Why Metformin Works Differently in Different Bodies: Multi-Organ Molecular Maps Point to an Answer

Metformin is the world’s most prescribed diabetes drug, yet it fails to deliver adequate blood sugar control in roughly 30 to 40 percent of patients who take it as monotherapy, and the magnitude of HbA1c reduction can vary two- to threefold from one person to the next. A new study in Frontiers in Endocrinology offers a tissue-level view of why this happens. By simultaneously profiling the transcriptome and proteome of liver, skeletal muscle, and fat in obese diabetic Mongolian gerbils treated with metformin, researchers at Capital Medical University traced the drug’s molecular fingerprints across three organs and linked them to how much each individual animal’s glucose actually improved. The result is a multi-organ framework that treats the variability in metformin response not as a mystery of a single tissue, but as the product of coordinated and sometimes opposing programs distributed across the body.

The team used a naturally obese diabetic gerbil line that develops insulin resistance, high blood sugar, and fatty liver without any dietary manipulation, providing a genetically coherent backdrop for studying drug response. Seven male animals were randomly assigned to receive either metformin at 100 milligrams per kilogram per day by oral gavage for five weeks, a dose the authors calculate corresponds to roughly 810 milligrams per day in a 60-kilogram adult, or a water vehicle. After the treatment period, metformin-treated animals showed significantly lower two-hour post-load glucose than controls, at 5.03 versus 6.90 millimoles per liter, and histological analysis revealed a striking 77.9 percent reduction in hepatic lipid droplet area. Crucially, body weight changes were nearly identical between the two groups, indicating that the glycemic improvement was independent of weight loss.

The liver emerged as the tissue with the clearest and most coherent molecular signature. Integrated analysis of RNA sequencing and data-independent acquisition mass spectrometry identified thirteen genes whose changes were concordant at both the transcript and protein levels, converging on two functional modules. The first comprised mitochondrial solute carriers of the Slc25 family, including Slc25a47, Slc25a29, and Slc25a30. The second comprised lipid-handling factors such as PLIN2, APOA5, APOC2, and LCAT. Quantitative PCR and Western blotting confirmed the downregulation of key candidates, including reduced Slc25a47 protein in treated livers. Because Slc25a47 encodes a mitochondrial NAD+ transporter whose suppression is known to activate AMPK, the cell’s central energy sensor, the authors propose that dampening these carriers could slow the mitochondrial fuel supply that supports gluconeogenesis and lipogenesis.

Notably, the principal component of hepatic gene expression correlated almost perfectly with endpoint glycemic burden, with a Spearman correlation of negative 0.964 against two-hour post-load glucose, a relationship that held up under bootstrap confidence intervals and leave-one-out analyses. Intriguingly, this hepatic signature tracked disease severity at the endpoint rather than the magnitude of improvement, suggesting the liver functions as a central integrator of metformin’s systemic effect. Benchmarking against two public metformin-treated datasets, one from human hepatoma cells and one from mouse liver, showed that 28 of 37 evaluable genes, or 76 percent, changed in the same direction, a statistically significant concordance despite vast differences in species, dose, and experimental design.

Skeletal muscle told a different story. At the level of individual genes, the overlap between RNA and protein changes was minimal, with only the lipid export gene Apob meeting strict dual-layer criteria out of 5,882 matched gene products. But at the pathway level, the picture sharpened considerably: thirteen metabolic pathways were enriched in the transcriptome with consistent proteomic trends, including PPAR signaling, bile secretion, retinol metabolism, and cytochrome P450 drug metabolism. The direction of these changes was the mirror image of the liver, where PPAR signaling was suppressed. When the researchers scored each animal’s muscle PPAR pathway activity by combining RNA and protein data, they found that treated animals with stronger PPAR activation numerically showed the largest glucose reductions, an exploratory correlation with a perfect rank ordering among the four treated animals.

