Integrative Blood-Based Characterization of Oxidative Mitochondrial DNA Damage Variants Implicates Mexican Americans’ Metabolic Risk for Developing Alzheimer’s Disease

Alzheimer’s Disease (AD) continues to be a leading cause of death in the US. As the US aging population (ages 65+) expands, the impact will disproportionately affect vulnerable populations, e.g., Hispanic/Latinx population, due to their AD-related health disparities. Age-related regression in mitochondrial activity and ethnic-specific differences in metabolic burden could potentially explain in part the racial/ethnic distinctions in etiology that exist for AD. Oxidation of guanine (G) to 8-oxo-guanine (8oxoG) is a prevalent lesion and an indicator of oxidative stress and mitochondrial dysfunction. Damaged mtDNA (8oxoG) can serve as an important marker of age-related systemic metabolic dysfunction and upon release into peripheral circulation may exacerbate pathophysiology contributing to AD development and/or progression. Analyzing blood samples from Mexican American (MA) and non-Hispanic White (NHW) participants enrolled in the Texas Alzheimer’s Research & Care Consortium, we used blood-based measurements of 8oxoG from both buffy coat PBMCs and plasma to determine associations with population, sex, type-2 diabetes, and AD risk. Our results show that 8oxoG levels in both buffy coat and plasma were significantly associated with population, sex, years of education, and reveal a potential association with AD. Furthermore, MAs are significantly burdened by mtDNA oxidative damage in both blood fractions, which may contribute to their metabolic vulnerability to developing AD.

