Mitochondrial DNA Sequence Variation Associated with Dementia and Cognitive Function in the Elderly

Abstract

Mitochondrial dysfunction is a prominent hallmark of Alzheimer's disease (AD). Mitochondrial DNA (mtDNA) damage may be a major cause of abnormal reactive oxidative species production in AD or increased neuronal susceptibility to oxidative injury during aging. The purpose of this study was to assess the influence of mtDNA sequence variation on clinically significant cognitive impairment and dementia risk in the population-based Health, Aging, and Body Composition (Health ABC) Study. We first investigated the role of common mtDNA haplogroups and individual variants on dementia risk and 8-year change on the Modified Mini-Mental State Examination (3MS) and Digit Symbol Substitution Test (DSST) among 1,631 participants of European genetic ancestry. Participants were free of dementia at baseline and incidence was determined in 273 cases from hospital and medication records over 10–12 follow-up years. Participants from haplogroup T had a statistically significant increased risk of developing dementia (OR = 1.86, 95% CI = 1.23, 2.82, p = 0.0008) and haplogroup J participants experienced a statistically significant 8-year decline in 3MS (β = −0.14, 95% CI = −0.27, −0.03, p = 0.0006), both compared with common haplogroup H. The m.15244A>G, p.G166G, CytB variant was associated with a significant decline in DSST score (β = −0.58, 95% CI −0.89, −0.28, p = 0.00019) and the m.14178T>C, p.I166V, ND6 variant was associated with a significant decline in 3MS score (β = −0.87, 95% CI −1.31, −3.86, p = 0.00012). Finally, we sequenced the complete ∼16.5 kb mtDNA from 135 Health ABC participants and identified several highly conserved and potentially functional nonsynonymous variants unique to 22 dementia cases and aggregate sequence variation across the hypervariable 2-3 regions that influences 3MS and DSST scores.

Introduction

Cognitive impairment has emerged as one of the greatest health threats to society. Cognitive decline and dementia, largely in the form of Alzheimer's disease (AD), affect approximately 10% of adults over the age of 65 rising exponentially to 50% of adults over the age of 85 in the United States [1]. Several conserved mechanisms underlie the changes observed in the aging brain including mitochondrial function and oxidative stress, autophagy, and protein turnover [2]. Considerable evidence suggests that the changes in mitochondria and oxidative stress levels precede plaque and tangle formation and the clinical manifestation of AD in humans [3]. Changes in mitochondrial function are causally linked to several early abnormalities that accompany AD including plaques and tangles [3]. Hence, the early alterations to mitochondria, which can induce multiple abnormalities, may present more desirable therapeutic targets than the reversal of the individual pathologies that occur in later cognitive decline and dementia.

Increased oxidative damage [4] and mitochondrial dysfunction are important early factors for accelerated cognitive decline and AD [5, 6]. Measures of reactive oxidative species (ROS) damage to proteins, nucleic acids, carbohydrates, and lipids are found in cerebrospinal fluid, plasma, and urine of subjects at very early stages of AD [7] and in autopsy brains from AD patients [8–12]. Different forms of mitochondrial impairment or oxidative stress can mimic AD-like changes to the brain [13–23]. For example, induction of oxidative stress by multiple approaches increases amyloid-β (Aβ) production and plaque formation [24–28]. In fact, oxidative damage in AD brain is more pervasive than plaques and tangles [29] and changes in mitochondrial function have been shown to precede Aβ and tangle formation [30–35]. Mitochondria-derived ROS [36–38] and Aβ formation [36, 39] may create a cycle that further exacerbates mitochondrial dysfunction [40] and accelerates cognitive impairment [39]. Organ-specific analysis of brain aging has revealed a progressive decline in mitochondrial gene expression in rats, rhesus macaques, and humans [14, 41, 42]. Several lines of evidence show that key enzymes responsible for mitochondrial energy metabolism are severely affected in AD [43–45] with some genes coding for respiratory chain subunits being differentially expressed in AD patients [46]. The brain is particularly susceptible to defective mitochondrial function related to mitochondrial DNA (mtDNA) mutations [2, 3]. In particular, oxidative phosphorylation (OXPHOS) defects resulting from somatic mtDNA mutations may play a role in AD pathophysiology [47].

Human mtDNA is a 16,569 base pair loop containing genes critical to mitochondrial energy production [48], and bioenergetic defects resulting from acquired and inherited mtDNA mutations are critical for both age-related dementia and associated neuropathological changes observed in AD [5, 47, 49–54]. OXPHOS enzyme activities decline with age in the human brain [55, 56] and this decline is correlated with the accumulation of somatic mtDNA deletions [57–62] and base substitutions [62, 63]. Sequence variation within the 13 mtDNA-encoded OXPHOS genes may impact superoxide production at OXPHOS complexes I and III and ATP generation efficiency through respiratory chain impairment [64, 65], apoptosis [66], and ATP supply [67]. MtDNA damage may also be a major cause of abnormal ROS production in AD [68] or increase neuronal susceptibility to oxidative injury during aging [37, 38, 69].

