Background
Mitochondria are complex multi-functional organelles involved in various pathways including fatty acid and cholesterol synthesis, apoptosis, calcium signaling, and adenosine triphosphate generation [
1,
2]. Dysfunctional mitochondria have been described in aging [
3] and in many neurodegenerative diseases such as Alzheimer’s disease (AD) and amyotrophic lateral sclerosis (ALS) [
4]. Mitochondria harbor their own circular genome of 16,569 base pairs encoding 13 proteins of the respiratory chain. Mitochondrial DNA (mtDNA) can be replicated independent of the cell cycle. Since mtDNA expression is required for respiratory activity, the mtDNA copy number (mtDNAcn) within a cell is regulated to meet the cell’s metabolic needs, resulting in a wide range of mtDNAcn in different tissues and conditions [
5,
6]. Like nuclear DNA, mtDNA can also carry mutations, which either affect all copies of the mtDNA in a cell (termed homoplasmy) or only a fraction of the mtDNA molecules (termed heteroplasmy). Heteroplasmic mutations are assumed to be somatically generated or inherited as low level variants, and they can clonally expand over an individual’s life-time [
7].
The mtDNAcn has become a popular potential marker of mitochondrial health in translational studies, because mtDNAcn can be measured in stored biospecimens at large scale using qPCR or DNA sequencing techniques. In AD, several studies have investigated mtDNAcn in tissue homogenates from different brain regions and found either a lower mtDNAcn in AD or no significant changes. One of the strongest reductions (50%) was reported by an early study of the frontal cortex [
8]. Smaller effect sizes or non-significant changes were reported for the hippocampus, cerebellar cortex, and cerebellum, indicating the possibility of brain region-specific effects [
9‐
11]. Similar results were reported for other neurodegenerative diseases [
5]. Interestingly, although mtDNAcn derived from whole blood is often confounded by variation in cell type composition between individuals, a recent study found an association with AD suggesting that the mtDNAcn in whole blood could potentially reflect metabolic health across tissues [
12]. While the mtDNAcn overall seems to be reduced in brain regions affected by neurodegenerative diseases, mixed results have been reported for the effect of aging on mtDNAcn. For example, two studies found no evidence for age-related changes of mtDNAcn in three brain regions, skeletal muscle, and heart muscle [
13,
14], whereas a more recent study reported a decrease in skeletal muscle and an increase in liver tissue with age [
15].
A higher burden of mtDNA heteroplasmy has been observed in tissues from aged individuals [
16]. Increased levels of heteroplasmic mtDNA deletions as well as heteroplasmic point mutations were also described in brains from AD and Parkinson’s disease (PD) patients [
8,
17,
18], which lead to the hypothesis that pathogenic mtDNA mutations, when they exceed certain thresholds, could contribute to the mitochondrial dysfunction observed in late-onset neurodegenerative diseases. However, a recent high-throughput sequencing study looking at heteroplasmic point mutations found no evidence for an association with AD or PD [
9].
In this study, we profiled mtDNAcn and mtDNA heteroplasmy levels in
n = 762 brain samples from the Religious Orders Study and the Rush Memory and Aging Project (ROSMAP) [
19] and implemented substantial improvements compared to previous studies: (i) The detailed pathologic characterization of ROSMAP samples facilitated the disentanglement of the effects of different brain pathologies and aging on mtDNA. Mixed pathologies are common in aged individuals including AD patients [
20] and unaccounted pathologies may have contributed to some of the mixed results in the literature. (ii) Standardized cognitive tests conducted proximate to death were employed to assess the association between cognitive functioning and mtDNAcn adjusted for pathologies. (iii) Using RNA-seq-derived estimations of cell type proportions, we accounted for neuronal loss as a major confounder of mtDNAcn analyses in AD brains. (iv) We profiled five different brain regions to assess brain-regional differences (three regions in ROSMAP, and two additional regions in two independent datasets with a total of
n = 599 additional samples). (v) Finally, we calculated a proteomic score representing mitochondrial content to investigate whether changes in mtDNAcn reflect changes of mitochondrial mass or whether they are specific to mtDNA maintenance.
