The mitochondrial epigenome and noncoding genome in aging and brain diseases

Abstract

The mitochondrial genome is a dynamic regulatory platform that extends far beyond its protein-coding capacity. Mitochondrial DNA harbors epigenetic modifications and produces diverse noncoding RNAs that influence mitochondrial function and cellular homeostasis. Evidence from animal models and human studies indicates that mitochondrial DNA methylation, mitochondrial-encoded noncoding RNAs, and cytosolic release of mitochondrial nucleic acids regulate transcription, metabolism, and stress signaling. Furthermore, these mechanisms are altered in aging, neurodegenerative disease, and acute brain injury. In this review, we synthesize emerging insights into how the mitochondrial epigenome and noncoding genome shape mitochondrial function and serve as regulatory interfaces linking cellular stress and neuronal vulnerability in brain aging and disease.

Layers of mitochondrial genome regulation

Mitochondria are essential regulators of cellular homeostasis, coordinating ATP production, calcium buffering, redox signaling and cell death pathways [1]. In addition to these well-established roles, increasing evidence further implicates mitochondria in the regulation of gene expression and cellular stress responses [2–4]. Central to these functions is the mitochondrial genome itself: although mammalian mitochondrial DNA (mtDNA) encodes only 13 core oxidative phosphorylation proteins, the mtDNA engages in regulatory functions that extend well beyond its canonical coding role.

mtDNA is subject to epigenetic modifications, including methylation [5–7], and gives rise to diverse classes of noncoding RNAs that influence mitochondrial and cellular function [8–11]. This indicates that the mitochondrial genome can be remarkably dynamic, with noncoding RNAs and epigenetic marks contributing to the fine-tuning of mitochondrial gene expression, protein homeostasis, and inter-organelle communication. Furthermore, cytosolic release of mtDNA and mitochondrial double-stranded RNA (mt-dsRNA) are significant noncoding modulators of inflammatory signaling [12,13]. Importantly, these noncoding regulatory features appear to be highly responsive to cellular context. Across aging, neurodegenerative disease, and acute brain injury, mtDNA modifications, mitochondrial-derived noncoding RNAs, and mitochondrial nucleotide cytosolic release are altered in ways that correlate with, and increasingly appear to drive, changes in mitochondrial function, metabolic flexibility, and pathogenic signaling.

Technological advances in mitochondrial sequencing, epigenomic profiling, and transcriptomics have been instrumental in uncovering this dynamic layer of mitochondrial genome regulation, enabling detection of mtDNA modifications and mitochondrial-encoded noncoding RNAs within distinct cellular compartments [14–16]. In this review, we synthesize current insights into the mitochondrial epigenome and noncoding mitochondrial genome, focusing on studies based on both animal model systems and human samples. We emphasize conceptual advances and highlight how the epigenetic and noncoding mitochondrial genome shapes vulnerability across aging and brain disease. By integrating these perspectives, we frame the mitochondrial epigenome and noncoding mitochondrial genome as key regulators of neural resilience and as promising targets for therapeutic intervention in brain aging and disease.

Diverse regulatory outputs of the mitochondrial epigenome and noncoding genome

Mitochondrial gene expression is regulated within highly organized intramitochondrial spatial architecture that gives rise to dynamic transcriptional and translational hubs. Studies in human cells indicate that mitochondrial transcription is spatially coordinated within discrete nucleoids where mtDNA packaging, transcriptional activity and RNA processing are tightly coupled, rather than occurring uniformly across mtDNA [16]. Complementary evidence further demonstrates that the long polycistronic RNAs transcribed from mtDNA are processed and translated within spatially restricted microenvironments, supporting efficient coordination between transcription, RNA maturation, and protein synthesis [2]. In addition to generating protein-coding transcripts, the mitochondrial genome gives rise to multiple classes of noncoding RNAs, including mitochondrial-encoded long noncoding RNAs (mt-lncRNAs), mitochondrial-encoded circular RNAs (mecciRNAs), and mitochondrial microRNAs (mitomiRs) (Fig. 1). These RNA species introduce regulatory flexibility into mitochondrial gene expression and expand the functional output of the mitochondrial transcriptome (Box 1).

Figure 1. Elements of the mitochondrial genome.

mtDNA consists of a small circular DNA molecule packaged into nucleoprotein structures termed nucleoids and transcribed as long polycistronic transcripts. A major regulatory region within mtDNA is the D-loop, which contains promoters and replication elements and serves as a key binding site for TFAM. The D-loop is also the primary region where mtDNA methylation has been reported including 5mC, 5hmC, and 6mA. mtDNA contains a heavy strand (H, outer ring) and a light strand (L, inner ring), which differ in nucleotide composition and transcriptional organization. The H strand encodes the majority of genes, while many noncoding RNAs are derived from the L strand, or from antisense transcription across the mitochondrial genome. mt-lncRNAs can derive from sense, antisense, intergenic or chimeric transcripts, while most characterized mecciRNAs derive from conventional coded transcripts. mt-dsRNA and mtDNA are produced and/or released into the cytosol following mitochondrial stress. mRNA: messenger RNA; rRNA: ribosomal RNA; tRNA: transfer RNA; miRNA: microRNA

Box 1: The mitochondrial epigenome and noncoding genome.

Mitochondrial DNA methylation

5-methylcytosine (5mC):

DNA methyltransferases (DNMTs), including isoforms with mitochondrial targeting sequences, catalyze the addition of methyl groups to cytosine residues of mtDNA [6,98]. 5mC is generally associated with transcriptional repression.

5-hydroxymethylcytosine (5hmC):

Ten-eleven translocation (TET) enzymes catalyze the oxidation of 5mC to 5hmC, a modification associated with transcriptional activation [6].

N6-methyldeoxyadenosine (6mA):

Enzymes such as N6-adenosine methyltransferases (METTLs) generate 6mA though the covalent addition of a methyl group to the nitrogen at the sixth position of adenine residues in DNA. 6mA is associated with gene repression [5].

See Box 2 for further controversies surrounding DNA methylation on mtDNA.