Identifying blood-based biomarkers capable of predicting disease onset and assessing disease progression (e.g., preclinical AD, mild cognitive impairment [MCI], and AD) is of great importance to understand the pathophysiological heterogeneity of AD which will inform ultimately more precise therapeutics.
Type-2 diabetes (T2D) is a considerable risk factor for AD due to its comorbid association with CI; however, the precise pathophysiological mechanisms connecting these two complex diseases are unclear.As the aging Hispanic/Latinx population is expected to exponentially increase compared to other ethnic/racial groups, the healthcare burden affecting this population is expected to worsen due to age-related diseases (e.g., AD and diabetes) 1,9 .It has been previously reported that the Ser(326)Cys polymorphism (rs1052133) in 8-Oxoguanine glycosylase (OGG1), an important glycosylase of the base excision repair pathway involved in the recognition and excision of 8oxoG within DNA, delays repair of oxidative DNA damage 10 and is associated with T2D risk in MAs 13 .Recent observations suggest OGG1 regulates cellular energy metabolism and thus shapes metabolic phenotypes during high-fat diet exposure, indicating DNA damage repair may be important to metabolic health [11][12][13] .Studies by Komakula et al., and Sampath et al., revealed that functional OGG1 prevented obesity and metabolic dysfunction upon induction in response to a high-fat diet potentially via direct or indirect expression changes of PGC-1a and fatty acid oxidation 11,12 .Similarly, reduced expression of PGC-1a is consistently observed in T2D patients 14,15 and is linked to elevated ROS levels and decreased levels of b-oxidation enzymes 16 .AD disproportionately affects Hispanics/Latinos compared to NHWs; this is thought to be due to the increased prevalence of metabolic syndrome, obesity, cardiovascular health risks, and diabetes 9,17,18 .These observations combined with our previous data indicating MAs have elevated levels of oxidative damage compared to NHWs, leads us to the premise that mitochondrial health may hold greater biological importance in the development of age-related disease for individuals with Hispanic/Latinx ancestry 19 .
Oxidation of guanine possesses unique mutagenic properties that when left unrepaired, can perturb cellular function through several mechanisms and effect protein-DNA binding (e.g., transcription factors) 21,22,25,26 .Furthermore, oxidized guanine can result in missense mutations and often modify the activity of downstream products, such as RNAs and proteins 20-22, 25,27 .Due to the nature of the mitochondrial genome within various cell types and tissues, oxidative DNA damage will in uence distinctive pathophysiological outcomes in tissue with differing function depending on the location, metabolic activity, and enzymatic processes 26 .In response to mitochondrial damage there are QC processes at the cellular, organellar, and molecular level that are employed to preserve mitochondrial integrity 28 .As mitochondria work in dynamic networks, fusion with nearby mitochondria can recover function; however, excessive damage undermines fusion activity resulting in fragmented mitochondria that are removed through mitophagy and/or apoptosis 28 .
Growing evidence implicates oxidative DNA damage as a primary and secondary contributor to pathology observed in the AD continuum [29][30][31][32] .Recent evidence from our lab analyzing blood-based indices of mtDNA copy number (CN) and cell-free mtDNA (cf-mtDNA) to investigate mitochondrial dysfunction in complex disease (T2D and CI) among MAs showed that mtDNA CN was signi cantly associated with both T2D and CI 18 .Also, cf-mtDNA was found to be higher in individuals with either disease, reaching signi cant levels in individuals with both diseases compared to normal controls 18 .Cellular mtDNA CN is an indicator of mitochondrial biogenesis and cellular energetics 18,33 , which can be used as a measurement of mitochondrial health 34 .
Circulating cell-free mitochondrial DNA (ccf-mtDNA) has been increasingly studied as a biomarker for systemic in ammation during cellular stress or apoptosis.In this process, mtDNA fragments are released by cells into the bloodstream, where their bacterial origins cause them to be recognized as a damage associated molecular pattern (DAMP), thereby eliciting an in ammatory response through activation of innate immune cells 18, [34][35][36] .There are numerous studies assessing ccf-mtDNA as a clinical diagnostic and predictive biomarker (e.g., in affective disorders, and mitochondrial, autoimmune, neurological, and cardiovascular diseases) 34,35,[37][38][39][40] , and accumulating evidence indicates that mitochondrial dysfunction and mtDNA damage are correlated with disease severity and levels of ccf-mtDNA 35,39 .
Our mtDNA CN and cf-mtDNA data can be considered as a proxy for our more recent data evaluating mitochondrial 8oxoG variant load since it does not capture if mutation has occurred.More recently in a similar population-based cohort, we discovered that mtDNA variants indicative of 8oxoG were signi cantly elevated in MAs compared to NHWs and was associated with sex, education, and suggestive for cognitive function 19 .Correspondingly, Miller, et al., revealed AD neurons compared to age-matched controls had signi cantly elevated levels of somatic single nucleotide variants (sSNVs) than anticipated when considering sSNVs are known to increase with age 32 , and the distribution of the variants are presumed to occur secondary to developing disease pathology 32 .Further analysis established potential mechanisms of oxidative DNA damage developed from 8oxoG (nonsynonymous mutations, e.g., C > A) that might contribute to the signi cant increase of sSNVs in AD, especially in protein-coding genes 32 .The increased substitution mutations could be a result of increased ROS and OS, a common feature observed in AD brains, the CNS, and periphery, which can contribute to in ammation and mitochondrial dysfunction 32,41 .This accumulating evidence may point at a mutational signature important to AD pathophysiology in uencing the differing endophenotype reported in AD, particularly of those with different mitochondrial capacity, metabolic health, and comorbidity risk such as those linked to ethnic/racial AD health disparities.
Oxidative transversion substitution mutations may be associated with increased mitochondrial dysfunction due to impaired mitochondrial capacity, metabolic health, and/or lifestyle affecting mitochondrial health, and contribute to the continuous progression of AD until a clinical endpoint, death.
In this current study, our objective was to use a blood-based measurement of 8oxoG sSNVs as an indicator of impaired mitochondrial function to investigate the role of mitochondria in pathophysiology of complex disease by (1) characterizing associations to population and sex, (2) highlighting burdened genomic regions that in uence mitochondrial function, and (3) determining differences in buffy coat or plasma on evaluating AD risk and/or endophenotype.

Results
Descriptive statistics for the cohort analyzed for cellular mitochondrial 8oxoG variants are displayed in Table 1.As anticipated, MMSE, CDR sum, and years of education in both populations had signi cantly different means based on cognitive status.Age signi cantly differed in MAs between cognitive groups and years of education was lower compared to NHWs.  3 and 4).

