Metformin Leaves No Robust Gene Signature in Small AMD Trial

An 18-month randomized trial sub-study found no statistically significant change in any individual gene after correction for multiple comparisons in whole-blood transcriptomes from older adults with age-related macular degeneration who received metformin. The analysis included only 15 nondiabetic participants: eight assigned to metformin and seven to observation without placebo. Although exploratory pathway analyses produced signals consistent with reduced inflammatory signaling, the researchers say the small sample, unadjusted screening thresholds and reliance on nominal associations make those findings hypothesis-generating rather than evidence that metformin altered the biology of aging or macular degeneration.

The blood analysis was embedded in the METformin FOR the MINimization of GA progression trial, known as METforMIN, which enrolled people aged 55 or older with geographic atrophy caused by nonexudative AMD. Across the full trial, 66 eligible participants were randomized at 12 US clinical sites, with 32 assigned to oral metformin and 34 to an observation group managed by their usual clinicians. Blood collection was optional, and only 15 of the 66 randomized participants consented and supplied usable samples for the reported sub-study. The investigators said logistics, including delayed collection supplies and limited time during some clinic visits, were the main reasons blood sampling was unavailable, rather than participants’ personal or medical objections.

Participants in the metformin arm received a standardized dose escalation designed to reduce gastrointestinal side effects: 500 mg once daily during the first week, 500 mg twice daily during the second week, and then 1,000 mg twice daily for a target maintenance dose of 2,000 mg per day. The control group received no placebo and participants were not masked to allocation. Five of the eight metformin-group participants reported 100% adherence, while the remaining three reported compliance between 75% and 99%. Three metformin recipients experienced self-resolving gastrointestinal distress. Blood was collected at enrollment and again at the 18-month follow-up, and laboratory personnel who handled the samples and conducted the sequencing analysis were masked to treatment assignment.

Researchers extracted total RNA from the stored blood, prepared sequencing libraries and used metagenomic sequencing on an Illumina NovaSeq 6000. Reads were aligned to the human reference genome, after which genes detected in fewer than 20% of samples were excluded. Differential expression was assessed with DESeq2, comparing metformin recipients with controls at 18 months while adjusting for baseline expression. The investigators applied a 5% false-discovery-rate threshold with Benjamini-Hochberg correction. Of approximately 1,100 genes that met an uncorrected P-value threshold of less than 0.05, none remained statistically significant after correction. That result is the study’s central finding and means the analysis did not establish a reproducible transcriptomic signature of metformin exposure at the individual-gene level.

Because the gene-level analysis was negative and the sample was too small for many individual changes to pass correction, the team then used QIAGEN’s Ingenuity Pathway Analysis as a secondary, hypothesis-generating exercise. The pathway tool draws on a curated knowledge base of mammalian genes, chemicals, microRNAs and biological processes, then assesses whether the direction of observed expression changes fits the activation or inhibition expected for known pathways and regulators. It combines overlap-based P values with a z score reflecting the consistency of expression direction; the researchers treated an absolute z score of at least 2 as significant. This approach is valuable for generating mechanistic candidates, but its dependence on nominally significant genes makes false-positive signals especially likely in small datasets.

Among the pathway signals, the metformin group showed nominal downregulation of interferon alpha/beta signaling, with a P value of 0.003 and z score of −2.4. Signals also pointed toward downregulation of cell-junction organization, regulation of epithelial-to-mesenchymal transition and autism signaling, each with an absolute z score above 2. Other trends included lower interferon gamma signaling, cGAS-STING signaling, interactions between lymphoid and non-lymphoid cells and WNT/beta-catenin signaling. The upstream-regulator analysis suggested inhibition of inflammatory molecules including STAT1, IFNL1, the interferon beta family, IFNAR, poly rI:rC-RNA, C1QA and IL1B. It also indicated possible activation of IL1RN, IRGM and DUSP11, which have established roles in limiting or resolving inflammatory signaling in other experimental and clinical contexts.

