Blood mitochondrial heteroplasmic variants and cognitive performance in late midlife: REGARDS study

Abstract

Background

Studies linking mitochondrial DNA (mtDNA) variants to cognition yielded inconsistent findings, and the underlying mechanisms remain unclear. We investigated whether mtDNA heteroplasmic variants were associated with cognitive outcomes, including the Montreal Cognitive Assessment (MoCA), in 197 late midlife adults from the Reasons for Geographic and Racial Differences in Stroke (REGARDS) cohort with complete data.

Methods

MtDNA was sequenced from blood using targeted deep sequencing. Adjusted linear and mixed-effects models examined the associations by functional regions, genes, total variant burden, nonsynonymous variants, and control regions.

Results

Heteroplasmic variants in the control region (β = -0.44, 95% CI: -0.83, -0.05, p = 0.027) and transfer RNA (tRNA) genes (β = -1.34, 95% CI: -2.58, -0.11, p = 0.034) were associated with MoCA baseline scores. Individual variants in cytochrome c oxidase subunit 1 (CO1) (β = -1.51, 95% CI: -2.54, -0.47, p = 0.005), NADH dehydrogenase subunit 1 (ND1) (β = -2.63, 95% CI: -4.56, -0.70, p = 0.008), and Displacement Loop (D-LOOP2) (β = -2.25, 95% CI: -4.20, -0.30, p = 0.025) was associated with reduced baseline MoCA scores. The ND6 (β = −1.23, 95% CI: −2.09, − 0.37, p = 0.006), ND4 (β = −1.11, 95% CI: −2.02, − 0.20, p = 0.018), ATP Synthase Membrane Subunit 8 (ATP8; β = −1.38, 95% CI: −2.63, − 0.13, p = 0.031), and D-LOOP1 (β = −0.61, 95% CI: −1.20, − 0.01, p = 0.045) genes suggested a potential association with executive function. Longitudinal Animal Fluency Test (AFT) scores were inversely associated with heteroplasmic variants in coding regions (β = -0.10, 95% CI: -0.19, -0.006, p = 0.049), the total number of variants (β = -0.06, 95% CI: -0.11, -0.003, p = 0.037) and total nonsynonymous variants (β = -0.11, 95% CI: -0.21, -0.01, p = 0.040). Variants in the control region were associated with the greatest decline in verbal fluency (β = −0.20, 95% CI: −0.39 to − 0.002, p = 0.049). No associations were observed between mitochondrial variants and verbal memory performance or the MoCA composite scores.

Conclusions

Our study indicates that mitochondrial variants measured in blood may provide insight into cognitive function during midlife. However, additional studies are needed to validate these associations and to address potential power limitations in our study.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12883-026-04857-x.

Background

Cognitive decline during aging has enormous implications for individual well-being [1]. Midlife cognitive performance can predict future dementia risk [2], including the risk of Alzheimer’s disease (AD), a common type of dementia [3]. Current projections indicate that the burden of AD in those aged 65 years or older will rise to 13.8 million by 2060 [4]. At this rate, the health system will struggle to cope with this growing burden unless effective treatments are developed [5]. Besides age, risk factors include sex, genetics, socioeconomic status, clinical factors, lifestyle, and some environmental exposures, such as air pollution [6, 7]. The Montreal Cognitive Assessment (MoCA) is a sensitive screening instrument with superior discriminative ability for identifying mild cognitive impairment (MCI) and AD [8].

Mitochondria are a vital and unique cell organelle, as they maintain their own extra-nuclear genomes containing key genes essential for cellular respiration, which is required for energy production, cellular signaling, metabolic regulation, and homeostasis, among other functions [9]. Human cells contain thousands of copies of mtDNA, which commonly undergo mutations, including point mutations, insertions, and deletions [10]. These mutations are classified as homoplasmic when they affect all copies of mtDNA, or heteroplasmic when they affect only a portion [11]. Importantly, mtDNA lacks crucial DNA repair pathways, rendering it particularly susceptible to oxidative damage [12].

Brain cells, including neurons, have a high energy demand but lack significant energy storage capabilities, increasing their susceptibility to mitochondrial dysfunction [13, 14]. More brain regions exhibit hypometabolism with aging due to a decline in mitochondrial function [7]. Furthermore, exposure to oxidative species and environmental toxicants further accelerates age-related mitochondrial mutation rates, contributing to reduced mitochondrial function [15–17]. As a result, a cause-and-effect relationship between the accumulation of mtDNA mutations and pathological changes in neurodegeneration has been proposed, including defective oxidative phosphorylation, increased oxidative stress, accumulation of amyloid-β, and apoptosis [18]. The mitochondrial cascade hypothesis suggests that mitochondrial dysfunction drives AD pathogenesis by inducing oxidative stress and energy deficits [19]. Baseline mitochondrial function and its rate of decline influence cognitive decline, which may begin years before AD diagnosis [20]. Also, previous studies have reported that heteroplasmic mutations are more frequent in the brains of AD cases than in controls [21]. However, other similar studies have not corroborated these findings [22].