Individual mtDNA mutations have been identified in patients with AD [45, 70–84]. However, these studies were small and most of the identified variants have not been confirmed [77, 79–84]. In the present study, we first set out to assess the role of common mtDNA haplogroups and individual variants in dementia risk and longitudinal cognitive change among 1,631 participants from the population-based Health, Aging, and Body Composition (Health ABC) Study. Because human mtDNA has a mutation rate that is 10–20 times higher than that of nuclear DNA [85–87] and up to one-third of sequence variants found in the general population may be functionally important [88], it is likely that most of the mtDNA variation that impacts function is rare in frequency and only detectable by direct sequencing. In order to address this possibility, we also sequenced the complete ∼16.5 kb mtDNA from 135 Health ABC participants to identify both highly conserved and potentially functional mutations unique to dementia cases and aggregate sequence variation that influences tests of cognitive function.

Materials and Methods

Population

Participants were part of the Health ABC Study, a prospective cohort study of 3,075 community-dwelling black and white men and women living in Memphis, TN, or Pittsburgh, PA, and aged 70–79 years at recruitment in 1996-1997 [89]. To identify potential participants, a random sample of white and all black Medicare-eligible elders, within designated zip code areas, were contacted. To be eligible, participants had to report no difficulty with activities of daily living, walking a quarter of a mile, or climbing 10 steps without resting. They also had to be free of life-threatening cancer diagnoses and have no plans to move out of the study area for at least 3 years. The sample was approximately balanced for gender (51% women) and 41% of participants were black. Participants self-designated race/ethnicity from a fixed set of options (Asian/Pacific Islander, black/African American, white/Caucasian, Latino/Hispanic, do not know, other). The study was designed to have sufficient numbers of black participants to allow estimates of the relationship of body composition to functional decline. All eligible participants signed a written informed consent, approved by the institutional review boards at the clinical sites. This study was approved by the institutional review boards of the clinical sites and the coordinating center (University of California, San Francisco).

Cognitive function testing

The Modified Mini-Mental State Examination (3MS) was administered to participants at baseline (year 1) and after 2, 4, and 7 years of follow-up (years 3, 5, and 8). The 3MS is a brief, general cognitive battery with components for orientation, concentration, language, praxis, and immediate and delayed memory [90]. Possible scores range from 0 to 100, with higher scores indicating better cognitive function. The Digit Symbol Substitution Test (DSST) was administered to participants at baseline (year 1) and after 4 and 7 years of follow-up (years 5 and 8). The DSST measures response speed, sustained attention, visual spatial skills, and set shifting, all of which reflect executive cognitive function [91, 92]. The test is reported to distinguish mild dementia from healthy aging [93]. The DSST score is calculated as the total number of items correctly coded in 90 s, with a higher score indicating better cognitive function. Participant-specific slopes of DSST scores were estimated from mixed-effects models with random intercepts and slopes [94]. The participant-specific slopes of 3MS scores were estimated by best linear unbiased predictions using a linear mixed model with random intercepts and slopes using STATA10.

Dementia incidence

All participants were free of dementia at baseline. Incident dementia was determined by the date of the first available record of a dementia diagnosis over 10–12 years of follow-up. Cases were identified through hospital records indicating a dementia related hospital event, either as the primary or secondary diagnosis related to the hospitalization, or by record of prescribed dementia medication (i.e., galantamine, rivastigmine, memantine, donepezil, tacrine).

Genotyping

Genomic DNA was extracted from buffy coat collected using PUREGENE^® DNA Purification Kit during the baseline exam. Genotyping was performed by the Center for Inherited Disease Research (CIDR) using the Illumina Human1M-Duo BeadChip system. This platform includes 138 mtDNA SNPs including the majority of haplogroup-defining variants [95]. Samples were excluded from the dataset for the reasons of sample failure, genotypic sex mismatch, and first-degree relative of an included individual based on genotype data as previously described [96]. Genotyping of 138 mtDNA SNPs (including 137 variant sites) was successful for 1,631 unrelated individuals of European genetic ancestry. The major European haplogroups (H, V, J, T, U, K, I, W, and X) were defined using PhyloTree [97].

