# The Genetic Test That Isn't Ready

By Nora Bissett, Pricing Editor (https://glpwatchdog.org/authors/nora-bissett) | Published: 2026-08-20 | Updated: 2026-10-27

Canonical page: https://glpwatchdog.org/research/the-genetic-test-that-isnt-ready

> A paper titled 'GLP1R and OCT1 variants modulate semaglutide response' reports in its own results that the GLP1R signal did not survive correction, in an arm of ten. Carries a correction: larger work has since found real predictors.

## Key points

- No variant survived correction for multiple comparisons.
- Groups of 10–16 are far too small to detect an ordinary effect.
- So it is not a refutation either. An underpowered null is not evidence of absence.
- CORRECTED: a later Nature GWAS did find real predictors, with effect sizes too small to act on.

Tests promising to identify your best GLP-1 are being sold. A 2026 study in Pharmacogenetics and Genomics is titled *GLP1R and OCT1 variants modulate semaglutide and metformin response in type 2 diabetes*, yet its own results report that the GLP1R signal in semaglutide users **did not survive correction for multiple comparisons**, in an arm of ten people.[1] The distance between that title and those results is the whole subject of this page.

> **Correction, August 2026.** This article originally concluded that nobody had shown a genetic test predicts GLP-1 response. That is no longer accurate. A genome-wide study of 27,885 people, published in Nature, has since found four variants associated with weight loss and side effects at genome-wide significance. We have corrected the conclusion below. The correct position is now a distinction rather than a denial: **the associations are real; the effect sizes are too small to build a useful test on.** The figures are in [the genetics are real, and small](https://glpwatchdog.org/research/the-genetics-are-real-and-small). Everything this page says about the pilot study below remains accurate.

## What the study did

Twenty-seven Bulgarian adults with type 2 diabetes and a BMI of 25 or above, averaging an HbA1c of 8.3%, were given either metformin XR 2000 mg (17 people) or oral semaglutide 14 mg (10 people) and followed for three months. Sanger sequencing genotyped three polymorphisms, and the endpoints were change in weight and change in HbA1c.

The gene list is sensible. *GLP1R* encodes the receptor semaglutide binds. *SLC22A1* (also called *OCT1*) and *SLC47A1* encode transporters relevant to metformin. Large consortia have linked variants in all three to antidiabetic response, and the stated purpose was to see whether those effects show up in a small real-world clinic.

## What it found

The clearest result has nothing to do with genetics. Semaglutide beat [metformin for weight loss](https://glpwatchdog.org/research/metformin-for-weight-loss) over three months: **−6.5 ± 3.6 kg against −1.6 ± 2.5 kg** (95% CI −7.6 to −2.2; P = 0.001), with BMI down 2.0 against 0.3 kg/m². That is a drug comparison, and it is the finding in this paper with the firmest support.

*The genetic findings, as the paper reports them. FDR is the false-discovery-rate correction; q is the corrected value.[1]*

| Variant | Finding | Statistics | Survived correction? |
| --- | --- | --- | --- |
| *GLP1R* rs6923761 (semaglutide users) | Nominal trends for weight and BMI change | P ≈ 0.06–0.07; q ≈ 0.29; N = 10 | **No** |
| *OCT1* rs34130495 (metformin users) | HDL cholesterol change, +0.340 mmol/L per minor allele | P = 0.0026; q = 0.063; N = 16 | Only at an *exploratory* 10% FDR |

Read the first row against the title. The variant in the receptor these drugs bind produced trends that did not reach significance after correction, in ten people. Read the second row and note two things: the association that cleared the bar is about **HDL cholesterol in metformin users** (not weight, not semaglutide), and it cleared a bar the authors describe as exploratory.

## What an exploratory 10% FDR means

When you test many things at once, some associations appear by chance. False-discovery-rate correction accounts for that by asking what proportion of your declared findings you are willing to have be false. The conventional setting is 5%. This analysis used 10%, which means up to one in ten declared hits is expected to be spurious. The OCT1 result came in at q = 0.063, which passes at 10% and would not pass at 5%.

> A finding that clears a loosened bar by a small margin is a reason to run a bigger study, which is exactly what the authors say.

To their credit they say it plainly: the HDL signal *warrants replication*, and the data are *hypothesis-generating*. Nothing in the paper oversells the result. The overselling risk sits in the title, and titles are what circulate.

