Platelet mitochondrial DNA methylation predicts future cardiovascular outcome in adults with overweight and obesity

Background The association between obesity and cardiovascular disease (CVD) is proven, but why some adults with obesity develop CVD while others remain disease-free is poorly understood. Here, we investigated whether mitochondrial DNA (mtDNA) methylation in platelets is altered prior to CVD development in a population of adults with overweight and obesity. Methods We devised a nested case-control study of 200 adults with overweight or obesity who were CVD-free at baseline, of whom 84 developed CVD within 5 years, while 116 remained CVD-free. Platelet mtDNA was isolated from plasma samples at baseline, and mtDNA methylation was quantified in mitochondrially encoded cytochrome-C-oxidase I (MT-CO1; nt6797 and nt6807), II (MT-CO2; nt8113 and nt8117), and III (MT-CO3; nt9444 and nt9449); tRNA leucine 1 (MT-TL1; nt3247 and nt3254); D-loop (nt16383); tRNA phenylalanine (MT-TF; nt624); and light-strand-origin-of-replication (MT-OLR; nt5737, nt5740, and nt5743) by bisulfite-pyrosequencing. Logistic regression was used to estimate the contribution of mtDNA methylation to future CVD risk. ROC curve analysis was used to identify the optimal mtDNA methylation threshold for future CVD risk prediction. A model was generated incorporating methylation at three loci (score 0, 1, or 2 according to 0, 1, or 2–3 hypermethylated loci, respectively), adjusted for potential confounders, such as diastolic and systolic blood pressure, fasting blood glucose, and cholesterol ratio. mtDNA methylation at MT-CO1 nt6807 (OR = 1.08, 95% CI 1.02–1.16; P = 0.014), MT-CO3 nt9444 (OR = 1.22, 95% CI 1.02–1.46, P = 0.042), and MT-TL1 nt3254 (OR = 1.30, 95% CI 1.05–1.61, P = 0.008) was higher at baseline in those who developed CVD by follow-up, compared with those who remained CVD-free. Combined use of the three loci significantly enhanced risk prediction, with hazard ratios of 1.38 (95% CI 0.68–2.78) and 2.68 (95% CI 1.41–5.08) for individuals with score 1 or 2, respectively (P = 0.003). Methylation at these sites was independent of conventional CVD risk factors, including inflammation markers, fasting blood glucose concentration, and blood pressure. Conclusions Methylations of MT-CO1, MT-CO3, and MT-TL1 are, together, strong predictors of future CVD incidence. Since methylation of these mtDNA domains was independent of conventional CVD risk factors, these markers may represent a novel intrinsic predictor of CVD risk in adults with overweight and obesity.


Background
Cardiovascular disease (CVD) is the single largest cause of death and is responsible for approximately 30% of all deaths worldwide [1]. Overweight and obesity are risk factors for CVD, attributed to insulin resistance [2], inflammation [3][4][5], and the hyperaggregability of platelets [6]. Subsequently, inflammation markers, such as C-reactive protein (CRP) [7], uric acid (UA) [8,9], and fibrinogen [10], are used for CVD risk prediction, as are markers of platelet activation such as lipoproteinassociated phospholipase A2 (Lp-PLA2) [11]. However, not everyone with obesity develops CVD, and the reasons why some individuals with obesity develop CVD while others remain CVD-free are poorly understood.
Mitochondrial dysfunction and damage have been implicated in obesity [12,13] and CVD [14]. In particular, platelet mitochondria are important in maintaining thrombosis and hemostasis [15]. Intriguingly, platelets show hyperaggregability in adults with obesity and are unresponsive to anticoagulant treatment [6,16]. Mitochondria contain a circular genome of approximately 17 kb in size with 37 genes encoding for proteins, ribosomal RNAs, and transfer RNAs related to oxidative phosphorylation. There is growing evidence for epigenetic regulation of mitochondrially encoded genes through DNA methylation, supported by the identification of DNA methyltransferase activity in mitochondria [17], and these epigenetic marks are altered in response to environmental exposures [18,19] and in disease states such as cancer [20]. It has recently been demonstrated that mitochondrial DNA (mtDNA) in platelets is aberrantly methylated in CVD patients [21], but whether this precedes disease development is not known. Supporting the hypothesis that such epigenetic changes in the mitochondrial epigenome may be early events related to CVD development, nuclear DNA methylation patterns in the liver are known to be modified by obesity [22], while epigenetic analysis of blood samples predicts future CVD risk [23][24][25][26]. DNA methylation is not only modified in CVD patients [27,28], but also with exposure to CVD risk factors [29][30][31][32][33]. Further, the associations between DNA methylation and CVD events are often stronger in individuals with pre-existing CVD risk markers, such as obesity [23,34]. Therefore, we hypothesized that aberrant platelet mtDNA methylation occurs in at-risk individuals, such as adults with obesity, prior to developing CVD and may therefore serve as a biomarker of CVD risk.
Here, we tested this hypothesis in a nested casecontrol study investigating the utility of platelet mtDNA methylation to predict future CVD events in adults with overweight or obesity who were CVD-free at baseline.
The threshold values that maximized sensitivity and specificity to predict CVD risk were used to create dichotomous variables "methylation level above the    Using these values, overall scores were calculated for each individual participant as follows: methylation not above the thresholds at any of the three loci (score 0), methylation above the threshold at any one locus (score 1), and methylation above the threshold at any two or all three loci (score 2) (Table 2 (b)). Compared with score 0, the hazard ratio (HR) for developing CVD for score 1 was 1.38 (95% CI, 0.68-2.78) and for score 2 was 2.68 (95% CI, 1.41-5.08) (Fig. 6a). During follow-up, 65% of the individuals with score 2 developed CVD, while only 21% of individuals with score 0 developed CVD ( Fig. 6a and Table  2 (b)). Participants with score 2 had a lower median time without-CVD (35.1 months) than participants with score 1 (54.8 months). More than half of the participants with score 0 were CVD-free at the end of the follow-up period.

