An evaluation of single-nucleus RNA sequencing data across four cohorts has found that microglial ferro-aging is not significantly elevated in Alzheimer’s disease when statistical analyses treat donors rather than individual nuclei as the unit of inference.

Both iron accumulation and cellular senescence represent recognized hallmarks of Alzheimer’s disease, yet their proposed intersection—termed ferro-aging—has remained uncharacterized at single-cell resolution. Researchers Da Li and Honggang Zhang set out to test whether microglial ferro-aging is genuinely increased in Alzheimer’s disease by applying a 128-gene ferro-aging set across transcriptomic datasets while adjusting the unit of inference to account for biological donor origins rather than isolated nuclei.

The study analyzed single-nucleus transcriptomic profiles from 123 donors, comprising 91 individuals with Alzheimer’s disease and 32 control donors, gathered across four datasets: GSE174367, GSE264648, GSE222494, and GSE243678 from the SEA-AD cohort. Meta-analysis was performed using a DerSimonian–Laird random-effects model combined with Hartung–Knapp–Sidik–Jonkman intervals to pool outcomes across the four independent study cohorts.

When pooled across all cohorts, the difference in microglial ferro-aging produced a Hedges g of 0.23, with a 95 percent confidence interval from -0.24 to 0.70 (p = 0.35; I² = 11.6 percent) and a Hartung–Knapp–Sidik–Jonkman interval of -0.54 to 1.00 (p = 0.42). None of the four cohorts showed a statistically significant difference individually or during leave-one-out re-analysis, and one cohort exhibited a negative point estimate of g = -0.38. While microglia showed the largest point estimate among seven evaluated cell types in the discovery cohort at g = 0.69, none of the cell types reached statistical significance.

A donor-level re-testing of a 28-gene immune checkpoint panel similarly yielded no statistically significant genes, resulting in a minimum false discovery rate of 0.64. Specific checkpoints showed broad confidence intervals, including VSIR with a Hedges g of 0.04 (95 percent confidence interval -0.62 to 0.70) and HAVCR2 with a Hedges g of -0.44 (95 percent confidence interval -1.25 to 0.37). Donor-level re-testing withdrew the reported VISTA and TIM-3 downregulation, with neither supported at the donor level, while wide intervals leave the panel unresolved rather than unchanged.

Diagnostic machine learning models weakly discriminated Alzheimer’s disease cases under nested cross-validation, achieving an optimal area under the curve of 0.662 (95 percent confidence interval 0.36 to 0.93), indicating that an earlier reported area under the curve of 0.833 derived from full-dataset feature selection could not be retained. In parallel, microglial pseudotime trajectory analysis resolved two cellular poles—an iron-export-competent low-ferritin state and an iron-retention state—but revealed no significant difference between Alzheimer’s disease and control donors along the pseudotime trajectory (Hedges g = -0.66, 95 percent confidence interval -1.63 to 0.31; p = 0.21).

Secondary exploratory observations identified a 3.5-fold increase in microglial SPP1 signaling to astrocytes at the nucleus level, though this interaction was not donor-tested, alongside attenuated transferrin to transferrin receptor uptake in oligodendrocytes. In silico molecular docking simulations prioritized liproxstatin-1, showing binding affinities of -7.90 kcal/mol for ALOX5 and -6.08 kcal/mol for GPX4, as well as minocycline at -7.52 kcal/mol. The authors concluded that microglial ferro-aging remains an unverified hypothesis rather than an established pathological mechanism, offering a corrected donor-level analytical framework accompanied by covariate tables to make residual confounding explicit in single-nucleus neurodegeneration research.

Clinical trial registration was not applicable to the study, and registration numbers cited in the text refer to trials conducted by other investigators, documented alongside their first registration dates on ClinicalTrials.gov: study NCT03234686, registered on July 31, 2017, and study NCT04063124, registered on August 21, 2019.

Subject of Research: Donor-level single-nucleus RNA sequencing assessment of microglial ferro-aging in Alzheimer’s disease

Article Title: Reassessing microglial ferro-aging in Alzheimer’s disease: unit-of-inference correction across four single-nucleus cohorts

Article References: Li, D., & Zhang, H. (2026). Reassessing microglial ferro-aging in Alzheimer’s disease: unit-of-inference correction across four single-nucleus cohorts. BMC Neurology. https://doi.org/10.1186/s12883-026-05550-9

Image Credits: AI Generated

DOI: 10.1186/s12883-026-05550-9

Keywords: microglial ferro-aging, Alzheimer's disease, single-nucleus RNA-seq, unit of inference, cellular senescence, iron accumulation, random-effects meta-analysis, Hartung-Knapp-Sidik-Jonkman, immune checkpoints, pseudotime trajectory, VSIR, HAVCR2, SPP1 signaling, liproxstatin-1, minocycline, SEA-AD cohort, nested cross-validation, molecular docking