Three Methods, One Mitochondrial Protein: A New Molecular Lead in Vascular Dementia

A Mendelian randomization screen, human brain gene expression data and a rat model of chronic reduced blood flow all point to the same mitochondrial protein, COX5B, as altered in vascular dementia, researchers report, while cautioning that three other candidate proteins did not hold up consistently across the three methods.

A study combining genetic causal inference, human brain tissue gene expression and a rat model of chronic cerebral blood flow restriction has converged on one mitochondrial protein, COX5B, as a consistent signal in vascular dementia, according to a paper published in Frontiers in Neurology on July 9, 2026. The researchers, based in China, screened 66 mitochondrial-related plasma proteins using a two-sample Mendelian randomization approach, a genetic method that uses naturally occurring gene variants to probe whether a protein's level might causally relate to disease risk, with plasma protein data from a Nature Genetics meta-analysis and vascular dementia data from a separate genome-wide association study.

The genetic screen flagged four proteins at a nominal significance threshold (p less than 0.05): AIFM1 and COX5B appeared protective, while NDUFV2 and NUDT5 appeared to raise risk. None of the four, nor a fifth candidate, COA3, survived statistical correction for testing 66 proteins at once; the strongest adjusted result, for AIFM1, still fell short of the standard threshold for significance. The authors then checked whether gene activity for these four candidates moved in the same direction in human brain tissue, using a public gene expression dataset (GSE122063) comparing people with vascular dementia to those without. COX5B was significantly reduced in vascular dementia brain tissue, matching the direction predicted by the genetic analysis.

To test whether these patterns reflect actual protein changes in brain tissue exposed to reduced blood flow, the team used rats with both carotid arteries surgically narrowed, a long-established model of chronic cerebral hypoperfusion meant to mimic the sustained blood flow reduction thought to drive vascular dementia. In a final group of 30 rats, split evenly among untreated controls, hypoperfusion model animals, and hypoperfusion animals given donepezil, a drug already used for dementia symptoms, hippocampal tissue measured by Western blot showed the reduced COX5B pattern held up, matching both the genetic and human tissue data. AIFM1 and NDUFV2, by contrast, moved in the opposite direction from what the genetics and transcriptomics had predicted, a discrepancy the authors describe plainly rather than explain away. NUDT5 showed partial agreement, supported by the genetic and animal data but not by the human brain tissue analysis.

Donepezil improved the rats' performance on a maze test of spatial memory and reduced visible hippocampal cell damage from the hypoperfusion surgery, but it did not change the expression levels of any of the four candidate proteins, the authors report, suggesting whatever benefit the drug provided in this model ran through a pathway separate from these particular mitochondrial proteins.

The authors call the convergence on COX5B across the three layers of evidence a "high-confidence" result for that one protein specifically, while flagging AIFM1, NDUFV2 and NUDT5 as unresolved. They frame the finding as a starting point for identifying mitochondrial targets in vascular dementia, not a validated biomarker or treatment target; no human clinical testing of COX5B measurement or intervention was part of this study.