infrared tags which require sophisticated confocal imaging modalities. On the other hand, reporter mouse strategies possess high fidelity, but cannot account for lineage-tracing artefacts, post-translational processing, protein stability and trafficking of native receptor 28 . Lastly, none of the aforementioned approaches are amenable to super-resolution imaging of endogenous GLP1R.
Given the wider reported roles of GLP-1 signaling in the heart 29 , liver 30 , immune system 2 , and brain 31 , it is clear that new tools are urgently required to help identify GLP-1 target sites, with repercussions for drug treatment and its side effects. In the present study, we therefore set out to generate a specific probe for endogenous GLP1R detection in its native, surface-exposed state in live and fixed tissue, without receptor activation. Herein, we report LUXendin645 and LUXendin651, Cy5-and silicon rhodamine (SiR)-conjugated far-red fluorescent antagonists with excellent specificity, live tissue penetration, and super-resolution capability. Using our tools, we provide an updated view of GLP1R expression patterns in pancreatic islets, brain, and hESC-derived β-like cells, show that endogenous GLP1Rs form nanodomains at the membrane, and reveal receptor subpopulations with distinct diffusion modes in their non-stimulated state. Lastly, installation of a tetramethylrhodamine (TMR) fluorophore allows in vivo multiphoton imaging. As such, the LUXendins provide the first nanoscopic characterization of a class B GPCR, with wider flexibility for detection and interrogation of GLP1R in the tissue setting both in vitro and in vivo.

Results
Design of LUXendin555, LUXendin645, and LUXendin651. Ideally, a fluorescent probe to specifically visualize a biomolecule should have the following characteristics: straightforward synthesis and easy accessibility, high solubility, relatively small size, high specificity and affinity, and a fluorescent moiety that exhibits photostability, brightness and (far-)red fluorescence with an additional two-photon cross-section. Moreover, the probe should be devoid of biological effects when applied to live cells and show good or no cell permeability, depending on its target localization. While some of these points were addressed in the past, we set out to achieve this high bar by designing a highly specific fluorescent GLP1R antagonist using TMR, Cy5, and SiR fluorophores. As no small molecule antagonists for the GLP1R are known, we turned to Exendin4(9-39), a potent antagonistic scaffold amenable to modification (Fig. 1) 32 . We used solid-phase peptide synthesis (SPPS) to generate an S39C mutant 21 , which provides a Cterminal thiol handle for late-stage installation of different fluorophores. As such, TMR-, Cy5-and SiR-conjugated versions were obtained by means of cysteine-maleimide chemistry, termed LUXendin555, LUXendin645, and LUXendin651, respectively, with spectral properties shown in Table 1 (characterization of and purity of compounds in Supplementary Figs. 1-11) (Fig. 1).
As a first assessment of GLP1R-labeling efficiency, we probed YFP-AD293-SNAP_GLP1R cells with increasing concentrations of LUXendin645. Maximal LUXendin645 labeling occurred at 250-500 nM (Fig. 2c), with no signal detected in control YFP-AD293 cells lacking GLP1R (Fig. 2d). We next examined whether LUXendin645 would allow labeling of endogenous GLP1R in primary tissue. Following 60 min application of LUXendin645, isolated islets demonstrated intense and clean labeling, which was restricted to the membrane (Fig. 2e). To minimize background fluorescence, slightly lower concentrations of LUXendin645 were used in islets (50-100 nM) vs. plated cells (250 nM). Using conventional confocal microscopy, we were able to detect bright staining even 60 µm into the islet (Fig. 2e). Given these results, we next attempted to penetrate deeper into the islet by taking advantage of the superior axial resolution of two-photon Maximal excitation and emission wavelengths, and quantum yields were acquired using probes dissolved at 10 µM in PBS, pH 7.4 at 21°C a For maleimide-conjugated fluorophores excitation (Fig. 2f). Remarkably, this imaging modality revealed LUXendin645 labeling at high resolution throughout the entire volume of the islet (170 µm in this case) (Fig. 2f). Consistent with the cAMP assays, GLP1R internalization was detected following co-application of LUXendin645 and BETP to MIN6 β-cells, which endogenously express the receptor (Fig. 2g, h).
LUXendin645 specifically binds the GLP1R. To further validate the specificity of LUXendin645 labeling in primary tissue, we generated Glp1r knock-out mice. This was achieved using CRISPR-Cas9 genome editing to introduce a deletion into exon 1 of the Glp1r. The consequent frameshift was associated with absence of translation and therefore a global GLP1R knockout, termed Glp1r (GE)−/− , in which all intronic regions, and thus regulatory elements, are preserved ( Fig. 3a, b). Wild-type (Glp1r +/+ ), heterozygous and homozygous littermates were phenotypically normal and possessed similar body weights (Fig. 3c). Confirming successful GLP1R knock-out, insulin secretion assays in islets isolated from Glp1r (GE)−/− mice showed intact responses to glucose, but absence of Exendin4(1-39)-stimulated insulin secretion (Fig. 3d). Reflecting this finding, the incretinmimetic Liraglutide was only able to stimulate cAMP rises in islets from wild-type (Glp1r +/+ ) littermates, measured using the FRET probe Epac2-camps (Fig. 3e, f). As expected, immunostaining with monoclonal antibody showed complete absence of GLP1R protein (Fig. 3g). Suggesting that LUXendin645 specifically targets GLP1R, with little to no cross-talk from glucagon receptors 34 , signal could not be detected in Glp1r (GE)−/− islets (Fig. 3g).