ntDNA sequencing

A total of 138 Health ABC participants of European ancestry that were part of an energetics sub study [98] were sequenced. MtDNA was extracted from platelets and sequenced with the Affymetrix Mitochondrial Resequencing Array 2.0 (MitoChip, Affymetrix, Santa Clara, CA)as previously described [99]. The MitoChip interrogates the forward and reverse strands of the 16.5 kb mitochondrial genome for a total of ∼30 kb sequence, enables the detection of known and novel mutations and has redundant probe tiling for detecting the major human mitochondrial haplotypes and known disease-related mutations. Built-in redundancy via independent probe sets also allows a test of within-chip reproducibility. Briefly, the entire mitochondrial genome was first amplified in two long-range PCR reactions using LAPCR Kit (Takara Bio U.S.A., Madison, WI) for each sample using two sets of overlapping primers. Mitochondrial fragments were amplified and prepared for array hybridization according to the Affymetrix protocol for GeneChip CustomSeq Resequencing Array. The resulting PCR products were assessed qualitatively by 1% agarose gel electrophoresis and purified using a Clonetech Clean-Up plate (Clonetech, Mountain View, CA). The purified DNA was quantified by PicoGreen and for selected samples, confirmed by NanoDrop measurements. The amplicons were pooled at equi-molar concentrations. Chemical fragmentation was performed and products were confirmed to be in the size range of 20–200 bp by 20% polyacrylamide gel electrophoresis with SYBR Gold staining. The IQ-EX control template, a 7.5 kb plasmid DNA, was used as a positive control. The samples were labeled with TdT and hybridized to the array in a 49°C rotating hybridization oven for 16 h. Finally, streptavidin phycoerythrin (SAPE) and then antibody staining was performed. The microarrays were processed in the GeneChip Fluidic Station and the GeneChip Scanner. Signal intensity data was output for all four nucleotides, permitting quantitative estimates of allelic contribution. The allelic contribution was assessed using the raw data from the individual signal intensities by deriving the ratio of expected allele (REA), which is the log ratio of the raw signal intensity of the expected allele at any site (as defined by the mtDNA reference sequence) to the average raw signal intensity of the other three alleles, at each site for every individual. DAT files with raw pixel data were generated and used as input for grid alignment. CEL files generated from DAT files were analyzed in batches using GSEQ. Samples with call rates of less than 95% were discarded. Ten samples were repeated for concordance testing. For samples passing initial filtering, ResqMi 1.2 [100] was used for re-analysis of bases originally called as “N” by GSEQ. Analysis was performed using custom Perl scripts. Data was extracted from gene regions as defined by NCBI annotations for the revised Cambridge Reference Sequence (rCRS; NC 012920.1).

Statistical analyses: Common mtDNA haplogroups and individual variants

We first analyzed differences in baseline 3MS and DSST scores and participant-specific slopes of 3MS and DSST scores for the European haplogroups and 137 mtDNA variants genotyped using the Illumina Human1M-Duo BeadChip system. The 3MS and DSST baseline and participant-specific slopes were compared between the common European haplogroup, H, and the remaining haplogroups (V, J, T, U, K, I, W, and X) using the generalized linear model. Unconditional logistic regression was used to obtain odds ratios (ORs) as estimates of relative risks (hereafter called risk) and 95% confidence intervals (CIs) for dementia involving haplogroups and common variants. Risk of dementia was examined for haplogroups V, J, T, U, K, I, W, and X as compared to the haplogroup H reference group. All analyses were adjusted for age and gender in simple models and age, gender, body mass index (BMI), diabetes, education, and APOEε4 allele carrier status in more inclusive models using SAS version 9.2 (SAS Institute Inc., Cary, NC). To avoid false positives due to population stratification, the 137 mtDNA variants were also adjusted for 6 eigenvectors of mitochondrial genetic ancestry derived from principal component analysis [101–105]. In our previous mtDNA sequencing work the first 6 eigenvectors have accounted for 71% of the variance in the mtDNA sequence dataset [96]. Mitochondrial PCA has been shown to outperform haplogroup-stratified or adjusted association analyses with no loss in power for the detection of true associations [105].

Statistical analyses: mtDNA sequences

Rare sequence variants and singletons unique to dementia cases were identified from the OXPHOS coding regions (both NS and synonymous [S]), ribosomal RNAs (rRNAs), transfer RNAs (tRNAs), and the hypervariable (HV) 2-3 regions. Several in-silico methods were employed to examine mtDNA nucleotide conservation (PhastCons [106] and PhyloP [107]) for all mtDNA variants and to predict the potential functional impact of NS substitutions on amino acid protein sequences (Sorting Intolerant From Tolerant (SIFT) [108, 109], MutPred [110], and PolyPhen2 [111]).

