Novo Nordisk built a trial to find out how semaglutide protects the kidneys, and all three of its headline measurements came back flat. The answer showed up in the backup measurements instead, in the blood vessels and the cells that line them.
The trial is REMODEL, published Thursday, October 1, in Nature Medicine ↗. It randomized 106 adults with type 2 diabetes and chronic kidney disease, 71 to a weekly 1 mg injection of semaglutide and 35 to placebo, and followed them for 52 weeks. The point was not to ask whether the drug helps kidneys. A large outcomes trial already answered that. The point was to ask why.
A mechanism trial that misses its primary endpoints
Most trials ask whether a drug works. A mechanism-of-action trial assumes it works and asks where the effect lives in the body. REMODEL picked three places to look, using repeated kidney MRI scans: oxygen levels in the kidney tissue, blood flow through the whole kidney, and a scan signal that tracks inflammation. Those were the three coprimary endpoints, the measures the trial was designed around.
None of them moved. After 52 weeks, semaglutide and placebo looked the same on kidney oxygenation, on perfusion, and on the inflammation signal. For a trial built on those three measures, that is a miss on everything it set out to prove.
A mechanism trial that misses its primary endpoints is not the same thing as a failed drug. It is a map with the X drawn in the wrong spot. The drug still protects kidneys. The three things the researchers bet the protection would show up in were the wrong three things.
Where the signal actually landed
The secondary measures tell a different story, and they point at the plumbing rather than the tissue.
The renal artery resistive index dropped. That index is a measure of how hard the kidney's arteries resist blood flowing through them, and a lower number means less stiffness and less resistance. Semaglutide pushed it down by a small but statistically clear margin (estimated treatment ratio 0.96, 95% CI 0.93 to 0.99, P=0.008). A second MRI measure that tracks scarring held steady on the drug while drifting on placebo, which the authors read as fibrosis being held off rather than advancing.
The protein leaking into patients' urine fell about 40 percent on semaglutide, a standard marker that kidney damage is slowing. Measured filtration ran about 12 milliliters per minute higher than placebo.
Then the biopsies. The researchers took kidney tissue and looked at gene activity cell by cell. The clearest effect was on glomerular endothelial cells, the thin cells lining the tiny blood vessels where the kidney filters blood. On semaglutide, those cells turned down genes tied to metabolic stress, inflammation, and scarring, and an exploratory spatial analysis found fewer immune cells crowded around them.
Why this matters
The working assumption for years has been that GLP-1 drugs help kidneys mostly by lowering weight, blood sugar, and blood pressure, with some direct anti-inflammatory effect on top. REMODEL's flat inflammation scan complicates the tidy version of that story. The scan meant to catch inflammation caught nothing, while the vascular measures and the endothelial-cell genes carried the signal. That shifts the leading explanation toward the blood vessels, specifically the cells lining the kidney's filtration units, rather than a broad dampening of inflammation across the tissue.
This is one small trial, 106 people, funded by the drug's maker, and the vascular findings are secondary endpoints that a mechanism trial is free to generate but not powered to confirm. The next trial gets to aim at the vessels on purpose. That is what a mechanism trial is for, including when its own headline measures come up empty.
On peptidemodel, semaglutide sits on the GLP-1 receptor card ↗ under the GLP-1R target ↗. The kidney thread runs through the section already. In July an oral GLP-1 peptide from a different company cut urinary protein loss on top of the two newest kidney drugs ↗, an outcome signal without a mechanism. REMODEL is the reverse: a mechanism hunt that found its clue one layer down from where it looked.