Together, these data provide strong evidence for a specific mode of LUXendin645 action via the GLP1R.
constitute~20% of the insulin-negative islet population 37 , this leaves~5% of GLP1R+ α-cells. This was not an artefact of optical section, since two-photon islet reconstructions showed similar complete absence of LUXendin645 staining in discrete regions near the surface (where α-cells predominate) (Supplementary Movie 1).
LUXendins reveal higher-order GLP1R organization. By combining LUXendin645 with super-resolution radial fluctuations (SRRF) analysis 38 , GLP1R could be imaged at superresolution using streamed images (~500) from a conventional widefield microscope (Fig. 5a). To image endogenous GLP1R at <100 nm lateral resolution, we combined STED nanoscopy with  LUXendin651, which bears SiR instead of Cy5. LUXendin651 displayed antagonist behavior, with no evidence of partial agonism, and produced bright labeling of wild-type but not Glp1r (GE)−/− islets, with an identical distribution to LUXendin645 (Supplementary Figs. 12 and 13). Incubation of MIN6 cells with LUXendin651 and subsequent fixation allowed STED imaging of the endogenous GLP1R with a FWHM = 70 ± 10 nm (mean ± s.d.; n = 15 line profiles measured on the raw data, from two independent repeats) ( Fig. 5b-d). STED snapshots of MIN6 βcells revealed detailed GLP1R distribution: receptors were not randomly arranged but rather tended to organize into nanodomains with neighbors ( Fig. 5b-d). This was confirmed using the F-and G-functions, which showed a non-random and more clustered GLP1R distribution (Fig. 5e, f). Differences in GLP1R expression level and pattern could clearly be seen between neighboring cells with a subpopulation possessing highly concentrated GLP1R clusters (Fig. 5g). LUXendin651 even allowed GLP1R to be imaged in living MIN6 cells using SRRF and STED, although nanodomains were more difficult to resolve due to the lateral diffusion of receptors ( Fig. 5h, i).

LUXendin645 and Luxendin651 label single GLP1R molecules.
To test whether LUXendin645 and LUXendin651 would be capable of tracking single GLP1Rs in live cells, we performed single-molecule microscopy experiments in which individual receptors labeled with either fluorescent probe were imaged by total internal reflection fluorescence (TIRF) microscopy 39,40 . Both probes allowed GLP1R to be tracked at the single-molecule level in CHO-K1-SNAP_GLP1R cells, but bleaching precluded longer recordings with LUXendin645 ( Fig. 6a and Supplementary Movies 2, 3). By combining single-particle tracking with LUX-endin651, we were able to show that GLP1Rs diffuse at the membrane in their non-stimulated or antagonized state ( Fig. 6a and Supplementary Movie 4). However, a mean square displacement (MSD) analysis 40 revealed a high heterogeneity in the diffusion of GLP1Rs on the plasma membrane, ranging from virtually immobile receptors to some displaying features of directed motion (superdiffusion) (Fig. 6b, c). GLP1R diffusion properties were not ligand-dependent, since similar profiles were detected for both LUXendin645 and LUXendin651 (Fig. 6c).
LUXendin645 allows visualization of central GLP1 targets. To further show the utility of LUXendin645 for visualizing endogenous GLP1R, we extended studies to the brain in which mAbs do not work reliably and where peripheral GLP1 targets still remain poorly characterized. Two hours following subcutaneous injection of LUXendin645, perfuse-fixed brains were retrieved for analysis. Intense labelling could be detected in the arcuate nucleus (ARC), area postrema (AP) and choroid plexus (CP) (Fig. 7a, b), all regions known to express GLP1R using reporter or fluorescent agonist approaches 6,22 . Notably, LUXendin645-labeled neurons overlapped with areas receiving innervation from GLP1producing neurons 41 , with GLU-YFP synaptic boutons closely abutting GLP1R+ areas ( Fig. 7a, b). LUXendin645 labeling colocalized with GLP1R-expressing neurons in the ARC/median eminence (ME) and AP/nucleus tractus solitarius (NTS), shown using GLP1RCre;LSL-GCaMP3 reporter mice ( Fig. 7c-e). Super-resolution imaging (~140 nm lateral resolution) revealed the presence of LUXendin645 on the cell membranes of GLP1R+ neuron cell bodies, as well as dendrites. Moreover, GLP1R were found to accumulate into nanodomains on the membranes of ARC and AP neuron membranes, as well as ependymal cells of the CP (Fig. 7f). Lastly, optical projection tomography allowed entire LUXendin645-labelled brains to be imaged and mapped in three-dimensions, confirming the above results and also extending probe localization to the subfornical organ (SFO), organum vasculosum of the lamina terminalis (OVLT), and ventricles ( Fig. 7g).
