Mitochondrial DNA Methylation and Human Diseases

Epigenetic modifications of the nuclear genome, including DNA methylation, histone modifications and non-coding RNA post-transcriptional regulation, are increasingly being involved in the pathogenesis of several human diseases. Recent evidence suggests that also epigenetic modifications of the mitochondrial genome could contribute to the etiology of human diseases. In particular, altered methylation and hydroxymethylation levels of mitochondrial DNA (mtDNA) have been found in animal models and in human tissues from patients affected by cancer, obesity, diabetes and cardiovascular and neurodegenerative diseases. Moreover, environmental factors, as well as nuclear DNA genetic variants, have been found to impair mtDNA methylation patterns. Some authors failed to find DNA methylation marks in the mitochondrial genome, suggesting that it is unlikely that this epigenetic modification plays any role in the control of the mitochondrial function. On the other hand, several other studies successfully identified the presence of mtDNA methylation, particularly in the mitochondrial displacement loop (D-loop) region, relating it to changes in both mtDNA gene transcription and mitochondrial replication. Overall, investigations performed until now suggest that methylation and hydroxymethylation marks are present in the mtDNA genome, albeit at lower levels compared to those detectable in nuclear DNA, potentially contributing to the mitochondria impairment underlying several human diseases.


Introduction
Mitochondria are double-membrane organelles which play a vital role in a variety of key biological functions, including production of ATP through oxidative phosphorylation (OXPHOS), apoptosis via caspase-dependent and independent mechanisms, regulation of calcium homeostasis and production of reactive oxygen species (ROS) [1]. Mitochondria contain their own DNA, a molecule of 16,569 bp inherited in a maternal, non-Mendelian fashion. Each mitochondrion contains multiple copies of mtDNA, which is distinctly different from nuclear DNA in several ways. In part, this can be explained by the endosymbiotic theory, which states that mitochondria evolved from an alpha-proteobacterium that invaded eukaryotic cells [2]. Indeed, similar to the DNA of prokaryotic cells such as bacteria, mtDNA is a circular, double-stranded DNA molecule formed by the heavy (H) and the light (L) strand, lacking histones and organized into tightly packed nucleoprotein complexes called nucleoids [3].
The mitochondrial genome comprises 37 genes, 13 of which encode for polypeptides required for the electron transport chain (ETC), including seven genes encoding for subunits within complex I (MT-ND1, MT-ND2, MT-ND3, MT-ND4, MT-ND4L, MT-ND5, MT-ND6), one for complex III (MT-CYB), three for complex IV (MT-CO1, MT-CO2, MT-CO3) and two for complex V (MT-ATP6, MT-ATP8), in addition to two ribosomal RNAs (MT-RNR1 and MT-RNR2) and 22 transfer RNAs [4]. Unlike nuclear DNA, which contains at least one promoter per gene, mtDNA contains only three promoter regions, including LSP for genes encoded by the L-strand, and HSP1 and HSP2 for the genes encoded by the H-strand, that transcribe multiple genes at once to produce polycistronic transcripts [5]. These promoters are located within or in the vicinity of a 1124-bp locus, known as the mitochondrial displacement loop (D-loop) region, which also contains the origin of replication of the Hstrand. Several proteins encoded in nuclear DNA regulate both mtDNA transcription and replication, including the mitochondrial RNA polymerase (POLRMT), the transcription and mtDNA maintenance factor (TFAM), the transcription specificity factors (TFB1M and TFB2M) and the transcription termination factor (mTERF) [6].
Growing evidence is showing that also epigenetic mechanisms could contribute to the regulation of both mitochondrial DNA transcription and replication, leading to coining the term "mitoepigenetics" [7]. Epigenetics refers to heritable changes in gene regulation that are not due to changes in the DNA sequence, including DNA methylation, histone modifications and non-coding RNA-mediated mechanisms, which are tissue-and cell-specific and may change overtime as a result of aging, disease or environmental stimuli [8]. In the last few years, there has been a growing interest in the potential role of epigenetic mechanisms in the pathogenesis of various human diseases, including cancer, cardiovascular diseases and neurodegeneration. Epigenetic alterations have started being considered as valuable biomarkers for the diagnosis and prognosis of certain types of cancer, as well as for neurological and autoimmune diseases, and, given their reversible nature, it is widely accepted that epigenetic mechanisms provide promising opportunities for interventions aimed to prevent or ameliorate disease symptoms by targeting lifestyles and using epigenetic-based therapy [9].
However, compared to the epigenetic regulation of nuclear DNA, the mechanisms of mtDNA epigenetic regulation have been less investigated, and only in the last few years has there been a growing interest in this field [7]. mtDNA is devoid of histones, and its structure is different from that of nuclear chromatin, meaning that one of the most investigated epigenetic modifications occurring in this molecule is DNA methylation [10]. In this review, we discuss the role of mitoepigenetics in the regulation of mitochondrial metabolism, highlighting the role of DNA methylation inside this organelle and its impact on human diseases. We also discuss the possible impact of the nuclear genetic background and of environmental factors in the regulation of mtDNA methylation.