The joint effects of all mitochondrial variants within each gene or region on cognitive function test scores were evaluated using several rare variant burden tests. Pooled associations of all sequence variants were run using VT test [112] in R and included the T1 (1% MAF threshold) [113], T5 (5% MAF threshold) [113], WE (weighted-sum) [114], and VT (variable threshold) approaches [115]. All analyses were adjusted for age at baseline, gender, and study site using residuals from linear regression and then normalized to Z scores prior to conducting analyses. We applied these approaches to baseline 3MS and DSST scores and computed statistical significance for each test using 10,000 independent simulations. Variant aggregations were tested across the following regions: 1) the individual OXPHOS complexes; 2) all rRNAs combined; 3) all tRNAs combined; and 4) each of the HV 2-3 regions.

Results

Among 1,631 genotyped Health ABC participants of European ancestry, 273 (17%) developed dementia (Table 1). In general, dementia cases were more likely to have had diabetes and be APOEε4 allele carriers but there were no major differences in age, gender, BMI, and level of education (Table 1). Among the subset of 135 sequenced participants with high quality data, 22 (16%) developed dementia (Table 1). As with the larger genotyped sample, dementia cases were more likely to have had diabetes and be APOEε4 allele carriers and were also more likely to be female and less likely to have a postsecondary education (Table 1). Haplogroup frequencies are consistent with mtDNA sequencing performed by us [116] and others [95] and were largely similar between the genotyped and sequenced samples (Table 1).

Table 1. Characteristics of dementia cases and controls among genotyped and sequenced Health ABC participants.

	Genotyped (n = 1631)		Sequenced (n = 135)
	No dementia	Dementia	No dementia	Dementia
n (%)	1358 (83)	273 (17)	113 (84)	22 (16)
Age, mean (SD)	73.6 (2.8)	74.8 (2.8)	73.2 (2.8)	74.7 (3.0)
BMI (kg/m2), mean (SD)	26.7 (4.1)	26.2 (4.5)	26.8 (4.6)	25.8 (5.8)
APOEε4 carrier, n (%)	281 (21)	97 (36)†	28 (25)	7 (32)‡
Prevalent diabetes, n (%)	267 (18)	69 (26)	21 (19)	5 (23)
Gender, n (%)
Male	717 (53)	146 (53)	55 (48)	8 (38)
Female	641 (47)	127 (47)	59 (52)	13 (62)
Education, n (%)
Less than HS	160 (12)	35 (13)	15 (13)	4 (19)
HS grad	465 (34)	96 (35)	35 (31)	9 (43)
Postsecondary	732 (54)	142 (52)	63 (56)	8 (38)
Haplogroup, n* (%)
H	600 (44)	110 (40)	51 (45)	9 (41)
U	173 (13)	33 (12)	9 (8)	2 (9)
K	152 (11)	30 (11)	13 (11)	3 (14)
T	127 (9)	41 (15)	14 (12)	4 (18)
J	99 (8)	21 (8)	10 (9)	3 (14)
V	46 (3)	11 (4)	2 (2)	–
I	36 (3)	7 (3)	5 (4)	–
W	31 (2)	2 (1)	1 (1)	–
X	27 (2)	3 (1)	3 (3)	–

^*Numbers do not add up to total due to missing information for haplogroups.

^†ApoE4 frequency significantly differs between dementia cases and controls, Fisher's Exact Test p-value < 0.0001.

^‡ApoE4 frequency does not significantly differ between dementia cases and controls, Fisher's Exact Test p-value=0.60.

Common mtDNA haplogroups and individual variants

Risk of developing dementia among 8 European sub-haplogroups is reported in Table 2. Carriers of haplogroup T had a statistically significant increased risk of developing dementia compared with the most common European haplogroup H (OR = 1.86, 95% CI = 1.23, 2.82, p = 0.0008) after adjustment for multiple comparisons (8 haplogroups, critical α = 0.006). Haplogroup J was associated with a statistically significant decline in 3MS (Table 3) after adjustment for multiple comparisons (β = −0.14, 95% CI = −0.27, −0.03, p = 0.0006). Two genotyped mtDNA variants were associated with statistically significant declines in 3MS and DSST after adjustment for multiple comparisons (137 variants, critical α = 0.0004). The m.15244A>G, p.G166G, CytB variant was associated with a significant decline in DSST score (β = −0.58, 95% CI −0.89, −0.28, p = 0.00019). The m.14178T>C, p.I166V, ND6 variant was associated with a significant decline in 3MS score (β = −0.87, 95% CI −1.31, −3.86, p = 0.00012). Both of these variants were observed at very low frequencies (<1%) and occurred in independent samples: m.15244A>G, n = 4; m.14178T>C, n = 3. All statistically significant results were from analyses adjusted for age, gender, BMI, diabetes, education, APOEε4 allele carrier status, and, in the case of the individual variants, 6 eigenvectors of mitochondrial genetic ancestry. There were no associations between haplogroups or genotyped variants and baseline measures of 3MS or DSST.

