Mitochondria signaling to the epigenome: A novel role for an old organelle

Abstract

Epigenetic modifications influence gene expression programs ultimately dictating physiological outcomes. In the past decades, an increasing body of work has demonstrated that the enzymes that deposit and/or remove epigenetic marks on DNA or histones use metabolites as substrates or co-factors, rendering the epigenome sensitive to metabolic changes. In this context, acetyl-CoA and α-ketoglutarate have been recognized as critical for epigenetics, impinging on histone marks and nuclear DNA methylation patterns. Given that these metabolites are primarily generated in the mitochondria through the tricarboxylic acid cycle (TCA), the requirement of proper mitochondrial function for maintenance of the epigenetic landscape seems obvious. Nevertheless, it was not until recently when the epigenomic outcomes of mitochondrial dysfunction were tested, revealing mitochondria’s far-reaching impact on epigenetics. This review will focus on data that directly tested the role of mitochondria on the epigenetic landscape, the mechanisms by which mitochondrial dysfunction may dysregulate the epigenome and gene expression, and their potential implications to health and disease.

1. Introduction

Mitochondria are centrally involved in energy metabolism within the cells. They are also recognized for their role in signaling, relaying to the nucleus information of their status for cells to adapt to a new environment. This type of communication was coined retrograde signaling because it initiated from the mitochondria and was proposed to be deployed in states of stress in which the organelle requires the nucleus to adjust gene expression to regain homeostasis. The canonical retrograde signals involve energetic or oxidative stress, calcium-dependent responses and, more recently, protein toxicity [1–3]. In the last decades, mitochondrial-derived signaling largely focused on reactive oxygen species (ROS) and their ability to impinge on physiological outcomes by acting over redox-sensitive protein switches. For example, superoxide anion (${\text{O}}_2^{.-}$) generated by mitochondrial complex III and metabolized by manganese superoxide dismutase (MnSOD) causes HIF1α (hypoxia inducible factor 1) stabilization and gene expression regulation [4,5]. Likewise, ROS produced by complex I in the forward or reverse directions modulate macrophage activation and myotube differentiation [6]. More recently, the role of mitochondria in signaling has further expanded to involve metabolites, including those derived from the tricarboxylic acid cycle (TCA) and their ability to impact epigenetics [7].

The past decade has seen an explosion on the role of metabolism in orchestrating different cellular programs through modulation of the epigenome. Data largely derived from the stem cell and cancer fields have demonstrated that changes in glucose or lipid metabolism impact post-translational modifications in transcription factors, kinases and histones that in turn influence gene expression and cell differentiation programs [8]. Similarly, levels of the metabolite acetyl-CoA or α-ketoglutarate (α-KG) were shown to regulate histone acetylation and DNA methylation, respectively, thereby impacting transcriptional outputs and cellular fate [9,10]. Given the main role of mitochondria in metabolism, it seems sensible to envision a direct relationship between mitochondria and epigenetics, which perhaps could explain - mechanistically - how retrograde signaling is effectively orchestrated. However, until recently little effort had been put to test this relationship given the idea that metabolite pools are compartmentalized [11]. As such, those generated in the mitochondria would not reach the nucleus, making changes in the mitochondrial pool irrelevant to the site of the epigenetic reactions. While this idea has been primarily associated with acetyl-CoA, this hypothesis fails to explain how other metabolites generated by mutant TCA cycle enzymes such isocitrate dehydrogenase 2 (IDH2), succinate dehydrogenase (SDH) or fumarate hydratase (FH) can reach the nucleus altering DNA and histone methylation [12–14]. Some TCA cycle proteins were proposed to moonlight in the nucleus to generate a nuclear pool of these epigenetic-relevant metabolites [15], but the data remains equivocal. Finally, it is known that cytosolic reactions can generate epigenetic-relevant metabolites independent of the mitochondria, including acetyl-CoA that can be produced through the metabolism of acetate and lipids (McDonnell et al., 2016 [16], or through the reductive carboxylation of glutamine in the cytosol, which also generates α-KG (Gaude et al., 2018).

