跳到正文
原文
Nature 健康护理主题·· 9 天前精选AI 评分80

罕见GLP1R变异对semaglutide和tirzepatide疗效的影响

Rare GLP1R variants attenuate weight loss with semaglutide but not tirzepatide

AI 导读

研究发现罕见GLP1R变异显著减弱semaglutide的减重效果,但对tirzepatide影响较小。变异类型和剂量对疗效有显著影响。

推荐理由

研究揭示了罕见GLP1R变异对semaglutide和tirzepatide疗效的影响差异,为个体化肥胖治疗提供了新的遗传学依据。

正文

Main

Glucagon-like peptide-1 receptor agonists (GLP-1RAs) have transformed obesity pharmacotherapy, with semaglutide and tirzepatide producing mean weight reductions of 10–21% over 68–72 weeks in pivotal trials1,2,3,4. Despite large average effects, individual responses are heterogeneous1,2. Pharmacogenetic variation in drug targets represents a potential source of this heterogeneity. The GLP-1 receptor (GLP-1R), encoded by GLP1R, is the shared molecular target of all GLP-1RAs5. Rare loss-of-function (LoF) and deleterious missense GLP1R variants occur at aggregate frequencies of approximately 0.1–0.4% in population cohorts and impair receptor signalling in functional studies6,7. The missense variant rs146868158 (p.Arg421Trp) has been specifically shown to reduce receptor signalling in response to both GLP-1 and GLP-1RAs6,8. Prior pharmacogenomic studies of GLP1R have focused on glycaemic outcomes or common variants8,9,10,11; a recent genome-wide study by Su et al.12 identified the common gain-of-function GLP1R variant p.Pro7Leu (P7L; rs10305420) as associated with greater GLP-1RA-induced weight loss. Whether rare GLP1R variants affect weight loss during GLP-1RA treatment, and whether any such effect differs between drugs, has not been evaluated.

Tirzepatide differs from semaglutide in that it engages both GLP-1R and the glucose-dependent insulinotropic polypeptide receptor (GIPR)13, a dual-receptor mechanism hypothesized to underlie its superior weight loss efficacy14. For GIPR, rare damaging variants occur in approximately 1% of individuals, and some have been associated with adiposity and cardiovascular traits15,16. This dual mechanism raises the possibility that tirzepatide may retain efficacy when GLP-1R function is impaired, whereas semaglutide, acting exclusively through GLP-1R, may not. Conversely, rare GIPR variants might selectively impair the response to tirzepatide but not to semaglutide. To test these hypotheses, we leveraged the Helix Research Network (HRN), a multi-health-system programme that integrates clinical-grade exome sequencing and a common variant backbone with longitudinal electronic health records (EHRs) and insurance claims across 13 North American health systems (n = 351,818). This network provided a sufficiently large treated cohort to identify enough rare variant carriers for pharmacogenetic analysis.

We studied 13,725 adults with a body mass index (BMI) of ≥25 kg m−2 who initiated treatment with semaglutide or tirzepatide administered subcutaneously in routine clinical care within the HRN. Eligible participants had a baseline weight recorded within 6 months before the first treatment record, either at least two initial insurance claims or at least two initial prescription records separated by 7–90 days, and body weight measurements spanning ≥6 months of on-treatment follow-up, censored at 12 months for the primary analyses. The analysis used the HRN data release with records through 31 December 2025.

Exome+ sequencing identified 16 common GLP1R and GIPR coding variants as well as 93 rare variants (minor allele frequency (MAF) < 0.1%). Genotype calls were filtered using per-call quality metrics from the Helix variant calling pipeline accredited by the Clinical Laboratory Improvement Amendments (CLIA) and the College of American Pathologists (CAP); only calls passing standard quality filters were retained. Common variants were analysed individually, while rare variants were grouped together by consequence type. Predicted LoF (pLoF) variants were defined as stop-gained, frameshift, and canonical splice-donor or splice-acceptor variants. Rare missense variants were classified by AlphaMissense and REVEL scores as either predicted damaging (AlphaMissense ≥ 0.564 or REVEL ≥ 0.75)17,18, predicted possibly damaging (AlphaMissense 0.34–0.564 or REVEL 0.25–0.75) or predicted tolerated (AlphaMissense < 0.34 and REVEL < 0.25; analysed as a negative control group). The most common rare GLP1R variant examined, rs146868158 (p.Arg421Trp; MAF 0.045% in gnomAD v4.1), falls within the predicted possibly damaging tier (AlphaMissense 0.150, REVEL 0.304) and additionally has experimental evidence of impaired GLP-1- and GLP-1RA-stimulated receptor signalling8,11. Given its larger carrier count (n = 15), this variant is also reported individually alongside the grouped tier results. Several additional variants have been experimentally characterized in cellular assays in prior publications, including p.Thr149Met (predicted possibly damaging), p.Arg190Gln (predicted damaging) and p.Thr353Met (predicted damaging)6,7,19; these variants were retained within their computationally assigned tiers (each with n < 5 carriers). For the primary analysis, three categories—pLoF, predicted damaging missense and predicted possibly damaging missense—were used, reflecting a more sensitive definition. Non-carriers were defined as individuals with no rare coding variants of any functional class detected in the respective gene. Rare variants and annotation scores are listed in Extended Data Table 1 and Supplementary Table 1, while predicted tolerated variants are listed in Supplementary Table 2.

Percentage weight change from baseline through 6–12 months was the outcome measure. Rare variant carriers were matched 1:10 to non-carriers based on drug, dosage, sex, type 2 diabetes status, prior GLP-1RA exposure, genetic similarity group, baseline BMI and follow-up duration. Weighted linear regression quantified the differences in weight loss between carriers and matched non-carriers.

Among the 13,725 treated individuals, 33 carried a qualifying rare GLP1R variant (6 pLoF, 5 predicted damaging missense, 15 rs146868158 and 7 other predicted possibly damaging missense; hereafter, GLP1R carriers), and 63 carried a qualifying rare GIPR variant (12 pLoF, 19 predicted damaging missense and 32 predicted possibly damaging missense; hereafter, GIPR carriers) (Extended Data Table 1 and Supplementary Table 1). All 33 GLP1R carriers and 60 of the 63 GIPR carriers were successfully matched to non-carriers with similar patient characteristics and treatment regimens, yielding 274 and 438 matched non-carriers, respectively. Following matching, the groups were well balanced on demographic, clinical and treatment characteristics (Table 1 and Supplementary Tables 3 and 4).