Table 2. Odds Ratios (ORs) and 95% Confidence Intervals (CIs) for dementia associated with haplogroup N subgroups.

Haplogroup, n* (%)	No dementia n = 1358	Dementia n = 273	OR (95% CI)a	p-value	OR (95% CI)b	p-value
H	600 (44)	110 (40)	1.0 (Ref.)		1.0 (Ref.)
U	173 (13)	33 (12)	0.94 (0.61–1.43)	0.93	0.87 (0.56–1.35)	0.81
K	152 (11)	30 (11)	1.15 (0.73–1.79)	0.41	1.02 (0.64–1.64)	0.64
T	127 (9)	41 (15)	1.83 (1.22–2.76)	0.0015	1.86 (1.23–2.82)	0.0008
J	99 (8)	21 (8)	1.15 (0.69–1.91)	0.45	1.09 (0.64–1.84)	0.51
V	46 (3)	11 (4)	1.24 (0.62–2.48)	0.42	1.24 (0.61–2.53)	0.36
I	36 (3)	7 (3)	0.98 (0.42–2.27)	0.96	0.98 (0.42–2.28)	0.87
W	31 (2)	2 (1)	0.41 (0.10–1.74)	0.20	0.42 (0.10–1.82)	0.24
X	27 (2)	3 (1)	0.59 (0.17–2.00)	0.39	0.50 (0.14–1.75)	0.29

^aAdjusted for age and gender.

^bAdjusted for age, gender, BMI, diabetes, education, and APOEε4.

^*Numbers do not add up to total due to missing information for haplogroups.

Table 3. Rate of change over 8 years on the 3MS and DSST tests for haplogroup N subgroups.

Haplogroup	3MSa			DSSTa
	Baseline (SE)	Slope (95% CI)	p-valueb	Baseline (SE)	Slope (95% CI)	p-valueb
H	92.9 (0.20)	0.01 (0.02)	Ref.	40.7 (0.44)	−0.02 (0.02)	Ref.
U	92.5 (0.37)	−0.03 (0.03)	0.23	39.4 (0.8)	−0.04 (0.04)	0.66
K	92.1 (0.42)	−0.01 (0.03)	0.65	41.5 (0.9)	−0.03 (0.04)	0.84
T	92.8 (0.42)	−0.03 (0.03)	0.26	39.1 (0.9)	−0.04 (0.04)	0.60
J	93.4 (0.50)	−0.14 (0.04)	0.0006	39.5 (1.07)	−0.04 (0.05)	0.65
V	93.3 (0.71)	0.1 (0.06)	0.13	40.3 (1.52)	0.03 (0.07)	0.54
I	93.6 (0.84)	0.06 (0.07)	0.48	42.7 (1.82)	−0.22 (0.08)	0.02
W	92.9 (0.94)	0.0 (0.08)	0.86	42.1 (2.02)	−0.01 (0.09)	0.90
X	92.2 (0.99)	0.1 (0.08)	0.28	39.7 (2.11)	0.12 (0.10)	0.17

^aAdjusted for age, gender, BMI, diabetes, education, and APOEε4.

^bp-value for 8-year change as compared to haplogroup H.

mtDNA sequences

Of the 138 sequenced Health ABC samples, a total of 135 yielded sequence data of sufficient quality for analysis. Sequencing of 16,544 mtDNA bases (positions 12–16,555) from 135 participants yielded a cumulative total of 449 variants including: 56 common (MAF ≥5%), 160 low frequency variants (MAF 1–5%), and 233 singletons. The 10 duplicate samples had >98% sequence concordance (the majority of discordant calls resulted from positions successfully called in one but called as “N” in another). The within-chip error rate was 0.0028%, which is comparable to previously published rates of 0.0025% and 0.0021% [117, 118].

Among 22 participants with incident dementia, we identified 10 NS variants with some occurring at highly conserved sites predicted to affect protein structure/function (Table 4 and Supplementary Table 1; available online: http://www.j-alz.com/issues/32/vol32-2.html supplementarydata01). MutPred predicted a gain of acetylation for the ATP8 E52K substitution (p = 0.004). The CytB p.A191T and p.T194M substitutions occur in a potentially functionally relevant site known as the Qi binding pocket. Among non-coding variants unique to dementia cases were two HV2 (m.114, C>T; m.238, A>T), two 16S rRNA (m.1700, T>C; m.2141, T>C), and three tRNA (m.5527, A>G; m.5567, T>C; m.5592, A>G) variants. Nominally significant pooled effects across HV2 were observed for 3MS (p = 0.04, T1 method) and HV3 for DSST (p = 0.04, VT method), however these tests were not significant after multiple test correction for 8 mtDNA regions (critical α = 0.006).