Epigenetic Mechanisms
Epigenetic mechanisms have been largely investigated in nuclear DNA and include DNA methylation, post-translational modifications on histone tails and nucleosome positioning that tightly regulate gene expression levels, chromatin folding and the threedimensional structure of the nuclear genome [11][12][13]. DNA methylation is the most widely studied epigenetic mechanism and consists in the addition of a methyl group to cytosine, forming 5-methylcytosine (5-mC) [14]. The reaction is mediated by enzymes called DNA methyltransferases (DNMTs) which use S-adenosylmethionine (SAM), produced during the one-carbon metabolism, as the methyl donor compound [15]. In the nuclear genome, DNA methylation occurs primarily on cytosines followed by a guanine, called CpG sites. Sites of CpG clusters are called CpG islands, and when a CpG island in the promoter region of a gene is methylated, the expression of that gene is usually repressed. By contrast, cytosine methylation in gene bodies could be related to both the active and the repressed transcriptional state depending on the tissue in which it occurs [16]. CpG dinucleotides are also located in repetitive or centromeric sequences, where their methylation is associated with the maintenance of chromosomal stability and with prevention of translocation events [17]. DNA methylation also occurs at non-CpG sites (CpA, CpT and CpC) in the nuclear genome, but non-CpG methylation is restricted to specific cell types, including neurons, glial cells and embryonic stem cells [18]. Hydroxymethylcytosine (5-hmC) is another modification of cytosine mediated by members of the ten-eleven translocation (TET) protein family, which is important for proper gene transcription as it directs the dynamic remodeling and organization of the chromatin structures [19].
Histones are the most abundant proteins associated with nuclear DNA and aggregate each other, forming the histone octamer around which DNA is wrapped, creating the nucleosome. The N-terminal tails of histones may undergo several post-translational modifications, including acetylation, methylation, phosphorylation, ubiquitination and ADP ribosylation. These changes influence the chromatin structure, facilitating or inhibiting gene transcription [11]. For example, acetylation of lysine residues leads to a more relaxed chromatin structure, allowing greater access of transcriptional activators to the underlying genomic sequence [20].
In addition to histone modifications and DNA methylation, a further layer of epigenetic regulation of gene expression and chromatin state exists at the level of short (<200 nt) and long (>200 nt) non-protein coding RNAs (ncRNAs) [21]. miRNAs (22-25 nt) are the most studied ncRNAs and regulate gene expression in a sequence-specific manner. In fact, they bind to the 3 untranslated region of target mRNA molecules and mediate their post-translational regulation, leading to either degradation or translational inhibition, depending on the degree of sequence complementarity [22].
The different structure, nature and complexity of the mitochondrial genome with respect to the nuclear one and the absence of histone proteins have led researchers to question, for several years, about the existence of mitochondrial epigenetic mechanisms [23]. However, mounting evidence is revealing a bidirectional crosstalk between the nuclear and the mitochondrial genome to tightly regulate metabolic reactions and gene expression levels according to the cellular demands, primarily acting through epigenetic mechanisms [15,24,25].

Mitoepigenetics
The term "mitoepigenetics" has been coined to indicate the epigenetic mechanisms that regulate the mitochondrial genome, but recently, it has more broadly been used to include the complex interactions between mitochondria and epigenetic mechanisms [7,26]. Indeed, mitochondria provide key metabolites to the nucleus, including β-nicotinamide adenine dinucleotide, ATP, α-ketoglutarate and acetyl coenzyme A that are co-substrates required for epigenetic processes [1]. Moreover, mtDNA variants may have effects on the transcription of genes in the nuclear genome through epigenetic mechanisms, as the removal of mtDNA or changes in mtDNA haplogroups have been found to be associated with differences in nuclear DNA methylation levels [27][28][29], suggesting an important epigenetic interplay between the two genomes. Recently, it has also been observed that changes in the mtDNA copy number influence nuclear DNA methylation at specific loci and result in differential expression of specific genes that may impact human health and disease via altered cell signaling [30]. On the other hand, nuclear epigenetics modulates mitochondrial function, as many of the mitochondrial proteins are nuclearencoded. For example, the POLGA nuclear gene, which encodes the enzyme polymerase gamma, a subunit of the polymerase responsible for mtDNA replication and repair, is regulated by epigenetic mechanisms, and its methylation status regulates the mtDNA copy number [31,32]. Overall, it is emerging that epigenetic mechanisms are fundamental for a bidirectional crosstalk between the nuclear and mitochondrial genomes that allows a coordinated gene expression and metabolic response to different cellular conditions and energy demands [15,24,25].
Concerning epigenetic mechanisms occurring within mitochondria, few studies have demonstrated the presence of ncRNAs inside these organelles, both nuclear-and mitochondria-encoded, which seem to be mainly involved in the communication between the nucleus and the mitochondria, at both anterograde and retrograde signaling levels [reviewed in 6]. Moreover, although mtDNA organization is not supported by histones, mtDNA interacts with multiple proteins, including TFAM, mitochondrial single-stranded DNA binding protein and Twinkle mtDNA helicase to create nucleoprotein structures called nucleoids, which could be regulated by epigenetic mechanisms, such as acetylation and phosphorylation [33]. However, many more studies have focused on the role of mtDNA methylation and hydroxymethylation as mitoepigenetic mechanisms, and these epigenetic marks will be the main focus of the present review.
Although some authors failed to find DNA methylation in the mitochondrial genome [34][35][36], the presence of methylation in mtDNA, as well as the identification of a DNA methyltransferase activity in mitochondria, has been observed for the first time in studies performed more than forty years ago in loach embryos, and mouse and hamster samples [37][38][39]. Subsequent studies showed the presence of mtDNA methylation in human and mouse fibroblasts [40,41], but not in gastric and colorectal cancer human tissues [42]. Investigations by Rebelo and co-workers [43] and Shock and coworkers [44] rekindled the interest in mitoepigenetics, since they not only identified the presence of mtDNA methylation marks in human and mouse cell cultures using more modern techniques than those used in previous studies but also suggested a functional role of these marks in mitochondrial metabolism. Indeed, Rebelo et al. [43] showed that DNA methylation associated with the accessibility of TFAM to mtDNA, thus regulating mtDNA replication, whilst Shock and collaborators showed that mtDNA methylation induced by the binding of the DNMT1 enzyme to the D-loop control region was able to regulate the expression of MT-ND6 and MT-ND1 genes [44]. Several researchers have subsequently corroborated these observations. For example, by using biophysical approaches, it has been recently demonstrated that CpG methylation in the D-loop region regulates TFAMdependent activities in vitro, strictly regulating mtDNA gene transcription [45]. Moreover, while the presence of DNMT1 in mitochondria has been further confirmed by other research groups [46,47], also the presence of the de novo DNMT3A and DNMT3B [48][49][50] and of the TET enzymes [51,52] has been identified in mitochondria. Several authors also observed a significant correlation between the mtDNA methylation levels and mitochondrial copy number, as well as between mtDNA methylation and mtDNA gene expression, thus further suggesting the important role of DNA methylation in mitochondrial regulation [53][54][55][56][57][58][59].
These studies highlighted that mtDNA could be subjected to DNA methylation and hydroxymethylation by the same enzymes of the nuclear genome. However, some differences between mtDNA and nuclear DNA methylation exist. Indeed, contrarily to genomic DNA, mtDNA showed to possess high levels of non-CpG versus CpG methylation, which is not symmetrical on both strands, being biased toward the L-strand, and it seems that depending on the cell type and context of cytosine methylation, mtDNA methylation can play a role in the regulation of mtDNA gene expression or mtDNA replication [46,50,54,[60][61][62]. Additionally, 5-hmC has been detected in mtDNA, and there is evidence that 5-hmC marks are dynamic in nature and are enriched in the upstream of gene start site regions and in the gene body, similarly to nuclear genes [63]. Moreover, recent studies reported that human mtDNA is particularly enriched in N 6 -methyldeoxyadenosine (6-mA), which is usually widespread in prokaryotes but less present in mammals' nuclear genome, an epigenetic mark which affects mitochondrial DNA transcription, replication and activity [64][65][66]. These studies clearly suggest that there is a specific regulation of mtDNA methylation, and it is of outmost importance to know how these regulatory mechanisms work in order to better understand how mitochondrial function is regulated.