This review will focus on studies in mammalian models that have directly tested the impact of changes in mitochondrial function to nuclear epigenetics, the means through which this crosstalk may be achieved and the outcomes for gene expression programs. Only methylation and acetylation changes in histones will be discussed as the field has yet to report on mitochondria-driven effects on other histone modifications. To more broadly explore the role of metabolism on the epigenome, readers are referred to other reviews including, but not limited to, [17–21]. Also, this review will not touch on methylation of the mitochondrial genome itself. Finally, I would like to propose that the effects of mitochondrial function on epigenetics are not only deployed as a response to stress but evolved as a means through which the organelle dynamically interacts with the nucleus to convey its metabolic state also under normal physiological ranges. As such, this communication would not fall into the classic stress-response associated with retrograde signaling, information on which readers are invited to explore through other comprehensive reviews [1,22–25].

2. Mitochondria and DNA methylation

The first evidence that mitochondria may play a role in epigenetics came from Keshav Singh’s group, who intrigued by the connection between altered gene expression and mitochondrial function in cancers [26] hypothesized a link between the two through DNA methylation. By performing restriction landmark genomic-scanning (RLGS) analysis to assess nuclear CpG island methylation in 3 different pairs of cancer cells harboring mitochondrial DNA (mtDNA, rho⁺) and their mtDNA-less (rho⁰) derivatives, the authors identified several nuclear loci in which DNA methylation differed between rho⁺ and rho⁰ cells. These changes involved both hypo and hypermethylation events and were confirmed, in some instances, by standard bisulfite sequencing. Notably, the authors showed that re-introduction of mtDNA in the rho⁰ cells reversed some of the nuclear DNA methylation changes back to that of rho⁺, suggesting a direct epigenetic consequence of modulation of mitochondrial function. Many loci did not reverse and some of the changes got more pronounced with the reintroduction of mtDNA, which the authors concluded reflected CpGs that were not aberrantly methylated consequent to mitochondrial impairment. It could be argued that the chemical treatment to generate the rho⁰ cells played a role in the nuclear phenotype, but this possibility was largely ruled out by the demonstration of similar changes in DNA methylation when mtDNA was depleted genetically. As mtDNA loss leads to changes in cellular redox, inner membrane potential and ATP levels, the mechanism linking the mitochondria to the epigenome remained unclear. Whether it relied on a single intracellular change associated with mitochondrial dysfunction or a combination of signals that impacted DNA methylation remained to be tested [27]. It was not until a decade later that more mechanistic details about this relationship started being unveiled [28].

The cancer field has contributed evidence on the ability of mitochondria to affect the epigenome by demonstrating that mutations on the TCA cycle enzymes IDH2, SDH and FH lead to the generation or accumulation of metabolites that impact nuclear epigenetic reactions. In this context, IDH2 mutations associated with leukemia have been shown in vitro and in vivo to generate 2-hydrogyxglutarate (2-HG), which in turn significantly inhibited the Ten Eleven Translocation (TET) dioxygenases massively decreasing levels of 5-hydroxy-methylcytosine (5-hmeC) on the nuclear genome. As genetic or pharmacological inhibition of IDH2 was shown to decrease levels of 2-HG sufficiently to change the epigenome and reduce leukemia maintenance, these studies concluded that mitochondrial production of 2-HG was causative for the DNA methylation changes associated with the cancer phenotype [12]. Similarly, data obtained for SDH and FH loss of function mutations also showed inhibition of the TETs although, mechanistically, this resulted from the accumulation of succinate and fumarate, respectively, that can out compete α-KG in the active site of these enzymes. A cause-effect relationship was demonstrated by reversal of TET inhibition through ectopic expression of the wild-type (WT) isoforms of the enzymes which rescued SDH and FH activities and normalized levels of succinate and fumarate [13].