LUXendin645 labels GLP1R in hESC-derived β-cells. Since GLP1R are expressed in mature β-cells 42 , we wondered whether LUXendin645 would serve as a useful surface marker for assessing differentiation of human embryonic stem cell (hESC)derived β-like cells. LUXendin645 was unable to label undifferentiated ES cells (Fig. 8a). Following differentiation and 21 days' culture, LUXendin645 labeling was clearly visible in spheroids, in line with increasing levels of GLP1R expression (Fig. 8b). As for mouse islets, LUXendin645 co-localized with insulin, with minimal signal in areas strongly positive for glucagon (Fig. 8c), shown using Manders' split coefficients (Fig. 8d). Not all insulincontaining cells stained for GLP1R, however (Fig. 8c). Confirming a β-like cell phenotype, spheroids were fixed and sliced before staining for insulin and NKX6-1 (Fig. 8e).
We next investigated whether LUXendin645 would allow βlike cells to be purified according to GLP1R expression. To this end, LUXendin645-labelled spheroids were subjected to fluorescence-activated cell sorting (FACS), before gene expression analyses of LUXendin645+ and LUXendin645− populations. Notably, the LUXendin645+ population expressed higher levels of GLP1R and NKX6-1, with a tendency toward increased INS (Fig. 8f). As expected from the imaging data, GCG expression was significantly decreased in the LUXendin645+ β-like cells (Fig. 8f).
termed LUXendin555. Labeling was detected in YFP-AD293-SNAP_GLP1R ( Fig. 9a) but not in YFP-AD293 cells (Fig. 9b), with max labeling at 600 nM (Fig. 9c). A slightly lower concentration (250 nM) of LUXendin555 was found to produce bright staining in both cells and islets, whilst minimizing background fluorescence. However, we noticed a more punctate LUXendin555 staining pattern when viewed at high-resolutions (Fig. 9a). To determine whether the appearance of puncta was due to receptor internalization, or alternatively accumulation of cleaved, charged TMR in organelles, labeling was repeated in islets co-stained with GLP1R monoclonal antibody (Supplementary Fig. 14). In these experiments, no differences in GLP1R surface expression could be seen between LUXendin555, LUX-endin645, and Ex9 (Fig. 9d), suggesting that puncta are unlikely to be internalized receptor. In line with this, LUXendin555 was found to display antagonist properties using HTRF-based cAMP assay (Fig. 9e). However, a second independent detection method (luciferase) showed the opposite result, raising the possibility that the installed TMR might influence assay readout (Fig. 9f). As for the other probes, LUXendin555 was unable to label Glp1r (GE)−/− islets ( Supplementary Fig. 14).
We thought that the relatively high quantum yield of TMR, coupled with good two-photon cross-section might suit LUX-endin555 well to in vivo imaging. Two-photon imaging was applied to anaesthetized mice to allow visualization of the intact pancreas, exposed through an abdominal incision (Fig. 9g). Vessels and nuclei were first labeled using FITC-dextran and Hoechst before injecting LUXendin555 intravenously. The following observations support that LUXendin555 displays antagonist activity in vivo: (1) labeling occurred rapidly within 5 min post-injection; (2) staining was confined to the cell membrane with no apparent internalization (Fig. 9h); and (3) normoglycemia was not significantly altered over 30 min (173.0 ± 21.1 vs. 215.3 ± 41.4 mg/dl, 0 and 30 min post-injection, respectively; mean ± s.d.; n = 3 mice; non-significant, paired t-test).

Discussion
In the present study, we synthesize and validate far-red fluorescent labels, termed LUXendins for the real-time detection of endogenous GLP1R in live cells. Nanomolar concentrations of LUXendin645 and LUXendin651 lead to intense membranelabeling of the GLP1R, with best in class tissue penetration and signal-to-noise ratio, as well as super-resolution capability. Notably, LUXendin645 and LUXendin651 do not activate the GLP1R unless agonist activity is conferred with the widely available PAM BETP. LUXendin645 and LUXendin651 are highly specific, as shown using a CRISPR-Cas9 mouse line lacking GLP1R expression. Lastly, the analogous compound LUX-endin555, bearing a different fluorophore, expands the color palette without changing the peptidic pharmacophore.
