Your DNA May Decide Whether Ozempic Makes You Sick: Inside the Largest Study Linking GLP-1 Side Effects to Genetics

A new study published in Nature has identified genetic variants that help predict who will lose the most weight on GLP-1 drugs — and who will spend weeks battling nausea and vomiting. The findings arrive as prescriptions for these drugs have exploded, but the path from genetic discovery to clinical use is far longer and more fraught than the headlines suggest.

The Study: What 28,000 Genomes Revealed

On April 8, 2026, researchers at the 23andMe Research Institute published results from a genome-wide association study (GWAS) of 27,885 people who had used at least one GLP-1 receptor agonist — Ozempic, Wegovy, compounded semaglutide, Mounjaro, Zepbound, or compounded tirzepatide [1]. It is the largest pharmacogenomic study of GLP-1 drugs to date.

The study identified two genes with statistically significant associations:

GLP1R (chromosome 6): A missense variant, rs10305420, was linked to greater weight loss. Each copy of the effect allele corresponded to an additional 0.76 kg (about 1.7 lbs) of weight loss. Individuals carrying two copies lost more than 3 additional pounds compared to non-carriers [2]. The variant is present in roughly 40% of people with European and Middle Eastern ancestry [3]. The main association reached a P value of 2.9 × 10⁻¹⁰ — well beyond the standard genome-wide significance threshold [4].

GIPR (chromosome 19): A variant, rs1800437, which alters the 354th amino acid of the GIP receptor protein from glutamic acid to glutamine, was associated with increased nausea and vomiting — but only in people taking tirzepatide (Mounjaro and Zepbound), not semaglutide [1]. This makes biological sense: tirzepatide is a dual GIP/GLP-1 agonist, while semaglutide targets only the GLP-1 receptor. People carrying two copies of both risk-associated variants were up to 15 times more likely to experience vomiting [3].

The findings were externally replicated in the All of Us research cohort, achieving a P value of 0.001 [4].

Source: PMC / GLP-1 RA Adverse Effects Review

Data as of Jan 1, 2025CSV

Context: A Market Under Pressure

These genetic findings land in the middle of an unprecedented pharmaceutical boom. GLP-1 receptor agonist prescriptions in the United States have grown from roughly 800,000 in 2020 to over 6.1 million by the third quarter of 2025 [5]. Between January 2024 and July 2025 alone, 1.6 million patients received a GLP-1 prescription for the first time [5].

Source: medRxiv / GLP-1 RA Prescribing Trends

Data as of Sep 1, 2025CSV

The financial stakes are enormous. Eli Lilly, which manufactures tirzepatide (Mounjaro and Zepbound), reported that Zepbound sales alone more than doubled year-over-year to nearly $4.3 billion in Q4 2025, while Mounjaro generated $7.41 billion in the same quarter [6]. Lilly's U.S. GLP-1 market share climbed to 60.5% by Q4 2025 [6]. Novo Nordisk, maker of semaglutide (Ozempic and Wegovy), saw its share fall to 39.1% and warned of a 5–13% decline in sales and profit for 2026 as U.S. prices fell and international exclusivity expired [6].

The global obesity epidemic provides the demand backdrop. Adult obesity prevalence stands at 42% in the United States, 30.8% in South Africa, 30.2% in Australia, and 28.1% in Brazil, according to the WHO [7].

Source: WHO Global Health Observatory

Data as of Dec 31, 2022CSV

What the Variants Mean for Side Effects — and Who Carries Them

Gastrointestinal side effects are the primary reason patients discontinue GLP-1 therapy. Nausea affects up to 50% of patients during initial treatment, vomiting affects roughly 25%, and diarrhea about 20% [8]. These effects are dose-dependent and generally subside over time, but for a substantial minority they are severe enough to end treatment.

The more serious safety signals — pancreatitis (occurring in fewer than 1% of patients), thyroid C-cell tumors (the basis for current black box warnings on most GLP-1 drugs), and gastroparesis — were not the primary focus of the 23andMe GWAS, which concentrated on self-reported nausea, vomiting, and weight loss [1]. GLP-1 receptor agonists carry a labeled black box warning for thyroid C-cell tumors, including medullary thyroid carcinoma, based on animal models and postmarketing case reports [9].

The specific side-effect SNPs identified were rs11760106 (vomiting signal) and rs9357296 (nausea signal) [10]. For the GIPR variant specifically, the association with vomiting was found in near-perfect linkage disequilibrium with the missense variant rs1800437 [10].

If roughly 40% of European-ancestry individuals carry at least one copy of the GLP1R efficacy variant [3], and there are currently over 6 million active GLP-1 prescriptions in the U.S. [5], a rough estimate suggests 2–3 million current users may carry variants that modestly influence their drug response. But this calculation carries major caveats — chiefly, that the variant frequencies in non-European populations are not well characterized by this study.