Comparison with existing risk prediction models
Conventional CVD risk prediction scores were calculated at baseline using the Framingham Risk Score [35] and the European HeartScore [36]. These scores were not different between those who remained CVD-free and the CVDdeveloped group (P = 0.636 and P = 0.232, respectively), demonstrating the potential utility of mtDNA methylation as a predictor of CVD development.

Sensitivity analysis
A sensitivity analysis was performed by excluding participants who developed CVD within a year from baseline, but this did not change the relationships previously observed (Fig. 6b). The HR for those who scored 2 remained significantly higher than those who scored 1 (HR = 2.17, 95% CI 1.06-4.47) and was even higher in comparison with those who scored 0 (HR = 2.53, 95% CI 1.12-5.72) (Fig. 6b). An additional sensitivity analysis was performed by stratifying the CVD cases into "Mild," such as hypertension (n = 51), and "Severe" events, such as ischemic heart diseases (n = 33) ( Table S2). The model was tested in the Mild subgroup and showed that the mtDNA methylation score was a significant (P < 0.001) predictor of future risk of developing CVD. The HR for those who scored 2 was significantly higher than for those scored 1 (HR = 2.27, 95% CI 1.13-4.44, P = 0.021) and those who scored 0 (HR = 4.34, 95% CI 1.76-10.73, P < 0.002). No such relationships were apparent in the Severe subgroup of CVD events (n = 33), due to lack of power (P = 0.086) (data not shown).