Evaluation of 8oxoG Variant Count in the Buffy Coat of MA and NHW TARCC Participants
In the MA population 8oxoG variant count was signi cantly reduced for subjects reporting depression compared to those without depression; mean = 6.548 and 7.704, respectively (Supplemental Fig. 1).Tobacco abuse demonstrated an approach for signi cance in association with 8oxoG demonstrating a higher variant load compared to non-smokers in MAs; mean = 7.935 and 7.048, respectively (Supplemental Fig. 2).These trends were not observed in the NHW cohort.
Multiple linear regression model predictions in the whole cohort were performed to assess the associations with 8oxoG and to determine if there are predictive interactions.Sex regarding females (p = 0.0007), years of education (p = 0.0055), BMI (p = 0.0288), and tobacco abuse (p = 0.0086) were signi cantly associated with 8oxoG variant count, and the population-sex interaction demonstrated a signi cant interaction effect (p = 0.0038) (Supplementary Table 5).Further analysis of 8oxoG variant count in both population and sex via two-way ANOVA indicates population is signi cantly associated (p < 0.0001), while sex was marginally signi cant (p = 0.0922).
In the subsequent multiple linear regression model, a diabetes × cognition interaction was evaluated and showed a signi cant association with population (p < 0.0001), sex (p = 0.0429), years of education (p = 0.0109), BMI (p = 0.0254), and tobacco abuse (p = 0.0155); although the interaction effect was not signi cant (Supplementary Table 6).Cognitive status with respect to AD displayed a suggestive association compared to controls (p = 0.0556).
Previously derived 8oxoG variant load for each subject corresponding to 8oxoG "hotspots" 19 were further analyzed via multiple linear regression prediction models assessing a population × sex and years of education interaction effect, and a diabetes × cognition interaction effect was less informative than analyzing total 8oxoG variant count (Supplementary Tables 7 and 8).Population strati cation lost signi cant statistical associations (Supplementary Tables 9 and 10) that were observed in Supplementary Tables 7 and 8.
Multiple linear regression modelling in MAs indicated that total 8oxoG was signi cantly associated with cognitive status, sex, years of education, and tobacco abuse (Table 2).BMI did not show signi cant association in MAs as compared to the regression models investigating interactive effects (Supplementary Tables 5-6).Modelling within NHWs did not demonstrate any associations (Table 3).Additional prediction modelling used cognitive status as a binary variable to combine the effects of AD and MCI compared to NCs and showed similar results to the models with greater resolution on cognitive status (Supplementary Tables 11-18).

Assessment of ccf-mtDNA 8oxoG Variant Count in MA and NHW TARCC Participants
The subset of participants included for the ccf-mtDNA 8oxoG variants were selected from subjects included in the buffy coat analysis to compare the blood fractions collected from the same visit (Table 4).
Age, sex, and years of education were considered confounding variables for CI and were utilized with the aim to pairwise match AD with NCs to help reduce the risk of confounders in uencing false associations to AD due to the smaller sample size.and 20).Testing for Hardy-Weinberg Equilibrium for APOE and OGG1 were insigni cant for the total cohort and within each population, indicating that the genotype frequencies are in equilibrium (Supplementary Tables 21 and 22).
Although our attempt to match samples based on age was unsuccessful; a Pearson's correlation was performed to exclude age as a potential cofounder and showed age does not need to be considered a covariate in our dataset as it was not correlated with total ccf-mtDNA 8oxoG variant count (Supplementary Fig. 3).

