Your Genes Might Predict How Well Ozempic Works — But the Effect Is Smaller Than Headlines Suggest

A landmark study links variants in two genes to GLP-1 drug response. The science is real, but the gap between statistical association and clinical utility remains wide.

The Promise and the Hype

GLP-1 receptor agonists — semaglutide (sold as Wegovy and Ozempic) and tirzepatide (Zepbound and Mounjaro) — have transformed obesity treatment. But response to these drugs varies enormously. In clinical trials, the average weight reduction on semaglutide was about 10.2%, yet roughly 32% of patients lost less than 5% of their body weight or even gained weight, while another 32–40% were "super-responders" who lost more than 20% [1][2].

Source: 23andMe / Nature (2026)

Data as of Apr 8, 2026CSV

That wide spread has made one question urgent: can we predict who will benefit most? On April 8, 2026, researchers at the 23andMe Research Institute published the largest genome-wide association study (GWAS) to date on GLP-1 drug response, analyzing self-reported data from 27,885 users. The results, published in Nature, identify genetic variants in two genes — GLP1R and GIPR — associated with differences in weight loss and side effects [3]. The findings are scientifically credible but clinically modest, and experts are urging caution about overinterpreting what the genes can tell us.

What the Study Found

The study's primary finding centers on a missense variant in the GLP1R gene (rs10305420), which encodes the receptor that GLP-1 drugs bind to. Each copy of the effect allele was associated with an additional 0.76 kilograms (about 1.7 pounds) of weight loss over a median treatment period of approximately 8 months. Individuals carrying two copies lost roughly 3 pounds more than those with none. The association reached genome-wide significance with a P value of 2.9 × 10⁻¹⁰ and was independently replicated in the All of Us research database (P = 0.001) [3][4].

The variant is present in approximately 40% of people with European and Middle Eastern ancestry [5]. Its proposed biological mechanism: it may improve the efficiency of transporting GLP-1 receptors to cell surfaces, increasing the number of receptors available for drug binding [5]. Adam Auton, senior author and vice president at 23andMe Research Institute, said the finding "makes perfect biological sense" because it directly affects the drug's molecular target [5][6].

A second set of findings linked variants in both the GLP1R and GIPR genes to GLP-1 medication-related nausea and vomiting. The GIPR association was restricted to tirzepatide users — consistent with the fact that tirzepatide is a dual GLP-1/GIP receptor agonist, while semaglutide acts only on GLP-1 receptors. According to the study, depending on genetics and clinical factors, a patient's predicted probability of experiencing nausea or vomiting ranged from 5% to 78%. Individuals carrying two copies of both the GLP1R and GIPR variants were estimated to be 15 times more likely to experience vomiting on tirzepatide [3][5].

The Effect Size Problem

The headline numbers deserve scrutiny. An additional 0.76 kg of weight loss per allele, over 8 months of treatment, is statistically significant but clinically small. For context, the average weight loss in the study cohort was approximately 25 pounds [5]. An extra 1.7 to 3.4 pounds, depending on genotype, represents a marginal increment.

Dr. Marie Spreckley, a Research Programme Manager at the University of Cambridge, said in a Science Media Centre response: "The magnitude of these genetic effects is small in clinical terms...a difference of less than 1kg per allele is modest" [4]. She added that "evidence is not yet sufficient to support using genetic information to guide treatment decisions in routine clinical practice" [4].

Giles Yeo, a geneticist at the University of Cambridge, offered a more concise framing: "Your genes do matter, but it's not only your genes." He noted that genetics explains only partial variation, while sex, age, comorbidities, and specific drug choice also predict response [5].

Genetics vs. Everything Else

The study itself quantifies how much genetics adds to prediction — and the answer is: not much, yet. A model built on non-genetic factors alone (sex, drug type, dose, treatment duration, age, diabetes status) explained about 21.4% of the variance in percentage BMI loss. Adding genetic variables to the model pushed that to approximately 25% [3][4]. In other words, genetics contributed roughly 3.6 percentage points of additional explanatory power.