Mitochondrial-encoded noncoding RNAs

mt-lncRNAs:

These long noncoding RNAs are generated from sense, antisense or intergenic regions of mtDNA. mt-lncRNAs can bind to DNA, RNA or protein to dynamically regulate a diverse array of processes including mtDNA stability, transcription, translation, and cell signaling. mt-lncRNAs may also localize to the cytoplasm and nucleus where they are implicated in processes such as mitophagy or chromatin states, respectively [18,99].

mecciRNAs:

These circular RNAs are generated from mitochondrial transcripts through back-splicing or RNA processing events that covalently join transcript ends. mecciRNAs can shuttle between mitochondria and the cytoplasm and participate in mitochondrial protein import [21].

mitomiRs:

These microRNAs arise from mitochondrial transcripts that undergo processing by components of the microRNA biogenesis machinery into mature ~20–22 nucleotide RNAs. The term “mitomiRs” often includes nuclear-encoded microRNAs localized to mitochondria. In this article, we refer specifically to putative mtDNA-encoded miRNAs. Similar to canonical microRNAs, mitomiRs may regulate gene expression through sequence-specific binding to target mRNAs, leading to altered mRNA stability or translational repression [22]. However, evidence for bona fide mtDNA-encoded mitomiRs in the brain remains limited.

Cytosolic mitochondrial genome-derived nucleic acids

Cytosolic mt-dsRNA:

mt-dsRNA arises from bidirectional transcription of complementary mtDNA strands and are normally degraded by the mitochondrial degradasome. Under conditions of mitochondrial stress, impaired RNA degradation, or epitranscriptomic modification, mt-dsRNA accumulate and escape into the cytosol, binding to innate immune sensors (such as RIG-I and MDA5) thereby triggering immune response [12] [23] [100].

Cytosolic mtDNA:

Mitochondrial injury, such as mitochondrial membrane permeabilization, can trigger release of mtDNA into the cytosol. Cytosolic mtDNA is detected by the cGAS-STING pathway, thereby activating inflammatory pathways [13].

The role of mitochondrial noncoding RNA species is not limited to transcriptional and translational processes within the mitochondria as they also participate in broader cellular regulatory networks. This is demonstrated by mt-lncRNAs that impact nuclear transcriptional activity. In human cell systems, the mt-lncRNA MDL1 inhibits translocation of p53 transcription factor from the mitochondria to the nucleus, thereby regulating nuclear p53-dependent transcriptional programs in response to cellular stress [17]. This establishes a direct retrograde signaling mechanism in which mt-lncRNAs may influence nuclear transcription factor activity from within the mitochondria. However, in various human cell lines including endothelial cells which may have implications in brain vasculature, several mt-lncRNAs were also shown to undergo active mitochondria-to-nucleus trafficking to regulate nuclear gene expression by interacting with chromatin and RNA processing machinery to modulate metabolic pathways in response to stress [18,19]. In parallel, the lncND5 mt-lncRNA regulates mitophagy following cytoplasmic transport in human arterial smooth muscle cells during hypoxic conditions [20], suggesting a potential mechanism relevant to ischemic cerebrovascular injury. mecciRNAs further expand this framework by regulating stress-responsive shuttling of nuclear-encoded proteins into the mitochondria through effects on mitochondrial permeability transition pore opening and chaperone machinery [21]. For example, in a mouse neuronal cell line, mecciATP6 was shown to maintain mitochondrial homeostasis by regulating the stress-responsive mitochondrial import and abundance of a nuclear-encoded RNA-binding protein, and mecciATP6 knockdown further reduced ATP production and increased intracellular reactive oxygen species [8]. Moreover, while specific roles for mitochondrial-encoded mitomiRs remain to be determined, mitomiRs may modulate mitochondrial- or cytoplasmic-localized transcripts from both the mitochondrial and nuclear genomes [22]. Finally, studies identifying that mt-dsRNA release into the cytosol coordinates immune signaling, independent of its translational coding ability, has expanded our understanding of mitochondrial-mediated inflammatory signaling (Fig. 2) [12,23]. Together, these studies demonstrate that mitochondrial noncoding RNAs not only function within mitochondria, but also engage nuclear and cytoplasmic regulatory processes, linking mitochondrial transcriptional output to cellular stress responses.

Figure 2. Mitochondrial nucleic acid cytosolic release drives inflammatory signaling in aging and neurological disease.

Mitochondrial dysfunction can promote the release of mtDNA and mt-dsRNA into the cytosol, where these nucleic acids engage innate immune signaling pathways. Cytosolic mtDNA activates cGAS–STING signaling, whereas cytosolic mt-dsRNA activates the RNA sensors RIG-I and MDA5, ultimately promoting activation of inflammatory transcription factors and neuroinflammatory gene programs. Altered PINK1 signaling and SEC61A1-mediated mitochondrial–endoplasmic reticulum interactions have been implicated in mitochondrial nucleic acid release. These pathways are increasingly associated with aging and neurological disorders, highlighting mitochondrial nucleic acid leakage as a potential mechanism linking mitochondrial dysfunction to neuroinflammation. ER: endoplasmic reticulum

The mtDNA is subject to epigenetic modifications related to DNA methylation, particularly in regulatory regions such as the mitochondrial displacement loop (D-loop). Current evidence supports a model in which mtDNA methylation influences mitochondrial transcription and function, thereby shaping downstream signaling pathways that can reprogram mitochondrial gene expression. However, some controversy exists on the prevalence of mtDNA methylation (Box 2). Further work using single-cell multi-omic approaches has revealed that the mitochondrial and nuclear genomes are co-regulated at the cellular level, where cell-to-cell variation in mtDNA content is coupled to changes in mitochondrial transcription and broader phenotypic adaptations [14,15]. These observations highlight a potential axis by which mitochondrial epigenetic mechanisms may shape cell-type-specific mtDNA regulatory features and mito-nuclear communication.

Box 2: Controversy of methylation in the mitochondrial genome.