Although these cancer studies provided clear evidence of the ability of mitochondrial function to impact epigenetics, surprisingly they did not lead to traction in the mitochondrial field at the time. This was perhaps because metabolic regulation of the epigenome was already a concept in the cancer field [9] or, alternatively, because these severe changes were rare events found only in cancer. Also, FH had been shown to be present in the nucleus, where fumarate could be generated [29] and IDH2 mutations are frequently accompanied by changes on the cytosolic IDH1, to some extent questioning the mitochondrial contribution to the epigenetic effects. It was back in 2016 when studies involving alterations of mitochondrial function through means beyond the TCA that the role of mitochondria in nuclear DNA methylation became evident. For example, a transgenic mouse model in which a mutant (Y955C) mitochondrial DNA polymerase gamma (PolG) was overexpressed specifically in the heart was used to interrogate changes in DNA methylation in this tissue [30]. The Y955C mutation was known to significantly impair PolG catalysis, causing mtDNA genome instability in different model systems, including mtDNA loss in animal models [31]. In mice, mtDNA depletion associated with this mutation impaired mitochondrial respiration and ATP production, caused oxidative stress and left ventricular (LV) hypertrophy [32,33]. As some of these phenotypes were previously linked to altered DNA methylation patterns in patients with cardiomyopathy [34], the authors hypothesized that the LV dysfunction resulted from altered nuclear DNA methylation caused by Y955C. To test this, they used a tiling array that covered the promoters of about 20 K genes and 15,963 annotated mouse CpGs to find differentially methylated sites between WT and mutant hearts; both young and aged littermates were analyzed. They found ~73 K differentially methylated peaks, which were primarily associated with aging, and another 4506 peaks that were deemed mutation-specific. Both hyper and hypomethylated peaks were identified, with the latter being more abundant. Because Y955C led to increased cardiac oxidative stress, the authors proposed ROS as potential mediators of the hypomethylated phenotype presumably by direct oxidation of 5-methyl-cytosine (5-meC). Alternatively, it was proposed that impairment of DNA methyltransferase (DNMT) activity or diminished S-adenosylmethionine (SAM) levels may contribute to the observed hypomethylation. However, because LV dysfunction was already present in the Y955C-expressing mutant animals at the time of the DNA methylation analysis, it remained a possibility that the epigenetic changes were a consequence of the cardiac not the mitochondrial phenotype.

Recently, we used HEK293 cells that express a catalytically dead PolG (DNPOLG) under doxycycline control to get insights into nuclear DNA methylation changes caused by progressive mtDNA loss in 9 days [35]. By concomitantly monitoring the metabolome, epigenome and transcriptome in a time-dependent manner during this 9-day period, we found an early transcriptional and metabolic response involving various amino acids, including methionine degradation, which occurred within 3 days of doxycycline addition. The methionine cycle provides SAM for DNA, RNA and protein methylation, and while not occurring within mitochondria it is linked to it by its crosstalk with one carbon (1C) metabolism. Previous studies had shown that mtDNA depletion using this same transgene led to significant remodeling of 1C metabolism, with carbons being channeled through transsulfuration and cysteine biosynthesis. However, the specific effects on the methionine cycle and methylation reactions were not studied [36]. We monitored DNA methylation at a single nucleotide resolution using the Illumina 450 K array, which covers 99% of the reference genes (RefSeq) and found that at day 3 the nuclear genome was mostly hypomethylated. MtDNA was depleted to ~40–60% and no measurable mitochondrial defects were apparent at this time, which we interpreted as indication that the metabolic and epigenomic changes were sensitive to subtle modulation of mitochondrial function that is not detectable using standard biochemical techniques. However, as the mtDNA became further depleted to 80–100% at days 6 and 9, respectively, the changes in the methylome became more pronounced and the changes in the epigenome and in the transcriptome also became pervasive, covering thousands of loci and hundreds of genes, respectively. By then, mitochondrial dysfunction was overt with significant loss of oxygen consumption, ATP generation and inability to maintain the inner membrane potential. Also, the nuclear DNA was primarily hypermethylated. Notably, methionine salvage linked to polyamine biosynthesis, which can generate SAM independent of the classic 1C-dependent methionine biosynthetic pathway, was also highly active [28].