Altered mtDNA Methylation and Human Diseases
The role of altered nuclear epigenetic mechanisms is now well established in the etiology of various human diseases [9]. On the other hand, information regarding a potential role of mitoepigenetic mechanisms in human diseases is only recently emerging [10,23,67]. This discrepancy is mainly due to the limited availability of methodologies for studying mtDNA methylation and hydroxymethylation until a few years ago. However, the application of different approaches, including 5-mC immunoprecipitation, bisulfite sequencing, pyrosequencing, next-generation sequencing, liquid chromatography tandem mass spectrometry and ELISA assays [68], has largely increased the possibility of studying mtDNA modifications in recent years, and several authors focused their investigations on the search for mitoepigenetic alterations in various human diseases, and particularly on differentially methylated and hydroxymethylated mtDNA regions.

Evidence of Altered mtDNA Methylation in Cancer
The association between mtDNA methylation and cancer has been largely investigated (Table 1). Whole-mtDNA genome methylation Differentially methylated sites between tumor and their matched adjacent tissues detected in head and neck, but not in liver [75] Although some authors did not identify the presence of mtDNA methylation in cancer cell lines, in malignant and non-malignant tissue from patients with gastric and colorectal cancer, in adenomas and in cervix cancer [42,69,72], findings of other authors suggested an involvement of mtDNA methylation impairment in tumorigenesis. Feng and co-workers analyzed D-loop methylation in 44 colorectal cancer tissues and in their corresponding adjacent non-cancerous tissues, finding that the D-loop of non-cancerous tissue was methylated, while in the majority of cancer tissues, the D-loop was unmethylated [58]. Moreover, ND2 protein expression was higher in cancer tissues compared to adjacent normal tissues, suggesting that D-loop methylation levels may play a role in regulating MT-ND2 expression. Gao and co-workers correlated methylation of the D-loop region with the mtDNA copy number and ND2 protein expression in 65 colorectal cancer specimens and their corresponding non-cancerous tissues [55]. The methylation rate of the D-loop region in all the 65 colorectal cancer tissues was markedly reduced when compared with that of their corresponding non-cancerous tissues. In addition, the mtDNA copy number and ND2 protein expression were increased in colorectal cancer tissues with respect to non-cancerous ones. Demethylation of the D-loop region was associated with an elevated mtDNA copy number and an increased ND2 expression. Furthermore, the mtDNA copy number and ND2 expression in Caco-2 cancer cells were significantly increased after treatment with the demethylating agent 5-aza-2 -deoxycytidine. Similarly, Lovo and Colo-205 colorectal cancer cell lines treated with 5-aza-2 -deoxycytidine showed lower D-loop methylation levels along with a higher content of the mtDNA copy number compared to non-treated cells [70]. Altered mtDNA methylation has also been associated with breast cancer. Indeed, in a study performed in the peripheral blood of female individuals belonging to five fam-ilies with one breast cancer patient, aberrant mtDNA methylation of the D-loop region was correlated with breast cancer risk [71]. Interestingly, the authors observed that the D-loop region demonstrated a familial-specific mtDNA methylation pattern among the five families and suggested that D-loop methylation was maternally inherited [71]. In a following study, a whole-mtDNA genome methylation analysis was performed in human model cells of breast cancer and hepatocarcinoma, as well as in the corresponding normal cells, finding that the methylation pattern was markedly different between normal and cancer cells primarily in a non-CpG context [61]. By using cellular models of glioblastoma and osteosarcoma, it was found that mtDNA methylation levels tended to decrease during tumor progression, potentially contributing to the increase in mtDNA [73]. These changes also correlated with transcriptional changes of MT-ND5 and MT-ND6 genes during tumorigenesis. Interestingly, after tumors had restored sufficient mtDNA to initiate tumorigenesis, higher levels of 5-mC over the D-loop were acquired and the authors suggested that this could occur to potentially restrict further replication of mtDNA [73]. Recently, whole-mtDNA genome methylation has been investigated by means of nanopore sequencing in oral squamous cell carcinoma cell cultures with different sensitivity to cisplatin, observing that enhanced cisplatin resistance was not influenced by the methylation status of the mitochondrial genome [74]. However, the authors suggested that mtDNA methylation in the gene bodies promoted the expression of the genes, presumably by affecting the post-transcriptional modifications of polycistronic mitochondrial mRNAs, since in one cell culture, they found hypermethylation of MT-CO1 and MT-CYB genes with concomitant high expression levels [74]. By using the same technique, increased CpG and CpH mtDNA methylation levels have been observed in head and neck tumor samples when compared to their matched adjacent normal tissues [75].

