Genetically modeled GLP1R and GIPR agonism reduce binge drinking and alcohol-associated phenotypes: a multi-ancestry drug-target Mendelian randomization study

To construct our primary instruments for European ancestry analyses, we applied cis-instrumentation strategies, selecting genetic variants within the locus of each drug target [22]. For GLP1R agonism, we used independent (linkage disequilibrium R² < 0.1), genome-wide significant (P < 5 × 10⁻⁸) SNPs located ±500 kilobases of the GLP1R locus (chromosome 6:39,016,574–39,055,519 on GRCh37/hg19) and associated with HbA1c levels (mmol/mol) in UK Biobank European ancestry participants. GIPR and GLP1R instruments were also separately constructed using BMI GWAS data, capturing their impact on appetite suppression and weight loss. For Europeans, we used a meta-analysis of GIANT and UK Biobank summary statistics [28]. Since no genome-wide significant BMI-associated variants were found within the GLP1R locus, we applied a relaxed P-threshold of 5 × 10⁻⁶, a strategy commonly used in cis-instrument MR studies when conventional genome-wide significant variants are absent [29, 30]. To model the effects of dual GLP1R and GIPR agonists (e.g., tirzepatide [31]), we combined the primary HbA1c and BMI instruments for GLP1R and GIPR into single instruments capturing both loci. The genetic models concatenating the glucose-lowering effects from both GLP1R and GIPR loci or BMI impact is denoted as "GIPR/GLP1R" for the remainder of the manuscript. To further validate the robustness of our findings, we used the Million Veterans Program (MVP) [32] GWAS summary statistics for BMI and HbA1c to perform replication analyses by instrumenting GLP1R, GIPR, and dual agonists in these biomarker data.

In addition to using two independent GWAS data sources per exposure to test robustness and consistency of the estimates, we also conducted sensitivity analyses using alternative instruments to evaluate potential pleiotropy, test for bias in instrument selection, and confirm that findings remain consistent across multiple instrument definitions [21, 22]. First, we constructed instruments for GLP1R, GIPR, and dual GIPR/GLP1R agonism by identifying SNPs that were associated with GLP1R and GIPR gene expression in the Genotype-Tissue Expression (GTEx) Project V8 eQTL dataset [33]. Lookups were performed separately for both GLP1R and GIPR, ensuring that selected variants were relevant for each receptor's gene regulation. Second, we created additional GLP1R and GIPR instruments from the HbA1c and BMI GWAS data using missense SNPs (rs10305420 for GLP1R and rs1800437 for GIPR [rs1800437 was only extracted from BMI data because it was not present in the HbA1c data), identified using the VariantAnnotation R package (version 3.20) [34]. Finally, we constructed cis-instruments for GLP1R and GIPR reflecting gene expression changes in the brain cortex [35], evaluating potential effects on central nervous system regulation.

For the exploratory non-European ancestry analyses, we leveraged ancestry-specific datasets to construct GLP1R and GIPR genetic instruments. For East Asian ancestry, we used Biobank Japan (HbA1c N = 42,790; BMI N = 158,284) [36], identifying cis-acting variants within the GLP1R and GIPR loci to model the glycemic and metabolic effects of agonism. For African ancestry, we used data from the UK Biobank Pan-UKB release (HbA1c N = 5,901; BMI N = 6,458) constructed by the Neale Lab [27], applying similar cis-instrumentation strategies to approximate pharmacological modulation of these targets. All instruments used ancestry-specific 1000 Genomes panels [37].

Drug-target MR instruments aim to capture the lifelong expected physiological impact of the target of interest (here, GLP1R and GIPR agonism through genetic variants associated with BMI and HbA1c levels) [21, 22]. Importantly, the reported association findings are not confined to individuals with specific BMIs or Hba1c levels. Rather, they reflect the relationships of GLP1R and GIPR activity as inferred from the instruments genetically proxied effects on BMI and glycemic control. These instruments are constructed using variants that are associated with BMI and HbA1c within the GIPR and GLP1R loci, which serve as proxies for the therapeutic effects of drug targets that act through these pathways. Therefore, the findings are linked to changes in BMI or HbA1c due to the metabolic and appetite-regulating mechanisms of GLP1R and GIPR agonism. This approach allows for the assessment of drug target activity over the lifelong genetic exposure, rather than short-term clinical intervention, providing insights into the broader implications of GLP1R and GIPR modulation beyond metabolic regulation [21, 22].

