Characterization of genetic variants of GIPR reveals a contribution of β-arrestin to metabolic phenotypes

Abstract

Incretin-based therapies are highly successful in combatting obesity and type 2 diabetes¹. Yet both activation and inhibition of the glucose-dependent insulinotropic polypeptide (GIP) receptor (GIPR) in combination with glucagon-like peptide-1 (GLP-1) receptor (GLP-1R) activation have resulted in similar clinical outcomes, as demonstrated by the GIPR-GLP-1R co-agonist tirzepatide² and AMG-133 (ref. ³) combining GIPR antagonism with GLP-1R agonism. This underlines the importance of a better understanding of the GIP system. Here we show the necessity of β-arrestin recruitment for GIPR function, by combining in vitro pharmacological characterization of 47 GIPR variants with burden testing of clinical phenotypes and in vivo studies. Burden testing of variants with distinct ligand-binding capacity, Gs activation (cyclic adenosine monophosphate production) and β-arrestin 2 recruitment and internalization shows that unlike variants solely impaired in Gs signalling, variants impaired in both Gs and β-arrestin 2 recruitment contribute to lower adiposity-related traits. Endosomal Gs-mediated signalling of the variants shows a β-arrestin dependency and genetic ablation of β-arrestin 2 impairs cyclic adenosine monophosphate production and decreases GIP efficacy on glucose control in male mice. This study highlights a crucial impact of β-arrestins in regulating GIPR signalling and overall preservation of biological activity that may facilitate new developments in therapeutic targeting of the GIPR system.

Fig. 1. cAMP production and binding properties of the 47 non-synonymous GIPR variants.

a, Localization of GIP-secreting cells in the gastrointestinal tract. b, GIP secretion in response to ingestion of low (25 g), moderate (50–75 g) or high (100–125 g) amounts of glucose according to previous human studies (N = 8)^,. The blue box highlights the range from low (10 pM) to high (100 pM) levels of GIP. c, The corresponding area of the physiological range of GIP on a WT GIPR dose-response curve for cAMP production. d, cAMP production at 100 pM GIP, shown as percentage of activity compared with WT GIPR. Grey, WT-like activity and dark red, below 50% activity of WT. The data represent the mean ± s.e.m. WT GIPR: N = 18; R12Q, S64P, Y73H, R164Q, P195R–A207V, S382T–S382Y, S415I–G449A: N = 6; L13F–R38L, V99I–R101H, R164W, T168A, L261H, E461*–E463Q: N = 4; R43G, M67R, C84P–W90L, I221M, R336P: N = 5; T116R–A150T, H166Y, R183Q–A185V, W233*–Y258S, E291K–R300W, E354Q–I378M, K397N–V399M: N = 3; R190Q: N = 9; E288G: N = 8; L304R: N = 7. Independent experiments were performed in duplicate. e,f, Homologous competition binding of GIP, showing maximum binding capacity (B_max) and affinity (K_d) compared with WT GIPR. The variants are arranged in the same order as in b. NB, no binding. The data represent the mean ± s.e.m. WT GIPR: N = 18; R12Q: N = 5; L13F–R101H, G144S–A207V, W233*, E288G, E354Q, S382T–G449A, E463Q: N = 3; T116R, I221M, Y258S–L261H, E291K–R336P, I378M, E461*: N = 4. Independent experiments were performed in duplicate. g, Illustration of the GIPR with variants located at functional sites (orange) and non-functional sites (blue). Green area: ligand binding site; yellow area: (micro) switches; red area: G protein interface (left). Mean difference in cAMP efficacy between variants at functional (N = 13) and non-functional (N = 34) sites (right). No significant difference was observed between the efficacy of 100 pM, P = 0.30. Statistical significance was assessed using the Wilcoxon rank-sum test (two sided). Source data

Fig. 2. β-arrestin 2 recruitment profiles of GIPR non-synonymous variants.

a, β-arrestin 2 recruitment potency (logEC₅₀) of GIP. Variants are arranged according to cAMP efficacy of 100 pM GIP. b, β-arrestin 2 recruitment maximal signalling (E_max), shown as a percentage of WT GIPR activity. NA, no activation. c, Dose-response curves of cAMP production (left) and β-arrestin 2 recruitment (right) of GIPR variants R12Q, I221M, M67R and S64P, respectively. Data represents the mean ± s.e.m. WT GIPR: N = 30; R12Q–S64P, Y73H–C84P, V99I–T168A, A185V, P195R: N = 5; M67R: N = 6; W90L, R183Q, A207V, R336P–E354Q: N = 4; R190Q, E288G: N = 6; I221M–L261H, E291K–L304R, I378M–E463Q: N = 3. Independent experiments were performed in duplicate. Source data