Altered mtDNA Methylation in Metabolic and Cardiovascular Diseases
Metabolic syndrome represents a cluster of clinical conditions, including high blood pressure, insulin resistance, lipid abnormalities and obesity that are associated with increased risk of diabetes and cardiovascular diseases (CVDs) [76]. Among the pathophysiological abnormalities underlying the development of metabolic syndrome, impaired mitochondrial oxidative phosphorylation and mitochondrial biogenesis seem to play a pivotal role, although the molecular mechanisms underlying the mitochondrial impairment remains largely unexplored [77]. Studies performed in the last few years are suggesting that mitochondrial impairment that characterizes obesity, insulin resistance, diabetes and CVDs could be derived from alterations in mtDNA methylation (Table 2).  Pyrosequencing D-loop region D-loop methylation levels were significantly higher in overweight than in lean female subjects, and a specific CpG located in the D-loop associated with impaired body composition. Body composition impairment was predicted by a combined variable including mtDNA copy number and the D-loop methylation (AUC = 0.785; p = 0.009). Retinal microvasculature from type 2 (T2D) and type 1 (T1D) diabetes rat models, and high-fat (HF) diet rats

Immunoprecipitation D-loop region
Increased D-loop methylation in T2D group, compared to T1D or HF groups [84] Vessel specimens from patients with carotid occlusive disease, in vitro and in vivo models of arterial stenotic/occlusive diseases Methylation-specific PCR D-loop region DNMT1-mediated D-loop hypermethylation in the intima media layer of mouse and human tissue. The impaired contractility of a ligated vessel could be restored by transplantation of DNMT1-deleted mitochondria. [85] Peripheral blood leukocytes from 50 patients with stable coronary artery disease (SCAD) and 50 with acute coronary syndrome (ACS) Methylation-specific PCR D-loop region and nuclear PPARGC1A gene Compared to patients with SCAD, those with ACS had significantly lower mtDNA copy number and higher D-loop methylation levels [86] Indeed, insulin resistance has been associated with a reduction in the mtDNA copy number due to increased D-loop methylation levels in the leukocytes from obese individuals [80]. Moreover, D-loop methylation levels have been linked to increased diabetes risk and have been proposed as indicators of early-stage prediabetes [81]. Furthermore, treatment of retinal endothelial cells with high levels of glucose increased D-loop methylation levels, and the retinal microvasculature from human donors with diabetic retinopathy presented a similar increase in D-loop methylation and a decrease in mtDNA transcription [79]. Increased D-loop methylation levels were also observed in buccal swab DNA collected from overweight female subjects, when compared to lean female subjects [82]. Moreover, a specific CpG site in the D-loop region associated with impaired body composition, evaluated by body mass index, waist to height ratio and bioimpedance measurements. Interestingly, body composition impairment was well predicted by a combined variable including the mtDNA copy number and D-loop methylation [82]. Increased D-loop methylation levels were also detected in retinal microvasculature samples from a type 2 diabetic rat model when compared to type 1 diabetic rats or high-fat diet rats [84].
The first evidence of a potential involvement of an impairment of mtDNA methylation in the etiology of CVDs was obtained by a study performed in platelet mitochondria of 10 CVD patients and 17 healthy individuals, in which higher methylation of MT-CO1, MT-CO2, MT-CO3 and MT-TL1 genes in CVD patients was observed [78]. In a following study performed in human samples, in cell cultures and in mouse models of arterial stenotic/occlusive disease, it was observed that in vascular smooth cells, the enzyme DNMT1 translocates to the mitochondria in response to pro-proliferative stimuli and induced D-loop hypermethylation [85]. The D-loop hypermethylation led to repression of mtDNA transcription, inducing mitochondrial dysfunction and reduction in ATP production, thus impairing vascular smooth cell contractility in the context of vascular restenosis or occlusion [85]. Recently, in an attempt to identify mitochondria activity differences between stable coronary artery disease (SCAD) and acute coronary syndrome (ACS), the mtDNA copy number and methylation levels of the nuclear peroxisome proliferatoractivated receptor gamma coactivator 1-alpha (PPARGC1A) and of the D-loop region were evaluated in peripheral blood leukocytes of 50 patients with SCAD and of an equal number of individuals with ACS [86]. The authors observed that SCAD patients had a higher content of the mtDNA copy number as well as lower D-loop methylation levels, thus suggesting that an altered mtDNA copy number and mtDNA methylation may affect the clinical phenotype of coronary artery disease [86].
To better understand why some individuals with obesity develop CVDs while others remain disease-free, methylation levels of several mtDNA genes, including MT-CO1, MT-CO2, MT-CO3, MT-TL1 and MT-TF, and of the D-loop and light-strand origin of replication (MT-OLR) regions were evaluated in platelet mtDNA from 200 adults with overweight and obesity, of whom 84 developed cardiovascular disease [83]. The authors found that methylation levels of MT-CO1, MT-CO3 and MT-TL1 were higher in subjects that developed CVDs, suggesting that methylation levels of these genes could be strong predictors of future CVDs incidence in adults with overweight and obesity [83].

Modulation of mtDNA Methylation in Aging and Senescence
The most common causes of the majority of diseases are senescence and aging, likely due to the accumulation of tissue and organ dysfunctions over time [87]. Decline in mitochondrial quality and activity has been associated with normal aging, but also with the development of a wide range of age-related diseases, and it is now well accepted that mitochondria contribute to specific aspects of the aging process including cellular senescence and age-dependent decline in cellular activity [88]. To better elucidate the molecular alteration underlying mitochondria involvement in cell senescence and aging, some authors investigated a potential role of mtDNA methylation (Table 3).  In order to explore a possible role of mtDNA methylation in relation to replicative senescence, D-loop region methylation was investigated in replicative as well as in senescent human and mouse endothelial cells [53]. Demethylation of the D-loop region with an increased mtDNA copy number was observed in senescent cells compared to proliferative endothelial cells [53]. Using senescent mesenchymal stem cells, three CpG sites located in various mitochondrial genes were found to be hypomethylated in senescent cells as compared to non-senescent ones. Of particular interest was the finding that methylation of one of these CpG sites, located in the COX1 gene, led to repression of COX1 gene expression in parallel with the onset of senescence [57]. More recently, also increased methylation levels of the COX2 gene, accompanied by decreased protein expression, have been associated with cell senescence [92]. Interestingly, cell treatment with 5-aza-2 -deoxycytidine inhibited COX2 methylation and downregulated COX2 expression, promoting cell proliferation and delaying cell aging [92].
While no presence of 5-mC in mtDNA has been detected during mouse oocyte maturation, aging and early embryo development [91], aging in adult individuals has been correlated with changes in mtDNA methylation and hydroxymethylation. A study performed in mouse brain samples found that levels of DNA hydroxymethylation decreased in the frontal cortex, but not in the cerebellum, during aging [52]. In a study performed in peripheral blood of 381 individuals ranging from 38 to 107 years of age, methylation levels of the MT-RNR1 gene were positively associated with increasing age [89]. Interestingly, subjects with higher methylation levels also exhibited a higher mortality risk than those with lower methylation levels, thus suggesting that the methylation of the analyzed CpG sites may reflect a condition of the cell or of the organism to survive. A later study investigated methylation levels of 133 CpG sites of mtDNA in the peripheral blood of 82 female individuals aged 18-91 years, finding detectable methylation in 54 CpG sites, two of which located within the MT-RNR1 gene showed a strong inverse correlation with subject age [90].