MAs, especially females, have a greater burden in total ccf-8oxoG variant count
In the whole cohort ccf-8oxoG variant count was signi cantly elevated in the MA population compared to NHWs; mean = 0.8500 and 0.7160, respectively (Fig. 1).Ccf-8oxoG variant count did not signi cantly differ based on cognitive status or sex (Supplementary Fig. 4 and Supplementary Fig. 5).A signi cant population ⋅ sex interaction was not observed (Fig. 2); however, ccf-8oxoG variant count was signi cantly elevated in MA females compared to NHW females (mean = 0.8702 and 0.7771, respectively).Despite cognitive phenotype, MAs had an elevated 8oxoG sSNV burden compared to NHWs (Fig. 3), yet there was not a signi cant difference for 8oxoG variant count in each population when assessing for sex and cognitive phenotype (Fig. 4).
Results of multiple linear regression modelling of ccf-mtDNA 8oxoG variant count in the whole cohort with respect to population-sex and education interactive effects while considering APOE status as a dosage effect (i.e., no ε4 allele, one ε4 allele, two ε4 alleles) indicated a signi cant positive association for APOE and the population ⋅ sex interaction (Supplementary Table 23).When assessing for a diabetes × cognition interaction, a signi cant association between ccf-8oxoG variant count and both population and APOE was determined (Supplementary Table 24).
Additional multiple linear regression models were performed to characterize associations with 8oxoG "hotspots" in the whole cohort (Supplementary Fig. 6 and Supplementary Tables 25-26).The regression modelling for both interaction effects, population × sex and education, and diabetes × cognition in the whole cohort did not observe the same associations for 8oxoG "hotspots", indicating these variants may not be informative in this context.

Population-speci c associations to ccf-8oxoG variant count
Population strati cation for the multiple linear regression models did not demonstrate any statistical signi cance for ccf-8oxoG count in MAs (Table 5); however, in the NHW population AD and diabetes was marginally signi cant (Table 6).Further, the model showed signi cant statistical association with sex and age, as well.Multiple linear regression models for 8oxoG "hotspots" in each population were performed and no associations were observed in the MA population; however, in the NHW population a signi cant negative association to age was discerned (Supplementary Tables 27-28).Although the signi cant association to age in the NHW population was previously observed in the non-"hotspot" strati ed regression model, generally the "hotspot" strati ed analyses were less informative.