Discussion
To the best of our knowledge, this is the first study investigating platelet mtDNA methylation in relation to the future development of CVD. In this nested case-  control study of 200 adults with overweight and obesity, higher mtDNA methylation at three loci (MT-CO1 nt6807, MT-CO3 nt9444, and MT-TL1 nt3254) in platelets was associated with higher risk of developing CVD within 5 years. Further, participants with score 2 (high methylation at two or three loci) developed CVD significantly sooner than the participants with score 1 and score 0. Thus, mtDNA methylation at the three loci may be a novel predictive biomarker for the future risk of developing CVD. We have previously demonstrated changes in the mitochondrial epigenome among individuals with CVD, including hypermethylation of MT-CO1, MT-CO3, and MT-TL1 [21]. Further, we have shown that mtDNA methylation modifies the effect of particulate matter exposure and heart rate variability, a prognostic marker of CVD [37]. We have built on our previous work to demonstrate that mtDNA methylation may serve as a predictor of CVD risk among individuals with overweight and obesity. However, the field remains at a nascent stage, with little understanding of the mechanisms underpinning how mtDNA methylation levels may be implicated in the etiology of CVD and/or platelet activation. Recently, it has been demonstrated that mtDNA methylation regulates expression of mitochondrial-derived peptides (MDP) with cytoprotective function [38] suggesting that mtDNA methylation level may be indicative of the overall stress to which the cell is exposed. Additionally, in vitro studies have shown that the presence of 5methylcytosine can alter mitochondrial transcription factor (TFAM) binding and transcription initiation [39].
MtDNA methylation levels in blood are associated with blood pressure and heart rate variability in individuals with CVD-related environmental and occupational exposures [18,19,37,40]. However, in platelets, we did not find any association between mtDNA methylation level and the most common CVD risk factors including age, BMI, blood pressure, blood glucose concentration, cholesterol, and uric acid in individuals with overweight and obesity. Therefore, our study supports the idea that altered mtDNA methylation in platelets precedes the development of CVD, and may serve as a non-invasive, Fig. 6 The Kaplan-Meier curves for probability of remaining CVD-free by methylation score. a Time CVD-free from baseline (months) among participants categorized by methylation at MT-CO1 nt6807, MT-CO3 nt9444, and MT-TL1 nt3254. b Sensitivity analysis in which all participants who developed CVD within 1 year of baseline were excluded easy-to-access biomarker to distinguish individuals with higher CVD risk. Adults with overweight or obesity may, therefore, benefit from identification to facilitate early primary prevention and monitoring to reduce their personal risk of CVD.
We observed low levels of mtDNA methylation in these participants and subtle, but detectable, differences between individuals who developed CVD during follow-up and those who remained CVD-free. Such subtle changes in methylation are not confined to the mitochondrial epigenome, as changes in methylation of < 5% are frequently reported in aging, in response to environmental exposures [41], and during disease initiation [42]. It is not known whether these small changes in DNA methylation reflect changes in gene expression. Regardless, they may serve as a biomarker of a cascade of other biological reactions [43][44][45], such as MDP regulation [38].
Our study has limitations that merit consideration. The outcome in our study was diagnosis of any of a heterogeneous group of CVDs that ranged from mild (e.g., hypertension) to more severe events. Our model remained strong in predicting the "mild" CVD events, but the lack of statistical power prevented examination of its ability to predict more "severe" cases. Further, replication of our findings is imperative. Such a validation would require access to data and samples from a cohort that had collected plasma or platelets and had follow-up data on CVD incidence as part of a prospective study of individuals with overweight and obesity. We utilized hospital discharge records, which are widely used for collection of data regarding clinical diagnoses (e.g., for Italian healthcare administrative databases and the WHO's European Health Information Gateway for classification), but which can potentially under-or overestimate the number of cases. The use of thoroughly validated administrative databases may strengthen future studies. We attempted a partial validation by dividing the population of 200 individuals into a test set (n = 150) and a validation set (n = 50) with the same proportions of CVD-free and CVDdeveloped at follow-up participants in both, which showed that the mtDNA methylation markers predicted CVD risk in both the test (P = 0.045) and validation sets (P = 0.034). Finally, as most of the participants were Caucasian, additional studies are needed to validate these findings in individuals with different ethnicities.
In conclusion, we have demonstrated that mtDNA methylation of MT-CO1, MT-CO3, and MT-TL1 in platelets from adults with overweight and obesity may predict CVD risk during the following 5 years. Our findings require confirmation in a larger, independent study.

Study design and sample selection
We utilized plasma samples and clinical data from the Susceptibility to Particle Health Effects, miRNAs and Exosomes (SPHERE) study in which 2000 participants with overweight (25 < BMI < 30 kg/m 2 ) and obesity (BMI ≥ 30 kg/m 2 ) were recruited in Milan, Italy [46]. We designed a prospective nested case-control study using samples and data from 200 participants within the SPHERE study without previous hospitalization for CVD at the time of enrolment (baseline; n = 200) for whom follow-up data for up to 5 years (median = 27 months) were available. For those who developed CVD, the follow-up stopped after the first CVD diagnosis; for those who remained CVD-free, the follow-up lasted until the last update from the Italian National Health Service. We selected 84 individuals who developed CVD in the follow-up period, and these were sex-and BMI-matched with 116 individuals who remained CVD-free. The demographic and clinical characteristics of these participants are summarized in Table 1. Ethical approval was provided by the Institutional Review Board, Fondazione IRCCS Cà Granda Ospedale Maggiore Policlinico at University of Milan. The ethnicity of the SPHERE study participants was predominantly Caucasian (95.8% of cases) [46].

Assessment of CVD risk at baseline and CVD events at follow-up
To estimate individual CVD risk at baseline, we calculated the Framingham Risk Score which uses information on sex, age, SBP, treatment for hypertension, smoking, type 2 diabetes, HDL, and total cholesterol [35]. In addition, we used the HeartScore to predict the incidence of fatal CVD within 10 years [36,47] using age, sex, SBP, cholesterol, HDL cholesterol, BMI, and smoking status.
Details of CVD events were obtained from the hospital discharge registry of the Italian National Health Service. A CVD event was defined as any principal or any 1 of 5 secondary diagnosis of diseases of the circulatory system (3digit ICD-9-CM codes from 390 to 459) [48]. A detailed list of the CVD events and antihypertensive medications of the participants by follow-up is summarized in the Supplementary Material (Table S2 and S3). Briefly, 47 participants (56%) were diagnosed with hypertension and 37 participants (44%) were diagnosed with other forms of CVD including atrial fibrillation and atherosclerosis (Table S2).