Altered mtDNA Methylation in Neurodegenerative Diseases
Multiple lines of evidence suggest that mitochondrial dysfunction is involved in the pathogenesis of neurodegenerative diseases, especially in Alzheimer's disease (AD), Parkinson's disease (PD) and amyotrophic lateral sclerosis (ALS), leading some researchers to investigate the potential involvement of mitoepigenetic mechanisms in the etiology of these diseases (Table 4). Table 4. mtDNA methylation studies in neurodegenerative diseases.

Experimental Model Method mtDNA Region Investigated Observation Reference
Brain and spinal cord motor neurons of mice, and post-mortem human cortex of 12 ALS patients Immunohistochemistry Global 5-mC Increased DNMT activity and 5-mC mtDNA content in motor neurons [48] Superior and middle temporal gyrus (SMTG) and cerebellum (CER) of 7 late-onset AD patients and 5 control subjects Immunohistochemistry Global 5-hmC A trend toward a significant increase in 5-hmC content in SMTG mtDNA of AD patients [93] Spinal cord and skeletal muscle of ALS mice and non-transgenic (non-tg) mice  Early in 2011, Chestnut and co-workers investigated the global 5-mC content and the DNMT protein levels in nuclei and mitochondria from both brain and spinal cord motor neurons of mice, as well as in cortical motor neurons of 12 ALS patients [48]. The authors revealed that motor neurons engaged epigenetic mechanisms to drive apoptosis, involving up-regulation of DNMTs that increased global DNA methylation in both nuclei and mitochondria [48]. Subsequently, another study from the same research group revealed that mtDNA methylation patterns and mitochondrial DNMT3A levels were abnormal in the skeletal muscles and spinal cord of pre-symptomatic ALS mice carrying mutations in the human superoxide dismutase 1 gene (SOD1), which included DNMT3A up-regulation, increased MT-RNR2 gene methylation and decreased D-loop region methylation [49]. Altered D-loop methylation levels have also been observed in the peripheral blood of sporadic and SOD1 ALS patients when compared to both ALS patients with mutations in FUS, TARDBP and C9orf72 and to control subjects who are noncarriers of ALS-linked gene mutations [96,100].
Regarding Alzheimer's disease, a non-significant increase in 5-hmC levels was observed in post-mortem mtDNA brain samples of seven late-onset AD patients with respect to five control subjects [93]. A later study revealed increased methylation levels of the mtDNA D-loop region and reduced MT-ND1 methylation levels in the enthorinal cortex of eight patients with AD-related pathology with respect to healthy control brains [94]. Interestingly, the degree of D-loop region methylation was higher in patients in the early disease stages than in later stages, and these results were corroborated by a dynamic pattern of methylation of this region that was observed in an AD mouse model along with the progression of the disease [94]. We evaluated D-loop methylation levels in the peripheral blood of 133 late-onset AD patients and 130 controls, observing a significant 25% reduction in DNA methylation levels in the patient group [95]. A decrease in D-loop methylation levels and an increase in mt-Rnr1, Cytb and Cox-2 gene methylation with a concomitant reduction in both the mtDNA copy number and the mitochondrial gene expression have been observed in the hippocampus of an AD mouse model when compared to wild-type mice [97,101]. An interesting paper reported that treatment of a blood barrier cell line (hCMEC/D3) with Aβ peptide induced global mtDNA hypermethylation, and that this mtDNA methylation status persisted after the removal of Aβ, inducing a cerebrovascular endothelial damage memory that likely contributes to AD progression [98].
Few studies investigated a potential involvement of altered mitoepigenetic mechanisms in Parkinson's disease. Decreased D-loop methylation levels were observed in the substantia nigra of 10 PD patients with respect to healthy matched controls [94]. More recently, no differences in platelet mtDNA methylation levels at MT-TL1 and MT-CO1 genes were observed between PD patients and control subjects [99].
Taken together, these studies suggest that mitochondria impairment characterizing neurodegenerative diseases could be related to altered mitoepigenetic mechanisms, with the potential to provide new insights into their etiopathogenesis and new biomarkers of diagnosis and progression of these diseases.

Altered mtDNA Methylation in Other Diseases
MtDNA methylation alterations have also been associated with other human diseases than cancer, cardiovascular diseases and neurodegeneration (Table 5).  Infantino and co-workers observed decreased SAM availability in Down's syndrome lymphoblastoid cells, with a consequent reduction in mitochondria methyl uptake, leading to mtDNA hypomethylation [102]. Analysis of mtDNA methylation in liver biopsies from patients with non-alcoholic steatohepatitis revealed that MT-ND6 methylation was significantly higher than that observed in individuals with simple steatosis [103]. Moreover, MT-ND6 mRNA expression was decreased in NASH patients. Interestingly, the methylation status of the MT-ND6 gene was inversely correlated with physical activity, suggesting that epigenetic changes in mtDNA are potentially reversible [103]. Methylation analysis in DNA extracted from fetal cord blood of newborns from mothers with placenta insufficiency showed that D-loop methylation levels were decreased in the cases with respect to controls and associated with poorer fetal outcomes, as indicated by their correlation with gestational age, fetal weight and umbilical vein oxygen partial pressure [104]. No alterations in mtDNA methylation were detected in individuals with major depressive disorder and attentiondeficit hyperactivity disorder when compared to healthy control subjects [105,106].