Table 1 Baseline characteristics of GLP1R and GIPR rare variant carriers and matched non-carriers

Full size table

Overall, GLP1R carriers lost a mean of 8.2% of body weight versus 12.5% among matched non-carriers (response ratio 0.65, 95% confidence interval (CI): 0.46–0.85; P = 0.0007), with a stronger effect among pLoF or predicted damaging missense variant carriers (ratio 0.46, 95% CI: 0.20–0.72) (Fig. 1). In contrast, GIPR carriers lost 10.3% of body weight versus 12.1% among matched non-carriers, showing modest overall attenuation (ratio 0.85, 95% CI: 0.70–0.99; P = 0.043), with null effects among pLoF and predicted damaging missense carriers (ratio 1.04, 95% CI: 0.87–1.21) (Extended Data Fig. 1).

Fig. 1: Drug-specific attenuation of GLP-1RA response by rare GLP1R variants.

Response ratios (mean weight loss in carriers versus matched non-carriers) by drug and rare variant type. Diamond, point estimate; horizontal line, 95% CI; leftward arrow, point estimate not calculable (NC); vertical dashed line, equal response (ratio = 1.0). Response ratios and 95% CIs are from two-sided weighted linear regression with cluster-robust standard errors at the matched-set level. Rows show results for the overall population (users of both drugs combined), semaglutide users and tirzepatide users. Within each drug, results are shown for all carriers of any qualifying rare variant (shaded grey) and for individual variant classes. Matched non-carriers have no rare GLP1R variants detected in exome sequencing and serve as the reference group for all associations shown.

Source data

Importantly, the GLP1R carrier effect differed markedly by drug. Among semaglutide users, GLP1R carriers (n = 17) lost 2.4% of body weight versus 8.8% in non-carriers (ratio 0.27, P = 0.0007), whereas tirzepatide-using carriers (n = 19) lost 13.3% versus 15.2% in non-carriers (ratio 0.88, P = 0.23) (Figs. 1 and 2a). A carrier-by-drug interaction test confirmed this differential effect (β = −4.4%, 95% CI: −7.7% to −1.1%; P = 0.009). At the individual level, nearly all semaglutide-treated GLP1R carriers (15 of 17, 88%) lost less weight than their matched non-carriers, whereas the same was true for a smaller proportion of tirzepatide-treated carriers (11 of 19, 58%) (Fig. 2b). GIPR carriers showed no consistent pattern with either drug and no drug-specific differences (semaglutide ratio 0.83, tirzepatide ratio 0.85; carrier-by-drug interaction, P = 0.68) (Extended Data Figs. 1 and 2). Results were robust to the choice of matching ratio (Supplementary Table 5) and to extending the follow-up window to 12–24 months (Supplementary Figs. 1 and 2).

Fig. 2: Individual-level GLP1R carrier weight loss responses by drug.

a, Distributions of percentage weight change after 6–12 months in GLP1R rare variant carriers versus matched non-carriers, stratified by drug (left, semaglutide; right, tirzepatide). Violin plots show density; box plots show the median (centre line), interquartile range (25th–75th percentiles, bounds of the box), and whiskers extending to the most extreme data point within 1.5× the interquartile range; diamonds show weighted means. Response ratios and P values are from two-sided weighted linear regression with cluster-robust standard errors at the matched-set level. b, Percentage weight change in individual GLP1R rare variant carriers versus mean weight change in their matched non-carriers, stratified by drug (left, semaglutide; right, tirzepatide) (n = 17 semaglutide-treated and n = 19 tirzepatide-treated GLP1R carriers, each matched to their own non-carrier set, totalling n = 149 and n = 154 matched non-carriers, respectively). Each line connects one carrier’s observed weight change (left endpoint) to the mean weight change of their set of matched non-carriers (right endpoint), coloured by variant class. Downward-sloping lines indicate carriers who lost less weight than the average of their matched non-carriers. Matched non-carriers have no rare GLP1R variants detected in exome sequencing.

Source data

The drug-specific GLP1R carrier effect persisted at high doses (semaglutide ≥ 1.7 mg: ratio 0.41, P = 0.044; tirzepatide ≥ 10 mg: ratio 0.84, P = 0.18) and across subgroups stratified by diabetes status, prior GLP-1RA exposure, sex and BMI (Supplementary Fig. 3). GIPR carriers showed no drug-specific differences in any subgroup (Supplementary Fig. 4).

Results were consistent with the primary matched analysis when using whole-genome regression in REGENIE to account for population stratification, restricting the sample to individuals genetically similar to European reference populations (n = 11,347), and adjusting for ten European principal components in addition to clinical covariates. The drug-specific pattern was preserved: the GLP1R carrier effect was significant among semaglutide users (ratio 0.42, P = 0.005), with no effect among tirzepatide users (ratio 0.96, P = 0.80; drug interaction P = 0.085) (Extended Data Table 2). For GIPR, no significant effects were observed for any variant category or either drug (Supplementary Table 6).

In the primary matched analysis, while GLP1R variants classified as pLoF, predicted damaging or predicted possibly damaging attenuated the response to semaglutide, only pLoF and predicted damaging variants showed a trend towards attenuation of the response to tirzepatide (ratio 0.67, P = 0.015) (Fig. 1). Drug-specific GLP1R differences appeared most pronounced for rs146868158 and other predicted possibly damaging missense variants (semaglutide ratios 0.38 and 0.44; tirzepatide ratios 1.13 and 0.97; interaction P = 0.027 and 0.015, respectively). Carriers of predicted tolerated GLP1R missense variants showed no attenuation (Fig. 1), even when stratifying by additional annotation scores such as CADD (Combined Annotation Dependent Depletion), SIFT (Sorting Intolerant From Tolerant) and PolyPhen-2 (Polymorphism Phenotyping v2) (Supplementary Fig. 5). For GIPR, carriers of pLoF and predicted damaging missense variants showed null effects across both drugs, consistent with the overall pattern. The modest overall GIPR attenuation was driven by lower-confidence variant tiers and was not robust to adjustment for population stratification (Extended Data Fig. 1 and Supplementary Table 6).

We additionally analysed common coding variants in GLP1R and GIPR using whole-genome regression in three separate populations: individuals genetically similar to reference populations from Europe (n = 11,337), Africa (n = 1,014) and the Americas (n = 859) (Extended Data Table 3). The GLP1R P7L T allele (rs10305420; MAF 36%, n = 1,918 carriers) was associated with greater weight loss in the European genetic similarity group (response ratio 1.05 per T allele, 95% CI: 1.02–1.09; P = 0.0047), replicating the gain-of-function association reported by Su et al.12, with directionally consistent estimates in the African (ratio 1.03, P = 0.82) and Americas (ratio 1.06, P = 0.39) groups. No interaction between rare GLP1R variant carrier status and P7L genotype was detected (Extended Data Fig. 3). Additionally, GLP1R p.Ala316Thr (rs10305492; MAF 1.7%, n = 365 carriers) was associated with attenuated weight loss in the European genetic similarity group (ratio 0.87 per alternative allele, 95% CI: 0.80–0.94; P = 3.6 × 10−4). No associations were observed for common variants of GIPR (Extended Data Table 3).