The hallmark biomarkers of cognitive disorders, including amyloid beta peptide (Aβ) and phosphorylated tau isolated from blood, have demonstrated clinical utility in diagnosing and predicting neurodegenerative conditions [23]. Based on this evidence, it is plausible that the interplay between aging and other risk factors may simultaneously influence both blood and brain tissues, especially given that peripheral blood biomarkers are widely used to diagnose various clinical conditions. Accordingly, we hypothesize that a portion of the broad range of physiological changes associated with neurodegeneration might be reflected in peripheral blood. However, there is currently limited consensus on the relationship between mitochondrial changes and cognitive function, mainly due to inconsistent findings from blood-based studies [24]. The primary objective of our study was to conduct a comprehensive analysis of the association between mitochondrial variants and cognitive performance in midlife from the Reasons for Geographic and Racial Differences in Stroke (REGARDS) cohort. This topic has not been studied previously. To build on existing knowledge, we also aimed to investigate how combined variations in specific functional regions and individual mitochondrial genes may influence overall cognitive ability and its component domains.

Methods

Population and sampling

The REGARDS is an ongoing study in the United States (US) of 30,239 non-Hispanic Black (42%) and White women and men aged 45 years or more [25]. Above half (56%) of the participants in this REGARDS were recruited from the Stroke Belt (Alabama, Arkansas, Georgia, Kentucky, Louisiana, Mississippi, North Carolina, South Carolina, Tennessee, and Virginia). Participants were enrolled between 2003 and 2007, which was considered the baseline period. In-home visits were part of the baseline assessment. Recruitment, protocols, and follow-up visits for REGARDS participants have been previously described [25]. Blood samples were collected at baseline and stored at the University of Vermont for future biobanking. For this study, we selected participants from the REGARDS cohort using stratified random sampling to ensure proportional representation from Stroke Belt and non-Stroke Belt regions across the US [25]. To achieve representation across geographic regions, the selection was stratified by area, with 30% of participants from the Stroke Belt (North Carolina, South Carolina, Georgia, Tennessee, Mississippi, Alabama, Louisiana, and Arkansas), 20% from the Stroke Buckle (coastal plain of North Carolina, South Carolina, and Georgia), and the remaining 50% from other areas within the continental US. Apart from ensuring geographical representation, it enables mitigating potential biases related to regional differences in environmental exposures, lifestyle factors, or health outcomes. Because of the nature of the study (a population-based study), the sample included healthy participants at baseline and was not selected based on a specific pathology; therefore, a separate control group was not included. No additional stratification by race or other factors was applied, though the selected sample included a balanced representation of Black (53.8%) and White (46.2%) participants.

Mitochondrial mtDNA sequencing

Sample preparation and DNA extraction

Blood samples were collected at baseline during the enrollment period from 2003 to 2007. The DNA was extracted from the packed cells remaining after centrifugation of a 10 ml EDTA tube containing whole blood. The extraction was performed using the Qiagen Puregene Blood Kit (QIAGEN, Germantown, Maryland, Cat. No. 158026. We quantified samples using a NanoDrop spectrophotometer (Thermo Fisher Scientific) and then shipped them to the Genomics, Epigenomics, and Sequencing Core at the University of Cincinnati.

mtDNA amplification

To amplify the 16,569 base pairs of the human mtDNA genome, the protocol Document No. 15,037,958 v01 (Illumina, San Diego, CA) was followed with modifications [26]. Genomic DNA was amplified using two long-range polymerase chain reactions (PCR; 27 cycles) that covered the entire mtDNA sequence, including negative controls. PCR amplification was conducted in 50-µL reactions using TaKaRa LA Taq (Takara Bio USA, Mountain View, CA), with two nanograms (ng) of DNA and one of two primer pairs (MTL-F1/MTL-R1 or MTL-F2/MTL-R2). Whole mtDNA PCR quality control was performed using 2.75% high-resolution agarose gel electrophoresis. Following amplification, PCR products were volume-adjusted and pooled to achieve roughly equivalent amounts of the two amplicons. For the initial test sequencing, equal volumes of each library were combined to generate low-read-count inputs. In the final sequencing run, library input volumes were adjusted based on test sequencing results to obtain roughly equal read counts per sample.