Discussion
We characterized the mtDNAcn in 1361 aged human brain samples from five regions and estimated a reduction of mtDNAcn by 7 to 14% in pathologic AD compared to non-AD samples in the cortical regions profiled in this study. We then leveraged the detailed pathologic and cognitive characterization of the ROSMAP study to identify the primary drivers of mtDNAcn loss and to assess the relation to cognitive function in the presence of mixed pathologies, which are frequently observed in the aged human brain [
20].
In the DLPFC, lower mtDNAcn was primarily associated with tau pathology. When accounting for ten common brain pathologies, tau was the only pathology that remained significantly associated with mtDNAcn. The mtDNAcn at the tissue level depends on the cell type composition of the studied tissue, as is well known in blood [
54]. We therefore estimated the neuronal proportion in the DLPFC samples, confirmed that the neuronal proportion is associated with mtDNAcn, and demonstrated that the association between tau pathology and mtDNAcn was still significant - albeit attenuated - when adjusting for neuronal proportion. Although previous studies showed that informative estimates of neuronal proportions can be obtained from DLPFC RNA-seq data [
60], we cannot exclude that changes in the complex composition of heterogeneous brain cells still contribute to the observed association. For example, a tau-mediated preferred depletion of a specific neuronal cell subtype could alter the composition within the neuronal cell population and cause changes in the mtDNAcn at the tissue level. Future single-cell studies will have to show to what extent the tau-related mtDNAcn loss observed in this study is a cell-intrinsic feature and which cell subtypes are affected.
Nevertheless, our analysis adjusted for neuronal proportions suggests that the relationship is not merely caused by tau-driven loss of mitochondria-rich neurons; other mechanisms link tau pathology to reduced mtDNAcn. Mechanisms that could potentially underlie our observation have been explored in model systems. For example, hyperphosphorylated tau has been shown to impair mitochondrial axonal transport [
61,
62] and to affect mitochondrial fission/fusion dynamics [
63,
64]. Conversely, reduced mtDNAcn has been shown to promote tau oligomerization in human neuronal cell lines [
65], suggesting complex interactions between tau and mitochondria. Tau is also known to be a strong predictor for cognitive decline. Interestingly, when studying cognitive function, mtDNAcn was a significant predictor in our model that included tau, nine other brain pathologies, neuronal proportion, and demographic variables.
In contrast to tau pathology, amyloid-β pathology was not significantly associated with mtDNAcn after accounting for other pathologies. This finding was supported by the results from the Mayo study where the mtDNAcn was not reduced in persons diagnosed with pathologic aging, which is defined by high amyloid-β burden but no or minimal tau pathology [
44]. Interestingly, numerous studies have demonstrated that the amyloid-β precursor protein (APP) as well as amyloid-β peptides localize at mitochondria and affect mitochondrial function and bioenergetics [
66,
67]. Further, C-terminal fragments resulting from the processing of APP (APP-CTFs) have been implicated in AD and may trigger morphological and functional changes of mitochondria [
68]. The absence of an association in our data could result from measuring only amyloid-β peptides (1–40) and (1–42) (both were detected by our antibody), which does not fully capture APP processing. Further, amyloid-β could still impair mitochondrial respiratory chain capacity or other non-energetic functions (e.g. calcium handling) without changes in mtDNAcn.
In the PCC, TDP-43 pathology was the most important factor and explained 23% of the mtDNAcn’s variance. The effect of TDP-43 on mitochondria has been mainly studied in model systems for ALS, where suppressing the localization of TDP-43 inside mitochondria reduces TDP-43 toxicity [
46]. When accumulating inside mitochondria, TDP-43 induces the release of mtDNA into the cytoplasm via the permeability transition pore [
47]. Whether any of these mechanisms underlie the correlation observed in our post-mortem brain data remains to be elucidated. Interestingly, the effect of TDP-43 pathology on mtDNAcn was moderate in the DLPFC, which could be caused by the distinct spatial pattern of TDP-43 progression in the aged brain or by a larger susceptibility of the neurons in the PCC to TDP-43 pathology.
The mtDNAcn has been suggested as a biomarker for aging because several studies of peripheral blood have reported an inverse correlation between age and mtDNAcn [
69,
70]. However, more recent studies demonstrated that the inverse correlation is likely caused by unaccounted age-related changes of cell type proportions in the blood [
71,
72]. In our study, we found no association between mtDNAcn and age or sex in the studied brain regions when we adjusted for pathologies. We also found no significant associations in the CB, which accumulates much less AD pathology than the DLPFC or the PCC. In summary, our results indicate that lower mtDNAcn is driven by certain pathologies rather than aging and restricted to brain regions directly affected by the respective pathologies. Similarly, a study of Parkinson’s disease brains found lower mtDNAcn in the vulnerable substantia nigra but not in the frontal cortex, which is less affected in Parkinson’s disease [
73].