Independent replication of the rare variant association was pursued in two settings. In the All of Us Research Program20, 4,432 eligible semaglutide users were identified; eight rare GLP1R carriers showed a mean weight change of −2.6% versus −5.7% in matched non-carriers (response ratio 0.46, P = 0.12), directionally consistent with the primary finding (Extended Data Fig. 4a). GIPR results in All of Us were consistent with the primary null finding (Extended Data Fig. 4b). The p.Ala316Thr association (common GLP1R variant) identified in the HRN was directionally consistent but underpowered in All of Us (n = 89 carriers with European genetic similarity; response ratio 0.77, 95% CI: 0.51–1.04; P = 0.091).

We also performed validation in an independent, non-overlapping cohort from the same 13 HRN health systems, comprising patients meeting relaxed eligibility criteria, as well as patients newly enrolled in the expanding programme and/or newly treated with sufficient follow-up since the primary analysis (3,969 semaglutide users and 4,144 tirzepatide users; updated records through 31 March 2026), replicating the drug-specific pharmacogenetic pattern. Among semaglutide users, the seven rare GLP1R carriers showed an attenuated response relative to matched non-carriers (−4.1% versus −7.5%; response ratio 0.54, P = 0.004), with a null tirzepatide effect among 11 carriers (response ratio 1.19, P = 0.43; drug interaction P = 0.036) (Extended Data Fig. 5a). Rare variants in GIPR showed no drug-specific effect in either treatment group (Extended Data Fig. 5b).

These results provide evidence that rare damaging GLP1R coding variants substantially attenuate semaglutide-induced weight loss but have a reduced impact on tirzepatide response. In contrast, rare GIPR variants did not meaningfully affect the response to either drug, showing null effects among pLoF and damaging missense carriers. Together, the GLP1R-specific and drug-specific patterns of attenuation support a pharmacogenetic mechanism rooted in receptor function.

Our findings extend and complement those of Su et al.12, who identified the common GLP1R P7L gain-of-function variant as associated with enhanced GLP-1RA-induced weight loss. For semaglutide specifically, P7L and rare pLoF and damaging missense variants delineate a bidirectional spectrum of pharmacological response: P7L T allele carriers show modest improvements, while rare damaging variant carriers show marked attenuation—an effect approximately 12 times greater in magnitude (Extended Data Fig. 6). GLP-1R functional variation may be a meaningful determinant of semaglutide efficacy, with independent contributions from both common and rare genetic variation.

Although rare damaging GLP1R variants were present in only approximately 0.2% of treated individuals, their effect on semaglutide response was large, with carriers achieving roughly one-quarter of the typical weight loss. Within GLP-1RA cardiovascular outcome trials, each 1 kg reduction in body weight has been associated with an approximately 7% lower risk of major atherosclerotic cardiovascular events21, suggesting that the attenuated weight loss observed in semaglutide-treated carriers may have long-term cardiometabolic consequences. More broadly, these findings illustrate that drugs with partially shared receptor targets can have different genetic determinants of response—a principle probably applicable to the growing class of multi-receptor agonists in development22.

Our findings extend prior GLP1R pharmacogenetic research, which has focused on common variants and/or glycaemic traits8,9,10. Although rare LoF variants have been described6, their impact on GLP-1RA weight loss response was previously unknown. Consistency of the attenuated semaglutide response across predicted damaging or possibly damaging GLP1R variant classes, the null effects for predicted tolerated GLP1R variants, the significant carrier-by-drug interaction and the absence of GIPR effects collectively support biological specificity. That the most severe GLP1R variants also appeared to attenuate the response to tirzepatide in the matched analysis, although to a lesser degree than the response to semaglutide, is a hypothesis-generating observation in a small drug-specific and variant class-specific subgroup. Although speculative, it is nonetheless consistent with the possibility of partial compensation through GIPR co-agonism, which may be insufficient at the extremes of GLP-1R impairment. Direct experimental evidence is required to establish whether GIPR co-agonism compensates for impaired GLP-1R signalling. It is also possible that combined variation in both GLP1R and GIPR could affect the efficacy of tirzepatide; however, no individuals in our study carried rare damaging variants in both genes.

This study has several limitations. The number of rare variant carriers is small, particularly in drug-specific subgroups, limiting variant-level inferences. We relied on real-world data in which medication adherence and weight measurements were imperfectly recorded. Residual confounding is possible, although matching achieved excellent covariate balance, and results were robust to whole-genome regression adjusting for population stratification. Our carrier group was predominantly of European genetic similarity and could be improved by including more treated carriers from diverse ancestries in future studies. Not all variants that affect gene function may have been included in the study; for example, non-coding variants, structural variants and coding variants with insufficient carriers or poor coverage (although, with only two exceptions—chromosome (chr.) 19:45671665:A:C/p.Asn2Thr and chr. 19:45671748:G:C/p.Gly30Arg—our dataset contained all coding variants in gnomAD with frequencies high enough for carriers to be expected in the dataset). Additionally, our variant classification relies on computational pathogenicity prediction tools, which can be less reliable for membrane proteins, including G-protein-coupled receptors. Where independent evidence exists, such as for rs146868158, we rely on it directly rather than on computational prediction alone6,8. Misclassification would be expected to bias our estimates towards the null, as including truly non-damaging variants in the carrier group would dilute, rather than inflate, the observed attenuation of the drug response. Functional characterization of all included variants remains an important direction for future work.

We identify GLP1R as a pharmacogenetic determinant that differentially affects responses to semaglutide and tirzepatide. These findings suggest that carriers of rare damaging GLP1R variants may be preferentially suited to tirzepatide or other agents engaging non-GLP-1R pathways. This study demonstrates how integrating clinical-grade sequencing with real-world treatment data can uncover actionable pharmacogenetic associations, enabling precision approaches to obesity pharmacotherapy.

Methods

Study population

We conducted a retrospective cohort study within the HRN, a multi-health-system genomic screening programme linking clinical-grade exome sequencing to longitudinal EHRs and insurance claims data. The 13 participating health systems and their programmes are as follows: DNA Answers (St. Luke’s University Health Network, Pennsylvania), DNA Discovery (TriHealth, Ohio), DNA Insights (Parkview Health, Indiana and Ohio), GeneConnect (Cone Health, North Carolina), the Genetic Insights Project (Nebraska Medicine), GenoWell (Rochester Regional Health, New York), the Healthy Nevada Project (Renown Health, Nevada), ImagineYou (Sanford Health, Upper Midwest), In Our DNA SC (Medical University of South Carolina), myGenetics (HealthPartners, Minnesota), Ohio State Genomic Health (The Ohio State University Wexner Medical Center), PreciselyYou (WakeMed, North Carolina) and The Gene Health Project (WellSpan Health, Pennsylvania). Study protocols were approved by institutional review boards (projects 956068-12 and 21143). All participants provided written informed consent, and all data were de-identified. Participants did not receive financial compensation for their participation in this research.

