Control of human pancreatic beta cell kinome by glucagon‐like peptide‐1 receptor biased agonism

Abstract Aim To determine the kinase activity profiles of human pancreatic beta cells downstream of glucagon‐like peptide‐1 receptor (GLP‐1R) balanced versus biased agonist stimulations. Materials and Methods This study analysed the kinomic profiles of human EndoC‐βh1 cells following vehicle and GLP‐1R stimulation with the pharmacological agonist exendin‐4, as well as exendin‐4–based biased derivatives exendin‐phe1 and exendin‐asp3 for acute (10‐minute) versus sustained (120‐minute) responses, using PamChip protein tyrosine kinase and serine/threonine kinase assays. The raw data were filtered and normalized using BioNavigator. The kinase analyses were conducted with R, mainly including kinase‐substrate mapping and Kyoto Encyclopedia of Genes and Genomes pathway analysis. Results The present analysis reveals that kinomic responses are distinct for acute versus sustained GLP‐1R agonist exposure, with individual responses associated with agonists presenting specific bias profiles. According to pathway analysis, several kinases, including JNKs, PKCs, INSR and LKB1, are important GLP‐1R signalling mediators, constituting potential targets for further research on biased GLP‐1R downstream signalling. Conclusion The results from this study suggest that differentially biased exendin‐phe1 and exendin‐asp3 can modulate distinct kinase interaction networks. Further understanding of these mechanisms will have important implications for the selection of appropriate anti‐type 2 diabetes therapies with optimized downstream kinomic profiles.


| INTRODUCTION
Type 2 diabetes (T2D) is the commonest form of diabetes mellitus, arising because of defective insulin production and insulin resistance. 1 Therefore, increasing insulin secretion from pancreatic beta cells to compensate for insulin resistance and maintain blood glucose levels is an important principle of T2D therapies. 2 Insulin secretion is a key physiological process involving the co-operation of multiple signalling pathways. 3 In particular, glucagon-like peptide-1 receptor (GLP-1R) targeting to potentiate this process is highly effective for the treatment of T2D, with GLP-1R agonists (GLP-1RAs) such as liraglutide, exenatide and semaglutide already in use successfully in the clinic. 4 There are, however, associated deleterious effects of GLP-1RA therapies linked to gastrointestinal disturbances 5 and, given its global prevalence, there remains a huge unmet need for more efficacious agents for T2D treatment. It is therefore necessary to identify possible directions to improve and prolong the beneficial effects of GLP-1RAs while reducing the negative ones. 6 In the context of deepening understanding of multidimensional networks in G protein-coupled receptor (GPCR) signal transduction, the concept of 'biased agonism', which describes ligand selectivity for specific downstream pathways, has emerged. 7 For the GLP-1R, the G α s subunit is the main G protein subtype binding to active receptors, triggering adenylate cyclase (AC) activation, and resulting in the production of cyclic adenosine monophosphate (cAMP). 8 Additionally, some studies highlight G protein-independent signalling elicited by the recruitment of β-arrestins to active GLP-1Rs. 9 Single amino acid changes in the reference GLP-1RA exendin-4 (exenatide) were found to result in pronounced GLP-1R-biased signalling. 10 Specifically, two exendin-4 derivatives, exendin-phe1 and exendin-asp3, were associated with preferential G α s or β-arrestin recruitment, respectively.
GLP-1R activation with these biased compounds was also linked to changes in receptor trafficking profiles, with exendin-phe1 triggering slow GLP-1R internalization followed by rapid recycling, while exendin-asp3 is associated with rapid GLP-1R internalization and preferential sorting of the receptor for lysosomal degradation. As a result, GLP-1R desensitization was reduced and insulin release prolonged with exendin-phe1 in rat INS-1832/3 beta cells, and in vivo glucoregulation improved in mice. 10 GLP-1R-biased agonists are therefore potentially attractive options to improve existing incretin therapies. 11 Indeed, the newly approved incretin mimetic tirzepatide may depend partly on biased GLP-1R agonism for its high therapeutic efficacy. 12 However, the complex molecular mechanisms leading to the divergent responses of biased GLP-1RAs remain unclear.
In this context, phosphorylation-regulated signalling pathways play an essential role in enabling cells to respond quickly and effectively to various cellular signals and stressors. 13 Protein kinases are an important group of intracellular enzymes involved in the regulation of cellular functions such as proliferation, apoptosis and metabolism. 14 In eukaryotic cells, serine/threonine (Ser/Thr) and tyrosine (Tyr) are the main amino acid phosphosites. 15 Analysis of cellular 'kinomic' profiles is a recent field of study whereby global kinase signalling responses are determined. This approach has been chiefly applied to cancer research, resulting in the development of specific kinase inhibitors used successfully in cancer therapy. 16 However, despite its potential to increase our understanding of the complex mechanisms linked to GLP-1R signalling, kinase network analysis has not been previously applied to pancreatic beta cells in the context of GLP-1RA stimulation, 17 or for the analysis of downstream effects of biased GLP-1R signalling.
In this report, we describe the acute (10-minute) versus long-term (120-minute) protein tyrosine kinase (PTK) and serine/threonine kinase (STΚ) activity profiles of human EndoC-βH1 pancreatic beta cells under the effects of balanced (exendin-4) versus oppositely biased (exendin-phe1, exendin-asp3) GLP-1RAs using PamChip technology. The major interest of this study is the identification of critical kinases activated by GLP-1R downstream signalling, including the determination of those differentially regulated by biased agonism.