Table 4.

Synonymous and nonsynonymous substitutions unique to Health ABC Study participants with dementia. Values in bold indicate sites that are predicted in-silico to significantly impact evolutionary conservation (PhastCons and PhyloP >0), protein stability (SIFT < 0.1) or protein function (PolyPhen ‘damaging’ prediction). Synonymous substitutions are indicated with the nucleotide change in lowercase and nonsynonymous substitutions are indicated with the amino acid change in uppercase

Complex	I																			III						IV								V
Gene		ND1		ND2					ND4L	ND4					ND5			ND6		CytB						COI					coII	coIII		ATP8		ATP6
Nt. position		3630c	3943a	4586t	4856t	4890t	5461c	5495t	10750a	10807a	10819a	10978a	11470a	11950a	12811t	12954t	13676a	14305g	14587t	14902c	14929c	15109t	15317g	15326a	15459c	5960c	6296c	6305g	7058t	7080t	7897g	9329g	9548g	8518a	8519g	8901a	9117t
	Haplogroup	108T	213I	39A	129L	141I	331A	342F	94N	16W	20K	73L	237K	397G	159Y	206A	447N	123S	129G	52A	61T	121L	191A	194T	238S	19Y	131P	134G	385A	393F	104W	41T	114G	151W	52E	125L	197I
AA Position
F, 70y	H	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·
F, 78y	J	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·
F, 72y	H	·	·	·	·	·	·	·	·	·	·	·	·	·	H	·	·	·	·	·	·	·	·	·	·	t	·	a	·	·	·	·	·	·	·	·	·
M, 69y	K	·	·	·	c	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	g	·
F, 71y	U	·	·	·	·	·	·	c	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	g	·	·	·
M, 73y	T	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	a	·	·	·	·	·	·	·	·	·	·	·	·	a	·	·	·	·	·	·
M, 77y	H	·	·	·	·	V	·	·	·	·	·	·	·	·	·	·	S	·	·	t	·	·	·	·	·	·	·	·	·	L	·	·	·	·	·	·	·
M, 79y	U	·	·	·	·	·	·	·	·	g	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	a	·	·	·	·
F, 79y	H	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·
F, 74y	H	g	V	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·
F, 77y	T	·	·	·	·	·	·	·	S	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·
F, 73y	H	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	A	·	·	·	·	·	·	·	·	·	·	·	·	·
F, 74y	I	·	·	·	·	·	·	·	·	·	g	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	K	·	·
F, 79y	J	·	·	·	·	·	D	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·
M, 76y	K	·	·	·	·	·	·	·	·	·	·	g	g	·	·	c	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·
M, 76y	H	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	t	·	a	·	·	·	·	a	·	·	·	·	·
F, 75y	K	·	·	·	·	·	·	·	·	·	·	·	·	g	·	·	·	·	·	·	t	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·
M, 73y	T	·	·	C	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	c
F, 72y	H	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	c	·	·	·	·	·	·	·	·	·	c	·	·	·	·	·	·	·	·
M, 76y	H	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	a	·	·	·	·	·	·	·	·	·	·
F, 75y	T	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	c	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·
F, 72y	J	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	T	·	·	·	·	·	·	·	·	·	·	·	·	·	·
PhastCons		0.13	0.97	0.00	0.00	0.00	0.00	0.00	1.45	0.16	0.00	0.00	1.00	0.00	0.00	0.00	0.11	0.00	0.00	0.02	0.24	0.00	0.67	0.00	0.00	0.00	0.00	0.00	0.00	0.89	1.00	0.00	0.00	0.99	0.41	0.67	0.00
PhyloP		−0.84	4.00	−3.43	−3.05	−0.07	0.20	−0.06	0.98	0.33	−6.64	−8.51	0.93	−1.18	−0.08	−2.44	1.23	−0.38	−2.19	−6.14	−0.83	−4.66	3.85	−1.08	−0.18	−1.47	−5.51	−6.14	−4.40	2.43	0.67	−6.23	−6.52	4.55	0.54	−0.21	−5.48
SIFT			1.00			1.00	0.33								0.71		0.05						0.11	0.83						0.25
PolyPhen			0.00			0.82	0.00								0.00		0.09						0.00	0.00						0.00

PhastCons and PhyloP identify evolutionarily conserved DNA elements in a multiple-species sequence alignment.