Compared to present chemical biology approaches, LUXendins possess a number of advantages for GLP1R labeling, which generally rely on Exendin4(1-39) or Exendin4(9-39) peptides labeled with for instance FITC, Cy3, Alexa594, Cy5, or VT750 19,[22][23][24][25]33 . Firstly, the use of an antagonist retains more receptor at the cell surface, which likely increases detection capability. Secondly, the GLP1R is not fully activated, meaning that results can be interpreted in the absence of potentially confounding cell signaling or internalization, such as that expected with agonists 25 . Thirdly, Cy5 and SiR occupy the far-red range, leading to less background fluorescence, increasing depth penetration in confocal microscopy due to reduced scatter, and avoiding the use of more phototoxic wavelengths 43 . Fourthly, LUXendin pharmacology and labelling specificity has been validated in-depth. Lastly, LUXendins are well-adapted for superresolution imaging through the use of optimally suited fluorescent moieties. Together, these desirable properties open up the possibility to image expression and distribution of native GLP1R over extended periods of time using multiple imaging modalities.
While LUXendins also allowed GLP1R trafficking to be monitored, this required the presence of a PAM to allosterically activate the receptor. Due to the probe-dependent nature of PAMs, LUXendins with a number of different pharmacophores would need to be generated to fully assess the ligand-dependency of GLP1R trafficking. In some experiments, we also noticed the presence of punctate LUXendin645 and LUXendin651 labelling. Suggesting that this staining pattern reflects cleaved fluorophore rather than internalized GLP1R are the following observations: (1) succinimide exchange with reactive thiols can lead to linker loss 44 , allowing free fluorophore to cross the membrane and accumulate in organelles; and (2) puncta were not apparent in the same samples co-stained with GLP1R mAb.
To test the specificity of LUXendins, we used CRISPR-Cas9 genome-editing to globally knock out the GLP1R in mice. Protein deletion was confirmed by absence of detectable GLP1R signal following labeling with monoclonal antibody, LUXendin555, LUXendin645, and LUXendin651. While Glp1r −/− animals already exist, and have made important contributions to our understanding of incretin biology, they were produced using a mutation to replace exons encoding transmembrane regions 1 and 3 (encoded by exons 5 and 7), presumably leading to deletion of the introns in between (~6.25 kb) 45 . By contrast, Glp1r (GE)−/− mice possess intact introns. Since introns contain regulatory elements, such as distant-acting enhancers 46 , miRNAs 47 , and lncRNAs 48 , their loss in transgenic knockouts could have wider influence on the transcriptome. GLP1R knock-out mice might therefore be useful alongside conventional approaches for validating GLP1R reagents, including antibodies, agonist and antagonist, and derivatives thereof.
Demonstrating the excellent sensitivity of the Cy5-linked LUXendin645 in particular, we were able to detect GLP1R expression in~5% of α-cells. Understanding α-cell GLP1R expression patterns is important because incretin-mimetics reduce glucagon secretion 49 , which would otherwise act to aggravate blood glucose levels. Previous studies using antibodies, reporter animals and agonist-fluorophores have shown that 1-10% of mouse and rat α-cells express GLP1R, in line with the low transcript abundance 7,25,35,50 , despite reports that GLP-1 can directly suppress glucagon release 36 . Our data are in general concordance with these findings, but demonstrate an increase in sensitivity compared to other approaches capable of detecting native GLP1R protein, namely mAb and agonist-fluorophore. This improvement is likely related to the superior signal-to-noise ratio (SNR) and selectivity of LUXendin645, increasing the Fig. 4 LUXendin645 reveals GLP1R expression in a subpopulation of α-cells. a-c LUXendin645 labeling is widespread throughout the intact islet, colocalizing predominantly with β-cells a and δ-cells b, but less so with α-cells c stained for insulin (INS), somatostatin (SST), and glucagon (GCG), respectively (n = 18 islets, seven animals, three separate islet preparations) (scale bar = 26 µm). d Following dissociation of islets into cell clusters, LUXendin645 labeling can be more accurately quantified (arrows highlight cells selected for zoom-in) (scale bar = 26 µm). e Zoom-in of d showing a LUXendin645− (left) and LUXendin645+ (right) α-cell (arrows highlight non-labeled cell membrane, which is not bounded by a β-cell) (scale bar = 26 µm). f Box-and-whiskers plot showing proportion of β-cells (INS) and α-cells (GCG) co-localized with LUXendin645 (n = 18 cell clusters, ten animals, three separate islet preparations) (box and whiskers plot shows range and median; mean is shown by a plus symbol). g Ins1Cre Thor ;R26 mT/mG dual fluorophore reporter islets express tdTomato until Cre-mediated replacement with mGFP, allowing identification of β-cells (~80% of the islet population) and non-β-cells for live imaging (scale bar = 26 µm). LUXendin645 (LUX645) highlights GLP1R expression in nearly all β-cells but relatively few non-βcells (n = 31 islets, six animals, three separate islet preparations). h A zoom-in of the islet in g showing GLP1R expression in some non-β-cells (left) together with quantification (right) (arrows show LUXendin645-labeled non-β cells) (scale bar = 12.5 µm) (scatter dot plot shows mean ± s.e.m.). White boxes show the location of zoom-ins. In all cases, LUXendin645 was applied at 100 nM. Source data are provided as a Source Data file.
ability to resolve relatively low levels of endogenous GLP1R. A recent report showed GLP1R expression in~80% of α-cells using an antibody raised against the N-terminal region, with both membrane and cytosolic staining evident 51 . While the reasons for this discrepancy are unknown, it should be noted that LUXendin645 binds the orthosteric site and so reports the proportion of GLP1R that is "signaling competent" 7,19,32 .