To validate the instruments, we assessed their associations with both type 2 diabetes (T2D) and obesity for each genetically proxied GLP1R, GIPR, and dual GIPR/GLP1R agonism exposure, ensuring they captured expected metabolic effects [31, 38, 39]. Our primary validation comparisons examined whether HbA1c-based instruments for GLP1R and GIPR agonism were associated with T2D risk and whether BMI-based instruments for GLP1R and GIPR agonism were associated with obesity risk, given these are the primary clinical effects of these pathways. However, we also tested all exposure-outcome pairs, including whether BMI-based instruments were associated with T2D and whether HbA1c-based instruments were associated with obesity, to fully characterize metabolic impacts. T2D effects were assessed using GWAS data from European, East Asian (88,109 cases/339,395 controls), and African (N = 50,251 cases/103,909 controls) populations [40]. Obesity associations were evaluated using European ancestry GWAS data from FinnGen release 11 [41], though comparable datasets were not available for East Asian or African populations. The expected associations confirm that these instruments appropriately capture the glycemic effects (via HbA1c) and weight-regulatory effects (via BMI) of GLP1R and GIPR agonism, supporting their validity for causal inference in alcohol-related and metabolic trait analyses.

Because effect‐allele frequencies at GIPR and GLP1R vary across populations, estimating ancestry-specific carrier rates of activation alleles helps anticipate differential baseline receptor engagement and guide trial design [42, 43]. We estimated the proportion of individuals in each 1000 Genomes super-population (EUR, EAS, AFR) carrying at least one "activation" allele at our GIPR and GLP1R cis-instrument loci. Per-SNP carrier probabilities P_carrier = 1 − (1−f)² were combined across independent instruments to yield locus-level estimates (P_locus), then averaged to produce an overall ancestry-specific prevalence. Full computational details and formulae are provided in the Supplementary Methods.

We curated a comprehensive set of alcohol-related outcomes to evaluate the therapeutic potential of GLP1R and GIPR agonism. The primary analysis focused on problematic alcohol use (PAU), defined by Zhou et al. through meta-analysis of AUD cases combined with Alcohol Use Disorders Identification Test (AUDIT) problem drinking scores (AUDIT-P, questions 7–10) from the UK Biobank [44]. Non-European ancestry analyses, restricted to AUD diagnoses, included East Asian (N = 7,364) and African (N = 122,024) populations. Additionally, we assessed distinct alcohol consumption behaviors to capture broader drinking patterns and misuse: weekly alcohol intake (grams/day; European N = 666,978; East Asian N = 90,852; African N = 8,078) [45], binge drinking frequency using AUDIT Question 3 ("How often do you have more than 6 drinks/occasion?") [27], and weekly drinking [45]. Self-reported drinking behaviors were further stratified by drinking status (current, former, or lifetime abstainer), changes in drinking behavior over the past decade (increased, decreased, or stable), and typical drinking frequency (days per month) [46]. We also incorporated alcohol misuse classes identified via latent class analysis of 410,961 UK Biobank participants [46]. This classification model identified four groups: low risk, internalizing-light/non-drinkers, heavy alcohol use–low impairment, and broad high risk [46]. These classifications allowed for detailed exploration of how GLP1R and GIPR activity may differentially impact drinking behaviors associated with both low-risk and problematic use. Finally, we examined relationships with other substance use disorders (SUDs), including tobacco use disorder (TUD) [46], cannabis use disorder (CUD) [47], and opioid use disorder (OUD) [48], to explore broader addiction-relevant effects of GLP1R and GIPR modulation. See Table S1 and Supplementary Methods for more details.

We analyzed six liver-related outcomes, including alcohol-related liver disease (ALD) from FinnGen (ICD10 code K70) [41], liver enzymes (alanine aminotransferase [ALT], aspartate aminotransferase [AST], alkaline phosphatase [ALP], and gamma-glutamyl transferase [GGT]) from the UK Biobank [27, 49, 50], and non-alcoholic fatty liver disease (NAFLD) from a recent meta-analysis [51]. NAFLD diagnoses were based on ICD-10 codes, excluding confounding conditions like ALD, liver transplantation, and hepatitis [51]. This GWAS included 8,464 NAFLD cases, with a prevalence of ~1%, below global estimates (~25%) [51, 52], likely reflecting underdiagnosis and some misclassification. To improve resolution, we integrated the NAFLD GWAS with liver fat percentage data (N = 33,235) [53] using Multi-Trait Analysis of GWAS (MTAG) [54], which jointly analyzed the traits. MTAG steps included filtering variants (MAF > 0.1%), merging datasets, calculating genetic covariance using LD Score regression (Supplementary Methods) [55], and annotating multi-trait GWAS results with Functional Mapping and Annotation (FUMA v1.5.2) [56] (Figures S1–S3). We refer to the combined multi-trait outcome as NAFLD throughout this manuscript.

We used GWASs from the UK Biobank food liking questionnaire, which assessed responses to 139 food and drink items using a 9-point Hedonic scale (1 = "Extremely dislike" to 9 = "Extremely like"), with additional options for "Have never tried it" and "Prefer not to answer" [57]. May-Wilson et al. conducted these GWASs and identified 45 multivariate food liking factors, capturing the genetic architecture of related traits (e.g., fatty food liking, vegetable liking, sweet food liking) that were used in our analyses (Table S1).