Library preparation and sequencing

DNA was purified with the Wizard SV Gel and PCR Clean-Up Kit (Promega), and about 100 ng was used as input for library preparation with the NEBNext Ultra II FS DNA Library Prep Kit (New England Biolabs, Ipswich, MA). Libraries were evaluated using the Agilent Bioanalyzer (Santa Clara, CA) and Qubit assay (Thermo Fisher Scientific, US), indexed, and sequenced on the Illumina NextSeq 2000 platform with 2 × 150 bp paired-end reads, achieving a median of 2.5 million read clusters per sample and a median sequencing depth of 5800x.

Mitochondrial variant calling

Initial quality control checks of the raw mtDNA sequencing data were performed using FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Paired-end reads were aligned to the mitochondrial reference genome (ChrM; revised Cambridge Reference Sequence [rCRS], GenBank accession NC_012920.1) using the Burrows-Wheeler Aligner (BWA-MEM v0.7.15).

Variants were called using the Genome Analysis Toolkit (GATK) Mutect2 variant caller (10.1101/861054) in “mitochondrial mode,” following the methods described by Laricchia et al. [10] Deletions and insertions were excluded due to technical limitations in accurately detecting and validating these variants in mtDNA sequencing data. Specifically, the short-read sequencing approach (Illumina NextSeq 2000, 2 × 150 bp paired-end reads) and alignment to the revised Cambridge Reference Sequence (rCRS) can introduce ambiguities in calling indels, particularly in repetitive or homopolymeric regions of the mtDNA genome, such as the D-loop. To ensure high-confidence variant calls and minimize false positives, we focused exclusively on single-nucleotide variants (SNVs), which are more reliably detected with the achieved median sequencing depth. To call variants in the control region (CR), the artificial break in the circular mitochondrial chromosome in positions ranging from (16,024 − 16,569 and 1–576), reads were realigned to a reference shifted by 8,000 base pairs. Coordinates were converted back to their original positions and combined with variants from the non-control region into a unified VCF file.

We excluded variants with ‘fail,’ ‘base-quality,’ ‘position,’ ‘backlisted’, and ‘strand’ flags in the filter column of the VCF file. We also filtered out deletions and insertions. We estimated mtDNA contamination using Haplocheck (v1.3.3) [27], and haplogroups were determined using HaploGrep2 (v2.2.9) [28]. The variant allele fraction (VAF) was calculated as the ratio of alternate allele reads to total reads for a specific variant in each sample. Variants with VAF between 0.05 and 0.98 were considered heteroplasmic, while those with a VAF between 0.98 and 1.00 were considered homoplasmic. To avoid false positive calls, variants with alternative allele fractions below 5% were filtered out [10, 29]. Repeated sequencing on 9% of the samples showed 100% variant call concordance in 18 out of 19 cases, with one sample exhibiting a single discordant call (83% concordance; Supplementary Table 1). As heteroplasmic variants are more likely to reflect mtDNA damage rather than germline mtDNA variation or haplogroup differences, only heteroplasmic variants were considered in the current analysis.

Cognitive function assessments

The complete details on the cognitive assessments in the REGARDS cohort are available in the Cognitive Data Manual (Version 3) [30]. In brief, the shorter versions of cognitive assessments were used for evaluation. They complied with the National Institute of Neurological Disorders and Stroke and the Canadian Stroke Network (NINDS-CSN) standards. Shorter versions of cognitive assessments were modified for data collection over the phone. All personnel involved in data collection were formally trained and certified. Among the assessments were a battery of three cognitive assessments from the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD): Word List Learning (WLL), Word List Delayed Recall (WLD), the Animal Fluency Test (AFT), and the MoCA. The test intervals were highly variable across participants because they were recruited at different times. Cognitive assessments were conducted every 18 months for a brief period at the beginning, and then biannually, beginning between 2006 and 2013, with approximately over 10 years of longitudinal follow-up [31, 32]. The MoCA 5-minute short version was included among the assessments 3 years after the CERAD was first measured in the cohort. It was scored using a specially modified 5-word recall and 6-item orientation components for telephone-based assessments, which were then piloted from April 7, 2008, to April 30, 2008. Subsequently, it was added to the short cognitive battery and is administered every 2-year intervals as part of the NINDS-CSN 5-minute protocol. Global cognition, word lists, and verbal fluency can be measured reliably and precisely over the telephone in middle-aged and older adults, with scores virtually identical to those obtained in person [33–35]. Experts adapted cognitive assessments to allow for accurate administration over the telephone. For example, visual stimuli were not assessed, and participants were not permitted to write until the tests were completed. The entire process was recorded for computer-assisted scoring and quality checks, and inter-rater reliability was ensured. The latest cognitive evaluations included in the current analyses took place in 2023. Global cognitive function in our study was evaluated using standalone MoCA test scores and by combining them with domain-specific scores. Verbal fluency was evaluated using the Letter F test (LF), in which individuals were asked to generate as many words as possible that begin with the letter “F” within 60 s [36]. The cognitive tests were consistent with the Vascular Cognitive Impairment Harmonization Standards [37], which have been validated in both Black and White individuals [38].