Effects of Environmental Exposure and of Nuclear Genetic Variants on mtDNA Methylation
It is well known that the individual genetic background and environmental factors can impact the health status by modulating nuclear epigenetic mechanisms [107]. Recent research clearly suggests that also mtDNA methylation and hydroxymethylation patterns can be modulated by exposure to various environmental agents and by nuclear genetic variants (Table 6).   Peripheral blood of individuals exposed to particulate matter (PM 1 ), air benzene and traffic-derived elemental carbon exposure Pyrosequencing D-loop region, MT-TF and MT-RNR1 genes MT-TF and MT-RNR1 methylation levels positively correlated with PM 1 , as well as with mtDNA copy number [108] Frontal lobe of Wistar rats exposed to the endocrine disruptor polybrominated dyphenil ether BDE-47 Pyrosequencing, Immunohistochemistry Mt-co1, Mt-co2 and Mtco3 mitochondrial genes; Bdnf and Nr3c1 nuclear genes. Global nuclear 5-hmC and 5-mC content.
Decreased methylation levels of Mt-co2, Bdnf and Nr3c1 genes, together with decrease in global DNA methylation in rats exposed to BDE-47 MT-TR and MT-NAD4L methylation higher in the OO and PO than in the FO, whereas MT-RNR1 methylation was lower in the OO group than in the FO group [111] Effect of maternal smoking during pregnancy in placenta and foreskin

Pyrosequencing D-loop region
Positive association between D-loop methylation and maternal smoking during pregnancy [112] Association between exposure to particulate matter (PM 2.5 ) and blood mtDNA methylation in relation to heart rate variability (HRV) markers Pyrosequencing D-loop region, MT-TF and MT-RNR1 genes D-loop methylation negatively associated with PM 2.5 levels. Participants with higher mtDNA methylation levels were more susceptible to the effect of PM 2.5 on HRV. [113] Polycystic ovaries (PCO) of gilts with hyperhomocysteinemia Bisulfite sequencing D-loop region and MT-RNR1, MT-RNR2 and ND4 genes mtDNA significantly hypermethylated in all regions analyzed in PCO with respect to healthy ovaries [114] Liver of large yellow croaker fed with one of three diets characterized by a low (6%), moderate (12%, the control diet) or high (18%) crude lipid content Pyrosequencing D-loop region and MT-ND6 and MT-RNR1 genes D-loop methylation levels higher in the high-lipid than the control group. Increased mtDNA copy number in the high-lipid group. [115] Liver of rats fed with high levels of fructose and with normal diet (control group) Enzyme-linked immunosorbent assay Global 5-mC and 5-hmC mtDNA content Global hypomethylation of mtDNA in fructose-fed rats as well as higher mtDNA content and transcription of several mtDNA genes [116]  The mtDNA methylation was lower in chrome-exposed compared to control subjects [117] Effect of maternal smoking during pregnancy on placenta Pyrosequencing D-loop region, and MT-RNR1 gene; nuclear CYP1A1 gene Higher MT-RNR1 gene methylation levels and lower mtDNA content in placenta samples of smoker pregnant mothers [118] Effect of cord blood thyroid hormones thyroxine (T4) and triiodothyronine (FT3) on placenta methylation levels MT-RNR1 and D-loop methylation inversely correlated with thyroid hormones levels. Positive association between thyroid hormones and placenta mtDNA content. [119] Peripheral blood of welders exposed to respirable dust Pyrosequencing D-loop region, MT-TF gene Decreased D-loop and MT-TF methylation and increased mtDNA copy number associated with welding fumes exposure [120] Hippocampus of rats that received iron in the neonatal period and cannabidiol (CBD) for 14 days in adulthood Enzyme-linked immunosorbent assay Global 5-mC and 5-hmC mtDNA content Iron induced decreased levels of both 5-mC and 5-hmC. CBD induced increase in 5-hmC but not in 5-mC. [121] Peripheral blood of individuals exposed to arsenic in drinking water and of control subjects Methylation-specific PCR D-loop region and MT-ND6 gene D-loop and MT-ND6 hypomethylation, increased MT-ND6, MT-ND4 and TFAM gene expression as well as increased mtDNA copy number in individuals exposed to arsenic [59] Glioblastoma multiforme HSR-GBM1 cell line exposed to the DNA demethylation agents 5-azacytidine and vitamin C

MeDIP-Seq Multiple mtDNA sequences
Reduced mtDNA methylation at most of the regions analyzed [122] Association between mtDNA methylation and mitochondrial-derived peptides (MDPs), including HN, MOTS-c and SHLPs in cord blood tissue from newborns exposed to air pollutants during gestation Pyrosequencing D-loop, MT-TF and MT-RNR1 genes Liver tissue from mice exposed for 10 weeks to particulate matter had higher MDP levels. DNA methylation of two distinct regions of the D-Loop was associated with levels of MDPs. [123] Frozen platelet pellets from 26 women peripheral blood samples (   Degree of D-loop methylation slightly decreased. Although only modest alterations were seen in the degree of mtDNA methylation, these strongly correlated with IL-6 and IL-10 expression. [127] Correlations among common polymorphisms of genes required for one-carbon metabolism (MTHFR, MTRR, MTR and RFC-1) and DNA methylation reactions (DNMT1, DNMT3A and DNMT3B) and mtDNA methylation in peripheral blood of 263 subjects