In addition to islets, LUXendin645 was also able to label brain tissue. 3D rendering of entire cleared brains using optical projection tomography showed strong LUXendin645 staining in the We also detected LUXendin645labelling in the CP, an epithelial-vascular structure that secretes cerebrospinal fluid (CSF). Rat CP has recently been shown to express GLP1R ex vivo, with Ex4 reducing intracranial pressure in hydrocephalic models 52 . Notably, super-resolution snapshots showed that GLP1R in the ARC, AP, and CP were organized as nanodomains at the membrane. Such higher organization has not been previously appreciated in the brain and it will now be interesting to understand the functional relevance for GLP1 signaling. A number of central GLP1R-expressing regions remained obscure, likely reflecting the pharmacokinetic properties of peripherally administered agonist/antagonist. However, we anticipate that LUXendin645 will be useful for the study of these a LUXendin645 allows super-resolution snapshots of MIN6 β-cells using widefield microscopy combined with super-resolution radial fluctuations (SRRF) (representative image from n = 8 images, three independent repeatss) (scale bar = 10 µm for full-field images, 2.5 µm for zoomed-in images). b-d Confocal and STED snapshots of endogenous GLP1R in LUXendin651-treated MIN6 cells at FWHM = 70 ± 10 nm (mean ± s.d.; n = 15 line profiles measured on the raw data, two independent repeats). Note the presence of punctate GLP1R expression as well as aggregation/clustering in cells imaged just away from b, close to c or next to d the coverslip using STED microscopy (representative image from n = 8 images, three independent repeats) (scale bar = 2 µm for full-field images, 1 µm for zoomed-in images). e, f Representative graph showing spatial analysis of GLP1R expression patterns using the F-function e and G-function f, which show distribution (red line) vs. a random model (black line; 95% confidence interval shown) (n = 6 from three independent repeats). g Approximately 1 in 4 MIN6 β-cells possess highly concentrated GLP1R clusters. h, i LUXendin651 allows GLP1R to be imaged in living MIN6 cells using SRRF h and STED i (representative image from n = 6 and 18 images, three independent repeats for SRRF and STED, respectively) (scale bar = 10 µm for full-field SRRF image, 2.5 µm for the zoomed-in image) (scale bar = 2 µm for STED images). White boxes show the location of zoom-ins. The following compound concentrations were used: 100 nM LUXendin645 (SRRF) and 100-400 nM LUXendin651 (STED). Mean ± s.e.m. are shown. Source data are provided as a Source Data file.    = 1 µm). c GLP1R molecules with diffusion coefficient D < 0.01 are classed as immobile (left), whilst those with D > 0.01 are further divided according to their anomalous diffusion exponent (α), which defines the type of motion followed (confined, normal, or directed) (right) (pooled data from n = 16 cells, 5057-8612 trajectories, six independent repeats). LUXendin645 and LUXendin651 were used at 100 pM. Source data are provided as a Source Data file.
regions following targeted injections. Taken together, these data show that LUXendin645 allows endogenous GLP1R to be visualized in extrapancreatic tissue using both conventional and super-resolution-imaging approaches.
We could also extend studies to hESC-derived β-like cells, where GLP1R expression was detected in spheroids following induction of differentiation and 21 days' culture. LUXendin645 labeling was seen almost exclusively in the β-like cell compartment, reflecting known distributions in mouse 7,25 and human 53 islets. While not all β-like cells were visibly labelled, this likely reflects GLP1R expression levels, which were~50% lower in spheroids than adult human islets. Pertinently, LUXendin645 allowed more mature/differentiated β-like cells to be identified and purified according to GLP1R, which increases almost 20-fold in adult vs. neonatal rat β-cells 42 . Thus, these studies show the potential of LUXendins to understand GLP1R expression patterns in regenerated β-like cells, as well as rapidly mark differentiation/maturity status using a simple one-step surface marker. Since LUXendin645 showed excellent SNR using conventional epifluorescence, it was highly amenable to SRRF analysis. As such, LUXendin645 and its congeners open up the possibility to image the GLP1R at super-resolution using simple widefield microscopy available in most laboratories. For stimulated emission depletion (STED) microscopy experiments, Cy5 was replaced with SiR to give LUXendin651. STED imaging showed that endogenous GLP1R possess a higher structural order in the presence of LUXendin651 binding: namely organization into nanodomains at the cell membrane. Since GLP1R plasma membrane distribution is ligand-dependent, for example via effects on palmitoylation and clustering into cholesterol-rich nanodomains 20 it will be interesting to repeat these experiments using a range of agonists/antagonists. Indeed, LUXendins provide an ideal template for the production of fluorescent ligands that would allow super-resolution examination of nanodomain architecture in response to different activation modes.