EHR data were transformed into a Helix-modified version of the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) v5.4 from Epic Clarity source systems. Medical and pharmacy claims data were obtained through Komodo Health and similarly transformed into the OMOP CDM to enable integrated queries across both data sources. EHR and claims data were linked using a de-identification process that complied with the Health Insurance Portability and Accountability Act and incorporated Datavant tokenization services. This analysis used the HRN data release with records through 31 December 2025 (351,818 individuals with merged EHR and genetic data; 98.9% with linked claims data; median of 10.4 years of longitudinal records per participant).

Adults aged ≥18 years with documented subcutaneous treatment with semaglutide or tirzepatide in routine clinical care were eligible if they had a baseline BMI of ≥25 kg m−2. Baseline weight was the most recent measurement taken within 6 months before the first treatment record. To increase confidence in medication initiation, we required either at least two insurance claims or at least two prescription records separated by 7–90 days. Participants were additionally required to have weight measurements spanning at least 6 months of on-treatment follow-up. For the primary analyses, follow-up was censored at 60 days after patients’ last claims or prescription record for the index medication; thus, the 6-month follow-up requirement effectively necessitated additional treatment record(s) extending ≥4 months beyond initiation (the 12-month cap is relaxed for an extended-follow-up sensitivity analysis; see below). This definition prioritized confidence in identifying sustained medication use, acknowledging that real-world treatment data imperfectly capture actual medication adherence.

Individuals were excluded if they had prior bariatric surgery, pregnancy within the previous 9 months or a malignancy diagnosis within the previous year, to mitigate the influence of other conditions independently affecting body weight. Follow-up was censored at the earliest occurrence of any of the following: (1) switching to a different GLP-1RA (including switching from semaglutide to tirzepatide or vice versa); (2) incident pregnancy, bariatric surgery or a malignancy diagnosis; (3) 60 days after the last prescription or claims record for the index medication; or (4) 12 months after initiation. When individuals had multiple eligible treatment episodes, we retained the earliest qualifying course for the overall analysis combining both drugs. Individuals who sequentially used both drugs could contribute independent episodes to each drug-specific analysis. This yielded 13,725 treated adults for the primary analysis. No statistical methods were used to predetermine sample sizes. This is a real-world pharmacogenetic study in which carrier counts were determined based on natural allele frequencies among all eligible treated individuals in the HRN.

Sequencing, variant identification and annotation

Participants provided blood or saliva samples that underwent Exome+ sequencing at Helix, a CLIA- and CAP-accredited laboratory. Exome+ is a clinical exome assay that combines comprehensive medical exome coverage with a microarray-equivalent single-nucleotide polymorphism backbone23. Sequencing was performed using Illumina technology with alignment to GRCh38 and variant calling following GATK (Genome Analysis Toolkit) best-practice guidelines. Genotype calls were filtered using per-call quality metrics from the Helix variant calling pipeline; only calls meeting standard quality thresholds (including sequencing depth ≥ 20× and genotype quality score ≥ 20) were retained. Variant annotation was performed with VEP 110 (https://www.ensembl.org/vep). Coding regions were defined according to GENCODE version 44 (https://www.gencodegenes.org), and the MANE transcript was used to determine variant consequences (https://www.ncbi.nlm.nih.gov/refseq/MANE). Genotype processing was performed using Hail 0.2.115 (https://hail.is) and PLINK v2.00a1LM (https://www.cog-genomics.org/plink). Genetic similarity to reference populations was determined using ADMIXTURE software with 1000 Genomes phase 3 reference data, classifying individuals into six ancestry groups (Europe, Africa, the Americas, East Asia, South Asia, other)24.

We extracted all variants in GLP1R (GRCh38 chr. 6:39,048,781–39,091,303) and GIPR (chr. 19:45,668,221–45,683,722) with coding consequences as annotated by VEP. Non-coding and structural variants were not assessed. We annotated missense variants using the AlphaMissense and REVEL v1.3 pathogenicity prediction algorithms17,18. Variants were defined as rare using a threshold of MAF < 0.1% in gnomAD and local ancestry-specific cohorts. Any other variant that reached at least ten carriers in an ancestry group was analysed individually. This yielded 16 common coding variants (8 GLP1R and 8 GIPR; Extended Data Table 3) and 93 rare coding variants (34 GLP1R and 59 GIPR; Extended Data Table 1 and Supplementary Tables 1 and 2). Rare variants were collapsed into predicted-consequence tiers for analysis (see below).

Variant classification (rare variants)

pLoF variants were defined by consequence type: stop-gained, frameshift, and canonical splice-donor or splice-acceptor variants. Rare missense variants were scored using AlphaMissense and REVEL17,18. A variant was classified as predicted damaging if AlphaMissense ≥ 0.564 or REVEL ≥ 0.75, and as predicted possibly damaging if AlphaMissense ≥ 0.34 or REVEL ≥ 0.25 but below the predicted damaging thresholds. Variants below the possibly damaging thresholds were classified as predicted tolerated and analysed as negative controls (Supplementary Table 2). For the primary analysis, the qualifying carrier group for each gene combined three tiers—pLoF, predicted damaging missense and predicted possibly damaging missense.

The most frequently observed rare GLP1R variant, rs146868158 (p.Arg421Trp; MAF 0.045% in gnomAD v4.1), falls within the predicted possibly damaging tier by score (AlphaMissense 0.15; REVEL 0.304) and additionally has published cell-based experimental evidence of impaired receptor signalling in response to GLP-1 and GLP-1RAs8,9. Given its comparatively high carrier count (n = 15 versus n < 5 for most other individual variants in the qualifying group) and the existing functional evidence, rs146868158 is included in the predicted possibly damaging tier for the collapsed analyses and is additionally reported as an individual variant. A small number of other qualifying variants with published cell-based functional characterization (each with n < 5 carriers) are noted in Extended Data Table 1 and Supplementary Table 1 but are retained within their computationally assigned tiers, as their low carrier counts provide insufficient statistical power for separate analysis; these include p.Thr149Met (predicted possibly damaging), p.Arg190Gln (predicted damaging) and p.Thr353Met (predicted damaging)6,7,19. Per-variant carrier counts fewer than five are suppressed in accordance with the HRN data privacy policy, which restricts the disclosure of small cell counts.

Non-carriers (the reference group for all rare variant comparisons) were defined as individuals with no rare coding variants of any functional class (including predicted tolerated variants) detected in the relevant gene, so that qualifying carriers are compared against a variant-free population.