Discussion
Ethnic/racial differences in developing cognitive impairment are known to exist, yet reports investigating biological, behavioral, and lifestyle factors that lead to differential mechanisms of neurodegeneration in populations more heavily burdened by cognitive decline are limited.Here, we investigated the predictability of 8oxoG variant count from two blood fractions in assessing risk for cognitive decline in two populations.Additionally, haplogroup-speci c associations to ccf-mtDNA 8oxoG variants were performed.We hypothesized that indirectly evaluating mitochondrial dysfunction through cellular mitochondrial 8oxoG sSNVs may serve as a better biomarker for MAs due to their observed metabolic endophenotype and burden, while ccf-mtDNA 8oxoG variants may provide improved utility as a biomarker for NHWs due to their in ammatory endophenotype.
Altogether, our results con rm MAs, especially females, show greater mtDNA oxidative damage compared to NHWs.Tobacco abuse trended for signi cance with increasing cellular 8oxoG mutational load in MAs; however, interestingly, non-depressed individuals showed elevated cellular 8oxoG variant burden.
Strati ed regression analysis by population in buffy coat PBMCs demonstrated an association with AD, sex, education, and tobacco abuse, while this was not observed in NHWs.Similarly, ccf-8oxoG variant load was signi cantly higher in MAs, and MA females had elevated levels compared to males.Ccf-8oxoG strati ed regression analysis for NHWs showed a suggestive association with normal cognition in younger aged males without diabetes.This may indicate that oxidative variants from ccf-mtDNA are reduced in older NHW females with AD and diabetes comorbidity; however, the potential reason for this is unclear.There is growing evidence indicating the protective effects of estrogen against OS related damage 42,43 .Estrogen de ciency post menopause is associated with increased levels of OS, higher blood levels of free fatty acids, and reduced antioxidant defense 42,43 , so it is expected that aging females would demonstrate elevated levels of oxidative damage.Nonetheless, due to our observation it begs the question if there are sex-related differences for the age-related decline in mitophagy.
Depression is a known risk factor for developing MCI and AD, and a study demonstrated a depressive endophenotype of MCI and AD in MAs 44 .Furthermore, numerous studies report MAs experience more depressive symptoms compared to other Hispanic/Latino subpopulations 45,46 , as well as NHWs [47][48][49][50][51][52] .
Accumulating evidence indicates that individuals with depression have higher levels of 8oxoG and oxidative damage, as OS encompasses a critical role in depression pathophysiology through the activity of ROS 53,54 .Substantial evidence indicates a higher prevalence of depression and depressive symptoms among MAs, yet our data denoted that non-depressed MAs exhibited elevated levels of 8oxoG variants in buffy coat PBMCs.However, depression appeared to be underrepresented because of limited reports of depression in MAs with MCI and AD.Subsequent regression models in the whole cohort and MA population show a negative trend between 8oxoG variant count in buffy PBMCs and non-depressed MAs.
Surveying for the presence or absence of depression may have poor resolution when investigating cognitive associations compared to assessing for a collection of depressive symptoms.Previous studies report distinct clustering of depressive symptoms is imperative when studying the connection between cognition and depression 55,56 .Our results for 8oxoG and depression among MAs warrants further investigation by implementing depressive symptoms and/or other indicators of depression.
Our results revealed increased levels of 8oxoG variants in buffy PMBCs of MAs with a history of tobacco abuse.Smoking tobacco and exposure to tobacco smoke has been shown to cause elevated levels of 8oxoG compared to non-smokers [57][58][59][60] because of the various carcinogens contained within 61,62 .As previously mentioned, carcinogens readily form DNA adducts and can lead to OS through the production of ROS.Additionally, cigarette smoke has been recognized to cause chronic in ammation leading to increasing OS which further results in accumulating oxidative damage 61,63 .A recent study reported that smoking tobacco increased risk for cognitive decline in aging MAs 64 .Following linear regression models including MAs all demonstrated a signi cant link with tobacco abuse.These results seem to indicate tobacco abuse as a strong modi able risk factor for increased mitochondrial oxidative damage in MAs and demonstrates the importance of addressing such behaviors to prevent increased risk for CI in this population.
Our higher resolution linear regression models in the whole cohort from buffy coat PBMCs established that population (MA), tobacco abuse, BMI, education, and population × sex were statistically associated with 8oxoG somatic variants.AD was suggestively associated with cellular 8oxoG sSNVs when assessing an interaction between diabetes and cognitive status.These results were not observed in the NHW population, and perhaps of interest is the fact that many of the coe cients were in the opposite direction (though not signi cant).Increasing evidence connects AD and T2D, showing a greater risk for cognitive decline due to T2D, and robust correlations indicate high blood sugar is associated with the presence of Ab plaques 65 .Brain dysfunction is frequently observed in earlier stages of T2D, and hemoglobin A1C (established biomarker for T2D) has been related to decline in functional memory and hippocampal size 65 .Links between T2D and AD implicate mitochondrial dysfunction as a participating factor in the development and/or progression of neurodegeneration and may be of exceptional importance for ethnic/racial differences in disease severity and manifestation.This evidence further suggests that mitochondrial health could be a contributor to the unexplained disparities in CI among MAs, especially since MAs are at great risk for metabolic disorders.
It is important to note that this study has its limitations including (1) measuring 8oxoG lesions indirectly, (2) small sampling for the plasma dataset, (3) lacking biochemical, metabolic, and in ammatory phenotypes, (4) missing most nDNA variants, (5) solely sampling blood tissue, (6) evaluating in one cohort, and (7) APOE in the NHW model could have reduced power due to a larger number of missingness compared to MAs.To better understand the biological and mechanistic roles of mitochondrial dysfunction and oxidative DNA damage in this context, future studies should incorporate larger sampling, include more biological markers indicative of metabolic health and systemic in ammation, determine if utilizing MMSE, CDR sum, and/or other neuropsychological tests for cognitive function strengthens our power to further support/validate our results, and characterize the nuclear genetic background associated with the mtDNA for each subject.Future studies should also investigate the mechanistic role of mitochondrial processes, such as mitophagy and mitochondrial quality control and sensing, play in contributing to the observed differences in cellular and ccf-mtDNA 8oxoG sSNVs between MAs and NHWs in relation to cognitive decline.Additionally, future studies will aim to validate the applicability of peripheral cellular and cell-free pathophysiological phenotypes as biomarkers for assessing brain pathology, disease risk, and/or disease stage.These studies will also investigate expression of DNA repair machinery, the role of sex hormones, and validate oxidative mtDNA load using alternative methods.
Uniquely, our data speci cally point to novel, population-based effects in 8oxoG damage in cellular and cell-free mtDNA.Overall, our results indicate both cellular and ccf-8oxoG variants mtDNA are signi cantly elevated in MAs compared to NHWs and indicate sex-differences (Fig. 5).Notably, our results from evaluating oxidized cellular mtDNA compared to ccf-mtDNA presents the possibility that these biomarkers may have signi cant predictive capability in MAs, especially females, compared to NHWs due to the observed statistical signi cance of assessing various independent variables.This evidence implies mitochondrial dysfunction in cellular mtDNA may be distinctly related to disease pathology in MAs with cognitive decline, whereas in ccf-mtDNA displayed poor associations in MAs compared to NHWs.This cumulative evidence supports the notion that (1) blood-based signatures of mitochondrial dysfunction differ between ethnic/racial populations, (2) cellular and ccf-mtDNA possess different functionality in potentially developing pathophysiological conditions, and (3) ethnic/racial differences exist in the manifestation of neurodegeneration through the assessment of mitochondrial oxidative DNA damage from different blood fractions.4) collected at the same visit were selected to match the distribution of subjects with respect to age, sex, and type-2 diabetes among both populations.The plasma subset did not include individuals diagnosed with MCI.These samples were analyzed to characterize cellular and circulating cell-free mtDNA (ccf-mtDNA) oxidative damage from blood.