The presence and extent of mtDNA cytosine methylation remain controversial, partly due to limitations of standard bisulfite sequencing, which cannot distinguish 5mC from 5hmC and may overestimate methylation due to bisulfite-induced DNA degradation and misalignment to nuclear mitochondrial pseudogenes. In light of these limitations, bisulfite-sequencing analyses with more rigorous controls and mapping approaches indicate that previous reports of robust mtDNA cytosine methylation may be an overestimation [63,101]. For example, targeted bisulfite sequencing at single base resolution in the human brain revealed low cytosine methylation levels across the mitochondrial genome [37]. However, cytosine methylation was observed at several sites, primarily in non-CpG contexts, with the highest methylation observed in the D-loop and ND5 regions. Furthermore, differential cytosine methylation was observed in anatomically distinct brain regions, suggesting regional differences in mtDNA methylation [37]. Similarly, ultra-deep whole genome bisulfite sequencing in the human cortex revealed overall low mtDNA cytosine methylation profiles confined to the D-loop [63]. This D-loop methylation was associated with mtDNA transcript levels [63], which continues to support a role for mtDNA methylation in regulating mtDNA gene expression in the human brain.

One method to distinguish methylation marks is the use of 5mC, 5hmC or 6mA antibody-based enrichment and detection methods of mtDNA, with the caveat of potential antibody cross reactivity and reduced sensitivity compared to sequencing [39,83]. Other studies have used nanopore long-read sequencing, which allows for simultaneous detection of DNA sequence and base modifications, with the advantage of being able to distinguish 5mC, 5hmC and 6mA separately, while avoiding biases introduced by bisulfite treatment. In the mouse brain and human neural cell lines, long-read sequencing revealed low levels of 5mC in discrete mtDNA loci [7]. Interestingly, 5mC levels were higher in the mouse brain tissue compared to the human neural cell lines, suggesting possible differences between in vivo and cell culture systems, or between mammalian species [7]. In non-neural cell lines and tissue (including liver, kidney, and heart), long-read sequencing failed to detect CpG methylation on mtDNA [102], potentially suggesting higher mtDNA methylation in the brain, although more studies are needed to confirm this.

The presence and abundance of 6mA in mammalian DNA have also been controversial, in part because its low abundance makes detection particularly susceptible to technical artifacts [103]. Bacterial DNA contamination, residual RNA, antibody cross-reactivity, and background signals in mass spectrometry can inflate 6mA levels, while long-read sequencing approaches may generate false-positive as neighboring 5mC modifications can alter polymerase kinetics during sequencing and potentially be misassigned as 6mA [104]. Nevertheless, studies using highly purified mtDNA and orthogonal detection approaches suggest that 6mA may be enriched in the mitochondrial relative to the nuclear genome [5], although its absolute abundance, locus-specific distribution, and functional significance remain incompletely resolved.

Despite these controversies, the presence of mitochondrial-localized DNMT, METTL, and TET enzymes helps to support the existence of 5mC, 6mA and 5hmC on mtDNA [5,6]. Further evidence indicates certain DNMT isoforms contain mitochondrial targeting sequences, bind to mtDNA, and promote de novo methylation [6,98]. Overall, mtDNA methylation is still an unresolved area in which the presence, abundance, and functional significance of these marks likely depend on cell type, physiological state, and the sensitivity and specificity of detection methodologies. Thus, the interpretation of mtDNA methylation must be weighed against the technological approach used in the study.

Mitochondrial epigenome and noncoding RNAs in brain aging and neurodegenerative diseases

Age-dependent alterations in the mitochondrial epigenome and noncoding genome occur against a broader decline in mitochondrial function that progressively compromises cellular resilience and increases brain vulnerability to neurodegeneration. This mitochondrial dysfunction is a defining feature of both brain aging and neurodegenerative disease and is characterized by impaired oxidative phosphorylation and ATP production, altered mitochondrial dynamics and calcium handling, increased reactive oxygen species generation, and reduced mitochondrial quality control [24]. These changes are particularly consequential in the brain where high energetic demands of neurons lead to sensitivity to even modest reductions in mitochondrial capacity. Human magnetic resonance imaging and spectroscopy imaging studies indicate greater mitochondrial vulnerability in the frontal cortex than in the occipital cortex, white matter and deep gray matter in the aged brain [25] [26], suggesting that regional energetic deficits and oxidative stress may contribute to the selective vulnerability observed in aging and neurodegenerative diseases. Furthermore, it should be noted that aging-induced mitochondrial decline is not unique to neurons. For example, aged mouse microglia exhibit impaired mitochondrial respiration and glycolysis which may promote pro-inflammatory phenotypes [27], aged mouse oligodendrocyte precursor cells display smaller mitochondria and reduced motility that may impair remyelination [28], and aged mouse brain micro-vessels show reduced mitochondrial respiration and ATP production that may compromise blood brain barrier integrity [29]. Because many of these cell- and region-specific deficits are also prominent features of neurodegeneration [30–33], the age-dependent erosion of mitochondrial fitness may progressively reduce cellular resilience and lower the threshold at which genetic, metabolic or environmental insults precipitate neurodegenerative disease.

Within this context, regulatory mechanisms embedded within the mitochondrial genome may shape aging-induced mitochondrial dysfunction, with evidence implicating mtDNA methylation and noncoding RNAs as modulators of age-associated mitochondrial decline. Studies of mtDNA abundance, integrity and heteroplasmy have provided important markers of mitochondrial genomic dysfunction across aging and neurological disease, although their interpretation can be influenced by factors such as tissue and cell-type composition [34–36]. The mitochondrial epigenome and noncoding RNAs may provide complementary mechanistic insight by defining how the mitochondrial genome is expressed and communicated to the rest of the cell, thereby linking mitochondrial genomic alterations to changes in bioenergetics, stress signaling and inflammation. Notably, diverse neurodegenerative pathologies including amyloid and tau pathology and α-synuclein aggregation, appear to converge on mitochondrial regulatory dysfunction of the noncoding mitochondrial genome, suggesting a shared vulnerability rooted in disrupted mitochondrial gene control. The following subsections examine how this layer of epigenetic and noncoding mitochondrial regulation contributes to aging and major neurodegenerative diseases, while considering whether these alterations act as primary drivers of neuronal dysfunction, or arise as adaptive and maladaptive responses to cellular stress.