CADD, SIFT and PolyPhen-2 scores were generated for all missense variants as secondary annotations and are reported alongside AlphaMissense and REVEL scores in Extended Data Table 1 and Supplementary Tables 1 and 2, but were not used for primary classification. As an empirical check on the predicted tolerated group (based on AlphaMissense and REVEL), we stratified it by whether variants met at least one secondary-tool threshold (CADD ≥ 20, SIFT < 0.05 or PolyPhen-2 ≥ 0.908) versus none, and compared variant associations between the two subgroups; neither subgroup showed a significant departure from the null.

Outcome definition and clinical variables

Percentage weight change from baseline was calculated as the lowest on-treatment weight recorded 6–12 months after treatment initiation minus the baseline weight, divided by the baseline weight, and multiplied by 100. As a secondary analysis, the primary comparison was repeated using on-treatment weight measurements taken between 12 and 24 months. Longitudinal body weight trajectories for all carriers and matched non-carriers with ≥25% weight loss or ≥5% weight gain were manually inspected to confirm data quality and biological plausibility. Semaglutide and tirzepatide treatment dosages were assigned as the highest recorded doses during follow-up (semaglutide: 0.5, 1.0, 1.7, 2.0 or 2.4 mg; tirzepatide: 5.0, 7.5, 10.0, 12.5 or 15.0 mg). For each drug, prior use of the other study drug (semaglutide or tirzepatide), a different GLP-1RA (albiglutide, dulaglutide, exenatide, liraglutide or lixisenatide) or another anti-obesity agent (orlistat, naltrexone–bupropion, phentermine, phentermine–topiramate or setmelanotide) within the past 12 months was also ascertained from claims and prescription records. Type 2 diabetes and hypertension were defined at baseline using diagnostic codes. Genetic sex was determined from the sequence data, and only individuals whose genetically inferred sex matched their self-reported sex were included.

Matching strategy

This is an observational pharmacogenetic study; randomization was not applicable to group assignment because carrier status is a naturally occurring genetic exposure rather than an intervention. To compare GLP1R and GIPR variant carriers and non-carriers under similar treatment and clinical conditions, we performed optimal variable-ratio 1:10 matching using the MatchIt package v4.5.5 in R, with carrier status as the exposure. We required exact matching on index drug, highest dose category, sex, genetic similarity group, type 2 diabetes status and prior GLP-1RA exposure (including prior use of the other study drug or another GLP-1RA in the previous year). We applied callipers of ±3 kg m−2 on baseline BMI and ±30 days on follow-up duration, and used the Mahalanobis distance on age, baseline BMI and follow-up duration to select up to ten non-carriers for each carrier without replacement (33 GLP1R carriers matched to 274 non-carriers, corresponding to a mean of 8.3 non-carriers per carrier, and 60 GIPR carriers matched to 438 non-carriers, corresponding to a mean of 7.3 non-carriers per carrier). Each matched carrier received a regression weight of 1, and each matched non-carrier received a weight of 1/n, where n is the number of non-carriers in that matched set. Standardized mean differences (SMDs) quantified covariate balance before and after matching, with absolute SMD < 0.10 considered excellent (Supplementary Tables 3 and 4). For sensitivity analyses, the primary analyses were repeated using optimal matching ratios of 1:5 and 1:20.

For the overall combined-drug analysis, each carrier’s earliest qualifying treatment course was used. For drug-specific analyses, matching was performed separately within semaglutide and tirzepatide initiators.

Statistical analysis: matched comparison

The primary analysis used weighted linear regression (incorporating weights from matched sets) with percentage weight change as the outcome and GLP1R or GIPR carrier status as the main predictor. Cluster-robust standard errors (HC1) were computed at the matched-set level to account for within-set correlations. The estimand was the mean difference in weight change (in percentage points) between carriers and non-carriers, with 95% CIs and two-sided P values calculated. P < 0.05 was considered statistically significant.

In the primary matched-cohort analysis for GLP1R (overall, n = 307), the outcome distribution was approximately normal, with a skewness of −0.23 and a kurtosis of 2.57 (values near 0 and 3, respectively, indicate minimal departure from normality); the mean (−12.3%) and median (−11.7%) weight changes were closely aligned, and the Shapiro–Wilk statistic was close to 1 (W = 0.99, P = 0.026). A Breusch–Pagan test confirmed significant heteroscedasticity between carriers and non-carriers (BP = 19.98, d.f. = 1, P = 7.8 × 10−6), justifying the use of cluster-robust standard errors.

We additionally report the mean weight change among matched non-carriers and express the carrier response as a response ratio (carrier mean weight change divided by non-carrier mean weight change), which is used as the primary effect-size metric throughout in place of regression coefficients. The response ratio was derived from the regression coefficient (β) as (non-carrier mean + β)/non-carrier mean, with the 95% CI propagated from the CI for β. Because the ratio is meaningful only when both groups experience weight loss, lower bounds of the ratio CI that fell below 0.01 were floored at 0.01 for display when the CI for the carrier group crossed zero.

Drug-specific analyses estimated the carrier effect separately for semaglutide and tirzepatide users. To formally test whether the GLP1R or GIPR carrier effect differed by drug, we included an interaction term between carrier status and drug (semaglutide versus tirzepatide) in the overall combined-drug analysis model. The significance of the interaction term was assessed using a Wald test. We also evaluated associations stratified by variant type, drug dosage, type 2 diabetes status, prior GLP-1RA exposure (past year), sex and baseline BMI category (<40 versus ≥40 kg m−2). Analyses were conducted using R v4.2.3.

Analysis was not performed blind to the conditions of the experiments. Outcome data (body weight measurements) were extracted from routine clinical records and were not influenced by knowledge of genotype, as treating clinicians were unaware of the research-related rare variant status. Variant classification criteria were prespecified before examining weight loss outcomes.

Whole-cohort regression analysis

As a complementary analysis that does not depend on matching and to formally assess confounding by population structure, we repeated the analysis using whole-genome regression in REGENIE v2.2.4. This analysis was restricted to individuals genetically similar to European reference populations (n = 11,347) and adjusted for the first ten ancestry-specific principal components in addition to age, sex, baseline BMI, type 2 diabetes status, hypertension, follow-up duration, drug and dose, prior GLP-1RA use and other anti-obesity medication use25. Rare variants were collapsed into the same categories used in the matched analysis.

Common variant analysis

This same approach in REGENIE, using ten ancestry-specific principal components, was used to analyse common variants, separately in the groups with genetic similarity to reference datasets from Europe (n = 11,337), Africa (n = 1,014) and the Americas (n = 859). The 16 common coding variants identified above were analysed individually, with each variant coded as an additive allele dosage (0, 1 or 2 copies of the alternative allele). Results are reported as a response ratio per allele copy, derived analogously to the rare variant response ratio. This includes the common GLP1R variant P7L (rs10305420), previously reported as a gain-of-function determinant of GLP-1RA-induced weight loss12.