Notably, a subpopulation of β-cells appeared to possess concentrated GLP1R clusters even in unstimulated conditions. It will be important in the future to investigate whether this is a cell autonomous heterogenous trait, or instead reflects biased distribution of receptors in membranes of specific β-cells. Lastly, both LUXendin645 and LUXendin651 allowed GLP1Rs to be imaged in live cells by single molecule microscopy, revealing variability in diffusion at the plasma membrane. Particle tracking analyses segregated GLP1R into four different populations based upon diffusion mode, in keeping with data from a class A GPCR, the beta adrenergic receptors 40 . Together, these experiments provide the first super-resolution characterization of a class B GPCR and suggest a degree of complexity not readily appreciated with previous approaches.
LUXendin555 showed antagonist activity in terms of cAMP generation using a HTRF approach. However, we could not confirm this result using a luciferase-based detection method. The reasons for this are unclear, but might include interference of the red fluorescent TMR with either assay, or alternatively different GLP1R coupling strength between cell lines. As such, use of LUXendin555 should consider the possibility that the ligand is an antagonist or agonist. Nonetheless, LUXendin555 retained GLP1R at the membrane, and possesses advantageous properties for in vivo imaging including good two-photon cross-section and high quantum yield.
In summary, we provide a comprehensively tested and unique GLP1R detection toolbox consisting of far-red antagonist labels, LUXendin645 and LUXendin651, an agonist/antagonist LUX-endin555, and knockout Glp1r (GE)−/− animals. Using these freely available probes, we provide an updated view of GLP1R organization, with relevance for the treatment of complex metabolic diseases such as obesity and diabetes. Thus, the stage is set for visualizing GLP1R in various tissues using a range of imaging techniques, as well as the production of peptidic labels and agonists.
Animals. Glp1r (GE)-/-: CRISPR-Cas9 genome-editing was used to introduce a single base pair deletion into exon 1 of the Glp1r locus. Fertilized eggs of female Cas9overexpressing mice (strain Gt(ROSA)26Sor tm1.1(CAG-cas9*,-EGFP)Fezh /J; JAX stock no. 024858) were harvested following super-ovulation. Modified single-guide RNA (Synthego) targeting exon 1 of Glp1r and a single-stranded repair-template were injected at 20 ng/µl into the pronucleus of embryos at the 1-cell stage. In culture, 80% of embryos reached the 2-cell stage and were transplanted into surrogate mice. Glp1r (GE)−/− mice did not integrate the repair-template (confirmed by genotyping PCR), but instead harbored a single nucleotide deletion leading to a frame-shift mutation and loss of GLP1R protein. Knock-in mice that integrated the repair template were not used in the present studies and will be described elsewhere. The targeted locus of Glp1r (GE)−/− offspring was analyzed by PCR and sequencing. Offtarget sites were predicted using the CRISPR Guide Design Tool (crispr.mit.edu). Loci of the top 10 off-target hits were amplified by PCR and analyzed via Sanger sequencing (Supplementary Table 1). Founder animals carrying alleles with small deletions were backcrossed to wild type animals (strain C57BL/6J) for 1-3 generations to outbreed affected off-targets and then bred to homozygosity (Supplementary Figs. 15 and 16). Animals were born in Mendelian ratios, genotyping was performed using Sanger sequencing or PCR. Genotyping PCRs were performed Fig. 7 LUXendin645 highlights GLP1R-expressing neurons in the brain. a LUXendin645 labeling is detected in the the median eminence (ME), arcuate nucleus (ARC), area postrema (AP)/nucleus tractus solitaris (NTS), and choroid plexus (CP), in close association with GLP1-producing neurons, identified using GLU-YFP reporter animals (3V, third ventricle) (representative images from n = 4 animals) (scale bar = 106 µm). b Z-projection of an image stack (~30 µm) showing direct contacts between LUXendin645-labelled and GLP1-producing (GLU-YFP) neurons in the ARC (representative image from n = 4 animals) (scale bar = 20 µm). c, d LUXendin645 labeling co-localizes with GLP1R+ neurons in the AP/NTS c and ARC d, identified using GLP1RCre;LSL-GCaMP3 reporter animals (representative image from n = 4 animals) (scale bar = 61 µm). e Super-resolution imaging using Airyscan shows that LUXendin645 labeling is restricted to the membrane of the cell body and dendrites of GLP1R+ neurons (arrows show cell body and dendrite from left to right, respectively) (representative images from n = 4 animals) (scale bar = 9 µm). f GLP1R form nanodomains in ARC and AP neurons, as well as ependymal cells of the CP (confocal image is shown on the left for comparison) (representative images from n = 8 animals) (scale bar = 9 µm). g Mapping of LUXendin645 distribution in cleared brains shows labelling of the ARC, AP/NTS, CP, lateral ventricles (LV), fourth ventricle (4V), subfornical organ (SFO), and organum vasculosum of the lamina terminalis (OVLT) (representative images from n = 4 animals) (scale bar = 1 mm). Note that, due to suspension of the brain, the coronal section is slightly offset in the dorsal-ventral plane; hence, the SFO appears above the ARC. In all cases, LUXendin645 was injected subcutaneously at 100 pmol/g.