Statistical analysis

All regression analyses and accompanying visualizations were conducted using R (version 4.4.2) [39]. Continuous data were presented as means ± standard deviations (SD), and categorical data as frequencies and percentages. Mitochondrial heteroplasmy was the main predictor in the models. We followed two approaches to define exposure: first, the total count of heteroplasmy burden across functional mtDNA regions; and second, the presence of variants within specific mtDNA genes. For the latter, genes were evaluated as “mutated” if two or more distinct variants were identified in each gene (e.g., CO1, which encodes cytochrome c oxidase subunit 1). The primary outcomes of the study were the two global cognition scores (MoCA alone, MoCA composite with CERAD subdomains). The MoCA and CERAD are both robust cognitive assessment tools compared to the widely used Mini-Mental State Examination (MMSE), as it is much more sensitive to subtle cognitive changes over time and less prone to ceiling effects [40, 41]. These properties make them appropriate measures for detecting long term cognitive changes in our cohort study. We created composite z-scores at each time point and used them as continuous outcomes [42, 43]. A minimal set of covariates was selected for analysis a priori—using Directed Acyclic Graphs (DAGs), given our limited sample size. All analyses began with assumption-free plots to inspect associations visually, and models were selected based on the nature of the variables and their distributions. Variables were transformed to z-scores as needed to satisfy the model assumptions. For each outcome-predictor combination, two types of analyses were conducted. The association between the burden of mtDNA heteroplasmic variants and baseline cognitive performance was evaluated using standard linear regression models. Longitudinal associations between baseline heteroplasmy and cognitive outcomes were assessed using linear mixed models in the lme4 package. No data was imputed in the analysis. Covariates were modeled as fixed effects, and slopes were allowed to vary for subjects to account for intra-individual variation over time. All models were adjusted for age at cognitive assessment, sex, race, income, education, physical activity, body mass index (BMI), smoking status, and residence (urban vs. rural). Positive coefficients indicated better cognitive performance, and vice versa. A two-tailed p < 0.05 was considered as evidence of suggestive significant associations.

Results

Participant characteristics

We analyzed 197 REGARDS participants with mtDNA data, longitudinal cognitive assessments, and complete covariate information. This sample size represents those with complete mtDNA sequencing data, longitudinal cognitive assessments (including MoCA and CERAD battery), and full covariate information after quality control. The median follow-up time was 10.6 years (IQR: 6.5–16.6 years). Overall, concordance between replicates in samples from REGARDS participants included in the study exceeded 99% (Supplementary Table 1). Participant characteristics are summarized in Table 1, and comparisons between the current sample and previous cognitive studies in REGARDS are provided in Supplementary Table 2. The mean age of the participants was 64 years, and the majority were women and Black. An average BMI exceeding 30 kg/m² in the sample reflected a higher prevalence of overweight/obesity. As regards education, over half of the participants attended college. Most of them lived in urban areas, and nearly 30% reported being more physically active. Roughly half were current or past smokers. Approximately 47% of participants reported a current or previous history of smoking. At baseline, 8 participants had a history of clinical stroke, and there was an additional case of stroke during follow-up.

Table 1.

