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

Iron accumulation and cellular senescence are hallmarks of Alzheimer’s disease (AD), yet their intersection — ferro-aging — remains uncharacterized at single-cell resolution. Our objective was to test whether microglial ferro-aging is elevated in AD. We applied a 128-gene ferro-aging set to single-nucleus RNA-seq from 123 donors (91 AD, 32 control) across four cohorts (GSE174367, GSE264648, GSE222494, GSE243678/SEA-AD), treating donors — not nuclei — as the unit of inference. Pooling used DerSimonian–Laird random-effects meta-analysis with Hartung–Knapp–Sidik–Jonkman intervals. Secondary analyses covered immune checkpoints, microglial pseudotime (a transcriptional ordering along a continuum) and docking. Pooled across cohorts, the microglial ferro-aging difference was Hedges g = 0.23 (95% CI − 0.24 to 0.70; p = 0.35; I² = 11.6%; HKSJ − 0.54 to 1.00, p = 0.42). No cohort was significant individually or in leave-one-out re-analysis, one was negative (g = − 0.38). In the discovery cohort, microglia gave the largest of seven point estimates (g = 0.69), but none was significant. Donor-level re-testing of the 28-gene checkpoint panel yielded no significant gene (VSIR g = 0.04 [− 0.62, 0.70]; HAVCR2 g = − 0.44 [− 1.25, 0.37]; minimum FDR = 0.64), but wide intervals leave the panel unresolved rather than unchanged. Diagnostic models discriminated AD weakly under nested cross-validation (best AUC 0.662, 95% CI 0.36–0.93); an earlier AUC of 0.833 (full-dataset feature selection) is not retained. Pseudotime resolved two poles — iron-export-competent low-ferritin and iron-retention — with no AD–control difference along it (g = − 0.66, 95% CI − 1.63 to 0.31; p = 0.21). Microglial SPP1 signalling to astrocytes rose 3.5-fold (nucleus level, not donor-tested); oligodendrocyte TF→TFRC uptake attenuated. Docking prioritised liproxstatin-1 (ALOX5 − 7.90; GPX4 − 6.08 kcal/mol) and minocycline (− 7.52). At the donor level we did not establish an AD-associated elevation of microglial ferro-aging; the previously reported cell-type-specific elevation was not reproducible with donors as the unit of inference. Ferro-aging remains a hypothesis, not an established mechanism. Donor-level re-testing withdrew the reported VISTA and TIM-3 downregulation; neither is supported at the donor level. The contribution is a corrected donor-level framework plus a covariate table making residual confounding explicit. Not applicable. The registration numbers cited in the text refer to trials conducted by other investigators and are given with their ClinicalTrials.gov dates of first registration: NCT03234686 (registered 31 July 2017) and NCT04063124 (registered 21 August 2019).