Aging

Alterations in mtDNA cytosine methylation have been observed in the aging brain, although the functional interpretation of these changes remain unclear. Single-base resolution bisulfite-sequencing in human post-mortem studies indicate that mtDNA CpG and non-CpG methylation increases with age, and cytosine methylation patterns vary across brain regions and between sexes, pointing to spatially distinct and sex-specific regulatory mechanisms [37,38]. While these studies appropriately mapped sequencing to the mitochondrial genome to exclude nuclear mitochondrial pseudogenes, standard bisulfite-sequencing approaches do not distinguish between 5mC and 5hmC. Thus, the specific contributions of each modification in mtDNA of the aging human brain are not clear. However, these changes point to the possibility that aging mitochondria may display aberrant DNMT or TET activity, or that aging mitochondria lose the ability to eliminate methylation on cytosines. In C. elegans, misregulation of 6mA on mtDNA detected via long-read sequencing, caused alterations in mtDNA copy number and transcript levels and led to aging phenotypes [39], suggesting aberrant mtDNA 6mA may be a driver of aging. A recurring theme in aging biology is the loss of transcriptional flexibility in response to stress. In the nucleus, aging is associated with reduced chromatin accessibility and impaired activity-dependent gene expression [40,41]. A similar principle may extend to the mitochondrial genome where mitochondrial gene expression becomes less adaptable to energetic or oxidative stress due to aberrant mitoepigenetics. Such a constraint could impair the ability of neurons to mount appropriate metabolic responses, thereby exacerbating vulnerability to degeneration. Furthermore, aging mice showed increased cytosolic mtDNA and activated cGAS–STING signaling in microglia, which led to aging-associated nuclear transcriptional programs linked to neuroinflammation and memory impairment [13]. This highlights the capacity of the mitochondrial genome to regulate both local mitochondrial transcription and retrograde nuclear gene expression in the aging brain (Fig. 2).

Noncoding RNAs may add an additional layer of regulation to influence aging-related phenotypes. Analysis of peripheral mononuclear blood cells in young and old patients revealed altered mecciRNAs expression profiles associated with aging, including decreased mecciRNA circMT-RNR2 [42]. Notably, circMT-RNR2 was shown to bind to tricarboxylic acid cycle (TCA) enzymes, promote TCA activity and mitochondrial homeostasis, and suppress senescence in human fibroblasts [42]. In parallel, cytosolic accumulation of mt-dsRNA may function as a key driver of senescence-associated inflammatory phenotypes in the aged mouse brain, through activation of the RNA sensors RIG-I and MDA5 [12]. In the post-mortem human prefrontal cortex, mt-dsRNA signatures increased beginning at the midlife transition and continued to rise with advancing age [43], suggesting that cytoplasmic mt-dsRNA accumulation may represent an early feature of age-associated inflammation in the brain. In mouse neurons, aging-associated mt-dsRNA release was regulated by endoplasmic reticulum subunit SEC61 translocon alpha 1 subunit (SEC61A1), which promoted mitochondrial RNA synthesis and cytosolic mt-dsRNA accumulation through modulation of endoplasmic reticulum–mitochondria contact sites [9]. Notably, neuron-specific SEC61A1 knockdown in aged mice reduced cytosolic mt-dsRNA and improved cognitive function [9], highlighting the therapeutic potential of targeting mitochondrial RNA dysregulation in brain aging.

Parallels between aging and neurodegenerative disease suggest that mitochondrial regulatory pathways may lie at the intersection of physiological aging and pathological processes. Just as nuclear epigenomic analyses of aging have identified pathways that predispose the brain to neurodegeneration, mitoepigenetic and mitochondrial RNA-based mechanisms may similarly define trajectories of neuronal vulnerability. Determining whether these noncoding mitochondrial genome changes represent early drivers of dysfunction or late-stage consequences of cellular stress will be critical for understanding the contribution of aging to neurodegenerative disease transitions.

Alzheimer’s disease

Alzheimer’s disease (AD) is characterized by the presence of amyloid-beta plaques and tau tangles leading to neurodegeneration and progressive cognitive decline. These pathogenic processes are accompanied by widespread metabolic dysfunction, with mitochondria emerging as central regulators of disease vulnerability [44]. Impairments in oxidative phosphorylation, increased reactive oxygen species, and altered mitochondrial dynamics are evident early in disease progression, preceding inflammation and overt neuronal loss [44]. Accumulating evidence in various mouse models of AD indicated that mitochondrial instability resulted in cytosolic release of mtDNA and cGAS-STING activation in microglia and endothelial cells, and was a major driver of neuroinflammation, neurotoxicity and AD deficits [45–47]. In particular, cGAS and STING were highly expressed in microglia associated with AD pathology (such as amyloid-beta plaque-associated microglia), which led to microglial senescence and neurodegeneration [46,47]. The early mechanistic drivers of mitochondrial dysfunction leading to mtDNA release remain unclear but could be the result of both upstream regulators and downstream consequences of mitochondrial damage.

In both human post-mortem brain tissue and animal models of AD, loss of TET activity and decreased 5hmC are highly associated with AD neuropathology [48,49], but whether these 5hmC changes specifically extend to mtDNA in AD is yet to be resolved. However, several studies suggest cytosine methylation of circulating mtDNA may serve as a biomarker for AD progression. Analysis of peripheral blood or plasma cell-free DNA indicated that D-loop cytosine methylation was increased in patients with mild cognitive impairments but remained unchanged or was reduced in patients with early or advanced stage AD [50–53]. While there is no consensus regarding the directionality of these methylation changes in AD, multiclass classification models using mtDNA cytosine methylation profiles predicted cognitive decline in AD patients with mild cognitive impairment [52]. These findings suggest that mtDNA methylation signatures may hold promise as biomarkers of AD progression, although further validation and mechanistic studies are required. Importantly, because these methylation studies used bisulfite-based approaches which do not distinguish between 5mC and 5hmC, differential regulation of these modifications, or methodological confounds could underlie some of the variability observed across studies (see Box 2).