SIFT (Sorting Intolerant from Tolerant) and PolyPhen (Polymorphism Phenotyping) predict whether an amino acid substitution affects protein function.

Discussion

In this study, we genotyped and sequenced the entire mtDNA in a large, population-based, longitudinal study of elderly participants to examine the role of common mtDNA haplogroups and genetic variants, rare sequence variants, and aggregate mtDNA genomic variation in determining dementia risk and cognitive decline. Among the major mtDNA haplogroups, we observed that haplogroup T participants were at a significantly increased risk of developing dementia and haplogroup J participants experienced a significant 8-year decline in 3MS, both compared to haplogroup H participants. We did not observe an increased risk of dementia among the haplogroup J participants. In two previous studies, European mtDNA haplogroups [119] and maternal lineages in the Old Order Amish [120] were not associated with an increased risk of AD. Determining which variants contribute to the associations observed for haplogroups J and K is complicated by the number of variants which are diagnostic for these haplogroups. Haplogroup T is defined by variants in 12S, 16S, tRNA Arg, tRNA Thr, ND2, ND5, CytB, and ATP6 [97]. Only one of these variants has apparent functional potential: m.4917A>G, which encodes amino acid substitution p.N150D, ND2. Haplogroup J is defined by variants in ND5 and HV2 [97]. The HV2 variant that defines haplogroup J, m.295C>T, has been shown to change mitochondrial transcription and copy number [121]. Cytoplasmic hybrids (cybrids) containing haplogroup J mtDNA had a greater than 2-fold increase in mtDNA copy number compared with cybrids containing haplogroup H mtDNA. This is one of the few examples demonstrating functional consequences for a variant underlying a specific haplogroup and is of particular interest since haplogroup J is over-represented in long-lived people and centenarians from several populations [122–124]. In this case, the impact of the haplogroup J regulatory region mutation on mtDNA replication or stability may partially account for population-based observations that haplogroup J is associated with human longevity. However, m.295C>T may not be the only potentially functional haplogroup J defining variant since m.13708G>A, which is also diagnostic for haplogroup J, encodes an amino acid substitution: p.A458T, ND5. There is a clear incongruity between our results suggesting that haplogroup J participants experience a significant 8-year decline in 3MS and other work demonstrating the over-representation of haplogroup J among centenarians and long-lived people [122–124]. However, earlier studies demonstrating associations between haplogroup J and lifespan did not examine other indices of health. Our haplogroup J results suggest that extending longevity-related genetic findings into specific age-related phenotypes (e.g., longitudinal change in cognitive function) is critical for revealing the roles of both mtDNA sequence variation and mitochondrial function in human aging.

Research to identify genetic factors that contribute to complex phenotypes should be sensitive to the ways that genes and genetic perturbations operate. For example, it is now widely recognized that common genetic variants play a much smaller role in mediating phenotypic expression and disease risk than initially thought [125–128] and that identification of causative variants requires comprehensive resequencing of genomic loci in multiple subjects [129]. It is notable that of the 137 common and rare mtDNA variants genotyped in 1,631 participants, the two variants associated with statistically significant declines in 3MS (m.14178T>C) and DSST (m.15244A>G) were observed at very low frequencies. This lends further support to the hypothesis that most of the mtDNA variation that impacts function is rare in frequency.

We further examined the role of mtDNA sequence variation in cognitive function and dementia risk by sequencing the entire mtDNA and identified several highly conserved and potentially functional variants that were unique to dementia cases. Of particular interest are the CytB p.A191T and p.T194M substitutions located in the complex III Qi binding pocket, where quinone is reduced by cytochrome b [130]. We have shown that these substitutions are found in Health ABC participants in the lowest extreme of free-living activity energy expenditure [131]. The p.T194M, CytB variant occurs at a residue that is noted to undergo significant conformational changes upon contact with antimycin A, a pharmacological inhibitor of the Qi site [130]. In the presence of antimycin A, complex III produces high quantities of superoxide indicating that inhibition at this site blocks electron transfer (from cytochrome b to quinone at Qi) causing a buildup of semiquinone at the Qo site. This buildup results in increased ROS production from complex III [65]. Free radicals (compounds with an unpaired electron) or ROS are a normal part of metabolism. As part of their role in normal metabolism, ROS serve as signaling molecules [132]. ROS communicate between metabolic pathways and have been shown to modify neurotransmitter receptors and ion transporters [133]. Oxidative stress occurs under conditions in which production of free radicals or ROS exceeds the ability of the cell to buffer them. Since the electron transport chain uses most of the oxygen in neurons, any interruption of its normal function increases ROS production and oxidative stress. The excess oxidative stress can modify DNA, RNA, lipids, and proteins.