To determine the extent to which these genomic effects were driven by or a consequence of the mitochondrial defects, we then monitored the same parameters in cells that ectopically expressed two non-mammalian proteins that could sustain a pseudo-electron transport chain in the absence of mtDNA. These enzymes, NDI1 (NADH dehydrogenase-like 1) and AOX (alternative oxidase), provided the ability to oxidize NADH and transfer electrons to oxygen, analogous to the activities of complexes I and IV, respectively. While their expression did not restore ATP generation or proton pumping activity, they allowed resumption of the TCA cycle [35]. More importantly, HEK293 cells expressing NDI1/AOX were able to maintain the epigenome and transcriptome when mtDNA was depleted like WT controls. The changes in methionine salvage were completely reversed by expression of NDI1/AOX, which in turn also normalized levels of SAM, while other metabolic changes previously associated with mtDNA depletion and the rewiring of 1C metabolism [36] were not rescued. Overall, we interpreted these data to reflect that methionine metabolism was responsive to changes in TCA cycle activity, and that SAM provided by the methionine salvage/polyamine biosynthetic pathway was the one supporting the increased nuclear DNA methylation phenotype [28]. It is noteworthy that normalization of NADH oxidation in the mtDNA depleted cells could impact the NAD⁺/NADH ratio and the cellular redox state. The sensitivity of some epigenetic enzymes, including those involved in DNA demethylation, to the redox environment is well described [37]. Thus, an additional explanation to the data involve changes in cellular redox, which remains to be studied.

This work provided unequivocal genetic evidence that changes in mitochondrial activity were drivers of the epigenomic effects, raising fundamental questions about the broader physiological outcomes associated with mitochondrial dysfunction. In particularly, whether similar effects would be seen in vivo and upon other types of changes in mitochondrial function became critical questions to be addressed. In the past 2 years, several new studies have started addressing these questions, and it has become evident that the effects of modulation of mitochondrial function to epigenetics are relevant in vivo For example, using a mouse model in which the mitochondrial complex III subunit Rieske iron sulfur protein (RISP) was deleted in regulatory T cells (Tregs), Weinberg and colleagues (2019) showed significant changes in nuclear DNA methylation. RISP had been previously shown to be key for complex III function and ROS signaling [38]. In Tregs. loss of RISP led to defects in mitochondrial respiration, changes in the NAD⁺/NADH ratio and significant dysregulated gene expression programs that blunted their suppression capacity. Reduced representation bisulfite sequencing (mRRBS) revealed mild hypermethylation of the DNA around the transcription start site (TSS) of genes that were differentially downregulated in the KO animals. Unsupervised clustering identified <17 K differentially methylated CpG sites between WT and KO cells; 150 of those occurred in differentially expressed genes (DEGs). It can be argued that the number of DEGs also showing altered DNA methylation is minimal, raising questions about the relevance of the epigenetic changes in terms of gene expression. This is noteworthy given in cell culture models this relationship reached ~80% [28]. However, mRRBS was used to detect differentially methylated loci (DML) in promoters, but other genomic regions that can get differentially methylated, such as enhancers, can also regulate gene expression and were not covered in this study. Also, the authors assumed that hypermethylation of CpG sites within the gene promoter would lead to transcriptional inhibition of the gene thereby only analyzing those downregulated; but this does not always apply [39, 40].