In contrast to the mitochondrial epigenome, studies in rodent models provide more direct evidence that mitochondrial noncoding RNAs actively modulate disease-relevant phenotypes in AD. For example, in AD mouse models and human AD patients, the mt-lncRNA lncMtDloop, which enhanced mitochondrial transcription by promoting mitochondrial transcription factor A (TFAM) binding to mtDNA promoters, was found at lower levels in the prefrontal cortex and hippocampus [10]. In AD mice, restoration of lncMTDloop enhanced mitochondrial function and improved amyloid-beta pathology, synaptic impairments and neurobehavioral deficits [10]. Finally, similar to aging, there is emerging evidence that mt-dsRNA may also modulate AD. mt-dsRNA signatures were increased in the post-mortem prefrontal cortex of AD patients compared to healthy controls, which correlated with exacerbated neuropathology and cognitive impairment [43]. AD patients also showed increased dsRNA levels in plasma and increased SEC61A1 expression in the cortices of AD patients compared to healthy controls [9]. Levels of SEC61A1 were further correlated with cognitive decline in AD patients. Similarly, in a mouse model of AD, SEC61A1 protein levels and RNA-sensing innate immune signaling was observed in the cortex [9]. Collectively, these findings suggest that mitochondrial-encoded RNA networks represent an active regulatory layer linking mitochondrial dysfunction to neurodegenerative progression in AD.

Parkinson’s disease

Parkinson’s disease (PD) is characterized by the degeneration of dopaminergic neurons in the substantia nigra and the accumulation of α-synuclein aggregates [54]. Mitochondrial dysfunction is a central feature of PD, with impairments in complex I activity, increased oxidative stress, and disrupted mitochondrial quality control contributing to neuronal vulnerability [54]. Emerging evidence further suggests that disruption of mitochondrial genome integrity, resulting from both genetic and environmental factors, may represent an important mechanistic link between mitochondrial dysfunction and neuroinflammation in PD [55–57]. In particular, PD mouse models revealed that mtDNA damage and cytosolic release activated the STING DNA-sensing pathway and that inflammasome signaling preceded the onset of neurodegeneration and PD-like pathology [57–60]. In mouse neurons and microglia, defects in mitophagy through loss of PINK1 signaling promoted accumulation of mtDNA in the cytosol, further amplifying inflammatory responses [11,61]. Consistent with these findings, increased cytosolic mtDNA has been detected in α-synuclein aggregates in the medulla oblongata of human postmortem PD brains [11], linking mitochondrial genome dysregulation to hallmark neuropathological features of the disease. In contrast, whether mitochondrial RNA species, including mt-dsRNAs and other mitochondrial-encoded noncoding RNAs, contribute to PD pathogenesis remains largely unexplored. Given the emerging role of mitochondrial nucleic acids as mediators of mito-nuclear communication and innate immune signaling, defining how mitochondrial noncoding RNA networks are altered in PD may reveal additional mechanisms linking mitochondrial dysfunction to neurodegeneration and neuroinflammation.

Studies in the post-mortem brain of PD patients have shown significant disease-associated alteration in nuclear 5mC and 5hmC [62], but the role of mtDNA cytosine methylation in PD is not well characterized. Ultra-deep bisulfite sequencing in the human prefrontal cortex of PD patients and healthy controls showed no association between mtDNA cytosine methylation profiles and PD [63]. However, biomarker studies in peripheral blood indicated increased mitochondrial D-loop cytosine methylation in PD patients versus controls, and mtDNA cytosine methylation could further differentiate PD from atypical parkinsonism using bisulfite-based methylation analyses [64]. This may be challenged by other work which found no differential mtDNA methylation levels in the peripheral blood or serum of PD patients versus controls using bisulfite-based sequencing [65,66]. As discussed previously (see Box 2), these differences may be the result of sequencing sensitivity or technical approach with bisulfite-based methods. Importantly, the relative contributions of 5mC and 5hmC to mitochondrial epigenetic remodeling in PD remain unresolved, as most existing studies lack the resolution to distinguish these modifications.

Huntington’s disease

Huntington’s disease (HD) is caused by expansion of CAG repeats in the huntingtin gene, leading to widespread transcriptional dysregulation and neurodegeneration, particularly within the striatum [67]. Mitochondrial dysfunction is a prominent feature of HD, characterized by impaired oxidative phosphorylation, increased oxidative stress, and altered mitochondrial dynamics [67]. Ultra-deep sequencing of mtDNA from primary myoblasts and fibroblasts from huntingtin CAG repeat expansion mutation carrier patients demonstrated compromised mitochondrial genome integrity, with accumulation of mtDNA mutations that impaired oxidative phosphorylation [68]. These changes were linked to high levels of mutant huntingtin fragments, which promoted excessive mitochondrial fission and aberrant mitophagy, ultimately destabilizing the mitochondrial genome and exacerbating bioenergetic deficits [68]. Indeed, in the cortex and striatum in mouse models of HD, release of mtDNA into the cytosol has been associated with neuroinflammation [69]. These findings highlight mtDNA release as an underappreciated contributor to HD pathogenesis and suggest that targeting mtDNA-driven innate immune signaling as a potentially targetable mechanism of disease progression.

Unlike AD and PD, where mtDNA methylation has been partially characterized, the mitochondrial epigenome in HD remains largely unexplored. Similarly, there is limited evidence for the role of mitochondrial noncoding RNAs. However, accumulation and cytosolic release of mt-dsRNA is observed in various experimental models of HD. For example, in both mouse striatal cells expressing mutant huntingtin and human primary fibroblasts from huntingtin CAG repeat expansion mutation carriers, mutant huntingtin impaired mitochondrial genome integrity and RNA homeostasis, in part by promoting mitochondrial fragmentation, defective mitophagy, and destabilization of mtDNA, which together increased the production and persistence of aberrant mitochondrial transcripts [68,70]. In parallel, cell-type–specific transcriptomic analyses in the human post-mortem HD brain and in a mouse model of HD revealed that mutant huntingtin drives the accumulation and release of mt-dsRNA in striatal projection neurons [71], linking mitochondrial transcriptional dysregulation to innate immune activation. As in other neurodegenerative diseases, it remains unclear whether mitochondrial RNA dysregulation initiates neuronal dysfunction or amplifies existing pathology. However, the ability of mt-dsRNA to directly activate immune pathways suggests that mitochondrial transcriptional disturbances may play a more prominent role in HD than previously appreciated.