Demographic characteristics of REGARDS participants included in the current study (N = 197)

		Mean/N	SD/%
Age, years		64.00	10.0
Race
Black		106	53.8%
Gender
Female		112	56.9%
Income
Less than $20K		43	21.8%
$20K-$34K		42	21.3%
$35K-$74K		51	25.9%
$75K and above		40	20.3%
Not available		21	10.7%
Education
Less than high school		27	13.7%
High school graduate		37	18.8%
Some college		64	32.5%
College graduate and above		69	35.0%
Physical activity
None		57	28.9%
1 to 3 time per week		79	40.1%
4 or more per week		56	28.4%
Not available		5	2.5%
Smoking
Current		25	12.7%
Past		68	34.5%
Never		103	52.3%
Not available		1	0.5%
Body mass index, Kg/m2		30.58	6.5
Residential area
Rural (≤ 25% urban)		11	5.6%
Mixed (26–75% urban)		14	7.1%
Urban (> 75% urban)		152	77.2%
Not available		20	10.2%

Sample composition: 30% from the Stroke Belt (North Carolina, South Carolina, Georgia, Tennessee, Mississippi, Alabama, Louisiana, and Arkansas), 20% from the Stroke Buckle (coastal plain of North Carolina, South Carolina, and Georgia), and the remainder from areas within in the continental U.S

SD Standard Deviation

Mitochondrial DNA variant profiling

After excluding replicate samples and applying variant filtering, 3,795 single-nucleotide variants (SNVs) remained with a median coverage of 5,800x and a variant allele frequency (VAF) threshold of 5%. Of these, 3,660 (96.4%) were homoplasmic, and 135 (3.6%) were heteroplasmic (Fig. 1, Panel A). Haplogroup characterization showed that the L haplogroup was the most frequent (51.9%), followed by the H haplogroup (21.3%; Fig. 1, Panel B). The L haplogroup was observed in most Black participants (93.4%) but was rare among White participants (2.2%) (Table 2). A 95.1% (3,610) of mtDNA variants were transitions. Of the 969 unique variants, 857 (88.4%) were homoplasmic, 46 (4.7%) were heteroplasmic, and 66 (6.8%) were identified as both homoplasmic and heteroplasmic. Compared to homoplasmic variants, heteroplasmic variants had a higher proportion of non-synonymous and rRNA variants. Conversely, synonymous variants were more frequent (50.4%) in the homoplasmics (Fig. 1, Panel C). The median number of homoplasmic variants was 17 per individual (range: 0–54), whereas most individuals had no detected heteroplasmies (median: 0, range: 0–20). Two participants had no homoplasmic variants detected relative to the rCRS mtDNA reference sequence. The mitochondrial Displacement Loop (D-LOOP) (~ 17%) and NADH dehydrogenase 5 (ND5) genes (~ 15%) showed the highest proportion of heteroplasmic variants (Fig. 1, Panel D). A mitochondrial variant annotation table detailing SNP positions, coverage, variant types, and nucleotide substitutions in the study sample is presented in Supplementary Table 3.

Fig. 1.

Mitochondrial variant characteristics in REGARDS participants. A Distribution of single nucleotide variants (SNVs) by heteroplasmy status. B Frequency of mitochondrial haplogroups across participants. C Functional annotation of homoplasmic and heteroplasmic variants. D Circular plot of the mitochondrial genome showing the distribution of heteroplasmic variants across mitochondrial genes (two inner circles show genomic locations of homoplasmic variants [red] and homoplasmic variants [orange]) and annotations on the outer circle

Table 2.

Distribution of the L and non-L haplotypes by race in REGARDS participants included in the current study (N = 197)

Race	Non-L haplotype		L haplotype		All		p-value*
	n	%	n	%	n	%
Black	7	6.6%	99	93.4%	106	53.8%	< 0.001
White	89	97.8%	2	2.2%	91	46.2%
Subtotal	96	48.7%	101	51.3%	197	100%

*Chi square test

Associations between heteroplasmic mitochondrial variants, MoCA, and global cognition composite scores

We did not observe any longitudinal associations between variant burden and MoCA across functional regions. However, we observed consistent negative associations with MoCA z-scores at baseline in the control regions (β = -0.44, 95% CI: -0.83, -0.05, p = 0.027), and transfer RNA (tRNA) genes (β = -1.34, 95% CI: -2.58, -0.11, p = 0.034) (Fig. 2, Panel A; Supplementary Table 4. Individual heteroplasmic variants in genes did not influence MoCA trajectories. We identified three suggestive associations involving genes and control region between individual variants and baseline MoCA scores, including cytochrome c oxidase (Complex IV) genes (CO1) (β = -1.51, 95% CI: -2.54, -0.47, p = 0.005), NADH dehydrogenase Complex I (ND1) (β = -2.63, 95% CI: -4.56, -0.70, p = 0.008), and D-LOOP2 (β = -2.25, 95% CI: -4.20, -0.30, p = 0.025) (Fig. 2, Panel B; Supplementary Table 4). The variant burden was not associated with the z-scores for trajectories or baseline global composite cognition scores. No associations were observed for individual mitochondrial variants in the control region or genes on the global cognition composite scores.