The Ancestry Gap

The study's participants were 78.3% of European ancestry, 12.9% Latino, and 4.2% African American [1]. They were 82.4% female with a median age of 52 [1]. This demographic skew is a well-known problem in pharmacogenomics research: variant frequencies, effect sizes, and even which variants matter most can differ substantially across ancestry groups.

This is not an abstract concern. Research has consistently shown that pharmacogenomic panels calibrated primarily on European-ancestry populations perform worse in other groups [11]. Historical and ongoing underrepresentation of marginalized populations in genomics research has produced variant databases that are poorly representative, increasing the chance that clinically important variants in non-European populations remain uncharacterized [11]. Given that obesity disproportionately affects Black and Latino Americans, the populations most likely to need GLP-1 drugs are precisely those least represented in the genetic data.

Ruth Loos, a geneticist at the University of Copenhagen, confirmed the weight differences found in the study are "not negligible" [3]. But Giles Yeo of the University of Cambridge offered a broader caution: "Your genes do matter, but it's not only your genes" [3].

How These Findings Compare to Other Pharmacogenomics

Pharmacogenomics — the study of how genetic variation affects drug response — is not new. For statins, variants in the SLCO1B1 gene can increase the risk of muscle damage (myopathy) by up to 17-fold in some populations, and the Clinical Pharmacogenetics Implementation Consortium (CPIC) has published dosing guidelines based on SLCO1B1 genotype. For SSRIs (selective serotonin reuptake inhibitors), CYP2D6 and CYP2C19 metabolizer status can shift drug levels by several-fold, and the FDA includes pharmacogenomic information in the labels of multiple antidepressants.

By comparison, the GLP-1 effect sizes reported here are modest. An additional 0.76 kg of weight loss per allele, against an average total weight loss of roughly 25 pounds among study participants [3], represents an incremental shift rather than a binary predictor of drug response. The side-effect associations, particularly the 15-fold increased vomiting risk for dual-variant carriers on tirzepatide, are more striking — but apply to a small subset of the population carrying two copies of both variants.

The Research Boom

Academic interest in GLP-1 pharmacogenomics has surged. According to OpenAlex data, 341 papers on the topic were published in 2025 — up from just 21 in 2011. The 2026 pace, with 124 papers published through early April, suggests continued growth [12].

Source: OpenAlex

Data as of Jan 1, 2026CSV

Previous work, including a meta-GWAS published in The Lancet Diabetes & Endocrinology, had identified common variants in GLP1R, TCF7L2, and SLC5A2 associated with glycemic response to GLP-1 receptor agonists [10]. The 23andMe study extends this to weight loss and side effects — a distinct and arguably more commercially relevant outcome given the obesity-treatment market.

What Genetic Screening Would Actually Cost

Pharmacogenomic testing prices in the U.S. range widely: basic single-gene tests run $100–$300, while comprehensive multi-gene panels typically cost $250–$2,000, with most clinically validated tests from reputable providers falling between $300 and $600 [13]. For common clinical tests like GeneSight (used primarily for psychiatric medications), the published out-of-pocket maximum is $330 for uninsured patients [13].

Insurance coverage remains inconsistent. Medicare Part B covers some pharmacogenomic tests when ordered by a treating physician. Medicaid generally covers testing at no out-of-pocket cost. Commercial insurance varies plan by plan, with many plans limiting coverage to specific scenarios like treatment-resistant depression or polypharmacy in elderly patients [13]. No major insurer currently covers pharmacogenomic testing specifically for GLP-1 prescribing decisions.

International cost-effectiveness studies suggest pharmacogenomic testing can save $3,000–$4,000 per patient per year in system costs, even when test prices are set around $2,000 [13]. But these savings estimates come from contexts like psychiatric polypharmacy, where wrong-drug choices lead to repeated hospitalizations — a different risk profile than GLP-1 side effects, which are mostly gastrointestinal and self-limiting.

A realistic timeline for integrating genetic screening into GLP-1 prescribing guidelines would require multiple replication studies, clinical utility trials demonstrating that genotype-guided prescribing improves outcomes over standard care, regulatory review, and guideline committee action. Based on precedent from other pharmacogenomic implementations, this process typically takes 5–10 years from initial discovery.

Methodological Limitations

Independent experts have raised several concerns about the study's methodology:

Self-reported data. Weight loss and side effect information came from participant surveys, not clinical records or electronic health data. As one expert noted at the Science Media Centre, "much of the data is self-reported, which can introduce bias" [4].

Not a randomized controlled trial. The study is observational and based on a consumer genetics platform's research cohort, not a pre-registered clinical trial [4]. Participants who choose to take 23andMe tests and respond to surveys may differ systematically from the broader patient population.

Modest effect sizes. Multiple experts emphasized that "the effect size remains modest" and that "most of the predictive capacity still derives from non-genetic factors such as sex, age, type 2 diabetes" [4]. Clinical variables — dose, adherence, comorbidities — currently explain more variance in drug response than the identified genetic variants.