GLU-YFP: Animals harboring YFP under the control of the glucagon promoter were generated and bred as previously described 55 .
CD1 wild-type animals were purchased from Charles River Laboratories UK.
All studies were performed with 6-12-week-old male and female animals,  Islet isolation. Animals were humanely euthanized before injection of collagenase 1 mg/mL (Serva NB8) into the bile duct. Following removal of the inflated pancreas and digestion for 12 min at 37°C, islets were separated using a Histopaque (Sigma-Aldrich) gradient. Islets were cultured in RPMI medium containing 10% FCS, 100 units/mL penicillin, and 100 μg/mL streptomycin.
Super-resolution microscopy. SRRF: MIN6 were treated with 250 nM LUX-endin645 before live imaging, or fixation and mounting on slides using Vectashield Hardset containing DAPI. Imaging was performed using a Crest X-Light spinning disk system in bypass (widefield) mode. Excitation was delivered at λ = 640/30 nm through a ×60/1.4 NA objective using a Lumencor SPECTRA X light engine. Emission was collected at λ = 700/75 nm using a Photometrics Delta Evolve EMCDD. A 250-500 frame raw image sequence was captured (~2 min) before offline super resolution radial fluctuation (SRRF) analysis to generate a single super-resolution snapshot using the NanoJ plugin for ImageJ (NIH) 38 .
STED microscopy: MIN6 cells were treated with 100, 200, and 400 nM LUXendin651 before fixation (4% paraformaldehyde, 20 min). Cells were mounted in Mowiol supplemented with DABCO and imaged on an Abberior STED 775/595/ RESOLFT QUAD scanning microscope (Abberior Instruments GmbH, Germany) equipped with STED lines at λ = 595 and λ = 775 nm, excitation lines at λ = 355, 405, 485, 561, and 640 nm, spectral detection, and a UPlanSApo ×100/1.4 oil immersion objective lens. Following excitation at λ = 640 nm, fluorescence was acquired in the spectral window λ = 650-800 nm. For live-imaging, MIN6 cells were seeded on 18 mm coverslips 24-48 h prior to treatment with 400 nM LUXendin651 for 30-45 min before washing once in full medium. Coverslips were transferred into a magnetic chamber (Chamlide CMB, Live Cell Instrument) and washed once with HBSS buffer (Lonza, with additional 5 mM HEPES bubbled with carbogen for 5 min and pH adjusted to 7.4 with NaOH), which was also used as imaging buffer at 24°C. Live imaging was performed within 45 min after mounting.
Deconvolution was performed with Richardson-Lucy algorithm on Imspector software. FWHM was measured on raw data and calculated using OriginPro 2017 software with Gaussian fitting (n = 15 profiles). Minimum and maximum intensity values refer to intensities after deconvolution for STED images and smoothing with a 1-pixel lowpass Gaussian filter for confocal images. Spatial GLP1R expression patterns were analyzed using the F-and G-functions, where F = distance between an object of interest and its nearest neighbor, and G = distance from a given position to the nearest object of interest (FIJI Spatial Statistic 2D/3D plugin) 57 . Both measures were compared to a random distribution of the same measured objects, with a shift away from the mean ± 95% confidence intervals indicating a non-random or more clustered organization (i.e. more space or smaller distance between objects). Cells possessing highly concentrated GLP1R clusters were identified based upon their fluorescence above a threshold of the population mean fluorescence +1 s.d.
Single-molecule microscopy: For single-molecule experiments, CHO-K1-SNAP_GLP1R cells were seeded onto 25 mm clean glass coverslips at a density of 3 × 10 5 per well. On the following day, cells were labeled in culture medium with 100 pM LUXendin645 or LUXendin651 for 20 min; this concentration avoids labeling all GLP1R, which would otherwise preclude single-molecule analysis in a stable cell line. At the end of the incubation, cells were washed 3 × 5 min in culture medium. Cells were then imaged at 37°C in phenol-red free Hank's balanced salt Fig. 8 LUXendin645 labels human ESC-derived β-like cells. a LUXendin645 (LUX645) labels β-like cells in intact spheroids, which were differentiated and cultured for 21 days. No signal is detected in undifferentiated human ES cells (day 0) or unlabelled β-like cells (-LUX645) (representative images from n = 6 spheroids) (scale bar = 100 µm). b GLP1R gene expression in day 0 undifferentiated cells, day 21 differentiated β-like cells, and human islets (n = 3 donors). c LUXendin645 labelling is localized to strongly insulin (INS)-positive but not strongly glucagon (GCG)-positive areas (representative images from n = 5-6 spheroids) (scale bar = 50 µm). d LUXendin645 (LUX) overlaps more with INS vs. GCG, as calculated using Manders' M1 co-efficient (n = 5-6 spheroids) (unpaired Student's t-test). e Day 21 spheroid sections (5 µm) showing expression of INS and NKX6-1, confirming a differentiated phenotype (representative images from n = 4 spheroids) (scale bar = 100 µm). f FACS plots of day 21 β-like cells with and without LUXendin645 (LUX645) incubation. LUXendin645+ (LUX+) and LUXendin645− (LUX−) cells were sorted for qPCR. g GLP1R, NKX6-1, INS, and GCG gene expression in sorted cells (n = 4 spheroids) (connecting bars indicate LUX+ and LUX− populations in the same samples) (paired Student's t-test). LUXendin645 was applied at 100 nM. Mean ± s.e.m. are shown. *P < 0.05, **P < 0.01 for all statistical tests. Source data are provided as a Source Data file.