To assess the effect of genetic variants on the mtDNAcn in the brain, we performed a targeted analysis of 81 lead SNPs that were recently reported by a large GWAS of mtDNAcn in blood [
52]. A missense variant in the protease
LONP1 demonstrated a moderate but significant effect size of 0.23 standard deviations larger mtDNAcn per dosage of the alternative allele in our meta-analysis of four brain regions. The original blood-based GWAS also detected a single variant at the
APOE locus, which harbors the strongest genetic risk factor for AD. We therefore investigated the effect of the
APOE ε4 haplotype (defined by two coding variants) on the mtDNAcn. While a large fraction of the
APOE ε4 effect was mediated via AD pathologies, we also found evidence for a direct effect on the mtDNAcn. These findings support the hypothesis that the
APOE ε4 allele exerts its risk not only via regulating amyloid-β aggregation and clearance but also through other pathways, including mitochondrial bioenergetics [
74,
75]. In summary, while this study is the first, to our knowledge, to report evidence for genetic regulation of the mtDNAcn in the brain, the sample size was a limiting factor and future non-targeted studies with much larger sample sizes will likely detect more loci.
Several tissues in addition to brain accumulate mtDNA mutations during aging [
76‐
78]. Here, we analyzed point mutations and small indels and found higher heteroplasmy levels with age in three cortical regions and estimated an association consistent with an accumulation rate of about 1.5% per year in this age group. We did not find a significant association with age in the CB where the overall frequency of mtDNA heteroplasmic mutations was very low (mean of 1.0) supporting the theory of region-specific accumulation of mtDNA mutations in the brain [
79]. Consistently, a previous study [
9] failed to detect an association between age and heteroplasmy levels in their samples which were mainly (87%) obtained from the CB. Thus, the CB seems to acquire less heteroplasmic mutations than the cortex or the heteroplasmic mutations in the CB have, on average, a lower relative frequency which may often not surpass the detection threshold of 3% used in this study. In contrast to the mtDNAcn, we found little evidence for the involvement of heteroplasmic mtDNA point mutations and small indels in AD or in the development of other brain pathologies. However, many of the previous studies reporting associations between mtDNA heteroplasmy and neurodegenerative diseases investigated structural variants, primarily large-scale mtDNA deletions, which were not considered in this study but are more likely to be pathogenic than small mutations [
17,
18]. Overall, our work showed that small mtDNA heteroplasmic mutations accumulate over an individual’s life-time in the cortex but are not related to neurodegenerative diseases. The seeming insignificance of low-level mtDNA heteroplasmy is consistent with a large reserve respiratory capacity, such that mtDNAcn is generally in excess (> 50%) of the minimum number of mtDNA copies required to sustain bioenergetic capacity [
80].
A lower mtDNAcn can be caused by lower mitochondrial mass in the cells or by a lower mtDNAcn per mitochondrion. We generated a proteomic measure of mitochondrial content for a subset of our DLPFC samples and showed that the mtDNAcn and mitochondrial content vary relatively independent of each other in the DLPFC. The decoupling of the two variables may be explained by distinct mechanisms that regulate mtDNAcn and mitochondrial content and that have been successfully manipulated in model systems to regulate one of the readouts, without affecting the other one [
5]. Similarly, the correlation between the abundance of mtDNA-encoded proteins and mtDNAcn was weak, and protein measures of the respiratory chain complexes were highly correlated with each other and with mitochondrial content protein markers, but showed only weak correlations with mtDNAcn. Thus, the moderate reduction of mtDNAcn observed in this study may not have a functional effect on the respiratory chain capacity. Finally, when we integrated the detailed pathologic and cognitive measures with the mitochondrial measures available for our DLPFC samples to disentangle their relationship, we found that both, mtDNAcn and mitochondrial content, were associated with cognitive function and neuronal loss indicating their involvement in neurodegeneration and the need for future studies to interrogate several mitochondrial measures to fully capture mitochondrial health.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.