The findings are compatible with the broader scientific interest in metformin as a possible geroscience intervention, but compatibility is not confirmation. Metformin is commonly used to lower blood glucose in type 2 diabetes and has been investigated for effects that may extend beyond glycemic control, including inflammation, cellular energy metabolism, autophagy, mitochondrial function and cellular senescence. Chronic low-grade inflammation associated with aging is often termed “inflammaging,” and the study’s interferon and upstream-regulator signals suggest that metformin might influence parts of that network in people without diabetes. The present dataset, however, did not directly test whether metformin reduced inflammaging, prevented retinal degeneration, affected cancer biology or altered lifespan, and the researchers explicitly warned against treating the exploratory patterns as causal or confirmatory.

Whole blood also limits what can be inferred about the eye. Peripheral blood can provide information about systemic circulating states, but its gene-expression profile overlaps only partly with processes in the retina, retinal pigment epithelium, choroid and resident microglia. A blood signal linked to AMD-related biology therefore does not demonstrate that the same change occurred in ocular tissue. Moreover, changes in the proportions of circulating cell types can resemble shifts in gene expression when no cell has changed its own transcriptional program. The investigators did not perform formal cell-type deconvolution, measured protein abundance or metabolites, or directly quantify metformin concentrations in plasma. Self-reported adherence was high, but drug exposure was not independently confirmed, and the study did not include measurements at intermediate time points.

Other design features also constrain interpretation. The optional blood draw created a risk of selection bias, although baseline demographic and AMD-related characteristics did not differ significantly between the 15 participants who donated blood and the 51 who did not, and the expected bias would apply to both treatment arms. Lack of masking could have influenced dietary or behavioral differences between groups. The population was deliberately limited to nondiabetic older adults with geographic atrophy, so the findings cannot be assumed to generalize to people with diabetes, other age-related conditions or younger adults. The path and upstream-regulator results also depend on the pathways represented in the Ingenuity database. With only eight metformin recipients and seven controls, the analysis had limited statistical power, and the investigators did not conduct an adherence-stratified sensitivity analysis because the metformin subgroup was too small.

Despite these limitations, the study offers a methodologically useful human dataset because it follows blood expression for 18 months within a randomized trial and uses unbiased transcriptome-wide sequencing rather than testing only selected biomarkers. It also provides a useful negative result: a plausible anti-inflammatory hypothesis did not yield a gene-level signal robust enough to survive false-discovery correction. The findings require validation in larger, adequately powered cohorts, ideally with formal analysis of blood-cell composition, direct drug-level measurements, additional sampling times and integration with proteomic or metabolomic data. Pairing systemic profiles with ocular-fluid biomarkers and retinal imaging could also test whether any circulating signal corresponds to local disease biology. Until then, the pathway patterns offer testable hypotheses about inflammation, lipid metabolism and cell junctions, not demonstrated clinical benefits from metformin for AMD, aging or cancer.

Subject of Research: Whole-blood transcriptome analysis of metformin in older adults with geographic atrophy due to age-related macular degeneration

Article Title: An exploratory study of the effect of metformin on the whole blood transcriptome of nondiabetic older adults with age-related macular degeneration

Article References: Chen, X., Zhang, Y. S., Yan, D., Doan, T., Shen, L. L., Taha, A. T., Yang, D., Psaras, C., Callander, J. K., Xin, N. Y., Shivkumar, A., Fawzi, A. A., Yiu, G., Telander, D. G., Leung, L.-S. B., Flaxel, C., Newman, J. C., Keenan, J. D., & Stewart, J. M. (2026). An exploratory study of the effect of metformin on the whole blood transcriptome of nondiabetic older adults with age-related macular degeneration. GeroScience. https://doi.org/10.1007/s11357-026-02593-1

Image Credits: AI Generated

DOI: 10.1007/s11357-026-02593-1

Keywords: metformin, age-related macular degeneration, whole-blood transcriptome, randomized trial sub-study, inflammaging, gene expression, RNA sequencing, interferon signaling, geographic atrophy, geroscience, exploratory, effect