The gain of lysine p.E52K, ATP8 can result in a new target for a diverse array of post-translational modifications which can impact protein function, including acetylation [134]. For example, the addition of an acetyl group to a lysine residue alters the positive charge of the ε-amino group and creates a region of local hydrophobicity. This substitution may facilitate acetylation-directed molecular interactions and in this case impact ATP8 activity and stability [135–137]. Two additional dementia-related variants include p.Y159H, ND5 which is a possible risk factors for Leber's Hereditary Optic Neuropathy [138, 139], and p.A331D, ND2 which was previously been associated with AD [70], although the role of this variant remains unclear [80, 81]. Several variants in the tRNA, rRNA, and HV2 regions were unique to dementia cases. The mitochondrial tRNAs and rRNAs are critical for protein synthesis and mitochondrial assembly and the HV2 region includes the priming site for mtDNA replication and the heavy-strand origin encoding 12 of the 13 OXPHOS genes [140].

Because collections of variants within genes or genomic regions do not work in isolation and are likely to influence phenotypic expression in important ways [128, 141], we considered how multiple sequence variants influence cognitive function and identified aggregated variation impacting the 3MS and DSST. Analytic approaches testing the combined effect of multiple variants have been used to resolve genetic associations for several complex traits [142–145] including the role of rare mitochondrial variants in disease [146]. Variant pooling effects suggest rare HV2 and HV3 sequence variation is associated with 3MS and DSST, respectively. It is not clear how HV2 and HV3 sequence variation impact cognitive function, however HV2 and HV3 are involved in regulating mtDNA copy number [121, 140]. As previously described for haplogroup J, cybrid cells carrying the m.295C>T variant allele have increased mtDNA transcription levels and a 2-fold increase in mtDNA copy number compared with cells carrying the more common m.295C>T allele [121]. Mitochondrial DNA copy number in peripheral blood has been inversely correlated with insoluble Aβ₄₀ and Aβ₄₂ levels [5] and is positively correlated with cognitive function in healthy elderly women [147].

This study had a number of strengths: a large sample size for assessing dementia risk and change in cognitive function among mtDNA haplogroups and genotyped variants; a well-characterized population-based longitudinal cohort with longitudinal assessment of 3MS and DSST; complete mtDNA sequencing allowing for an unbiased assessment of mitochondrial genomic variation; an analytic approach including both individual and pooled sequence variants; and in-silico prediction and structural modeling that enabled detailed interpretation of sequence-based findings. Small sample size for sequence-based analyses and limited power to detect individual variant effects of rare variants are acknowledged. It is possible that the mtDNA variants identified in this study may not be causally related to cognitive decline or dementia thus the lack of a replication cohort is also a limitation. We also do not have information on dementia subtypes. The complex mitochondrial genetic architecture suggests that there might be a complex set of gene interactions (epistasis) involving genetic variation in the nuclear and mitochondrial genomes [148–154] and future studies of mitochondrial genetic variation will therefore need to account for a complex set of interactions involving both genomes [155]. Indeed, mitochondrial-nuclear epistasis has important fitness effects [148, 149, 155–160] and is evolutionarily important [151].

These results help to uncover specific mtDNA sequence variants that suggest mitochondrial functions which may impact dementia risk and differences in longitudinal changes in cognitive function. The 13 mtDNA-encoded OXPHOS genes are essential to mitochondrial energy production [48] and sequence variation may impact superoxide production at OXPHOS complexes I and III, ATP generation efficiency through respiratory chain impairment [64, 65], apoptosis [66], and ATP supply [67]. Individual and collective variation in the HV2, HV3, tRNA and rRNA regions may affect mitochondrial function by impacting the rate or efficiency of mitochondrial biogenesis (increase in mitochondrial number and/or mass). An important aspect of mitochondrial biogenesis is turnover rate, which is thought to decline with age [161]. Impaired ability to turnover may allow the accumulation of defective mitochondria resulting from oxidative injury [37, 38], Aβ accumulation [36, 39], and/or mtDNA damage [68] in neurons leading to impaired respiratory capacity [40]. Several lines of evidence show that mitochondrial biogenesis is affected by pharmacologic agents [162–167], natural compounds [168], and behavioral interventions such as caloric restriction and exercise [169-172]. By identifying mitochondrial genetic variants that influence cognitive function and dementia risk we provide evidence that additional molecular mechanisms (e.g., mitochondrial protein synthesis and assembly) or targets (e.g., Qi binding pocket of complex III) could be involved. Such findings might lead to the development of interventions or new clinical strategies for improving mitochondrial function and delaying the onset of cognitive decline.