| PTK activity profiling
PTK profiles were determined using the PTK PamChip Array, a flowthrough microarray assay that contains 196 unique peptide sequences consisting of 13-15 amino acids with putative endogenous phosphorylation sites that are substrates of Tyr kinases. The principle of the assay is to determine the phosphorylation of the substrate peptide by detection of fluorescence following binding to a fluorescein isothiocyanate (FITC)conjugated PY20 antiphosphotyrosine antibody. PTK mixtures were prepared using standard protocols provided by Pamgene. Sample incubation, detection and preliminary analysis were performed in the PamStation 12 machine according to the manufacturer's instructions.
The microarray analysis was run for 94 cycles, while the CCD camera of the PamStation 12 recorded images at kinetic read cycles 32-93 and at the end level read cycle at 10, 20, 50, 100 and 200 ms. 17 Instrument manipulation as initial sample/array processing and image capture was performed using Evolve (Pamgene) software. In the final step, data were normalized with the BioNavigator analysis software tool (Pamgene) and the mean fluorescence signal intensity changes (log 2 [FC]) of four replicates per experimental group were obtained and normalized to the corresponding vehicle control.

| STK activity profiling
STK profiles were determined using the STK PamChip Array, a flowthrough microarray assay that contains 144 unique peptide sequences, which, as for the PTK array, consist of 13-15 amino acids containing putative endogenous phosphorylation sites that are substrates of Ser/Thr kinases. The phosphorylation of these substrate peptides was detected by fluorescence using a two-step assay 17

| Matching substrates to upstream kinases
Kinase-substrate information was obtained from PhosphoNET, Uniprot and PhosphoSite databases. The top 50 Ser/Thr kinases and 25 Tyr kinases with prediction V2 scores higher than 300 were selected for each phosphosite from PhosphoNET (Supplementary Files 1 and 2; see the supporting information). Kinase-substrate mapping was performed in R.

| Upstream kinase analysis
We modified the kinase-substrate enrichment analysis (KSEA) formula 19 to use the average phosphosite log 2 (FC) weighted by the prediction V2 scores for each phosphosite of a given kinase. The modified z kinase score formula is as follows: where s w represents the weighted mean log 2 (FC) of the substrate subset for a given kinase, p denotes the mean log 2 (FC) of all phosphosites in the raw dataset, m is the number of phosphosite substrates identified for each kinase, SD represents the standard deviation of the log 2 (FC) across all phosphosites and x represents the PhosphoNET predictor V2 score. The scores were assumed to be normally distributed, therefore the P values were determined by one-tail tests to estimate the probability that the real scores were at least as extreme as the measured scores. Data sorting, analysis and calculation were performed in Excel and R. Kinase scores were visualized in R mainly using the 'pheatmap' package (v. 1.0.12; https://github.com/raivokolde/ pheatmap).