Mitochondrial epigenome and noncoding RNAs in acute brain injury

Both stroke, which is caused by an interruption in cerebral blood flow and oxygen delivery to the brain, and traumatic brain injury (TBI), which results from external mechanical force followed by secondary inflammatory and metabolic injury cascades, produce profound energetic failure within vulnerable neural tissue [72,73]. Stroke and TBI impose abrupt and severe metabolic stress on neurons, leading to rapid disruption of mitochondrial function. Unlike the gradual mitochondrial decline observed during aging, these insults trigger immediate deficits in oxidative phosphorylation, accumulation of ROS, loss of mitochondrial membrane potential, and activation of cell death pathways [74]. The capacity of neurons to recover from such injury depends, in part, on their ability to dynamically reprogram mitochondrial function [75,76]. Several lines of evidence indicate that mtDNA instability is an early and defining feature of acute brain injury pathophysiology. Across clinical and experimental studies, alterations in mtDNA abundance, copy number, and integrity are associated with stroke susceptibility, disease severity, and functional outcomes, suggesting that mitochondrial genome dysfunction accompanies acute neurological injury at multiple levels [77–80]. Beyond serving as a biomarker, mtDNA may actively contribute to pathogenesis. For example, in both a rat model of stroke and mouse model of TBI, mitochondrial damage and subsequent release of mtDNA activated innate immune pathways, promoted microglial inflammatory responses, and amplified secondary brain injury cascades [81,82]. In addition to structural instability, the mitochondrial epigenome undergoes dynamic remodeling following acute brain injury. Following repeated mild TBI in rats, DNA immunoprecipitation using 5mC- and 5hmC-specific antibodies revealed increased levels of 5mC, and unchanged levels of 5hmC at the TFAM promoter, which was associated with reduced TFAM expression and binding [83]. These alterations further correlated with decreased mitochondrial gene expression, mitochondrial biogenesis and ATP production [83]. This suggests that mtDNA methylation may regulate mitochondrial stress adaptation and further influence the capacity for mitochondrial recovery following acute brain injury. A recent study in humans using bisulfite-based sequencing linked mtDNA cytosine methylation directly to stroke risk, with higher levels of mtDNA cytosine methylation in patient peripheral blood samples associated with increased risk of stroke [84], suggesting that systemic mitochondrial epigenetic states may reflect or contribute to cerebrovascular vulnerability. In human endothelial and macrophage cell models of atherosclerosis, antibody-based analysis of isolated mtDNA revealed preferential enrichment of 6mA in macrophages, where METTL4-dependent regulation of mtDNA 6mA was linked to altered mitochondrial function and oxidative phosphorylation and activation of pro-inflammatory signaling pathways [85]. METTL4 activation was further linked to cytoplasmic mtDNA release and inflammasome activation in macrophages [85]. Although 6mA on mtDNA has not yet been established in acute brain injury, these findings provide an important conceptual framework suggesting that diverse epigenetic modifications within the mitochondrial genome may dynamically regulate mitochondrial metabolism and inflammatory signaling relevant to cerebrovascular injury and brain injury pathophysiology.

Despite growing interest in the roles of noncoding RNAs in pathophysiology or biomarkers in stroke and TBI, evidence of mitochondrial-encoded noncoding RNAs in acute brain injury remains extremely limited. To date, the LIPCAR mt-lncRNA is one of the few mitochondrial-derived noncoding RNAs associated with stroke, where elevated circulating levels correlated with adverse neurological outcomes and mortality in stroke patients [86]. However, whether LIPCAR or other mitochondrial noncoding RNAs have functional roles beyond their utility as biomarkers, including regulation of mitochondrial dysfunction, neuroinflammation, or recovery following stroke or TBI, remains largely unexplored.

A common theme emerges in TBI and stroke studies in which disruption of mitochondrial genome integrity contributes directly to acute injury pathophysiology through altered mtDNA abundance, instability, and cytosolic nucleic acid signaling. In contrast, the roles of mitochondrial epigenetic modifications and mitochondrial genome-derived noncoding RNAs remain less well defined, with current evidence largely limited to emerging methylation studies and a small number of noncoding RNA biomarker reports. An important challenge for future work is to determine whether these regulatory mechanisms function as coordinated components of a broader mitochondrial stress-response network that shapes inflammation, tissue repair, and neurological recovery following acute brain injury.

Therapeutic targeting of the mitochondrial epigenome and noncoding genome

Growing recognition that mitochondrial epigenetic and noncoding regulatory mechanisms contribute to aging and neurodegeneration (Table 1) raises the possibility that these pathways could be therapeutically manipulated to preserve mitochondrial function. Mitochondrial genome editing provides a strategy for directly correcting pathogenic or functionally detrimental noncoding or epigenetically modified mtDNA. Mitochondrial-targeted nucleases, including transcription activator-like effector nucleases and zinc finger nucleases, can selectively eliminate mtDNA, with demonstrated efficacy in the mouse brain [87]. RNA-independent DddA-derived cytosine base editors and transcription activator-like effector-linked deaminases have further enabled targeted C-to-T and A-to-G mtDNA editing, respectively, including in mouse neural cells [88,89]. Although conventional CRISPR-Cas approaches have been limited by inefficient mitochondrial import of guide RNAs, recent strategies to facilitate mitochondrial targeting of single-guide RNAs have demonstrated genome-editing efficacy in HEK293T and HELA cell lines [90,91]. Despite these advances, whether mitochondrial genome-editing technologies can be adapted to therapeutically manipulate the mitochondrial noncoding genome or epigenome in the aging brain or neurodegenerative disease remains to be determined.

Table 1.

