Stanford April 2026: Why Ozempic works less well in 10% of people. The genetics behind it

On April 10, 2026, an international research team led by Anna Gloyn (Stanford Medicine) and Markus Stoffel (ETH Zurich) published a study in Genome Medicine that anchors a long-discussed clinical phenomenon in genetics for the first time: roughly one in ten people carries a variant in the PAM gene that measurably reduces the effectiveness of GLP-1 agonists such as semaglutide and liraglutide. Lead authors Mahesh Umapathysivam (Adelaide) and Elisa Araldi (Parma) describe it as "GLP-1 resistance": the body produces normal or even elevated GLP-1 levels, but the signal does not reach its full biological effect.

For researchers and users wondering why their glycemic and weight response lags behind study medians, this work is an important piece of the puzzle. It directly complements the 23andMe GWAS in Nature that we discussed in early April, adding a second genetic axis to the picture of response variability in GLP-1 therapies.

Disclaimer: This article is for informational purposes only and does not constitute medical advice. Genetic variants are currently not tested in routine care. Decisions about therapy or dosing belong in the hands of a qualified physician. All substances offered by PeptidesDirect are exclusively intended for research purposes.

What Stanford and ETH Zurich found

The study combines genomic data from large population cohorts with a meta-analysis of three clinical GLP-1 studies (a total of 1,119 participants). The focus is on two rare but, combined, common missense variants in the PAM gene (peptidyl-glycine alpha-amidating monooxygenase): p.S539W and p.D563G. Both alter an enzyme that normally "activates" peptide hormones biochemically by replacing the C-terminal glycine with an amide group. Without this amidation, many peptide hormones lose their full biological activity.

The key numbers from the meta-analysis, six months after start of therapy:

Genotype	Patients who reached their HbA1c target
Wild type (no PAM defect)	around 25%
p.D563G carriers	around 18.5%
p.S539W carriers	around 11.5%

This is clinically significant. Carriers of the more strongly affected variant had less than half the success rate of a wild-type carrier under GLP-1 therapy. Importantly, the effect was specific to GLP-1 agonists. Other antidiabetic drugs in the comparison arms did not show this genotype-dependent spread. This is therefore not a general "severe diabetes phenotype" but a drug-class-specific response weakness.

The combined frequency of both variants in European populations is about 10%. This is not a fringe phenomenon. Stanford's data applies to one in every ten users of semaglutide, liraglutide, or related substances.

The mechanism: elevated GLP-1, but the signal does not arrive

PAM is not a GLP-1 receptor and not a GLP-1 gene. It is an enzyme that plays a role in the maturation of various peptide hormones. Here lies the seemingly paradoxical result: in carriers of the defective PAM variants, endogenous GLP-1 plasma levels are not reduced, but rather elevated. The signal is there. But, presumably due to incomplete amidation and altered receptor pharmacodynamics, it is biologically processed inefficiently.

The authors emphasize that the exact downstream mechanism beyond receptor binding and signal transduction has not yet been clarified. The hypothesis: an incorrectly amidated or otherwise processed peptide form alters the bioactivity not only of GLP-1, but possibly also of related incretins and paracrine peptides in the pancreatic islet cells. This would explain why the resistance affects exogenous semaglutide in a similar way to endogenous GLP-1.

In practice, this means: even a dose escalation to 2.4 mg semaglutide weekly or comparably high liraglutide doses cannot be "up-titrated" arbitrarily to overcome the genetic defect. The receptor does get more ligand, but the downstream system remains dampened. A classic picture from receptor pharmacology: when the effector cascade behind the receptor is limited, a higher concentration of agonist at the receptor only helps up to a saturation threshold, beyond which side effects predominate over additional efficacy.

The evolutionary aspect is also interesting: the fact that a variant associated with weaker incretin response has been preserved in 10 percent of the population suggests it had no substantial selective disadvantage in the past. In a historical dietary context shaped by periodic food scarcity, a "less sensitive" satiety and insulin secretion axis may even have been advantageous. In today's environment with constant calorie availability, this advantage tips into a clinically relevant disadvantage.

How this was identified: GWAS plus functional validation

The methodology combines several steps:

1. Population GWAS: Prior work had already shown that PAM variants are associated with type 2 diabetes risk, impaired insulin secretion, and altered incretin levels.
2. Clinical cohort analysis: The Stanford group pulled together three clinical studies on GLP-1 therapies, genotyped participants for p.S539W and p.D563G, and stratified the response.
3. Functional tests: Reduced amidation efficiency of the mutants was confirmed in cell models and with recombinant enzyme variants.
4. Mendelian randomization: Genetic data were used to distinguish causality from mere correlation.