Fig. 3. Binding- and signalling-based grouping of GIPR variants and their respective burden association on cardiometabolic phenotypes in the UK Biobank.

a, The overlap between non-synonymous GIPR variants found in the Danish population and the UK Biobank. N, total cohort sample size. b, The pharmacological profile for all GIPR variants in terms of cAMP accumulation (efficacy (Eff.) at 100 pM GIP), binding ability and β-arrestin recruitment compared with WT receptor clustered into six in vitro phenotype groups. c, A forest plot showing burden test statistics in the UK Biobank cohort (N = ~440,000). Carriers represent the number of individuals for each phenotype carrying any of the GIPR variants within that GIPR variant group. q, FDR-adjusted P values of the burden tests (performed using ACAT-O). pLoF denotes variants predicted to cause loss of function with VEP and/or CADD score > 30, that is, pLoF variants were not functionally characterized. ^aThe WT-like_cAMP/WT-like_arr variant, E354Q, was excluded from burden testing due to high MAF (~20%). Beta represents the effect size as standard deviation of the phenotype with error bars representing the 95% CI. Source data

Fig. 4. Impact of β-arrestin on endosomal signalling of GIPR variants and glucose excursion in male mice.

a, Gs pathway engagement at early endosomes of representative GIPR variants from each of the six in vitro-based groups. The delta (Δ) BRET values are calculated by subtracting the vehicle value from the top values (1 µM) and evaluated by a one sample t-test (two-sided) performed on the mean of individual experiments with hypothetical value set to 0. The data represent mean ± s.e.m. of WT GIPR, N = 8 and GIPR variants, N = 4. Independent experiments were performed in duplicates. Exact P values are shown above the data points for variants significantly different from 0. b, Pearson correlation analysis of endosomal signalling (ΔBRET) with β-arrestin 2 recruitment (E_max % of WT) of the variants. LoF/LoF (LoF_cAMP/LoF_arr), LoF/WT (LoF_cAMP/WT-like_arr), WT/LoF (WT-like_cAMP/LoF_arr), WT/GoF, (WT-like_cAMP/GoF_arr), GoF/WT (GoF_cAMP/WT-like_arr) and WT/WT (WT-like_cAMP/WT-like_arr). The data represent the mean ± s.e.m. Independent experiments were performed in duplicates. The number of replicates is the same as Figs. 4a and 2. c, Workflow of the pharmacologic studies in Arrb2 knockout (KO) mice. t₀, t₁₅, t₃₀, t₆₀ and t₁₂₀ represent timepoints (in minutes) after initiating an intraperitoneal glucose tolerance test (IPGTT). d,e, Glucose excursion during an IPGTT in WT male mice (N = 10–14 per group) (d) and male Arrb2 KO mice (N = 12–13 per group) (e) 15 min following vehicle (black circles), d-Ala GIP (orange triangles) or GLP-1 (teal squares). Mouse data are represented as mean ± s.e.m. Statistical significance was determined by two-way analysis of variance, and Holm-Šidák testing was used to correct for multiple testing. The P value for either GLP-1 or GIP compared with vehicle is shown. Figure 4c was created with BioRender.com. Source data

Extended Data Fig. 1. Maximal signalling and potency of the 47 GIPR variants in cAMP production, and the relation between WT GIPR gene expression and signalling.

a, cAMP production E_max, shown as percentage of WT GIPR activity. The variants are arranged according to the cAMP production (efficacy at 100 pM GIP). Grey, WT-like; dark red, below 50% activity. b, cAMP production potency (logEC₅₀). c, E_max (cAMP production) vs B_max (homologues competitive binding) of WT GIPR, expressed at DNA doses ranging from 10 ng to 10 µg. d, GIP binding affinity (K_d) of WT GIPR at DNA doses ranging from 10 ng to 10 µg. e, Potency (logEC₅₀) of WT GIPR at DNA doses ranging from 10 ng to 10 µg in cAMP production. f, Linear correlation between WT GIPR expression and signalling of the 47 GIPR variants. Expression was measured as B_max, and signalling was measured as cAMP production efficacy at 100 pM. The data represent the mean ± s.e.m. of a minimum of N = 3. Independent experiments were performed in duplicate. Source data

Extended Data Fig. 2. Variant effect predictor (VEP) correlations to laboratory-tested GIPR variants.