We separately tested whether the effect of rare GLP1R variant carrier status on weight loss was modified by the P7L genotype among rare variant carriers, using an interaction term (rare variant carrier status × P7L genotype (0, 1 or 2 T alleles)) in the matched weighted-regression framework for the European genetic similarity cohort. Significance was assessed using the Wald test.

Replication cohorts

All of Us Research Program

Independent replication was pursued in the All of Us Research Program20, which is approved by the All of Us Institutional Review Board. Whole-genome sequence data restricted to exonic regions were used for rare variant analyses (using the v8 Hail MatrixTable resource). Eligibility criteria were relaxed relative to the primary HRN analysis as follows: (1) requiring at least one prescription record, with no censoring at 60 days after the last prescription; (2) including oral semaglutide users; (3) allowing participants with unknown doses; (4) expanding the baseline weight window to span from −365 to +30 days relative to treatment start; and (5) requiring at least one follow-up measurement at any time within 0–12 months after initiation. Primary analysis exclusions (prior bariatric surgery, pregnancy within 9 months, malignancy within 1 year) and the baseline BMI requirement (≥25 kg m−2) were retained. Rare variant definitions and categories were identical to those used in the primary analysis. GLP1R and GIPR carriers were matched 1:10 to non-carriers using the same matching strategy, with Euclidean distance substituted for Mahalanobis distance to accommodate the smaller sample size. Among 332 eligible tirzepatide users, only one rare GLP1R carrier was identified, precluding a tirzepatide-specific rare variant analysis in this cohort. Individual-level data for fewer than 20 participants are reported under an exception to the All of Us Data and Statistics Dissemination Policy granted by the Resource Access Board.

Independent HRN replication

We separately identified all individuals initiating semaglutide or tirzepatide in a more recent HRN data release (records through 31 March 2026) who were not included in the primary analysis—comprising both patients newly enrolled or eligible after the original data cutoff and patients from the original study period who did not meet the stricter eligibility criteria of the primary analysis. Eligibility criteria were intentionally relaxed to maximize the number of available carriers by using a follow-up window of 3–12 months (versus 6–12 months) and requiring at least one prescription or claims record for the index drug (versus the stricter primary ascertainment criteria); all other eligibility, exclusion and censoring rules remained unchanged. The same variant annotation pipeline, matching strategy and statistical analysis used in the primary analysis were applied.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

The data supporting the findings of this study were generated through the Helix Research Network (HRN), a multi-centre research programme governed by a Joint Steering Committee and a Science and Research Subcommittee. The HRN database is not publicly available due to participant privacy protections, institutional data governance requirements and ethical restrictions. Qualified researchers affiliated with HRN member sites or approved third-party institutions may request access to de-identified data through HRN’s established governance process, which includes the submission of a research proposal, subcommittee review, completion of human subjects research training and execution of a data use agreement. Inquiries regarding data access should be directed to the corresponding author.

All of Us Research Program data are publicly available to registered researchers at https://www.researchallofus.org/. All of Us Research Program data are subject to privacy protections including the Data and Statistics Dissemination Policy, which restricts the reporting of cell sizes below 20 participants. Source data are provided with this paper.

Code availability

Code used for the analyses is available on GitHub at https://github.com/myhelix/glp1r-pharmacogenetics.

References

  1. Jastreboff, A. M. et al. Tirzepatide once weekly for the treatment of obesity. N. Engl. J. Med. 387, 205–216 (2022).

    Article  CAS  PubMed  Google Scholar 

  2. Wilding, J. P. H. et al. Once-weekly semaglutide in adults with overweight or obesity. N. Engl. J. Med. 384, 989–1002 (2021).

    Article  CAS  PubMed  Google Scholar 

  3. Davies, M. et al. Semaglutide 2·4 mg once a week in adults with overweight or obesity, and type 2 diabetes (STEP 2): a randomised, double-blind, double-dummy, placebo-controlled, phase 3 trial. Lancet 397, 971–984 (2021).

    Article  CAS  PubMed  Google Scholar 

  4. Garvey, W. T. et al. Tirzepatide once weekly for the treatment of obesity in people with type 2 diabetes (SURMOUNT-2): a double-blind, randomised, multicentre, placebo-controlled, phase 3 trial. Lancet 402, 613–626 (2023).

    Article  CAS  PubMed  Google Scholar 

  5. Drucker, D. J. Mechanisms of action and therapeutic application of glucagon-like peptide-1. Cell Metab. 27, 740–756 (2018).

    Article  CAS  PubMed  Google Scholar 

  6. Melchiorsen, J. U. et al. Rare heterozygous loss-of-function variants in the human GLP-1 receptor are not associated with cardiometabolic phenotypes. J. Clin. Endocrinol. Metab. 108, 2821–2833 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  7. Gao, W. et al. Human GLP1R variants affecting GLP1R cell surface expression are associated with impaired glucose control and increased adiposity. Nat. Metab. 5, 1673–1684 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  8. Lagou, V. et al. GWAS of random glucose in 476,326 individuals provide insights into diabetes pathophysiology, complications and treatment stratification. Nat. Genet. 55, 1448–1461 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  9. Dawed, A. Y. et al. Pharmacogenomics of GLP-1 receptor agonists: a genome-wide analysis of observational data and large randomised controlled trials. Lancet Diabetes Endocrinol. 11, 33–41 (2023).

    Article  CAS  PubMed  Google Scholar 

  10. Phan, A. et al. A GLP1R gene variant and sex influence the response to semaglutide treatment in patients with severe obesity. Obesity (Silver Spring) 33, 1237–1242 (2025).

    Article  CAS  PubMed  Google Scholar 

  11. German, J. et al. Association between plausible genetic factors and weight loss from GLP1-RA and bariatric surgery. Nat. Med. 31, 2269–2276 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  12. Su, Q. J. et al. Genetic predictors of GLP1 receptor agonist weight loss and side effects. Nature 653, 770–775 (2026).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  13. Willard, F. S. et al. Tirzepatide is an imbalanced and biased dual GIP and GLP-1 receptor agonist. JCI Insight 5, e140532 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  14. Aronne, L. J. et al. Tirzepatide as compared with semaglutide for the treatment of obesity. N. Engl. J. Med. 393, 26–36 (2025).

    Article  CAS  PubMed  Google Scholar 

  15. Kizilkaya, H. S. et al. Loss of function glucose-dependent insulinotropic polypeptide receptor variants are associated with alterations in BMI, bone strength and cardiovascular outcomes. Front. Cell Dev. Biol. 9, 749607 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

  16. Kizilkaya, H. S. et al. Characterization of genetic variants of GIPR reveals a contribution of β-arrestin to metabolic phenotypes. Nat. Metab. 6, 1268–1281 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  17. Cheng, J. et al. Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science 381, eadg7492 (2023).