The mitochondrial epigenome and noncoding genome in aging and brain diseases

Condition	Regulatory layer	Model	Main findings	Ref.
Aging	mtDNA methylation	Human postmortem brain	mtDNA CpG and non-CpG methylation increases with age	[37,38]
		C. elegans	mtDNA 6mA dysregulation alters mtDNA copy number and transcript levels	[39]
	Noncoding RNAs	Human peripheral mononuclear blood cells	Altered mecciRNA profiles including reduced circMT-RNR2	[42]
		Human fibroblasts	CircMT-RNR2 promotes mitochondrial function and suppresses senescence	[42]
	Cytosolic mtDNA / mt-dsRNA	Aged mouse brain/primary microglia	Cytosolic mtDNA activates microglial cGAS–STING signaling	[13]
		Aged mouse brain	Cytosolic mt-dsRNA activates RIG-I/MDA5 signaling	[12]
		Aged mouse brain/primary neurons	SEC61A1 promotes neuronal mt-dsRNA release and cognitive decline	[9]
AD	mtDNA methylation	Human AD peripheral blood/plasma cell-free DNA	Circulating mtDNA D-loop cytosine methylation profiles are associated with mild cognitive decline and AD progression	[50–53]
	Noncoding RNAs	Human postmortem AD brain; AD mouse model and primary neurons	lncMtloop levels are reduced in AD, lncMtloop restoration improves mitochondrial function and AD-related pathology	[10]
	Cytosolic mtDNA / mt-dsRNA	Microglia and endothelial cells in AD mouse models	Cytosolic mtDNA activates cGAS–STING signaling and inflammation	[45–47]
		AD patient and human postmortem AD brains	Cytosolic mt-dsRNA is increased and correlated with worsened neuropathology and cognitive outcomes	[43]
		Human postmortem AD brain/serum; AD mouse model brain	Increased cytosolic double-stranded RNA and SEC61A1 are increased in AD and associated with immune signaling	[9]
PD	mtDNA methylation	Human postmortem PD brain	No association between prefrontal cortex mtDNA cytosine methylation and PD	[63]
		Human PD peripheral blood	D-loop cytosine methylation is increased in PD peripheral blood	[64]
		Human PD peripheral blood/serum	No differential mtDNA cytosine methylation in PD peripheral blood/serum	[65,66]
	Cytosolic mtDNA / mt-dsRNA	Human PD postmortem brain/cell line; PD mouse models	mtDNA damage/release activates STING and inflammasome-associated signaling	[57–60]
		Human PD postmortem brain/ neuronal cell line; PD mouse model; PD zebrafish model	PINK1-related mitophagy defects promote cytosolic mtDNA accumulation	[11,61]
		Human PD postmortem brain	Cytosolic mitochondrial dsDNA is detected and associated with α-synuclein-associated pathology	[11]
HD	Cytosolic mtDNA / mt-dsRNA	Human HD postmortem brain; HD mouse model brain/microglia	Cytosolic mtDNA release is associated with neuroinflammation in HD	[69]
		Human postmortem brain / HD mouse model	Mutant huntingtin promotes neuronal mitochondrial RNA cytoplasmic release and innate immune activation	[71]
Stroke	mtDNA methylation	Human peripheral blood	Higher mtDNA methylation is associated with increased stroke risk	[83]
	NoncodingRNAs	Human plasma	Elevated circulating LIPCAR is associated with neurological impairment and adverse stroke outcomes	[86]
	Cytosolic mtDNA / mt-dsRNA	Rat stroke model; Mouse neuronal cell line	Oxidized mtDNA activates cGAS–STING signaling in the brain and neurons after ischemia-reperfusion	[81]
TBI	mtDNA methylation	Rat TBI model	Increased 5mC disrupts TFAM expression and DNA binding in the brain	[83]
	Cytosolic mtDNA / mt-dsRNA	Mouse TBI model	Neuron-derived mtDNA activates microglia through ZBP1 signaling	[82]

Beyond approaches to modify the mitochondrial genome, pharmacological targeting of mitochondrial epigenetic regulation may involve modulation of DNA methylation pathways or epigenetic enzyme activity. Mitochondrial noncoding RNAs likewise represent attractive therapeutic targets as their abundance or localization could potentially be manipulated using antisense oligonucleotides, small interfering RNA, or replacement RNAs to suppress pathogenic transcripts or restore protective mitochondrial RNA signaling. A major barrier to translating these approaches in the brain, however, is achieving efficient delivery across the blood-brain barrier, into specific neural cell populations, and ultimately across mitochondrial membranes. Viral-based gene delivery systems, such as the adeno-associated virus (AAV) PhP.B serotype offers one potential approach due to its ability to cross the blood-brain barrier and preferentially target astrocytes and neurons [92]. For example, systemic delivery using the AAV PHP.B vector to restore expression of NDUFS4 in NDUSF4-deficient mice delayed the onset of neurodegeneration and prolonged lifespan [93], demonstrating the potential use of AAVs to prevent neurodegeneration by enhancing mitochondrial function. Furthermore, AAV vectors can be engineered to incorporate mitochondrial-targeting sequences, enabling preferential localization of their cargo to the mitochondria.

Mitochondrial-targeted nanocarriers may provide additional flexibility by accommodating diverse cargos, including oligonucleotides and gene-editing components, while permitting modification of their composition and surface chemistry for tissue-, cell-, and mitochondrial-targeting [94–97]. Peptide-based nanocomplexes offer particularly programmable subcellular targeting as mitochondria-targeting sequences can be incorporated directly into the peptide architecture; such systems have been used to successfully deliver exogenous plasmid DNA to mitochondria in human astrocytes in vitro and in zebrafish embryos in vivo [94]. Lipid-based platforms extend this concept toward systemic delivery. In a mouse model of cerebral ischemia, intravenous delivery of TCA cycle enzymes to the brain via mitochondrial-targeted lipid nanoparticles promoted neuronal survival and enhanced performance on motor and cognitive assessments [95]. Additional engineering strategies, including fluorinated ionizable lipid nanoparticles that enhance cellular uptake and endosomal escape following local or systemic delivery [97], have further expanded the potential for mitochondrial nucleic acid delivery through mitochondrial targeting [96]. In an ND4-associated optic neuropathy mouse model, intraocular administration of mitochondrial-targeted fluorinated lipid nanoparticles enabled mitochondrial delivery of therapeutic plasmid DNA, which improved mitochondrial bioenergetics and visual dysfunction [96]. Together, these approaches illustrate the ability of mitochondrial-targeted nanocarriers to deliver both protein and nucleic acid therapeutics in vivo. Although efficient delivery across the blood-brain barrier and into specific neural cell populations remains a major challenge, continued development of brain- and mitochondria-targeted delivery systems, coupled with increasingly precise mitochondrial epigenetic and transcriptomic technologies (see Box 3), may ultimately enable selective manipulation of mitochondrial regulatory networks disrupted during aging and neurodegenerative disease.

Box 3: Current technologies for studying the mitochondrial epigenome and noncoding RNAs.