Estimates reflect a 1 SD decrease in HbA1c (mmol/L) or BMI (kg/m²), or increased GIPR/GLP1R expression. Binary and categorical outcomes (e.g., AUD, T2D, alcohol groups) are reported as ORs with 95% CIs, where ORs <1 indicate reduced odds and ORs >1 indicate increased odds. Continuous outcomes (e.g., PAU, binge drinking, liver enzymes, liver fat, food liking) are reported as β coefficients, reflecting changes per 1 SD decrease in BMI or HbA1c. For instance, a β = −0.34 for liver fat indicates a 0.34% reduction per 1 SD decrease in HbA1c, while an OR = 0.62 indicates a 38% lower likelihood of heavy drinking vs. light drinking from genetically proxied dual agonism.

We applied a Bonferroni-corrected threshold of 0.0025 (0.05/20) to account for testing across 20 alcohol, liver, and food liking outcomes. Given the use of independent exposure data for both BMI and HbA1c, we also highlight findings where there was evidence of at least P < 0.05 and consistent directional effects across both primary and replication analyses. While we caution against over-reliance on strict P-value thresholds [63], we use this correction as a heuristic to enable follow-up on a manageable number of findings. Our study leverages two biomarkers—BMI and HbA1c—to capture distinct physiological effects of GLP1R and GIPR agonism, with each biomarker instrumented using two separate GWAS summary statistic sources (primary and replication) of European ancestry (Table 1). By evaluating each receptor's effects on BMI and HbA1c separately and replicating findings across these independent datasets for each exposure, we provide consistent, cross-validated evidence supporting their metabolic benefits. This dual-source approach strengthens the robustness of our findings and underscores the reproducibility of genetic estimates.

Given the strong protective relationships observed with heavy drinking behavior and the well-documented links between these drinking behaviors and cardiovascular disease [64 –66], we investigated whether reductions in alcohol consumption mediate the cardioprotective effects of GLP1R and GIPR agonism on coronary artery disease (CAD) risk using two-step MR [67] and the product of coefficients mediation method [67]. Given evidence that the cardioprotective effects of GLP1R agonism primarily operate through BMI reduction rather than glycemic modulation, we hypothesized that BMI-related instrument sets would show the strongest protective effects on CAD. We screened the main GLP1R and GIPR instruments for relationships with CAD risk and alcohol consumption traits, identifying lowered BMI via GIPR and combined GIPR/GLP1R as relevant instruments. Using the Aragam et al. CAD meta-analysis [68] and genome-wide independent SNPs for binge drinking frequency, we estimated causal relationships with CAD risk through inverse-variance weighted MR, MR-Egger, weighted median, and MR-LASSO to account for heterogeneity and pleiotropy. Mediation was quantified using the product of coefficients method, revealing that reductions in binge drinking explained a proportion of the protective effects of GIPR/GLP1R agonism on CAD risk. See the Supplementary Methods for full methodological details.

The SNP-based instruments targeting BMI and HbA1c lowering via GIPR, GLP1R, and their combined modulation were robustly powered with F-statistics ≥18 (Table S11). Notably, genetically reduced HbA1c via GIPR/GLP1R was associated with reduced T2D risk, with both loci demonstrating consistent, directionally aligned effects consistent with their intended therapeutic targeting of metabolic pathways (Table S12). Unexpectedly, BMI lowering via GIPR was associated with an increased risk of T2D (OR = 2.27, 95% CI [1.32, 3.91], P = 3.14 × 10⁻³). This result suggests that the GIPR instrument may inadequately capture GIPR activity in this population or may reflect confounding pathways unrelated to direct GIPR modulation. Colocalization analyses further supported these findings, with strong evidence of colocalization in the GIPR locus between BMI lowering and T2D (PP.H4 = 0.754) and for the GLP1R locus between HbA1c lowering and T2D (PP.H4 = 0.996) (Table S13). While the positive control outcomes aligned well with expected effects on T2D, there was little to no evidence of protective effects for alcohol-related outcomes, including AUD and drinks per week, with all estimates including the null (Table S12), suggesting divergence between metabolic-related and alcohol-related outcomes, particularly within this population.

Twelve SNPs were identified for analysis, including three for GIPR HbA1c, four for GLP1R BMI, and five for GLP1R HbA1c, all with F-statistics ≥10 (Table S14). HbA1c lowering via GIPR/GLP1R reduced T2D risk, consistent with the GLP1R HbA1c instrument but not GIPR (Table S15). No evidence was found for the three SUDs; however, HbA1c lowering by GIPR/GLP1R reduced DPW (β = −0.31, [−0.46, −0.15], P = 7.97 × 10⁻⁵), driven by GLP1R (Table S15). Colocalization signals were absent across all exposures (Table S16).