In trying to get insights into how RISP loss could lead to DNA hypermethylation, the authors performed a serious of elegant experiments. Using metabolomics analysis, Weinberg and colleagues found elevated levels of 2-HG, α-KG and succinate in RISP KO cells, which were also observed when WT cells were treated with the complex III inhibitor antimycin A but not other mitochondrial inhibitors. Additionally, under some physiological conditions decreases in the NAD⁺/NADH ratio allows cells to produce L(S)-2HG from promiscuous substrate usage of α-KG by lactate dehydrogenase or malate dehydrogenase. Notably, L(S)-2HG is oxidized in the mitochondria and converted back to α-KG by L(S)-2HG dehydrogenase, a FAD-containing enzyme that is dependent on an active mitochondrial complex III [41]. Given RISP-deficient cells have impaired complex III and lower NAD⁺, the authors concluded that L2HGDH activity was also diminished, favoring the high production of 2-HG that can inhibit the TETs [42]. Taken together, these mechanistic studies revealed that complex III dysfunction affected the DNA methylation landscape by altering the cellular redox state as well as by increasing levels of a competitive inhibitor of α-KG. Nevertheless, data demonstrating that decreasing levels of −2HG, restoring NAD⁺ or the activity of L2HGDH could reverse the DNA hypermethylation phenotype in the absence of RISP were not provided. It would be interesting to know whether simple modulation of the cellular redox state through NAD⁺ would suffice to reverse the epigenetic changes. Also, determination of 5-hmeC levels using genome-wide arrays [43] could provide evidence of TET inhibition. This would be important given bisulfite conversion-dependent methods cannot efficiently differentiate between the 5-meC and 5-hmeC [44]. Despite these remaining outstanding questions, this work fundamentally showed that complex III deficiency leads to significant changes in nuclear DNA methylation, adding to the scope of biochemical changes in mitochondrial function that impact the epigenome.

The epigenetic landscape of an organism is established very early in development. In mammals, it is in the pre-implantation stage that the first wave of genome demethylation and re-methylation occurs, setting DNA methylation patterns that are associated with life-long health effects [45,46]. Environmental exposures to many different agents in this critical developmental period have been shown to alter the epigenome and influence late life health outcomes, but the mechanisms underlying such effects remain poorly understood (Baccarelli and Bollati, 2009, [46]. As mitochondria undergo functional and structural changes in the embryo with robust activation of oxidative phosphorylation (OXPHOS) capacity also at the pre-implantation stage [46], it follows that modulation of mitochondrial function may mechanistic link early developmental exposures to changes in DNA methylation at the organismal level. We recently tested this concept by taking advantage of the viable yellow agouti (A^vy) mouse. In this mouse model, the DNA methylation status of the A^vy locus influences the coat color of the animals, which can be used as a visual screen for changes in this epigenetic mark in vivo. Specifically, hypomethylation of the locus leads to animals with yellow coat while, in the other extreme, hypermethylation leads to pseudoagouti animals. Mottled animals are also part of the population, representing cellular mosaics for the DNA methylation status of A^vy within the same tissue [47]. Because DNA methylation at the A^vy locus is established before the 3-germ layer separation [48–50], its status in the skin can be used as surrogate for the DNA methylation state of the locus in other tissues as well. Earlier work using this animal model showed that maternal exposure of A^vy animals to different environmental agents alter the DNA methylation state of the A^vy locus in the offspring from the exposed mothers [48,49,51], providing an unique model system and study design to interrogate the effects of modulation of mitochondrial function early in development to the organismal epigenetic landscape.