Fig. 2.

Associations between heteroplasmic mtDNA variants at baseline and cognition among REGARDS participants. A Associations between the number of mtDNA heteroplasmic variants grouped by functional regions, including protein-coding regions, transfer RNA (tRNA), ribosomal RNA (rRNA), and the control region (CR), and the first z-score for the Montreal Cognitive Assessment (MoCA). B Associations between the presence of heteroplasmic variants within individual mtDNA genes or control regions and the first z-score for MoCA. C Associations between the number of mtDNA heteroplasmic variants and the composite z-score for verbal fluency (Animal Fluency and Letter F tests). D Associations between the number of mtDNA heteroplasmic variants and the z-score for the Animal Fluency test. The blue dotted line indicates the nominal threshold for statistical significance (p-value = 0.05). Individual variants showing suggestive associations are highlighted in blue, and those with no associations are shown in red. Estimates were derived from multivariable-adjusted models controlling for age at cognitive assessment, race, gender, income, education, physical activity, smoking status, and urban/rural residence

Associations between heteroplasmic mitochondrial variants and LFT, AFT, and related composite scores

The detailed analyses of the associations between heteroplasmic mitochondrial variants and LFT are shown in Supplementary Table 5. Individual heteroplasmic variants in mitochondrial genes showed suggestive associations with executive function over time. We found that ND6 (β = −1.23, 95% CI: −2.09, − 0.37, p = 0.006), ND4 (β = −1.11, 95% CI: −2.02, − 0.20, p = 0.018), ATP8 (β = −1.38, 95% CI: −2.63, − 0.13, p = 0.031), and D-LOOP1 (β = −0.61, 95% CI: −1.20, − 0.01, p = 0.045) were associated with LF. As shown in Supplementary Table 6, longitudinal AFT scores were inversely related to the presence of heteroplasmic variants in coding regions (β =-0.10, 95% CI: -0.19, -0.006, p = 0.049), the total number of variants (β =-0.06, 95% CI: -0.11, -0.003, p = 0.037) and total nonsynonymous variants (β =-0.11, 95% CI: -0.21, -0.01, p = 0.040). By comparison, the same analysis for baseline executive function score revealed no regional involvement. Supplementary Table 7 shows the associations between heteroplasmic mitochondrial variants and verbal fluency composite scores. Longitudinal associations with heteroplasmic variant burden were observed for the verbal fluency composite scores for total number of variants (β = -0.05, 95% CI: -0.10, -0.004, p = 0.036), coding region (β = -0.08, 95% CI: -0.16, -0.005, p = 0.040), and control regions (β = -0.20, 95% CI: -0.39, -0.002, p = 0.049), with each showing a reduction in verbal fluency over time. At baseline, heteroplasmic variants in the regions were not associated with differences in verbal fluency.

Associations between heteroplasmic mitochondrial variants, verbal memory, and MoCA memory score

In a more in-depth analysis, statistical models did not detect associations between the burden of variants or individual variants within genes or regions and verbal memory, as measured by the WLL and WLD (Supplementary Tables 8 and 9). Both WLL and WLD primarily assess memory and attention and are designed to detect memory decline characteristics of pathological dementia. The association between the ND2 gene and WLD showed a trend toward statistical significance (β = 0.73, 95% CI: −0.04, 1.50, p = 0.068). Likewise, we did not detect any associations between the number of heteroplasmic variants and MoCA memory composite scores (Supplementary Table 10). Similarly, analyses conducted separately for individual genes and control regions revealed no associations between individual heteroplasmic variants and overall cognitive scores in either cross-sectional or linear mixed model analyses.

Discussion

In this study, we reported suggestive associations between blood heteroplasmic mtDNA variants and changes in cognitive performance during late midlife. Although only baseline mtDNA variants were available, we could still confirm associations with multiple cognitive domains and the MoCA. Even the baseline associations between heteroplasmy and MoCA scores may indicate that blood-derived mitochondrial measures could, to some extent, correlate with brain pathology [44]. However, given the limitations inherent to our study design, we were unable to formally examine the correspondence between mitochondrial alterations in blood and in the brain. Examining associations separately for global and domain-specific cognitive functions may be more informative than relying solely on a single composite summary score [45]. We found that increased heteroplasmic variant burden in mitochondrial genes, control regions, and tRNA was suggestively associated with baseline MoCA scores. In a study examining brain samples from AD patients, heteroplasmic mtDNA variants in the control regions were increased by 63% compared with healthy controls [21]. mtDNA variants usually increase with age, and a substantial proportion of these changes is likely to cause an amino acid change, which could be among the mechanisms contributing to our findings [46]. Another widely proposed mechanism is that the human body may lose its ability to physiologically cope with the rapid accumulation of variants, thereby derailing timely repair and brain neurogenesis [47]. However, mitochondrial heteroplasmy burden was measured at only a single time point in our study, and we were therefore unable to assess longitudinal changes or temporal relationships.

