https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.71469
chatGPT:
Summary
The paper by Chatterjee et al. asks whether mitochondrial DNA copy number — mtDNAcn, measured in blood — has a causal relationship with Alzheimer’s disease, AD/dementia, and Parkinson’s disease. mtDNAcn is used here as a proxy for mitochondrial abundance or mitochondrial functional state, although the authors correctly note that it is an imperfect marker.
The authors combine several genetic approaches:
- Genetic correlation to see whether mtDNAcn and disease share genetic architecture.
- Polygenic risk scores to test whether genetic liability to higher mtDNAcn predicts AD risk in the Alzheimer’s Disease Genetics Consortium.
- Mendelian randomization, including standard univariable MR, multivariable MR adjusting for platelets, and latent heritable confounder MR, LHC-MR, to infer causal direction while trying to handle confounding and sample overlap.
They use four mtDNAcn GWAS datasets: Longchamps, Hägg, Gupta, and Chong. These differ in sample size, ancestry mix, platelet adjustment, and mtDNAcn measurement method: microarray, WGS, qPCR, or combinations of these.
The main findings are:
- mtDNAcn measures were generally negatively genetically correlated with AD, AD/dementia, and PD, especially the Longchamps dataset.
- mtDNAcn polygenic risk scores did not significantly predict AD in ADGC, although the Longchamps PRS showed a near-significant protective trend.
- Standard univariable MR and platelet-adjusted multivariable MR found no robust causal effect of mtDNAcn on AD or PD.
- LHC-MR, however, found that genetically higher mtDNAcn was associated with lower risk of AD/dementia and PD.
- Reverse-direction analysis suggested that genetic liability to AD may be associated with higher mtDNAcn, which the authors interpret as possible compensatory upregulation in response to mitochondrial dysfunction.
So the central conclusion is: higher blood-derived mtDNAcn may causally reduce risk of AD/dementia and PD, but disease liability may also drive compensatory increases in mtDNAcn.
What is novel
The novelty is not simply the idea that mitochondria matter in AD or PD. That is already well established. The novelty is methodological and comparative.
First, the paper triangulates across four separate mtDNAcn GWAS datasets, rather than relying on a single exposure dataset. This matters because mtDNAcn estimates are highly sensitive to measurement method and blood-cell composition.
Second, it explicitly tests whether study design features — especially platelet contamination, sample overlap, and measurement method — distort causal estimates.
Third, the use of LHC-MR is important. The authors show that conventional MR gives null results, whereas LHC-MR suggests protective causal effects. Their argument is that standard MR may be diluted or biased by latent confounding, while LHC-MR can model a hidden heritable confounder.
Fourth, the bidirectional analysis is useful. The paper does not simply say “low mtDNAcn causes neurodegeneration”; it suggests a more subtle model: low mitochondrial reserve may increase risk, while early disease processes may induce mtDNAcn upregulation as compensation.
Critique
The paper is interesting, but its strongest claim depends heavily on LHC-MR. Standard MR and multivariable MR are null, and PRS results for AD are also non-significant. That means the headline conclusion — “higher mtDNAcn causally reduces AD/dementia and PD risk” — is not equally supported by all methods. It is mainly supported by the more complex model.
A second issue is that blood mtDNAcn is not brain mtDNAcn. AD and PD are brain diseases, and blood-based mtDNAcn may reflect immune-cell composition, inflammation, platelet contamination, systemic metabolism, or general health rather than neuronal mitochondrial function. The authors acknowledge this, but it remains a major biological limitation.
Third, mtDNAcn is ambiguous. Higher copy number could mean healthier mitochondrial abundance, but it could also mean compensatory replication of damaged mitochondria. Conversely, lower mtDNAcn could mean mitochondrial depletion, or simply altered blood-cell composition. The paper treats mtDNAcn as a useful proxy, but it is not a direct measure of mitochondrial membrane potential, oxidative phosphorylation efficiency, mitophagy, mitochondrial mutation burden, or citrate export capacity.
Fourth, the AD outcome definitions are heterogeneous. The paper uses both clinically diagnosed AD and broader AD/dementia or proxy-based dementia datasets. The LHC-MR finding differs depending on outcome: higher Longchamps mtDNAcn was associated with increased clinically defined AD risk, but reduced AD/dementia risk. That inconsistency is important and suggests that the result may be sensitive to case definition.
Fifth, most datasets are predominantly European ancestry, limiting generalisability.
Sixth, the platelet issue is only partly solved. Platelets contain mitochondria but no nuclear DNA, so they can distort blood mtDNAcn estimates. Some GWASs adjust for cell composition better than others. The authors use multivariable MR for platelets, but blood-cell composition remains a difficult confounder.
Bottom line
This is a useful and technically sophisticated paper. It strengthens the case that mitochondrial biology is upstream of AD/dementia and PD risk, but it does not prove that simply raising mtDNA copy number would be therapeutic.
The most cautious interpretation is:
Genetic factors linked to higher blood mtDNAcn are associated with lower AD/dementia and PD risk in the authors’ most confounder-adjusted MR framework, but mtDNAcn is an indirect and biologically ambiguous marker. The findings support mitochondrial dysfunction as part of causal neurodegenerative biology, while leaving open whether mtDNAcn is a cause, compensation, biomarker, or mixture of all three.