| Pathway enrichment analysis
To conduct a representative analysis of kinase lists, the 'enrichKEGG' function in R package ClusterProfiler v. 4.0 was applied to determine whether the kinases identified are enriched within specific pathways. 20 Statistical thresholds of P less than .05 and q-value less than 0.05 were chosen to select significantly enriched Kyoto Encyclopedia of Genes and Genomes (KEGG) terms and pathways. A dot plot of the top 10 KEGG terms and pathways, ranked by kinase ratio and coloured by adjusted P value, was created. A Venn diagram was generated to identify the kinases shared within these pathways using Venn webtools (https://bioinformatics.psb.ugent.be/webtools/Venn/), with a list of kinases illustrated.

| RNA-seq database analysis
High-throughput sequencing data were retrieved from Gene Expression Omnibus (GEO) databases to assess the level of expression of specific kinase genes. Gene expression data from the EndoC-βH1 cell line were obtained using GEO accession GSM3333914. Additionally, raw read count values of 18 non-diabetic and 39 T2D islet samples from the GSE164416 series were analysed using DESeq2 21 in order to compare differentially expressed kinase genes, followed by integration with our kinomic analysis results.

| Generation of kinase tree maps and kinase networks
An annotated phylogenetic kinase tree was generated using the Kin-Map online tool (beta version, http://kinhub.org/kinmap/). 22 The data source was selected as 'Disease associated CTTV', and specific

| Source code
All the source code from the present study can be found at https:// github.com/JiannanX/kinomic-assay.

| Kinomic effects of acute versus sustained GLP-1RA stimulation in human EndoC-βh1 pancreatic beta cells
The kinomic assays were performed following the workflow summarized in Figure 1 to analyse active kinases for two treatment duration groups: (a) 10-minute (acute) and (b) 120-minute (sustained) in EndoC-βh1 cells, a cell line that we have previously shown capable of displaying both cAMP and insulin secretion responses to GLP-1R stimulation. 23 We chose to assess acute versus sustained effects because differences in glucoregulatory responses between biased GLP-1RAs are exacerbated over time. 10 For each group, four treatment conditions were established: vehicle, exendin-4, exendin-phe1 and exendin-asp3 (at 100 nM each). we focused on the first three pathways from each kinomic dataset.
The mitogen-activated protein kinase (MAPK) signalling pathway was highlighted as the most enriched pathway for both PTK and STK analyses, in agreement with previous reports linking MAPK family activities to pancreatic beta cell function. 27 For instance, PKA-activated MAPKs play a crucial role in accelerating the cell cycle, which is associated with maintenance of beta cell mass. 28 According to the Venn diagram analysis of the kinases implicated ( Figure 3C,D)