Among the mitochondrial genes and control region suggestively associated with MoCA performance in our study, CO1 and ND1 are integral components of the mitochondrial electron transport chain (ETC). They belong to the enzyme complexes Complex IV and Complex I, respectively. ETC dysfunction leads to increased oxidative stress and, consequently, inflammation, both of which adversely affect neural health [48, 49]. Our findings support initial reports that an increased burden of mtDNA variants suppresses CO1 activity, consequently reducing brain metabolism and potentially contributing to neurodegeneration and cognitive decline [46]. Regarding ND1, mitochondrial complex I (NADH: ubiquinone oxidoreductase) is a large enzyme that catalyzes the first step of the ETC, oxidizing NADH, and must be precisely assembled from 45 protein subunits to function correctly [50]. Experimental studies have shown that ND1 variants, as they accumulate, disrupt complex 1 enzymes and play a role in mitochondrial dysfunction [50]. Apart from interfering with ETC function, ND1 variants alter protein structure and cellular energy homeostasis and are implicated in many disorders, including AD [51]. A study by Alkhaldi and colleagues provided experimental evidence that ND1 variants can have serious health consequences and that variants with mild effects may become more pathogenic when combined with other disease-related factors [52].

The D-loop, a non-coding region in mtDNA, is the primary control site for mtDNA expression and is involved in mtDNA replication and transcription. Prior studies demonstrated that the D-loop region is highly susceptible to variants and copy number reductions under specific pathological conditions [53, 54]. Reduced mitochondrial transcription was also linked to age-related cognitive decline in rodents [55]. D-loop associations with cognitive decline in our data appear to be connected mainly through oxidative stress [56]. In the Alzheimer’s Disease Neuroimaging Initiative (ADNI) cohort, mtDNA analysis revealed that the D-loop region interacts with other mitochondrial genes and influences protein biomarkers of neurodegeneration [57].

Our results suggest that mtDNA variants in blood may help identify early signs of dementia. Blood-based tests are less invasive, cost-effective, and suitable for large-scale screening than Cerebrospinal fluid (CSF) sampling or advanced neuroimaging techniques [58]. Medications approved by the U.S. Food and Drug Administration (FDA), such as Donanemab (Kisunla, Eli Lilly), have shown promise by slowing cognitive decline, with even greater benefits observed in patients treated earlier in the disease progression [59, 60]. To our knowledge, this is one of the first epidemiological studies to investigate the impact of blood mtDNA variants on cognitive performance in late midlife adults, using high-resolution targeted mtDNA sequencing data. Multiple comparison testing is a valuable tool to reduce the possibility that observed associations may be false positives due to multiple testing. We have therefore explored our results through this lens, and even with False Discovery Rate (FDR) adjustment for multiple comparisons (FDR < 0.2), we have observed that many associations remain significant. Our major limitation was that the sample size (N = 197) is relatively small and may limit statistical power, particularly for detecting smaller effect sizes or rarer mutations, and could increase the risk of Type II errors. Linear mixed models are generally robust to missing values [61], but bias may still arise even under the missing-at-random assumption in observational studies. The absence of associations for composite scores, apart from verbal fluency in the present study, may be attributed to statistical dilution due to ceiling effects and other methodological limitations [62]. We also advocate a more cautious approach to interpreting the observed associations between heteroplasmies and cognitive measures within specific cognitive domains. While these findings do provide some useful insights, they may have limited sensitivity for isolating the full spectrum of effects suggestively associated with discrete cognitive domains. The sample size was constrained by the cost of targeted deep sequencing of mtDNA from stored blood samples, which required specialized library preparation, high-depth sequencing, and rigorous variant calling to accurately detect heteroplasmic mutations (with a variant allele frequency threshold of 5%). Therefore, the findings are purely exploratory and need to be replicated in independent cohorts. Disentangling mitochondrial effects attributable to neurodegeneration from those related to cerebrovascular injury is challenging in this cohort. The REGARDS cohort does not collect information on vascular dementia diagnoses. It is certainly possible that some participants had mixed dementia, as neurodegenerative and cerebrovascular conditions are known to co-exist in dementia [63].