a, Cluster heatmap of the VEP normalized predicted deleteriousness scores, ranging from benign (0) to deleterious (1). The x-axis shows the included VEPs, and the classification of the laboratory-tested variants according to cyclic adenosine monophosphate (cAMP) accumulation (left). b, Prediction of variants potentially leading to a loss-of-signalling (LoS) molecular phenotype with four VEP masks (M1–M4). The 47 GIPR variants and their allele frequencies in the Danish population (all four study cohorts combined) are shown in the left panel, sorted according to their cAMP production (from low (bottom) to high (top)) compared with that of WT (Fig. 1d). M1, the Loss-of-Function Transcript Effect Estimator (LOFTEE) high-confidence (HC) mask with predicted LoS variants (red). M2–M4, grouped VEPs. A variant was considered a LoS variant if all VEPs in one of the four masks predicted the variant as a LoS variant (red). ‘pLoS’ represents the variants predicted to have a LoS molecular phenotype (bright red). ‘in vitro LoF’ represents the tested GIPR LoF variants with < 50% cAMP production (bright red). NAs are white. In total, 17 variants were pLoS variants, of which 15 overlapped with the in vitro cAMP LoF variants, and two were in vitro cAMP WT-like variants. Source data

Extended Data Fig. 3. Internalization abilities of the eight LoF_cAMP/WT-like_arr variants and two LoF_cAMP/LoF_arr.

a, The donor signal represents an estimate of receptor surface expression of the cells transfected with SNAP-WT GIPR or variants. N = 4. b, Internalization ratios of the SNAP-GIPR variants over time following stimulation with GIP in doses ranging from 1 nM to 1 µM. N = 4. c, Area under the curves of the SNAP-WT GIPR vs. SNAP-GIPR variants in response to GIP stimulation in doses ranging from 1 nM to 1 µM. N = 4. d, Dose-response curves of maximal internalization of the SNAP-GIPR variants, shown as percentage of maximal SNAP-WT GIPR internalization. N = 4. Data are shown as the mean ± s.e.m. Independent experiments were performed in triplicates. Source data

Extended Data Fig. 4. Burden association of the GIPR variant groups on cardiometabolic phenotypes in the Danish cohorts.

Forest plot showing burden test statistics in the Danish population (Inter99). N_C, number of individuals for each phenotype carrying any of the variants within that variant group. N_N-C, number of individuals not carrying any of the variants within that variant group. q, false discovery rate (FDR) adjusted p-values of the burden tests (performed using SKAT/SKAT-O). pLoF denotes variants predicted to cause loss of function with VEP. pLoF variants were not functionally characterized but include some of the LoF_cAMP/LoF_arr variants. The E354Q variant was excluded from burden testing of the WT-like_cAMP/WT-like_arr group due to high minor allele frequency (MAF; ~20%). Beta and error bars represent effect size as standard deviation and the 95% confidence interval, respectively, derived from a linear model. Source data

Extended Data Fig. 5. Forest plots of burden tests of cardiometabolic phenotypes in the Danish population (Inter99).

Association results of metabolic-associated biochemistry measures, estimates of beta cell function and insulin sensitivity, and cardiovascular-associated phenotypes (the lipid profile) for all variant groups are shown. Synonym, WT/WT, LoF/WT, LoF/LoF, and pLoF represent the groups of synonymous, WT-like_cAMP/WT-like_arr, LoF_cAMP/WT-like_arr, LoF_cAMP/LoF_arr, and predicted LoF variants, respectively. The E354Q variant was excluded from burden testing of the WT-like_cAMP/WT-like_arr group due to high minor allele frequency (MAF; ~20%). Abbreviations: FPG, fasting plasma glucose levels; PG, plasma glucose levels; FSI, fasting serum insulin levels; SI, serum insulin levels; FS C-pep, fasting serum C-peptide levels; HbA1c, hemoglobin A1c; AUC30ig, area under curve 0–30 min insulin/glucose; BIGTT-AIR, beta cell function insulin sensitivity glucose tolerance test (BIGTT)-acute insulin response (insulin secretion); BIGTT-SI, BIGTT-sensitivity index (insulin sensitivity); Insu index, insulinogenic index (insulin secretion); HDL, high-density lipoprotein; LDL, low-density lipoprotein; VLDL, very low-density lipoprotein. N_C represents the number of individuals for each phenotype carrying any of the variants within that variant group. N_N-C represents the number of individuals not carrying any of the variants within that variant group. q, FDR adjusted p-values of the burden tests (performed using SKAT/SKAT-O). Beta and error bars represent effect size as standard deviation and the 95% confidence interval, respectively, derived from a linear model. Source data

Extended Data Fig. 6. Forest plot of binary phenotype burden tests in the UK Biobank population.