Sample Acquisition and Description
Measurement of mtDNA Mutational Load from Buffy Coat and Plasma artifact generation and detection 74 .A text le was generated for each sample and were subjected to manual review to exclude technical oxidative artifacts with a Phred score below 30.
Identi cation of Variants Indicative of Oxidative Damage: The variant call les were manually assessed to identify 8oxoG transversions within the mitochondrial genome.The process of identifying these speci c oxidative transversions has been previously described 19 .Variants indicative of 8oxoG damage for each subject from the buffy coat portion were summed and normalized by accounting for read depth (variant count per 1000 read depth) to evaluate group differences based on the following variables: population, cognition, sex, type-2 diabetes, comorbidity (cognitive impairment and diabetes), both APOE and OGG1 genotype, and lifestyle factors.Variants indicative of 8oxoG damage from plasma was summed and normalized by accounting for read depth and were log10 transformed to evaluate group differences with the same variables described above.8oxoG variant "hotspots" were identi ed as variant locations with at least 25 participants in the cohort observed.27.0), R software (v.4.2.0), and GraphPad Prism software (v.9.4.0).Welch's t-test (two-tailed) and twoway ANOVA were performed on 8oxoG mutational load to compare between both population groups.
Multiple linear regression analysis was performed to evaluate the relationship between cognition, sex, age, education, diabetes and depression status, and tobacco abuse with 8oxoG variant count both within the whole study cohort and in strati ed analyses of MAs and NHWs.
Cohort: The Texas Alzheimer's Research and Care Consortium (TARCC) is a population-based collaborative longitudinal research initiative that has expanded between several Texas medical research institutions 66 .TARCC explores factors that may attribute to the development and progression of cognitive impairment due to AD in the MA population compared to their NHW counterparts.Participants: The study received institutional review board approval under the University of North Texas Health Science Center IRB #1330309-1 and all experiments were performed in accordance with relevant guidelines and regulations.Informed written consent was obtained from participants and/or their legally authorized proxies to take part in the study and allow publication of ndings before data collection.Volunteer aging participants enrolled in TARCC annually complete a medical evaluation, clinical interview, neuropsychological testing, and blood draw.Eligible participants obtained categorical clinical diagnoses of 'Alzheimer's disease', 'Mild Cognitive Impairment', and 'Normal Control' based on the criteria provided by the National Institute for Neurological Communicative Disorders and Stroke-Alzheimer's Disease and Related Disorders Association 67 .Additional information regarding the inclusion and exclusionary criteria of TARCC has been discussed elsewhere 44 .This study included NHW and MA subjects (N = 559; Table 1) diagnosed with Alzheimer's Disease (n = 104), Mild Cognitive Impairment (n = 127), or normal cognition (n = 328).Obtained buffy coat samples from NHWs (n = 261) and MAs (n = 299) and a subset of plasma samples from 62 NHWs and 57 MAs (N = 119; Table APOE and OGG1 Genotyping Imputation: Genome-wide SNP pro les were generated using the Illumina In nium Multi-Ethnic Global Array which types 1.7 million SNPs.Standard ltering based on SNP missingness, individual missingness, and minor allele frequency (5%) was conducted according to Anderson et al., 201075 .Genetic imputation of APOE (rs7412 and rs429358 for individuals missing APOE genotypes) and OGG1 (rs1052133) was performed using Impute2 based on the 1000 Genomes Project Phase 3 data; probabilistic genotypes for were called at a threshold of 0.8.Statistical Analyses: Statistical analyses were performed using Microsoft Excel, IBM SPSS software (v.

