23andMe study finds common gene variants modestly affect weight loss and nausea on GLP‑1 drugs

A Nature paper and a company report published April 8 tie common DNA differences to modest changes in how people respond to GLP‑1 weight‑loss drugs — and highlight how quickly research built from crowdsourced data can be turned into consumer products.

Study overview

Researchers at the 23andMe Research Institute analyzed survey and genetic data from 27,885 customers who reported using at least one GLP‑1 medication, including branded products such as Ozempic and Wegovy (semaglutide) and Mounjaro and Zepbound (tirzepatide). The paper, "Genetic predictors of GLP1 receptor agonist weight loss and side effects," led by Adam Auton, focuses on two outcomes: percentage change in body‑mass index (BMI) after starting a GLP‑1 drug and the occurrence of gastrointestinal side effects such as nausea and vomiting. All authors are current or former 23andMe employees, a conflict disclosed in the paper.

After quality control, the genome‑wide association analysis for BMI change included 15,237 participants; a stricter subset with both pre‑ and post‑treatment BMI passed checks for 4,889 people, and 3,948 had full clinical covariates used in prediction models. The cohort was predominantly middle‑aged women (82.4% female; median age 52) and mainly of European genetic ancestry (78.3%), with a median baseline BMI of 35.1 kg/m2 and a median reported treatment duration of roughly 8.3 months.

Key genetic findings

• GLP1R and weight loss: The strongest genetic association for weight loss was a missense variant in GLP1R (rs10305420, p.Pro7Leu). Each copy of the T allele was associated with an additional 0.641% reduction in BMI — about 0.76 kilograms of extra weight loss per allele — with a P value of 2.9 × 10−10. That variant was the only coding change in the study's 99% credible set at that locus.

• Side effects linked to GLP1R: Several signals mapped to GLP1R for gastrointestinal adverse effects. For nausea the leading variant was rs9357296; for vomiting the index variant rs11760106 had a P value of 2.5 × 10−27 and the T allele increased odds of vomiting by about 1.57‑fold.

• GIPR and tirzepatide: Among tirzepatide users, an association mapped to GIPR (rs71338792, P = 4.2 × 10−9; OR ≈ 1.84). That signal is in near‑perfect linkage disequilibrium with a missense variant rs1800437 (p.Glu354Gln). The C allele (Gln) was associated with higher vomiting risk, whereas the G allele was protective (reported OR ~0.546). The GIPR association appeared specific to tirzepatide and was not detected in semaglutide users.

Prediction models: clinical vs genetic contributions

The authors combined genetic variants and clinical variables (age, sex, baseline BMI, diabetes status, treatment duration) into predictive models. The combined model explained roughly 25% of variance in percentage BMI change in both training and held‑out test sets; most predictive power came from clinical features alone, which produced an adjusted R2 of about 21.4%.

For side effects the predictive performance was modest: AUC in the held‑out test set was 0.654 for nausea and 0.680 for vomiting. The paper emphasizes that genetic predictors added relatively little to discrimination beyond non‑genetic factors.

Replication and data‑source differences

The GLP1R efficacy signal replicated in the NIH All of Us Research Program but was not reproduced in an analysis of UK Biobank data. The authors note that UK Biobank collected data earlier, before newer GLP‑1 obesity drugs were widely available, and may have had limited statistical power and different exposure patterns.

The study also highlights how outcome measurement matters. In 909 participants with prescriptions recorded through Apple HealthKit EHR integration — and in 195 people with both EHR and self‑reported records — self‑reported weight loss was substantially larger than EHR‑based measures. Median self‑reported BMI change was −11.8% overall versus −5.79% in the EHR subset. Among those with both sources the median self‑reported change was −14.14% versus −8.43% in EHRs (paired P = 1.1 × 10−9). The authors suggest self‑report bias, incomplete medical records and use of telehealth or compounded prescriptions as possible explanations.

Access to data and conflicts

Summary statistics for variants with P < 1 × 10−4 are included in the paper's supplements; broader summary data are available to qualified researchers under 23andMe's dataset access program. Individual‑level data are not publicly released under the company’s IRB protocols. The paper discloses that all authors are current or former 23andMe employees.

Commercial rollout

Within hours of the Nature publication, 23andMe announced a new "GLP‑1 Medications: Weight Loss and Nausea" report for subscribers to its Total Health service, which the company says includes an interactive tool and clinician guidance. In the company's press release, Auton highlighted the value of the crowdsourced research community, while 23andMe medical officer Noura Abul‑Husn said the reports aim to support a "more informed and personalized approach to weight management." The product launch illustrates how quickly findings from a large consumer genetic database can be translated into retail offerings.

Expert reactions and limitations

Independent experts described the study as well designed and biologically plausible, while underscoring its limitations. Effect sizes for individual variants are small — the lead GLP1R variant amounts to roughly 0.76 kg extra weight loss per allele — and most variation in response is explained by clinical and demographic factors.

Critics also point to the study's sample composition and outcome measures: the cohort is majority female and of European ancestry, and the primary efficacy measure relies heavily on self‑reported weight and height. Those features may limit the generalizability of the findings to broader and more diverse patient populations.

What this means for patients and clinicians

The study shows that common genetic differences can nudge the odds of slightly greater weight loss or a higher chance of nausea and vomiting on GLP‑1 drugs, but the effects are modest compared with clinical predictors. For most patients, age, sex, baseline BMI, diabetes status and treatment duration will remain the dominant factors shaping response. Whether genotype‑based consumer reports will meaningfully change prescribing or outcomes remains uncertain and will depend on further validation in diverse, clinically measured cohorts.

Bottom line

23andMe's Nature paper adds to our understanding of biological variation in response to GLP‑1 therapies and identifies plausible receptor‑level signals in GLP1R and GIPR. But the genetic effects reported are small, the study relies on self‑reported outcomes from a nonrepresentative sample, and clinical variables still account for the bulk of predictive power. The rapid conversion of these research results into a paid consumer report underscores the increasing speed at which crowdsourced genetic research can move from academic publication to commercial product.

Methodological note: All statistics, P values and variant identifiers are taken from the paper and its supplementary materials published April 8.