MS-HRM D-loop region
MTRR 66A > G and DNMT3A −448A > G polymorphisms were significantly associated with D-loop methylation levels [128] Placental tissue from women who smoked during pregnancy, women with high air pollutant exposure and a control group with low air pollutant exposure Pyrosequencing D-loop region; PINK1, DNA2 and POLG1 nuclear genes D-loop methylation higher in mothers that smoked extensively or highly exposed to air pollutants. D-loop methylation levels inversely correlated with placental mtDNA content and associated with birth weight.
The main mitochondrial region analyzed in those studies was the D-loop region, which has been found to be negatively associated with PM 2.5 blood levels [113], welding fumes [120], arsenic [59], maternal smoking during pregnancy in placenta and foreskin [112], 5-azacytidine [70] and cord blood thyroid hormones levels in placenta [119]. On the other hand, increased D-loop region methylation has been associated with PM 2.5 levels together with maternal smoking in placenta [56,129], hyperhomocysteinemia [114], high glucose levels [79], high lipid levels [115], L-carnitine [124] and mitochondrial-derived peptides [123]. Another mtDNA region frequently investigated is the MT-RNR1 gene whose methylation levels have been found to be positively associated with PM 1 [108], PM 2.5 [56] and maternal smoking [118] and negatively associated with chrome exposure [117], thyroid hormones [119] and olive oil consumption [111]. However, many other mtDNA regions have been found to be sensitive to environmental cues, thus suggesting that mtDNA could be used as a sensor of environmental stressors.
There is also evidence that some nuclear genetic variants could impact mtDNA methylation. For example, a strong influence from the nuclear genome was observed on mtDNA methylation patterns in glioblastoma and osteosarcoma cells, since the same mtDNA genotype under different nuclear genomes associated with differential mtDNA methylation patterns [73]. Indeed, by using cell cultures with the same mitochondrial genome but with different nuclear genomes, derived from osteosarcoma or glioblastoma cells, several differentially methylated CpG sites located across the whole genome have been detected [73]. More recently, polymorphisms of genes involved in one-carbon metabolism, namely, MTRR 66A > G and DNMT3A −448A > G, have been found to be significantly associated with D-loop methylation levels [128]. The importance of one-carbon metabolism in mitochondria metabolism has also been shown by a study performed in individuals belonging to three families with segregate mutations in the SLC25A26 gene, which encodes for the mitochondrial S-adenosylmethionine (SAM) transporter, required for SAM uptake from the cytosol [130]. Probands of these families showed high clinical heterogeneity, which ranged from neonatal mortality resulting from respiratory insufficiency and hydrops to childhood acute episodes of cardiopulmonary failure and slowly progressive muscle weakness. The authors also showed that SLC25A26 mutations caused severe abrogation of SAM transport capacity, leading to mitochondrial methylation insufficiency, which led to altered mitochondria RNA stability and altered protein modifications [130]. Additionally, lymphoblastoid cells of individuals with Down' syndrome, a genetic disorder caused by the presence of all or part of a third copy of chromosome 21, have been found to be depleted in mitochondrial levels of SAM, with concomitant global mtDNA hypomethylation [102]. The authors suggested that this could be due to the overexpression of cystathionine-betasynthase, located on chromosome 21, which modifies the levels of several intermediates of the cellular one-carbon metabolism in Down's syndrome, including SAM. Interestingly, two rare hereditary neurodegenerative diseases, the autosomal dominant cerebellar ataxia, deafness and narcolepsy (ADCA-DN) and the hereditary sensory neuropathy with dementia and hearing loss (HSN1E), whose phenotypes share some clinical features typical of mitochondrial diseases, including mitochondrial dysfunction, optic atrophy, peripheral neuropathy and deafness, have been linked to dominant pathogenic mutations in the DNMT1 gene [131].
Methylation levels of mtDNA have also been found to be sensitive to mutations in genes related to ALS pathogenesis [96,100]. Indeed, in individuals belonging to families with segregate ALS-linked mutations, we observed that individual carriers of SOD1 and C9orf72 mutations had significantly higher levels of the mtDNA copy number than noncarriers of ALS-linked mutations, but only SOD1 mutation carriers showed a significant reduction in D-loop methylation levels [96]. On the other hand, TARDBP and FUS carriers showed no significant increase in the mtDNA copy number and no significant decrease in D-loop methylation compared to noncarriers of ALS-linked mutations. In a later study, we observed that both SOD1-mutant and sporadic ALS patients had lower D-loop methylation levels compared to controls, while C9orf72 ALS patients showed similar D-loop methylation levels to control subjects [100].