solution, using a custom built TIRF microscope (Cairn Research) based on an Eclipse Ti2 (Nikon, Japan) equipped with an EMCCD camera (iXon Ultra, Andor), 637 nm diode laser, and a ×100 oil-immersion objective (NA 1.49, Nikon). Image sequences were acquired with an exposure time of 60 ms.
Image sequences were analyzed with an automated particle detection software (utrack) in the MATLAB environment 40,58 . To analyze the motion of receptors, the time-averaged mean-squared displacement (TA-MSD) 59  analyzed (n traj = 5057 for Cy5 and 8612 for SiR). Trajectories were then categorized according to the diffusion parameters D and α. Particles with D < 0.01 μm 2 s −α were considered to be immobile. Normal diffusion was assigned to particles that had D ≥ 0.01 μm 2 s −α and 0.75 ≤ α ≤ 1.25. Sub-diffusion and super-diffusion were assigned to particles with D ≥ 0.01 μm 2 s −α and α < 0.75 or α > 1. 25, respectively.
Brain labelling. Mice were injected subcutaneously with 100 pmol/g of LUX-endin645 and left for two hours before terminal anaesthesia and transcardial perfuse fixation with 4% fresh formalin. Brains were serially sectioned at 30 µm and mounted on slides before imaging, as above. Super-resolution snapshots (~140 nm lateral resolution) were acquired using a Zeiss LSM880 equipped with an Airyscan module and a ×63/1.2W objective. Brain clearing was carried out using the 3DISCO protocol 60 . Samples were suspended on a needle, before imaging using a custom-built optical projection tomography (OPT) platform, with images collected after excitation at λ = 470 and 660 nm. Images were reconstructed using customwritten MATLAB scripts and visualized in Volocity (Perkin-Elmer).
Two-photon in vivo imaging. Female and male C57BL/6J mice 7-12 weeks of age were used. Each mouse was anesthetized with isoflurane and a small, 1 cm vertical incision was made at the level of the pancreas. The exposed organ was orientated underneath the animal and pressed against a 50 mm glass-bottom dish for imaging on an inverted microscope. Body temperature was maintained using heat pads and heating elements on the objective. The mouse received Hoechst 33342 (1 mg/kg in PBS) to label nuclei, a 150 kDa fluorescein-conjugated dextran (1 mg/kg in PBS) to label vasculature, and 75 μL of 30 µM LUXendin555 via retro-orbital IV injection. Images were collected using a Leica SP8 microscope, equipped with a ×25/0.95 NA objective and Spectra Physics MaiTai DeepSee mulitphoton laser. Excitation was delivered at λ = 850 nm, with signals collected with a HyD detector at λ = 460/50, λ = 525/50, λ = 624/40 nm for Hoechst, FITC, and LUXendin555, respectively. Blood was collected from the tail vein prior to and 30 min after LUXendin555 injection, and glucose was measured using an AlphaTrak2 glucometer. All in vivo imaging experiments were performed with approval and oversight from the Indiana University Institutional Animal Care and Use Committee (IACUC). Fig. 9 LUXendin555 allows in vivo labeling of islets. a-c LUXendin555 labels YFP-AD293_SNAP-GLP1R a but not YFP-AD293 b controls with max labeling at 600 nM c (n = 3 independent assays) (×10 scale bar = 213 µm; ×100 scale bar = 21 µm). d Surface GLP1R expression is similar in LUXendin555 − (LUX555), LUXendin645− (LUX645), and 250 nM Exendin4(9-39)-treated islets (100 nM Ex4, +ve control) (representative images shown above each bar) (one-way ANOVA with Bonferroni's test; F = 173.3, DF = 3) (n = 12 islets, seven animals, three separate islet preparations) (scale bar = 17 µm). e LUXendin555 behaves as an antagonist in HEK-SNAP_GLP1R cells using HTRF-based assays (n = 4 independent assays in duplicate). f LUXendin555 displays agonist activity in CHO-K1-SNAP_GLP1R cells, as assessed using luciferase-based detection (GLO) (n = 3 independent assays) (positive allosteric modulation was achieved using 25 µM BETP). g Schematic depicting the two-photon imaging set up for visualization of the intact pancreas in mice.