For each GIPR variant group, the odds ratios (OR) are shown for binary traits with corresponding confident intervals. ‘Carriers’ represent the number of individuals for each phenotype carrying any of the GIPR variants within that GIPR variant group. We excluded the E354Q variant from the burden testing of the WT-like_cAMP/WT-like_arr group due to high minor allele frequency (MAF; ~20%). ‘N’ represents the total number of individuals for each phenotype. q, the false discovery rate (FDR) adjusted p-values of the burden tests (performed using ACAT-O). Error bars represent the 95% confidence interval of the OR. Source data

Extended Data Fig. 7. Comparison of p-values between with all UK Biobank samples or only individuals from the United Kingdom.

-log₁₀(P) for each phenotype/grouping combination when testing either all samples passing quality control or limited to those with United Kingdom (UK) ancestry. The x-axis shows p-values (derived from burden testing using ACAT-O) for testing of all 465,000 individuals (Source Data of Fig. 3 and Extended Data Fig. 6). The y-axis shows p-values (derived from burden testing using ACAT-O) for testing of only 428,000 UK individuals. a, Binary traits. Circles represent type 2 diabetes (T2D), triangles represent obesity, and squares represent hypertension. Error bands represent the 95% confidence interval (CI) of the regression line. b, Quantitative traits. Circles, triangles, squares, pluses, crossed squares, and asterisks represent body fat percentage, body mass index, diastolic blood pressure, hip circumference, systolic blood pressure, and waist circumference, respectively. Error bands represent 95% CI of the regression line. Source data

Extended Data Fig. 8. Meta-analysis regression of single-variant phenotypic association estimates on cAMP signalling and β-arrestin recruitment.

UK Biobank single-variant phenotypic association effect sizes (see Fig. 3) regressed on each variant’s in vitro effect on cAMP production (see Fig. 1) and β-arrestin recruitment (see Fig. 2) compared to WT GIPR. Analysis performed as a linear interaction model of phenotypic effect size estimates (y-axis) as response of single-variant (points) β-arrestin recruitment (B-arr, blue) and cAMP signalling (red) impact (x-axis). We used Bayesian Gaussian model which accounts for uncertainty in both individual phenotype estimates (vertical bars as standard error of the beta (β) from the linear regression model) and in vitro uncertainty (not shown). Error bands represent 95% credible interval and grey dashed lines represent no effect. Source data

Extended Data Fig. 9. In vitro comparison of parental HEK293A cells and β-arrestin 1/2 KO cells transfected with GIPR or GLP-1R.

a1, cAMP accumulation measured in parental HEK293A cells transfected with 10 µg or 0.5 µg of receptor DNA upon cognate ligand stimulation. GLP-1R stimulated by GLP-1 mediates a more efficacious cAMP response at both receptor concentrations compared to GIPR stimulated by GIP. a2-a3, Receptor surface expression (B_max) determined by homologous radio-ligand competition binding shows that GLP-1R a2, and GIPR a3, are expressed to the same extent on the cell surface. The cAMP formed by GLP-1R is approximately double the magnitude compared to GIPR, considering the number of receptors on the surface. b, Potency (pEC₅₀) of GIP in cells transfected with GIPR. In the absence of β-arrestin, the potency of GIP is significantly decreased (p < 0.01). c-d, Corresponding GIPR cAMP formation upon 100 c, and 10 pM d, GIP stimulation. The absence of β-arrestin leads to a significant decrease in cAMP formation (p < 0.01). e, Potency (pEC₅₀) of GLP-1 in cells transfected with GLP-1R. A slightly higher, yet nonsignificantly, potency of GLP-1 on the GLP-1R was observed in β-arrestin KO cells compared to parental cells. f-g, Corresponding cAMP formation upon 100 pM f, and 10 pM g, GLP-1 stimulation. A nonsignificant reduction in cAMP formation was observed between WT and β-arrestin KO cells. Statistical significance was assessed using a two-tailed paired t-test. The data represent the mean ± s.e.m. of N = 3 performed in duplicates. Source data

Extended Data Fig. 10. Graphical illustration of the molecular and cellular pharmacological consequences of the GIPR variant groups.

GIP binds to the GIPR, leading to distinguishable patterns of Gαs signalling and β-arrestin recruitment, hereunder receptor internalization and endosomal signalling, depending on the GIPR variant groups. Only the LoF_cAMP/LoF_arr variant group is significantly associated with lower adiposity in humans. Intracellular red arrows, including red blunted arrows, indicate which pathways are affected in the variant receptor function. A red cross indicates diminished activation of the pathway. Adapted from ‘Cellular Environment (background)’, by BioRender.com (2023). Retrieved from https://app.biorender.com/biorender-templates.