This multi-layered approach makes the work more robust than a pure association study. The authors also derive a concrete clinical recommendation: before starting a GLP-1 therapy, genotyping the PAM variants would be cost-effective once genotype tests are established in routine diabetes care.

How the Stanford data relate to the 23andMe study

In April 2026, two genetic views of GLP-1 response appeared practically simultaneously. They measure different things and complement each other:

• The Nature 23andMe study (27,885 participants) identified, among other things, a variant in the GLP1R gene itself (p.Pro7Leu) and a GIPR variant. The GLP1R variant correlated with weight loss and appeared particularly often in European populations. The GIPR variant was linked to increased nausea under tirzepatide.
• The Stanford study, by contrast, focuses on the glycemic effect (HbA1c) and on a different gene (PAM), which affects an upstream enzyme.

Reading the two data sources together reveals two different biological layers of GLP-1 pharmacogenetics: a ligand processing layer (PAM) and a receptor layer (GLP1R, GIPR). Both can independently lead to a poor response. Clinically, this means a "non-responder" can have a PAM problem, a GLP1R problem, or both.

What it means for GLP-1 users who do not respond

Anyone who, after three months of semaglutide or liraglutide with good dose titration and lifestyle discipline, sees less than 3 to 5 percent weight loss or only a minimal HbA1c reduction belongs statistically to a relevant subgroup. The Stanford data shift the view of these users away from "doing something wrong" toward "possibly having a biological predisposition".

In practice this means:

• Realistic expectation management. Study medians (around 10 to 15 percent weight loss with semaglutide 2.4 mg) are averages across a very heterogeneous population. The spread is large. A known proportion below the median is biologically expected, not only motivation-related.
• Patience during the ramp-up phase. Most clinical recommendations call for a minimum of 12 to 16 weeks on target dose before talking about non-response. Anyone who has lost 1 kg after four weeks is not necessarily a non-responder but has simply not yet exhausted the effective range.
• Genotyping is currently not standard. Outside specialized diabetes centers, PAM is not routinely tested. Consumer gene tests (23andMe, Nebula) provide raw data in which the Stanford variants can sometimes be identified, but their interpretation is difficult without clinical support.

Options for non-responders

Stanford itself phrases things cautiously because the translation into a clear therapeutic decision is still pending. From the available evidence, however, three practical paths can be derived:

1. Check the dose, but do not escalate endlessly

If you are still below target dose, get there first. For semaglutide that means 2.4 mg weekly for Wegovy or 2.0 mg for Ozempic, for liraglutide 3.0 mg daily. Anyone at that level who does not respond usually gains little from further escalation to "off-label" doses. PAM resistance cannot be compensated for with more ligand.

2. Switching to a dual or triple agonist

This is where the Stanford study becomes strategically interesting. If PAM resistance primarily affects the GLP-1 axis, substances that also target GIP or glucagon receptors could compensate for part of the gap:

• Tirzepatide (GLP-1 plus GIP) acts via two receptors. A purely PAM-related GLP-1 dampening should not affect the GIP portion to the same extent. Details in the semaglutide versus tirzepatide comparison.
• Retatrutide (GLP-1 plus GIP plus glucagon) adds a third axis. The data here are still thinner, but the mechanism of action suggests that PAM carriers could benefit relatively more from the glucagon component than from a pure GLP-1.

For more background on the three substances, see the GLP-1 agonist comparison for a structured overview.

3. Critically reconsider microdosing

Microdosing protocols, popularized in part by Tyna Moore and described in our Microdosing Protocol Guide 2026, aim to achieve metabolic benefits with fractions of the therapeutic dose while minimizing side effects. For a PAM-resistant person, two sober questions arise:

• If even the full therapeutic dose only arrives in dampened form, a microdose in the range of 0.1 to 0.25 mg weekly may be too weak to even reach a biological threshold.
• Conversely, this does not mean that "dose higher" is the solution. Above target dose, the rate of side effects rises rather than efficacy linearly.

Realistically: microdosing is more for users who respond strongly to normal doses and want to reduce side effects, and less of a strategy for non-responders. With suspected PAM resistance, a switch of substance or supplementation of other pathways (exercise, sleep, insulin sensitizers like metformin) is likely more productive than micromanaging the dose.

A practical path for suspected non-responders

A pragmatic sequence emerges from the study data and clinical practice:

1. Define realistic expectations. Not all patients reach 15% weight loss. 5 to 10% is biologically a success, even if social media suggests other numbers.
2. At least 12 weeks on target dose. Before this threshold, the question of non-response versus slow response cannot be answered cleanly.
3. Verify the dose. Has the target dose actually been reached and consistently taken? Compliance gaps are more common than genetic resistance.
4. Check accompanying factors. Caloric balance, protein intake, sleep, stress, thyroid, cortisol status. GLP-1 does not replace metabolic hygiene.
5. Discuss a switch of substance. With the treating physician: switch to a dual agonist (tirzepatide), possibly with the PAM hypothesis as background.
6. Only then optional genotyping. In specialized centers or through validated consumer tests, always with medical interpretation.