    Article  CAS  PubMed  Google Scholar 

  18. Ioannidis, N. M. et al. REVEL: an ensemble method for predicting the pathogenicity of rare missense variants. Am. J. Hum. Genet. 99, 877–885 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  19. Beinborn, M., Worrall, C. I., McBride, E. W. & Kopin, A. S. A human glucagon-like peptide-1 receptor polymorphism results in reduced agonist responsiveness. Regul. Pept. 130, 1–6 (2005).

    Article  CAS  PubMed  Google Scholar 

  20. All of Us Research Program Genomics Investigators. Genomic data in the All of Us Research Program. Nature 627, 340–346 (2024).

    Article  Google Scholar 

  21. Gomes, D. A. et al. Association between the magnitude of glycemic control and body weight loss with GLP-1 receptor agonists and risk of atherosclerotic cardiovascular disease: a systematic review and meta-analyses of randomized diabetes cardiovascular outcomes trials. Cardiovasc. Drugs Ther. 39, 337–345 (2025).

    Article  CAS  PubMed  Google Scholar 

  22. Kusminski, C. M. et al. Transforming obesity: the advancement of multi-receptor drugs. Cell 187, 3829–3853 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  23. Cirulli, E. T. et al. Genome-wide rare variant analysis for thousands of phenotypes in over 70,000 exomes from two cohorts. Nat. Commun. 11, 542 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  24. Luo, S. et al. Comprehensive allele genotyping in critical pharmacogenes reduces residual clinical risk in diverse populations. Clin. Pharmacol. Ther. 110, 759–767 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  25. Mbatchou, J. et al. Computationally efficient whole-genome regression for quantitative and binary traits. Nat. Genet. 53, 1097–1103 (2021).

    Article  CAS  PubMed  Google Scholar 

Download references

Acknowledgements

We acknowledge the Helix bioinformatics and laboratory teams for their contributions to the production of the exome sequencing pipeline, as well as the research administration team for coordinating the project. We thank all the participants of DNA Answers, DNA Discovery, DNA Insights, GeneConnect, the Genetic Insights Project, GenoWell, the Healthy Nevada Project, ImagineYou, In Our DNA SC, myGenetics, Ohio State Genomic Health, PreciselyYou and The Gene Health Project. We gratefully acknowledge All of Us participants for their contributions, without whom this research would not have been possible. We also thank the National Institutes of Health’s All of Us Research Program for making the participant data examined in this study available.

Funding

The authors received no specific funding for this work.

Author information

Authors and Affiliations

  1. Helix, San Mateo, CA, USA

    Matthew E. Levy, Kelly M. Schiabor Barrett, Natalie Telis, Kent Shefchek, Lisa M. McEwen, Nicole L. Washington, William Lee & Elizabeth T. Cirulli

  2. Department of Pediatrics, WakeMed Health and Hospitals, Raleigh, NC, USA

    William H. Lagarde

  3. Rochester Regional Health, Rochester, NY, USA

    Pradyumna D. Phatak

  4. TriHealth Cancer Institute, Cincinnati, OH, USA

    James F. Maher

  5. Cone Health Precision Health, Greensboro, NC, USA

    Chad Haldeman-Englert

  6. University of Nevada School of Medicine, Reno, NV, USA

    Joseph J. Grzymski

  7. University of Nebraska Medical Center, Omaha, NE, USA

    Douglas Stoller

  8. The Ohio State University Wexner Medical Center, Columbus, OH, USA

    Amy C. Sturm

  9. Division of Cardiology, Medical University of South Carolina, Charleston, SC, USA

    Daniel P. Judge

  10. Center for Inherited Cardiovascular Diseases, WellSpan Health, Ephrata, PA, USA

    C. Anwar A. Chahal

  11. Parkview Health, Fort Wayne, IN, USA

    Jamie L. Renbarger

  12. Department of Pathology & Laboratory Medicine, St. Luke’s University Health Network, Bethlehem, PA, USA

    Christopher N. Chapman

  13. Sanford Health, Sioux Falls, SD, USA

    Steven F. Powell

  14. HealthPartners, Minneapolis, MN, USA

    Douglas A. Olson

Authors

  1. Matthew E. Levy
  2. Kelly M. Schiabor Barrett
  3. Natalie Telis
  4. William H. Lagarde
  5. Pradyumna D. Phatak
  6. James F. Maher
  7. Chad Haldeman-Englert
  8. Joseph J. Grzymski
  9. Douglas Stoller
  10. Amy C. Sturm
  11. Daniel P. Judge
  12. C. Anwar A. Chahal
  13. Jamie L. Renbarger
  14. Christopher N. Chapman
  15. Steven F. Powell
  16. Douglas A. Olson
  17. Kent Shefchek
  18. Lisa M. McEwen
  19. Nicole L. Washington
  20. William Lee
  21. Elizabeth T. Cirulli

Contributions

M.E.L., K.M.S.B., N.T. and E.T.C. designed the study. W.H.L., P.D.P., J.F.M., C.H.-E., J.J.G., D.S., A.C.S., D.P.J., C.A.A.C., J.L.R., C.N.C., S.F.P. and D.A.O. provided study oversight and access to patient populations and clinical data through their respective health systems’ participation in the Helix Research Network. M.E.L., K.M.S.B., N.T., K.S., L.M.M., N.L.W., W.L. and E.T.C. contributed to data curation and management. M.E.L. and E.T.C. performed the statistical analyses and drafted the paper. All authors revised and approved the final draft. M.E.L. and E.T.C. act as guarantors of this article.

Corresponding author

Correspondence to Matthew E. Levy.

Ethics declarations

Competing interests

M.E.L., K.M.S.B., N.T., K.S., L.M.M., N.L.W., W.L. and E.T.C. are employees of Helix. The other authors declare no competing interests.

Peer review

Peer review information

Nature Metabolism thanks Dipender Gill, Ruth Loos and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. Primary Handling Editor: Yanina-Yasmin Pesch, in collaboration with the Nature Metabolism team.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Table 1 Rare GLP1R coding variants identified in the GLP-1 RA-treated cohort

Full size table

Extended Data Table 2 Whole-genome regression analysis of rare GLP1R variants

Full size table

Extended Data Table 3 Whole-genome regression analysis of common GLP1R and GIPR variants

Full size table

Extended Data Fig. 1 Drug-specific associations between rare GIPR variants and GLP-1 receptor agonist response.