Limited diversity. The 78.3% European-ancestry composition limits generalizability [1].

Need for replication. Experts stressed the findings need to "be replicated in other studies before being incorporated into clinical practice" [4]. While the All of Us replication is promising, it represents a single additional cohort.

Adam Auton, VP of Human Genetics at the 23andMe Research Institute and the study's senior author, acknowledged the limitations while defending the approach: "I think we have proof of concept here that genetics is playing a role in terms of GLP-1 efficacy and side effects" [2].

The Industry Angle

23andMe's interest in this research is not purely academic. The company's Total Health platform offers consumers access to genetic information about drug response, and findings linking specific genotypes to GLP-1 outcomes directly enhance the commercial value of that product [2]. The study did not report external industry funding from pharmaceutical manufacturers, but the institutional interest is transparent: 23andMe monetizes genetic data and genetic insights.

Eli Lilly and Novo Nordisk, the dominant GLP-1 manufacturers, have not publicly endorsed or opposed pharmacogenomic screening for their drugs. Their financial incentives are mixed. On one hand, screening that identifies patients most likely to tolerate treatment could improve adherence and reduce negative publicity from side effects. On the other, any screening that restricts the eligible patient pool threatens the volume-driven growth model that has propelled both companies — Lilly toward projected 2026 revenue of $80–$83 billion and Novo Nordisk to approximately $48.9 billion in 2025 sales [6].

Competitors developing next-generation obesity drugs — including oral GLP-1 formulations and novel targets — might benefit if genetic screening raises the barrier to prescribing existing injectables.

The Case Against Genetic Gatekeeping

The strongest argument against routine genetic screening before GLP-1 prescribing is that it could create a two-tiered access system. If insurers began requiring genetic tests and using risk scores to determine coverage, patients carrying unfavorable variants could be denied drugs that, on average, produce substantial weight loss and cardiovascular benefit — even if their individual risk of side effects is manageable.

This concern is not hypothetical. In psychiatric pharmacogenomics, some insurers have begun requiring GeneSight testing before approving certain medications, a practice that has drawn criticism from physicians who argue it delays treatment for patients in crisis. Applied to GLP-1 drugs, where cost is already a major barrier — monthly retail prices range from $900 to $1,349 before insurance [14] — genetic gatekeeping could compound existing inequities.

The ancestry gap compounds this risk. If screening tools are calibrated on European-ancestry data, they may produce less accurate predictions for Black and Latino patients, who already face higher obesity rates and more barriers to care [11]. A genetic scoring system that works less well for these populations could systematically disadvantage them.

There is also a philosophical question about who bears responsibility when coverage decisions are based on probabilistic genetic risk rather than demonstrated individual harm. A patient denied GLP-1 coverage based on a variant associated with a 15-fold increase in vomiting risk might still have tolerated the drug without incident. The population-level statistic does not determine the individual outcome.

Regulatory Implications

The FDA's current black box warnings on GLP-1 drugs are one-size-fits-all: they warn about thyroid C-cell tumor risk across the entire drug class, without reference to individual genetic susceptibility [9]. If genetic variation meaningfully predicts who is at higher risk for specific side effects, a case exists for updating drug labels to reflect this — as the FDA has already done for dozens of drugs with pharmacogenomic information in their labeling.

However, no regulatory body — not the FDA, the European Medicines Agency, or any other — has signaled intent to act specifically on the 23andMe findings. The FDA recently completed a review finding no increased risk of suicidal ideation or behavior with GLP-1 drugs [9], and its current enforcement focus is on unapproved compounded GLP-1 products, not pharmacogenomic labeling [9].

In Europe, the European Medicines Agency has historically been receptive to pharmacogenomic evidence but has not issued guidance specific to GLP-1 receptor agonists. Any regulatory action would require substantially more evidence than a single GWAS, including prospective clinical trials demonstrating that genotype-guided prescribing reduces adverse events without reducing access to effective treatment.

What Comes Next

The 23andMe study is a proof of concept, not a clinical tool. It demonstrates that common genetic variation influences GLP-1 drug response in statistically detectable ways. But the gap between statistical association and clinical utility is wide.

For patients currently taking or considering GLP-1 medications, the practical implications are limited. No clinical guideline recommends genetic testing before GLP-1 prescribing. The identified variants explain only a fraction of the variation in drug response. And the most important predictors of who will benefit — starting weight, adherence, dose titration, and comorbidities — remain non-genetic.

For researchers, the study opens productive lines of inquiry: larger, more diverse cohorts; integration of electronic health record data to replace self-reported outcomes; functional studies of how the identified missense variants alter receptor signaling; and clinical trials testing whether genotype-guided prescribing improves outcomes.

For policymakers and insurers, the findings are a preview of decisions that will recur with increasing frequency as pharmacogenomic data accumulates for blockbuster drug classes. The question is not whether genetics influences drug response — it clearly does — but whether acting on that knowledge prematurely will help patients or create new barriers to care.