Some of the changes in cognitive performance over time may be random. Importantly, heteroplasmy was measured at a single time point (baseline), which precluded the application of the Reliable Change Index in the present analyses. We also acknowledge that other genetic factors, such as the APOE ε4 allele, can influence cognitive outcomes, and that nuclear and mitochondrial genomes can interact in complex ways, which we have not investigated here. Although APOE4 carriers are at risk for reduced mtDNA copy number (mtDNAcn), mediation analysis conducted in an earlier study estimated that this accounts for slightly less than 50% of the risk for AD [22]. Thus, there is a possibility that the mtDNA variants may operate independently of APOE4 and may contribute to dementia risk in at least a subset of cases, as reported in elderly Afro-Americans [64]. Finally, using the revised Cambridge Reference Sequence, which is based on individuals of European descent, may have limited our ability to detect a higher variant burden in Black participants.

In closing, we conducted this research to generate preliminary evidence that blood mitochondrial heteroplasmic variants are associated with cognitive performance during late midlife. Together, our findings suggest that the observed associations with cognition may be driven by pathological changes arising from systemic processes involving both the brain and blood, including increased reactive oxygen species and loss of regulatory control. However, this hypothesis deserves further research. We plan to include major risk factors (e.g., APOE, social determinants) [65–67] to validate and extend these results in future studies. Heteroplasmy profiling is being conducted in participants several years after baseline which, will allow us to apply measures like the Reliable Change Index and more reliably capture nuanced cognitive changes over time.

Conclusions

Our findings suggest that mtDNA variants were suggestively associated with cognitive impairment in late midlife. Studying mitochondrial dynamics in blood and brain during the early stages of dementia may help improve our understanding of disease mechanisms and develop strategies to delay dementia in old age.

Abbreviations

Aβ

Amyloid beta peptide

AD

Alzheimer’s disease

ADNI

Alzheimer's Disease Neuroimaging Initiative

AFT

Animal fluency test

ATP8

ATP Synthase Membrane Subunit 8

CERAD

Consortium to establish a registry for Alzheimer’s disease

CO1

Cytochrome c oxidase subunit 1

CR

Control region

CSF

Cerebrospinal fluid

ETC

Electron transport chain

FDA

U.S. Food and Drug Administration

FDR

False Discovery Rate adjustment for multiple comparisons

GATK

Genome Analysis Toolkit

LF

Letter F test

MCI

Mild cognitive impairment

MoCA

Montreal Cognitive Assessment

mtDNA

Mitochondrial DNA

mtDNAcn

mtDNA copy number

ND1

NADH dehydrogenase 1

REGARDS

REasons for geographic and racial differences in stroke cohort

ROS

Reactive oxygen species

SD

Standard deviations

SNV

Single-nucleotide variants

tRNA

Transfer RNA

US

United States

WLD

Word list delayed recall

WLL

Word list learning

Authors’ contributions

All authors contributed to the study design and interpretation of results. DP, AAB, and RP designed the original idea and led the overall project. AK and CL ran the bioinformatics analyses. HI and EBL developed the statistical model, which was implemented by DP. MC, JM, and SJ contributed data about the participants and to the design of the proposal. EM, DC, AR, JG, ALR, and CH contributed extensively to the discussion of results. DP, ALR, and RP drafted the manuscript. All authors contributed to manuscript revisions and approved the final manuscript.

Funding

This research project was supported by cooperative agreement U01 NS041588, co-funded by the National Institute of Neurological Disorders and Stroke (NINDS) and the National Institute on Aging (NIA), National Institutes of Health, Department of Health and Human Service. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NINDS or the NIA. While NINDS representatives participated in the review of the manuscript, they were not involved in the collection, management, analysis, or interpretation of the data.

Additional funding was provided by the US National Institutes of Health - National Cancer Institute (U54CA267776), the National Institute of Aging (U01AG088684), and the Consejo Nacional de Humanidades, Ciencias y Tecnologías, Mexico (A3-S-49533).

Data availability

The primary datasets used to generate the data are available upon request from the REGARDS Consortium ( regardsadmin@uab.edu ). This study used existing de-identified data from the REGARDS cohort. The mtDNA heteroplasmy datasets generated and analyzed during the current study are publicly available in the Zenodo repository at 10.5281/zenodo.18964372 (accession number: 18964372).

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the principles of the Declaration of Helsinki. All participants provided informed consent for both baseline and follow-up visits. The study was approved by the Institutional Review Board of the Icahn School of Medicine at Mount Sinai (Protocol ID: STUDY-25-01086).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.