Platelet mtDNA preparation and DNA methylation measurement
Plasma samples from 200 participants were used to isolate platelet mtDNA as described previously [21]. Briefly, platelet pellets obtained by centrifugation of 200 μL plasma at 1400×g were treated with DNaseI (30 U, ROCHE) to eliminate cell-free nuclear DNA containing nuclear mitochondrial DNA segments (NUMTs). The EZ DNA Methylation Direct kit (Zymo Research) was used for extraction and bisulfite conversion of mtDNA simultaneously. To maximize bisulfite conversion efficiency [49], mtDNA was linearized with BamHI (NEB) following proteinase K treatment. The bisulfite-converted mtDNA (20 μL) was stored at − 80°C until analysis.

Statistical analysis
For normally distributed demographic and clinical characteristics and for DNA methylation levels, data are expressed as mean and standard deviation, otherwise by median and range. Frequencies and percentages were calculated for categorical variables. Data for CVD-free and CVD-developed participants at follow-up were compared using the χ 2 test for categorical data and Student's t test for continuous variables. Multivariate logistic regression, adjusted for age, BMI, fasting blood glucose, cholesterol ratio (TC/HDL), SBP, and DBP, was performed to investigate the association between DNA methylation at each locus (CpG site) and the risk of CVD during follow-up. Estimated effects were reported as ORs and 95% confidence intervals (CI) associated with an increase in 5methylcytosine (5mC) at each locus.
ROC curves were generated to evaluate the diagnostic ability of the cholesterol ratio and mtDNA loci to distinguish between participants who were CVD-free and those in whom CVD-developed at follow-up. The optimum threshold was selected by the Youden Index as the one that maximized sensitivity (SE) + specificity (SP) − 1. The area under the ROC curve (AUC) and corresponding 95% CI, SE, SP, and threshold were reported for cholesterol ratio and for the MT-CO1 nt6807, MT-CO3 nt9444, and MT-TL1 nt3254 positions. For each CpG site, a dichotomous variable was created viz. "methylation level above the threshold" for the specific locus and "methylation level below the threshold." In addition, we tested the utility of a score built as the sum of the index value (0, 1) for each locus (MT-CO1 nt6807, MT-CO3 nt9444, and MT-TL1 nt3254) in predicting CVD. The score has three categories: none of three loci display mtDNA methylation above threshold (score 0), any one of the three loci has mtDNA methylation above threshold (score 1), and any two or all three loci display mtDNA methylation threshold (score 2).
The Kaplan-Meier survival curves and log-rank tests were calculated by stratifying CVD cases by each locus below or above the methylation threshold. To evaluate the independent prognostic value of each single locus and of their combination on future CVD cases, we calculated hazard ratios (HRs) with Cox multivariable regression models adjusted for DBP, SBP, fasting blood glucose concentration, and cholesterol ratio. The Cox multivariable regression was performed on a total of 193 participants, for whom the methylation percentage of all the three genes was available. The same model was used to evaluate the prognostic value of cholesterol ratio, when evaluated as the predictor. The assumption of proportional hazard was checked with the log [log(survival)] plot and by the time-dependent covariate test. Cox multivariable regression models were also used to evaluate the potential prognostic value of the Framingham Risk Score and of the European HeartScore on CVD risk.
A sensitivity analysis was performed by excluding the participants who developed CVD within a year from baseline in all Cox multivariable regression models. An additional sensitivity analysis was performed by stratifying the CVD cases into "Mild," such as hypertension (n = 51), and "Severe" events, such as ischemic heart diseases (n = 33). However, the Severe event (n = 33) category did not provide enough power to be tested reliably (data not shown). All reported P values were two-tailed, and those less than 0.05 were considered statistically significant. Statistical analyses were performed with SAS software, version 9.4.
Additional file 1 Table S1. Multivariate analysis for the association between potential risk factors and mitochondrial DNA methylation. Table S2. Diagnostic and procedure codes for CVD events at Follow-up. Table S3. List of antihypertensive medication. Table S4. PCR primers and pyrosequencing assays used to analyze mtDNA methylation. Table  S5. Correlation matrix of mtDNA methylation at CpG positions included in the study.