Rapid technological advances are expanding the ability to resolve and experimentally interrogate the mitochondrial epigenome. Long-read DNA sequencing platforms (e.g., Oxford Nanopore and PacBio), offer important advantages over conventional short-read approaches by enabling analysis of complete or near-complete mtDNA molecules. Further, long-read sequencing can directly detect 5mC, 5hmC or 6mA modified bases, circumventing limitations associated with bisulfite conversion (see Box 2) or potential non-specific binding from antibody-dependent methylation assays [105,106]. More recently, enzymatic sequencing approaches have enabled 5mC and 5hmC to be distinguished at single base resolution, with demonstrated applications in human glial cells and the mouse brain [107,108], but not yet for mtDNA. Additional current approaches for 5mC, 5hmC and 6mA include mass spectrometry, which can provide sensitive global quantification but generally lacks locus-specific information [39].

At the RNA level, long-read RNA sequencing similarly provides opportunities to resolve full-length mitochondrial transcripts and transcript isoforms that may be difficult to reconstruct from fragmented short-read data, offering improved resolution of mitochondrial transcript boundaries and noncoding RNAs [109–111]. Oxford Nanopore direct RNA sequencing analyzes RNA molecules without PCR amplification, preserving transcript-level information at the single-molecule level, enabling simultaneous assessment of mitochondrial RNA sequence, processing, and RNA modifications [112,113]. Finally, RNA interactome mapping approaches can further define the functional networks of mitochondrial noncoding RNAs by identifying their protein-, RNA, and DNA-binding partners, providing mechanistic insight into their regulation of cellular processes within the brain [8,10]. These epigenetic and transcriptomic approaches complement single-cell and spatial sequencing technologies with the potential to map mitochondrial alterations within neural subtypes and brain regions obscured by bulk analysis [2,15,16,35].

For cytoplasmic accumulation of mitochondrial nucleotides, subcellular fractionation followed by quantitative PCR remains a widely used strategy to quantify mitochondrial RNA and DNA within mitochondria- and nucleus-free cytosolic fractions. For mt-dsRNA, RNA immunoprecipitation using the dsRNA-selective J2 antibody, followed by qPCR or sequencing, can enrich dsRNA species and confirm the presence of mitochondrial-derived transcripts. Complementary immunofluorescence using the J2 antibody combined with mitochondrial markers enables spatial visualization and quantification of dsRNA localized outside the mitochondrial network [12,114]. Cytoplasmic mtDNA can similarly be visualized with high- or super-resolution microscopy by combining DNA staining or mtDNA-specific probes with mitochondrial/nucleoid markers, complementing fractionation-based measurements [11]. Finally, integration of these approaches with downstream immune signaling provides evidence for mt-dsRNA and mtDNA release and their biological consequences [13,100].

Concluding remarks

Mitochondrial dysfunction appears to be a recurring theme across neurodegenerative disorders and brain injury. A growing body of work increasingly positions mitochondrial epigenetic remodeling and noncoding RNA dysregulation as regulatory mechanisms linking mitochondrial dysfunction to brain disease. It is plausible that early mitochondrial regulatory dysfunction predisposes neurons to degeneration, whereas later-stage changes reflect a combination of compensatory and pathological processes. Similar to nuclear epigenetic aging, where reduced chromatin flexibility limits transcriptional adaptability, mitochondrial epigenetic and noncoding regulatory systems may progressively lose the ability to dynamically respond to energetic and inflammatory demands. Concomitantly, many mitochondrial regulatory mechanisms may initially serve adaptive functions aimed at preserving bioenergetics and cellular survival yet become maladaptive when chronically sustained. Thus, an important challenge for future work will be to define the temporal, regional, and cell-type-specific dynamics of these pathways across disease progression and determine whether these mitochondrial regulatory changes are causative drivers of pathology or secondary responses to accumulated cellular stress. Distinguishing these possibilities will be critical for understanding disease pathogenesis and optimizing therapeutic timing.

Despite substantial advances, fundamental questions remain regarding the regulation and functional significance of both the noncoding mitochondrial genome and mitochondrial epigenetic modifications in the nervous system (see Outstanding Questions). Nevertheless, recent evidence demonstrating modulation of mitochondrial epigenetic pathways and mitochondrial noncoding RNA signaling can ameliorate neuronal dysfunction highlights considerable therapeutic potential for neurodegeneration and brain injury. Collectively, the current evidence positions the mitochondrial epigenome and its noncoding regulatory mechanisms as important and still evolving dimensions of mitochondrial biology in the aging and diseased brain.

Outstanding questions.

• What are the precise mechanisms by which mitoepigenetic mechanisms regulate mitochondrial transcription, replication, and oxidative phosphorylation in the brain?

• What is the functional relevance of mtDNA 5mC or 5hmC across cells in neurodegeneration and brain injury?

• Does mitochondrial epigenetic remodeling impact mtDNA and mt-dsRNA production and signaling during acute and chronic stages of brain injury and neurodegeneration?

• Do bona fide mtDNA-encoded mitomiRs exist in the brain, and what functional roles do they play in neuronal metabolism, inflammation, and neurodegeneration?

• To what extent are mitochondrial epigenetic and noncoding RNA changes adaptive responses that promote recovery, or maladaptive drivers of chronic neurodegeneration?

• Can mitochondrial epigenetic pathways or mitochondrial-derived noncoding RNAs be therapeutically targeted to restore mitochondrial function and improve neurological recovery after brain injury?

HIGHLIGHTS.

• The mitochondrial genome contains epigenetic and noncoding regulatory layers that are increasingly implicated in brain aging, neurodegeneration and brain injury.

• Epigenetic modifications of mitochondrial DNA may regulate mitochondrial gene expression and function in response to cellular stress.

• Mitochondrial DNA and noncoding RNAs provide an additional layer of regulation, influencing mitochondrial function, cellular stress signaling, and inflammatory responses.

• During aging and brain disease, disruption of these regulatory mechanisms may shift mitochondrial responses from adaptive signaling toward inflammation and cellular dysfunction.

• Understanding how mitochondria regulate gene expression and communicate with the cell may reveal new therapeutic opportunities for neurodegenerative disease and brain injury.