A word on hair loss, nausea, and gastrointestinal side effects frequently reported by GLP-1 users: in the 23andMe study, these correlate more with GLP1R/GIPR variants than with PAM. So anyone suffering heavily from side effects probably belongs to a different genetic subgroup than a classic PAM non-responder. Background in the article Hair loss under GLP-1 and in the Reddit AI analysis of side effects.

Limitations of the Stanford study

As with any pharmacogenetic work, it is worth reading the methodology critically. Users and researchers should keep the following limitations in mind:

• Sample size. 1,119 participants in the meta-analysis is solid, but not large enough to reliably capture rare interactions or subgroup effects (for example ethnic stratification).
• HbA1c endpoint, not weight. The Stanford work primarily measures the glycemic effect. The extent to which PAM variants also influence the weight loss effect is plausible but not directly demonstrated.
• Incomplete mechanism. That amidation is reduced has been shown. Which exact peptide signals (GLP-1 alone? GLP-2? glucagon? others?) are affected and how strongly remains open.
• European population dominant. Allele frequencies differ by ancestry. Transferring the findings to non-European patient groups is possible but needs its own data.
• No prospective therapy decision. There is no randomized study in which patients were randomized to different substances based on their PAM genotype. The recommendation "PAM carriers should rather receive tirzepatide" is hypothesis-driven, not evidence-based.
• More complex than a single gene. Clinical response to GLP-1 is polygenic and multifactorial. PAM explains part of the spread, not the whole.

For the next wave of pharmacogenetic studies, it will be interesting whether a combined risk score from PAM, GLP1R, GIPR, and other loci can significantly improve the prediction of individual response. Initial models in the works linked above suggest this is the case.

What this means for PeptidesDirect customers in a research context

Anyone working with GLP-1 analogs in a research setting should read the Stanford data as a reminder that biological responses to incretin mimetics are not an average but a distribution with real outliers on both sides. A single negative self-experiment is no proof of ineffectiveness of the substance class, just as a single success anecdote post on Reddit is no proof of special efficacy.

Concretely useful for the interested reader:

• Scientific background on the substance class: Semaglutide Science 2026 and Tirzepatide Science 2026.
• Direct comparison of the three substances: Retatrutide versus Tirzepatide versus Semaglutide.
• Genetic context from the Nature study: GLP-1 Genetics 23andMe.

Frequently asked questions

Summary

The Stanford study of April 10, 2026 in Genome Medicine provides the clearest genetic explanation to date for why about 10 percent of the population respond significantly more weakly to GLP-1 therapies than expected. Responsible are two missense variants in the PAM gene (p.S539W and p.D563G) that impair the amidation of peptide hormones. The consequence: elevated GLP-1 levels, but biologically dampened signal. In the meta-analysis, carriers of the more strongly affected variant only reached their HbA1c target in about 11.5 percent of cases after six months, compared with 25 percent in wild-type carriers.

For users with a weak response, this means: not necessarily "dose higher", but consider switching to dual (tirzepatide) or triple agonists (retatrutide). Microdosing is probably not a sensible strategy for this subgroup. Genotyping is not yet clinical routine but could become part of personalized diabetes therapy in the coming years.

Together with the simultaneously published 23andMe Nature study, it is becoming clear that GLP-1 response is polygenic, layered, and highly variable from individual to individual. The medians of registration trials describe averages, not guarantees.

Sources

• Umapathysivam, M., Araldi, E., et al. (2026): Genetic variants in PAM modify response to GLP-1 receptor agonists. Genome Medicine, April 10, 2026. DOI: 10.1186/s13073-026-01630-0
• Stanford Medicine News, One in 10 people may have resistance to GLP-1 diabetes drugs
• Stanford Report, GLP-1 resistance: diabetes patients research
• ScienceDaily, April 11, 2026
• Nature, Genetic predictors of GLP-1 receptor agonist weight loss and side effects (23andMe GWAS, 27,885 participants)

---

This article reflects the state of knowledge as of May 11, 2026. Pharmacogenetic research evolves quickly. Clinical recommendations may change with new prospective data. Consult a qualified physician for therapy decisions.

All substances offered by PeptidesDirect are exclusively intended for laboratory and research purposes. They are not intended for human consumption, self-medication, or therapeutic use.