> **One inconsistency, reported without accusation.** The methods name *SLC22A1* rs628031 among the three genotyped polymorphisms; the results report *OCT1* rs34130495. OCT1 *is* SLC22A1, so those are two different variants of the same gene. We cannot tell from the published abstract which is intended, and it bears on how the replication the authors ask for should be designed.

## Why this is not a refutation either

> **An underpowered null is not evidence of absence, and we are not going to use it as one.** Ten people cannot detect an effect unless it is enormous. A real, ordinary-sized association between *GLP1R* variants and semaglutide response would be invisible in this study. “This did not find it” and “it is not there” are different statements, and only the first is available here.

So the claim this *particular study* supports is narrow: it did not detect an association, in an arm of ten. The authors were right that the consortium data were a reason to keep looking, and a much larger study has since looked and found something. What it found is real and small: variants that shift expected weight loss by under a kilogram per copy, and side-effect models that do not reach the usual threshold for a clinical tool. **Real associations, not yet a useful test.**

## What to ask someone selling you a test

A test marketed as identifying your optimal GLP-1 is still making a claim that runs ahead of the evidence, though for a different reason than this pilot suggested. The associations exist. What does not yet exist is an effect large enough to change a decision. A variant that shifts expected weight loss by under a kilogram cannot tell you which drug to take.

- **Which published study supports the specific recommendation** being made to me (not the gene, the recommendation).
- **How many people were in it**, and how many were on the drug I am asking about.
- **Did the finding survive correction for multiple comparisons**, and at what threshold?
- **Was it replicated** in a separate group of people?

If the answer is a mechanism story rather than an outcome study, that is the answer. What does predict response, as far as anyone has shown, is early response itself. The first weeks on the drug carry real information about the year ahead, covered in [losing fast early](https://glpwatchdog.org/research/losing-fast-early-what-it-predicts). It is a cruder instrument than a gene panel, and it is the one with evidence behind it.

## Frequently asked questions

### Can a genetic test tell me which GLP-1 will work best for me?

Not usefully, on current evidence. A large genome-wide study has since found real associations at overwhelming significance, but the efficacy variant shifts expected weight loss by roughly 0.76 kg per copy, and the side-effect models fall below the usual threshold for a clinical tool. In the small pilot discussed here, nothing survived correction at all.

### The paper's title says these variants modulate response. Does it not?

The title says that; the results are narrower. The only association that cleared even an exploratory 10% false-discovery threshold was between an OCT1 variant and HDL cholesterol in metformin users, not weight and not semaglutide.

### Does this mean genetics are irrelevant to GLP-1 response?

No. Larger work since has shown they are not irrelevant. Ten people in this pilot's semaglutide arm could only have detected a very large effect. A subsequent genome-wide study found real associations whose effect sizes are nonetheless small.

### What does 'did not survive FDR' mean?

Testing many variants at once means some will look associated by chance. False-discovery-rate correction adjusts for how many questions were asked. The GLP1R weight trends had raw P values around 0.06 to 0.07; after correction they were around 0.29.

### Did the study find anything solid?

Yes, but not about genetics. Semaglutide produced substantially more weight loss than metformin over three months (6.5 kg against 1.6 kg), which is a drug comparison rather than a pharmacogenomic one.

## References

1. Tourtourikov I, Kalinkova M, Ivanov P, Mileva-Popova R, Tafradjiiska-Hadjiolova R, Handjieva-Darlenska T, Kadiyska T. GLP1R and OCT1 variants modulate semaglutide and metformin response in type 2 diabetes. Pharmacogenetics and Genomics. 2026. PMID: 40996853. https://pubmed.ncbi.nlm.nih.gov/40996853/

---

Cite as: GLP Watchdog, Nora Bissett, "The Genetic Test That Isn't Ready", published 2026-08-20, updated 2026-10-27. https://glpwatchdog.org/research/the-genetic-test-that-isnt-ready

Canonical HTML page: https://glpwatchdog.org/research/the-genetic-test-that-isnt-ready

Medical disclaimer: Nothing here is medical advice. We report on this subject; we do not treat anyone. Whether to begin a medication, supplement, or treatment, whether to adjust it, and whether to stop are questions for a clinician licensed to answer them, who knows your history.