Debate on the Existence and Function of mtDNA Methylation
Although the presence of different DNMTs and epigenetic marks, including 5-mC and 5-hmC, inside mitochondria has been reported by several authors, the presence and biological function of mtDNA methylation have been hotly debated in recent years. Indeed, some researchers reported that mtDNA methylation is a very rare event, or even absent, suggesting that the literature reporting the presence of methylation marks in mtDNA could be biased by technical limitations [34][35][36]132]. Overall, these studies suggest that, although the possibility of the presence of low levels of mtDNA cannot be ruled out, it is doubtful that a biologically significant function should be considered at the extremely low frequency of CpG methylation detected.
Evidence of the absence of DNA methylation marks in mitochondria was reported about forty years ago [133,134]. However, the first study conducted in more recent years reporting the absence of methylation in mtDNA and performed using modern techniques, including bisulfite sequencing and next-generation sequencing, was conducted in 2013 in a human colon cancer cell line (HCT116) and in primary human cells [34]. The authors found a lack of an appreciable amount of methylated cytosines, arguing that it is highly unlikely that CpG methylation plays any role in the direct control of the mitochondrial function, confirming this evidence by analyzing ten genome-wide bisulfite sequencing published datasets. Of note, van der Wijst and collaborators investigating mtDNA methylation in different human cell lines, including HCT116 cells, found evidence of methylation levels, although low, in all the cells investigated except the HCT116 cell lines in which no evidence of 5-mC was detected [54]. Moreover, by reanalyzing the datasets used by [34], a recent study confirmed that CpG methylation was low, but when the analysis was carried out in a strand-specific manner, high non-CpG methylation was observed in the L-strand [50], meaning that the existence of mtDNA methylation cannot be ruled out in these datasets. In later studies, it was suggested that the circular structure of mtDNA could interfere with bisulfite conversion, which is a preliminary sample preparation step for several DNA methylation techniques, thus leading to overestimation of mtDNA methylation. During DNA treatment with bisulfite, unmethylated cytosines are converted to uracil residues, while methylated cytosines remain unchanged; the subsequent PCR amplification or extension step introduces thymine to the positions of uracil and cytosine to those of 5-mC. Liu and co-workers showed that that the methylation values obtained from linear mtDNA were significantly lower than those obtained from circular mtDNA and reported that CpG methylation in human mtDNA is a very rare event at most DNA regions and that it seems that such low level of mtDNA methylation would have limited or absent functional significance in the control of mitochondrial gene expression [132]. However, the authors also reported that average methylation levels of the 83 CpG sites investigated were less than 2% with the exception of two CpG sites within the D-loop region, which showed 5% of methylation, concluding that the higher methylation status at this region might play a role in the control of mtDNA replication [132]. Additionally, other studies have shown that the circular structure of mtDNA can influence bisulfite treatment and the detection of mtDNA methylation [35,135]. In a later study, the mtDNA methylation status was investigated in mouse embryonic stem cells and liver and brain tissues by means of three distinct methods for detecting 5-mC, namely, bisulfite sequencing, McrBC (an approach that uses enzymes sensitive to CpG methylation) and liquid chromatography mass spectrometry (LC/MS), finding that methylated cytosines are fairly low, thus arguing that it is unlikely that mtDNA methylation plays a role in mtDNA gene expression or mitochondrial metabolism [36]. It should be noted that by means of LC/MS, which is more sensitive than the other two methods in detecting 5-mC, weak methylation signals, corresponding to 18-30 5-mC residues per molecule of mtDNA, were detected [36]. Collectively, these studies suggest that it is doubtful that a biologically significant function should be considered at the extremely low CpG methylation levels detected in mtDNA. Moreover, they highlight that methodological precautions should be adopted when studying mtDNA methylation, as the mtDNA circular structure, the procedures used to extract DNA samples, the DNA bisulfite treatment, primer design and the use of inadequate control templates could lead to erroneous methylation and hydroxymethylation quantification [132,136].
However, several other studies that have carefully followed the necessary precautions for the study of mtDNA methylation, or that used techniques for which bisulfite treatment was not necessary, successfully identified the presence of methylation in the mitochondrial genome [61,73,84,110,127,137]. For example, Patil and co-workers performed a whole-genome bisulfite sequencing after linearization of mtDNA, finding a higher degree and frequency of methylation, particularly in the L-strand when compared to the H-strand in various human cell cultures [61]. Similarly, after linearizing mtDNA and using mtDNA methylation-negative and positive controls for bisulfite sequencing, an average CpG methylation of 10% was detected over the mtDNA genome of tumor cell cultures [73]. To overcome the possibility that the mtDNA circular structure might also affect immunoprecipitation-based methods that do not require bisulfite treatment, some authors performed a denaturation step before the incubation of mtDNA with 5-mC or 5-hmC antibodies, finding, however, the presence of methylated and hydroxymethylated cytosines across the D-loop region [73,110,122]. Moreover, a study investigating mtDNA methylation in human cells by means of a third-generation sequencing platform, namely, the nanopore sequencer (MinION, Oxford Nanopore), which is a high-throughput nanopore-based single-molecule device that can directly sequence the entire mtDNA molecule, minimizing potential nucleotide errors introduced by PCR amplification, and preserving epigenetic modifications by avoiding the use of bisulfite treatment, revealed clearly detectable levels of mtDNA methylation [74]. Likewise, by using high-coverage nanopore sequencing at the single-molecule level, mtDNA methylation was recently investigated in different sample types and biological conditions [75]. Methylation was overall higher in tissues compared to cell lines, and despite that mtDNA methylation levels were generally low, global and single-base differences were found between cancer tissues and the adjacent healthy tissues. Moreover, although low, the methylation levels detected in mtDNA have been frequently related to changes in both mtDNA gene transcription and mitochondrial replication (see Tables 1-6). Furthermore, the application of recent technologies able to overcome the limits for the detection of mtDNA methylation revealed that knockdown of DNMTs perturbs mtDNA methylation and gene expression levels, as well as the mtDNA copy number and oxygen respiration [50,61].
Overall, mitoepigenetic investigations performed until now suggest that methylation and hydroxymethylation marks are present in the mtDNA genome, albeit at lower levels compared to those detectable in nuclear DNA, and are altered in different human diseases and sensitive to environmental factors and nuclear DNA genetic variants ( Figure 1). Although there are several indications that mtDNA epigenetic modifications could play a role in the biological function of mitochondria, further studies are needed to better elucidate to which extent mtDNA gene expression and replication are regulated by those modifications.
In this regard, it should be outlined that from an epigenetic point of view, the mitochondrial genome is largely different compared to the nuclear one, as it lacks nucleosomal chromatin and CpG islands, meaning that the well-known epigenetic regulation of the nuclear genome could be different in mitochondria [138]. A better understanding of the interplay between mtDNA modifications and nucleoid post-transcriptional modifications is pivotal to fully appreciate the biological role of mitoepigenetic mechanisms in the regulation of mitochondrial metabolism. For example, Rebelo and co-workers reported that unbalanced levels, either low or high, of TFAM result in decreasing mtDNA methylation, and this scenario could explain why mtDNA methylation has been observed in some biological models but not in others [43]. Furthermore, it may also be possible that other as yet unidentified epigenetic mechanisms may regulate mtDNA replication and gene expression. Mitochondrial DNA regions whose methylation levels were found to be associated with different human diseases, environmental factors and nuclear DNA genetic variants. In the red box, the meaning of the symbols used is reported.

Conclusions
Alterations in mtDNA gene expression and replication underlie several human pathologies, and a better understanding of the regulation of these mechanisms is desirable to understand the pathophysiology of associated disorders. The discovery of mitoepigenetic mechanisms has opened a new window of research that could provide new knowledge on the mitochondria regulation in both physiological and pathological conditions, with the potential to yield specific biomarkers for several human diseases.
Although the literature reported in the current review indicates that mitoepigenetics has been poorly investigated compared to nuclear epigenetics in human diseases, it is clear that researchers are increasingly interested in this research field. One of the reasons that have delayed mitoepigenetic investigations compared to nuclear epigenetic ones has been the limitation in methodologies until a few years ago [68]. For example, one of the main techniques used in the study of nuclear DNA methylation, the Illumina methylation array, does not cover the mtDNA sequence, so information regarding mtDNA methylation patterns in different human tissues and diseases is still largely missing.
Although some reports suggested the absence of mtDNA methylation in human and mouse mitochondria [34][35][36]132], several other studies showed the presence of consistent levels of mtDNA methylation, and, in particular, the regulatory D-loop region seems to represent one of the mtDNA loci that most frequently undergo methylation and hydroxymethylation, and its methylation levels have been associated with several diseases, including colorectal and breast cancer [55,58,71], obesity [80], insulin sensitivity [81], diabetes [84], arterial stenotic/occlusive disease [85], AD [94,95,97,101], ALS [49,96,100] and PD [94]. Moreover, some studies also showed that D-loop methylation is sensitive to several environmental factors, as well as to nuclear DNA genetic variants (Table 6).
In conclusion, although our knowledge on the involvement of the mitoepigenetic mechanisms in human diseases is still in its infancy, data obtained until now encourage further addressing this research field, which could shed new light on the pathogenic mechanisms underlying human diseases, potentially providing new biomarkers of disease diagnosis and progression as well as new molecular targets for therapeutic interventions.