Adipose tissue provided perhaps the most provocative finding. Unlike the other two tissues, the inguinal white adipose tissue response was dominated by protein changes rather than transcripts, with 178 differentially abundant proteins but only 22 differentially expressed genes, indicating a largely post-transcriptional response. Transcriptomic enrichment pointed squarely at immune programs: chemokine signaling, Th17-cell differentiation, and cytokine receptor interactions. When the team computed an integrated chemokine module score for each treated animal, higher activity correlated with less glycemic improvement, or in other words, more residual hyperglycemia. Quantitative PCR confirmed upregulation of Ccr7, Ccl19, Cd40, and Cd40lg in treated fat tissue. This pattern is compatible with the idea that residual adipose inflammation acts as a brake on metformin’s therapeutic benefits, though the authors are careful to note it is equally consistent with the drug recruiting anti-inflammatory or regulatory immune subsets, and their data cannot distinguish between these possibilities.

The gut microbiome, by contrast, was a notable null result. Despite intense interest in intestinal microbes as mediators of metformin action, metagenomic sequencing of fecal samples revealed no significant changes in alpha diversity, beta diversity, or community structure after five weeks of treatment. No taxon of established metabolic relevance differed between groups. The authors stress this negative finding comes from a single endpoint sample with a small cohort and should not be taken as evidence that host tissue responses dominate the drug’s early glycemic effects.

Multi-tissue correlation analysis added a layer of coordination to the story. Across all seven animals, the hepatic candidate-gene module was tightly coupled between its RNA and protein layers, with a correlation of 0.86. At the protein level, the hepatic module was inversely correlated with the muscle PPAR module, with a correlation of negative 0.75 and a P value of 0.052, reinforcing the liver-down, muscle-up metabolic mirror. The muscle PPAR module was similarly inversely associated with the adipose chemokine module. These correlations are explicitly exploratory given the sample size, but they sketch a system in which metformin redistributes metabolic and inflammatory signals across organs rather than acting on any single tissue in isolation.

The study’s limitations are considerable and openly acknowledged. The cohort was small, with only four treated and three control animals, and response associations were estimated within just four treated animals, where the smallest achievable exact P value is 0.083. Protein-level testing used nominal P values without multiple-testing correction, making those findings hypothesis-generating by design. Overall RNA-protein fold-change correlations were weak across all tissues, ranging from 0.037 in muscle to 0.161 in liver, restricting cross-layer integration to candidate core genes and pathway modules. The single-endpoint design cannot distinguish primary drug actions from secondary adaptations, only male animals were studied, and fasting insulin was not measured. The drug was also prepared from commercial tablets containing trace excipients not matched in the vehicle control.

Still, the framework is a template for how to study drug response variation. By anchoring matched transcriptomic and proteomic signatures to individual glycemic outcomes in the same animals, the approach distinguishes treatment-associated changes from response-associated variation, a distinction cross-sectional human studies struggle to make. The organ-specific module scores, particularly the hepatic Slc25a-lipid axis and the adipose chemokine signature, are candidate features that the authors argue warrant validation in independent cohorts before any consideration as predictive biomarkers. If larger studies, ideally including both sexes and eventually human biopsy or liquid biopsy data, confirm these signatures, they could inform a future in which patients are stratified by their likely metformin response, or in which those with strong adipose inflammatory signatures receive adjunctive anti-inflammatory strategies. For now, the gerbil multi-omics map shows that metformin’s uneven efficacy is written in the coordinated molecular language of liver, muscle, and fat, with each tissue casting its own vote on how well the drug works.

Subject of Research: Organ-specific transcriptomic and proteomic signatures associated with inter-animal variation in glycemic response to metformin in obese diabetic Mongolian gerbils

Article Title: Tissue-resolved transcriptomic and proteomic profiling reveals organ-specific molecular signatures associated with inter-animal variation in glycemic response to metformin in obese diabetic Mongolian gerbils

Article References: Gao, Z., Fang, H., Li, F., Wang, M., Zhang, Z., Huo, X., & Du, X. (2026). Tissue-resolved transcriptomic and proteomic profiling reveals organ-specific molecular signatures associated with inter-animal variation in glycemic response to metformin in obese diabetic Mongolian gerbils. Frontiers in Endocrinology, 17, Article 1954533. https://doi.org/10.3389/fendo.2026.1954533

Image Credits: AI Generated

DOI: 10.3389/fendo.2026.1954533

Keywords: metformin, type 2 diabetes, multi-omics, transcriptomics, proteomics, liver, skeletal muscle, adipose tissue, Mongolian gerbil, PPAR signaling, mitochondrial carriers, glycemic response