Table 1
Descriptive statistics of NHW and MA participants categorized by population and cognitive phenotype in the Texas Alzheimer's Research and Care Consortium for buffy coat mitochondrial DNA oxidative mutational load.
a.The mean difference is signi cant at 0.05 .b. NC vs.MCI 0.016, NC vs. AD <0.001,MCI vs. AD <0.001 c.NC vs.MCI <0.001, NC vs. AD <0.001,MCI vs. AD <0.001 d.NC vs.MCI 0.003, NC vs. AD 0.042, MCI vs. AD 0.542 e. NC vs.MCI 0.028, NC vs. AD <0.001,MCI vs. AD 0.017 f.NC vs.MCI <0.001, NC vs. AD <0.001,MCI vs. AD <0.001 g.NC vs.MCI <0.001, NC vs. AD <0.001,MCI vs. AD <0.001 h.NC vs.MCI 0.001, NC vs. AD 0.273, MCI vs. AD 0.540 Directly genotyped and imputed APOE frequencies for each population is shown in Supplementary Table 1.Genotype frequencies for APOE and OGG1 by cognitive status in each population are shown in Supplementary Table 2. Hardy-Weinberg proportions for APOE and OGG1 in both populations separately and together indicates the genotype frequencies are in Hardy-Weinberg equilibrium and both the observed and expected genotype frequencies are not signi cantly different (Supplementary Tables

Table 2
Multiple linear regression prediction model in the Mexican American population considering total cellular 8oxoG variant count.Italics and bolding indicate a p-value of signi cance, while italics alone indicate a p-value approaching signi cance.

Table 3
Multiple linear regression prediction model in non-Hispanic Whites considering total cellular 8oxoG variant count.Italics and bolding indicate a p-value of signi cance, while italics alone indicate a p-value approaching signi cance.

Table 4
Descriptive statistics of NHW and MA TARCC participants classi ed by population and cognitive phenotype for plasma (i.e., ccf) mitochondrial DNA oxidative mutational load..The mean difference is signi cant at 0.05 .Genotype frequencies obtained and imputed for APOE in each population and genotype frequencies for APOE and OGG1 distributed by cognitive status in each population are shown (Supplementary Tables19 a

Table 5
Multiple linear regression results for ccf-8oxoG variant count within Mexican Americans.Italics and bolding indicate a p-value of signi cance, while italics alone indicate a p-value approaching signi cance.

Table 6
Multiple linear regression results for ccf-8oxoG variant count within non-Hispanic Whites.Italics and bolding indicate a p-value of signi cance, while italics alone indicate a p-value approaching signi cance.