| Biased agonist-mediated changes in downstream kinase signatures
Besides comparing kinase activity changes caused by acute and sus-  Table 1 for PTKs and Table 2 for STKs). Interestingly, our results suggest that there are some kinases being differentially regulated by exendin-phe1 compared with exendin-asp3 stimulation. For example, for PTKs, acute stimulation with exendin-phe1 leads to increased activity of lemur tyrosine kinase 2 (LMR2), while exendin-asp3 stimulation for the same time causes a reduction in LMR2 activity compared with exendin-4 ( Figure 4A). Moreover, although the change trends in Tyr kinase activity were similar for both biased agonists following sustained exposure, the effect of exendin-phe1 was more pronounced than that of exendin-asp3 at this time point for several kinases, including cJun Nterminal kinases (JNKs) and insulin receptor (INSR) ( Figure 4A). Results for STKs showed more dramatic differences between both biased agonists, especially after sustained treatment, with prolonged exendin-phe1 responses again leading to significantly increased effects compared with exendin-asp3 ( Figure 4B). Examples of differentially activated kinases include cyclin-dependent kinases (CDKs) and dual-specificity tyrosine-regulated kinases (DYRKs), which showed increased activity following acute exendin-phe1 stimulation, while PKCs showed higher acute activation with exendin-asp3 ( Figure 4B). Additionally, MAPK kinases (MAP2K/MEKs: MEK3, MEK6) and p21-activated kinases (PAKs: PAK4, PAK5) showed greater discrepancies between exendin-phe1 and exendin-asp3 at the sustained time point. According to a further KEGG analysis, the Ras, MAPK and PI3K-Akt signalling pathways were again the top three enriched pathways related to Tyr kinases ( Figure S2A). However, only the MAPK signalling pathway remained as one of the three most enriched for Ser/Thr kinases, this being the only one from the top three clearly related to pancreatic beta cell function ( Figure S2B). To provide more detailed evidence for the regulation of the pathways involved, a systems biology approach was conducted and depicted using STRING v. 11.0 (https://string-db.org) to analyse interactions amongst the three top-related kinases pathways for Tyr kinases ( Figure 4C) and within the MAPK signalling pathway for Ser/Thr kinases ( Figure 4D). These analyses showed that the selected pathways were highly overlapped and correlated, although some kinases may act independently from others. Specifically, most PTKs can be linked to JAK2, FGFR2 and JAK3 ( Figure 4C), while BRAF and PRKCB were the most centred among the STKs ( Figure 4D), suggesting that they may play important roles in regulating beta cell-biased GLP-1RA responses.
As pancreatic beta cells constitute more than 50% of the mass of human islets, 3 we additionally confirmed expression of the kinases assessed in this study in a human islet RNA-seq database that con- 10 min 120 min 10 min 120 min 10 min 120 min 10 min 120 min Family Kinase phe1 asp3 10 min 120 min 10 min 120 min  19 Here, we used the PhosphoNET database, which predicted more than 300 kinases for our substrates. By weighting the prediction V2 score, we were able to assess the influence of the different kinases in the degree of phosphorylation of each substrate. Although our substrate peptides are comparatively well conserved, public databases are mainly generated based for human kinases. Therefore, currently this method of analysis may only be suitable for human cell lines such as EndoC-βh1.
Our study particularly aimed to determine signalling differences elicited by biased GLP-1RAs. The KEGG pathway analysis allowed the identification of specific kinase families involved in this process, including JNK, implicated in the regulation of most cellular processes as one of the main MAPK subfamilies. 27 The main functional role of activated JNK is to phosphorylate c-Jun, with Ser63 and Ser73 phosphorylations required for c-Jun transcriptional activation. 32 All the JNKs (JNK1, 2 and 3) were identified in our kinase analysis as displaying increased activity following exendin-phe1 stimulation over a sustained period. However, JNK inhibition has been suggested to prevent beta cell dysfunction and apoptosis in both human islets and beta cell lines. 27 Note: z scores: """ (+3.63 to +4.84); "" (+2.42 to +3.63); " (+1.21 to +2.42); -(À1.21 to +1.21); # (À2.42 to À1.21); ## (À3.63 to À2.42); ### (À4.84 to À3.63); ✕: kinase not found. beta cell function by controlling glucose coupling by directing glucosederived carbons into the tricarboxylic acid cycle, an important process for the maintenance of mitochondrial structure and function. 41 Studying the effect of GLP-1R agonism in the control of this pathway will therefore be particularly important to understand the effects of incretin action on the regulation of beta cell function.
Our analysis of kinomic profiles provides evidence for the importance of factors related to the growth, proliferation and function of pancreatic beta cells to explain the effects of GLP-1R action. Besides some well-known kinase families, we have identified some kinases whose precise roles in beta cells are still unknown. For example, the DYRK family, which has previously been hinted at being closely related to the control of beta cell function, 42 has been shown here to be differentially modulated by biased GLP-1R agonism, with exendin-phe1 being more effective than exendin-asp3 at acute DYRK activation. To