The remaining 75% of variance is unexplained by either model. Among the known non-genetic predictors: women lose more weight than men (14–16.2% vs. 8–9.3% of body weight on semaglutide); people without type 2 diabetes respond better (14.9% weight loss vs. 9.6% for those with diabetes); each additional decade of age corresponds to about 0.5% less BMI reduction; and drug type matters, with tirzepatide generally producing larger responses than semaglutide [2][3].

Ruth Loos, a professor at the University of Copenhagen who studies obesity genetics, argued that even small per-allele effects can matter at the population level because they apply to many people. She noted that small weight losses still confer health benefits such as lower cholesterol [5]. That is true — but it is a different argument than saying genetic testing should guide individual prescribing decisions.

The Ancestry Gap

The 23andMe study cohort was predominantly female and largely of European ancestry [4][7]. This is a familiar limitation in genomics research. Historically, GWAS databases have been built disproportionately from European-descent populations, and the predictive accuracy of polygenic findings often degrades when applied to other ancestry groups [8].

This matters because the frequency of pharmacogenomically relevant variants can differ substantially across populations. For example, a variant in the ARRB1 gene (rs140226575), identified in earlier GLP-1 pharmacogenomics work, has a frequency of just 0.05% in White Europeans but 6% in Hispanic populations and 11% in American Indian or Alaska Native populations [9]. Whether the GLP1R variant (rs10305420) identified in the current study has equivalent predictive value across Black, Latino, Asian, and Indigenous populations remains unknown.

Source: WHO Global Health Observatory

Data as of Dec 31, 2022CSV

With adult obesity prevalence reaching 42% in the United States, 30.8% in South Africa, 28.1% in Brazil, and 26.8% in the United Kingdom, the global demand for effective obesity pharmacotherapy is enormous [10]. A pharmacogenomic framework that works primarily for European-ancestry genomes risks deepening existing health disparities — a concern raised by researchers who have called for "well powered, diversity-focused, multi-ethnic studies" [9].

The Biological Mechanism: How Much Do We Actually Understand?

The GLP-1 receptor (GLP1R) belongs to the G protein-coupled receptor (GPCR) family and is expressed in pancreatic beta cells, the central nervous system, and the gastrointestinal tract [11]. When GLP-1 drugs bind to this receptor, they trigger intracellular signaling cascades that stimulate insulin secretion, suppress appetite, and slow gastric emptying [11].

The proposed mechanism for the rs10305420 variant — improved receptor trafficking to cell surfaces — is biologically plausible but not yet confirmed through functional studies [5]. The distinction matters. A statistical association in a GWAS establishes that a variant correlates with an outcome in a population. It does not prove that the variant causes the outcome through a specific molecular pathway. Functional validation — showing, for instance, that cells carrying the variant actually express more surface receptors — would strengthen the causal claim considerably.

For the GIPR-related side effect findings, the biological story is more straightforward: tirzepatide activates both GLP-1 and GIP receptors, so variants affecting GIP receptor function would logically influence tirzepatide-specific responses but not semaglutide-only responses [3][11]. That pharmacological specificity adds credibility to the association.

From Lab to Clinic: A Long Road

No regulatory agency has approved or been formally petitioned for a pharmacogenomic test to guide GLP-1 prescribing. No validated diagnostic panel exists for this purpose. The research sits squarely in what translational scientists call the T1 phase — moving from basic discovery toward a candidate clinical application — with three more translational phases (evidence-based guidelines, clinical practice implementation, and population health impact) still ahead [12].

23andMe has announced that its Total Health platform will provide consumers with information about these genetic variants and their predicted impact on GLP-1 outcomes [6]. This is a direct-to-consumer offering, not a clinically validated diagnostic. The distinction is significant: a consumer report that says "you carry a variant associated with slightly more weight loss" is different from a physician-ordered test that changes prescribing.