Response ratios (carrier versus matched non-carrier mean weight loss) by drug and rare variant type. Diamond, point estimate; horizontal line, 95% confidence interval (CI); vertical dashed line, equal response (ratio=1.0). Response ratios and 95% CIs are from two-sided weighted linear regression with cluster-robust standard errors at the matched-set level. Rows show results overall (both drugs combined), for semaglutide users, and for tirzepatide users. Within each drug, results are shown for all carriers of a qualifying rare variant and for individual variant classes. Matched non-carriers have no rare GIPR variant detected in exome sequencing and are the reference group for all associations shown.

Source data

Extended Data Fig. 2 Individual-level GIPR rare variant carrier weight loss responses by drug.

a, Distributions of percentage weight change after 6–12 months in GIPR rare variant carriers versus matched non-carriers, stratified by drug (semaglutide, left; tirzepatide, right). Violin plots show density; box plots show the median (center line), interquartile range (25th–75th percentile, bounds of box), and whiskers extending to the most extreme data point within 1.5× the interquartile range; diamonds show weighted means. Response ratios and P-values are from two-sided weighted linear regression with cluster-robust standard errors at the matched-set level. b, Individual GIPR rare variant carrier percentage weight change versus mean weight change of their matched non-carriers, stratified by drug (semaglutide, left; tirzepatide, right) (n = 33 semaglutide-treated and n = 30 tirzepatide-treated GIPR carriers, each matched to their own non-carrier set, totaling n = 242 and n = 234 matched non-carriers, respectively). Each line connects one carrier’s observed weight change (left endpoint) to the mean weight change of their set of matched non-carriers (right endpoint), colored by variant class. Downward-sloping lines indicate carriers who lost less weight than the average of their matched non-carriers. Matched non-carriers have no rare GIPR variant detected in exome sequencing.

Source data

Extended Data Fig. 3 Interaction between rare GLP1R variant carrier status and P7L genotype.

Mean percentage body weight change at 6–12 months by GLP1R p.Pro7Leu (P7L) genotype, stratified by drug (semaglutide, left; tirzepatide, right) and qualifying rare variant carrier status (carriers, pink; non-carriers, gray). Points show unadjusted group means ± 95% confidence interval (CI), except for genotype strata with fewer than 3 rare variant carriers; numbers below each point indicate rare variant carrier count per genotype stratum (semaglutide: n = 4 Pro/Pro, 9 Pro/Leu, 2 Leu/Leu; tirzepatide: n = 9 Pro/Pro, 8 Pro/Leu, 3 Leu/Leu). Corresponding non-carrier counts per genotype stratum were: semaglutide, n = 2,515 Pro/Pro, 3,147 Pro/Leu, 927 Leu/Leu; tirzepatide, n = 2,089 Pro/Pro, 2,496 Pro/Leu, 744 Leu/Leu. The carrier-status × P7L-dosage interaction was tested within each drug using weighted linear regression adjusted for dose category, sex, age, baseline BMI, follow-up duration, type 2 diabetes status, prior GLP-1 RA use, and the first 10 ancestry principal components; interaction P-values (two-sided Wald test with cluster-robust standard errors) are shown for each drug. Analysis restricted to individuals genetically similar to European reference populations with an available P7L genotype call; for this analysis, missing genotypes were imputed using 1000 Genomes Phase 3 data to maximize genotype completeness, retaining only calls with a maximum genotype probability exceeding 0.9 (semaglutide n = 6,604; tirzepatide n = 5,349).

Source data

Extended Data Fig. 4 Replication in the All of Us Research Program.

Percentage body weight change at 6–12 months in a, GLP1R rare variant carriers (n = 8) and b, GIPR rare variant carriers (n = 23) compared to matched non-carriers among semaglutide users. Individual points represent participants; boxes show the median (center line) and interquartile range (25th–75th percentile, bounds of box), with whiskers extending to the most extreme data point within 1.5× the interquartile range; white diamonds indicate weighted means. GLP1R carriers included 3 predicted damaging missense and 5 predicted possibly damaging missense carriers. GIPR carriers included 7 predicted loss-of-function, 11 predicted damaging missense, and 5 predicted possibly damaging missense carriers. Response ratio and two-sided P-value from weighted linear regression with cluster-robust standard errors at the matched-set level.

Source data

Extended Data Fig. 5 Internal validation in an independent Helix Research Network cohort.

Percentage body weight change at 6–12 months in a, GLP1R rare variant carriers and b, GIPR rare variant carriers compared to matched non-carriers, stratified by drug (semaglutide, left; tirzepatide, right). Individual points represent participants; boxes show the median (center line) and interquartile range (25th–75th percentile, bounds of box), with whiskers extending to the most extreme data point within 1.5× the interquartile range; white diamonds indicate weighted means. Drug interaction P-value (two-sided) from weighted linear regression with a carrier × drug interaction term. Semaglutide-treated GLP1R carriers (n = 7) included 1 predicted loss-of-function, 2 predicted damaging missense, and 4 predicted possibly damaging missense carriers; tirzepatide-treated GLP1R carriers (n = 11) included 4 predicted damaging missense and 7 predicted possibly damaging missense carriers. Semaglutide-treated GIPR carriers (n = 20) included 3 predicted loss-of-function, 3 predicted damaging missense, and 14 predicted possibly damaging missense carriers; tirzepatide-treated GIPR carriers (n = 15) included 2 predicted loss-of-function and 13 predicted possibly damaging missense carriers.

Source data

Extended Data Fig. 6 Comparison of GLP1R P7L and rare variant effects on weight loss.

Response ratios and 95% confidence intervals for the common GLP1R p.Pro7Leu (P7L; blue circles, per T-allele) and any qualifying rare GLP1R variant (carrier vs. non-carrier; red squares), both estimated from whole-genome regression models in regenie restricted to individuals genetically similar to European reference populations, so that common- and rare variant effects shown are directly comparable. Circles and squares indicate response ratio point estimates; horizontal lines indicate 95% confidence intervals. Response ratios and P-values are from two-sided linear regression in regenie. Results are shown separately by drug: for the P7L analysis, semaglutide n = 1,905 and tirzepatide n = 1,597, restricted to individuals with a passing P7L genotype call, including 1,116 and 915 T-allele carriers, respectively; for the rare variant analysis, semaglutide n = 6,601 (6,586 non-carriers, 15 carriers) and tirzepatide n = 4,719 (4,703 non-carriers, 16 carriers).

Source data

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Levy, M.E., Schiabor Barrett, K.M., Telis, N. et al. Rare GLP1R variants attenuate weight loss with semaglutide but not tirzepatide. Nat Metab (2026). https://doi.org/10.1038/s42255-026-01656-z

Download citation

  • Received: 02 March 2026

  • Accepted: 06 October 2026

  • Published: 11 November 2026

  • Version of record: 11 November 2026

  • DOI: https://doi.org/10.1038/s42255-026-01656-z

来源:Nature 健康护理主题 · nature.com