The cost question is unresolved. Comprehensive pharmacogenomic panels currently range from roughly $200 to $500 in the United States, though prices vary widely depending on scope and provider. Whether insurers would cover pre-prescribing genetic screening for obesity drugs depends on evidence that such screening improves outcomes enough to justify the cost — evidence that does not yet exist [12]. The simpler alternative, which most clinicians already follow, is to prescribe the drug, monitor response over several months, and adjust if needed.

Source: OpenAlex

Data as of Jan 1, 2026CSV

Academic interest in obesity pharmacogenomics has surged, with publications on the topic rising from 357 papers in 2011 to over 1,600 in 2025 [13]. But publication volume is not the same as clinical readiness. The gap between identifying a statistically significant variant and deploying a validated, cost-effective test in clinical workflows is measured in years to decades, not months.

The Discrimination Question

Pharmacogenomic data introduces a less-discussed risk: could genetic stratification give insurers or employers a reason to deny coverage or penalize predicted "poor responders"?

The Genetic Information Nondiscrimination Act (GINA), enacted in 2008, prohibits health insurers and employers in the United States from using genetic information for coverage or employment decisions [14]. However, GINA has significant gaps. It does not cover life insurance, disability insurance, or long-term care insurance [14][15]. And while GINA bars insurers from requiring genetic tests, it does not require them to cover the cost of genetic testing or to pay for treatments that testing might recommend [14].

A more subtle risk: GINA protects against discrimination based on genetic predisposition but not against discrimination based on manifested conditions [14]. If a patient takes a pharmacogenomic test, is classified as a likely poor responder, and an insurer uses that classification to deny coverage for a GLP-1 drug — arguing that the expected benefit does not meet a cost-effectiveness threshold — the legal protections are unclear. Some states have enacted additional protections beyond GINA, but coverage varies [15].

The American Medical Association has flagged genetic discrimination as a growing concern as pharmacogenomics expands, and patient advocacy groups have called for updating GINA to address drug coverage decisions explicitly [15].

What the Remaining 75% Tells Us

The most important number in this study may be the one that gets the least attention: approximately 75% of the variance in GLP-1 drug response remains unexplained by the combined genetic and clinical model [3][4]. This means that even with the best available predictors — age, sex, drug type, dose, treatment duration, diabetes status, and now these genetic variants — three-quarters of why one patient loses 30 pounds and another loses 3 pounds remains a mystery.

Some of that unexplained variance likely comes from factors that are hard to measure in a GWAS: dietary adherence, physical activity, gut microbiome composition, psychological relationship with food, socioeconomic access to the drug, and consistency of dosing. Some may come from rare genetic variants with larger effects that were not captured in this analysis — a point Dr. Spreckley raised specifically [4].

The researchers themselves acknowledge these limitations. But the press coverage and 23andMe's consumer-facing communications have, in some cases, emphasized the genetic findings more than the caveats [6]. The risk is that patients and the public come away believing that a genetic test can meaningfully predict whether a $1,000-per-month drug will work for them, when the evidence supports a far more modest claim.

The Bottom Line

The 23andMe study is good science. It identified a biologically plausible genetic signal, replicated it in an independent cohort, and published it in a top-tier journal. The side-effect findings, particularly for tirzepatide-related nausea and vomiting, could eventually have practical value if validated in prospective clinical trials.

But the distance from here to a pharmacogenomic test that changes clinical practice is considerable. The effect sizes are small. The study population lacks ancestral diversity. The functional mechanisms are proposed but unconfirmed. Non-genetic factors dominate prediction. No regulatory pathway has been initiated. And the legal framework for protecting patients from genetic discrimination in drug coverage decisions has known gaps.

For now, the most evidence-based approach to prescribing GLP-1 drugs remains what most physicians already do: start the medication, monitor response, and adjust based on observed outcomes rather than predicted genotypes. The genetics are interesting. The clinical utility is not yet proven.