We exposed C57B6 females to two distinct concentrations of rotenone, a widely studied mitochondrial complex I inhibitor. We chose rotenone because in mammals it is metabolized by the liver cytochrome P450s, leading to the formation of two major byproducts that also inhibit complex I [52–54]. Additionally, absorption and distribution rates indicate that the pregnant uterus, the embryo and the fetus are affected by rotenone exposure through the maternal diet [53–58]. In our design, female animals were exposed to dietary rotenone for two weeks prior to conception, when they were mated with A^vy mutant males; exposure was maintained throughout pregnancy and lactation. Maternally exposed offspring were switch to a regular diet at weaning (post-natal day 21), when analysis of DNA methylation changes through coat color distribution was done. We found a dose-dependent increase in the frequency of animals with yellow coat, an indication of hypomethylation of the A^vy locus in the skin of the offspring of females exposed to rotenone relative to those exposed to the control diet. Next, we performed whole genome bisulfite sequencing (WGBS) at a single nucleotide resolution in the liver of the offspring to define if effects went beyond the skin. We sequenced animals at post-natal day 22 (PND22), 6–12- and 18-months of age to determine the extent to which the epigenetic effects would be long-lasting. We found that thousands of loci were differentially methylated with age in the control cohort, but many of those loci were not changed in the animals perinatally exposed to rotenone. Hundreds of other loci were differentially methylated only in the animals that had been exposed to rotenone through the mothers, even 18 months after that exposure ceased. Collectively, these results provided unequivocal evidence that early life exposure to a mitochondrial toxicant had pervasive and long-lasting effects on organismal DNA methylation [59].

We then performed RNA-seq in the same livers to determine the extent to which the changes in DNA methylation affected the transcriptome. While we could not determine cause-effect relationships given our study design, we did find a strong correlation between the two phenotypes at the gene (based on promoter coordinates) and at the genome level. We also found statistically significant associations between differentially methylated loci and enhancer regions, which we were able to further correlate to the promoters of genes differentially expressed - overall providing a broad view of how those mitochondrial-derived DNA methylation changes may impact transcription. Physiologically, we found that antioxidant enzymatic activity, specifically that of MnSOD and Cu,ZnSOD, was impaired in 12-month old animals that were perinatally exposed to rotenone. Mitochondrial complex I and complex II dysfunction were also observed in the livers of the same animals. These functional changes were seemingly associated with the transcriptional levels of co-factors of these enzymes, which were both differentially methylated and expressed. However, experiments to test this are still required.

Finally, we considered the mechanisms through which rotenone could affect DNA methylation. Because by inhibiting complex I rotenone increases the production of ${\text{O}}_2^{.-}$, we started by interrogating the potential role of ROS as not only the TETs are redox sensitive but ${\text{O}}_2^{.-}$ or its dismutation byproduct hydrogen peroxide (H₂O₂) can directly oxidize 5-meC [60]. Also, ${\text{O}}_2^{.-}$ has been shown in vitro to increase SAM’s affinity for DNA, leading to DNA hypermethylation [61]. Mitochondrial complex I inhibition was clear in the liver of pregnant mothers exposed to rotenone, and MnSOD activity was also increased. However, use of an anti-DMPO (5,5-Dimethyl-1-pyrroline N-oxide) antibody that traps free radical intermediates in biological samples did not reveal increased signs of oxidative damage in the same livers, suggesting that ${\text{O}}_2^{.-}$ generated as byproduct of rotenone exposure was efficiently decomposed by MnSOD. Nevertheless, H₂O₂ derived from ${\text{O}}_2^{.-}$ dismutation could leave the mitochondria potentially affecting the nucleus. To address whether physiological levels of H₂O₂ would affect DNA methylation, we then crossed A^vy males with females overexpressing catalase in the mitochondria (mCAT) but found no changes in coat color distribution in the offspring resulting from the A^vy vs mCAT cross. Collectively, these results suggest that redox changes are unlikely to have contributed to the effects of rotenone on nuclear DNA methylation under our experimental conditions [59]. However, additional experiments are required to fully rule out the effects of ROS and to determine the potential contribution of other mechanisms, including those associated with changes in SAM or the accumulation of succinate, another byproduct of complex I inhibition [6].

3. Mitochondria and histone methylation

Lysine demethylases (KDMs), like the TETs, are Fe²⁺ and α-KG-dependent dioxygenases. Hence, metabolites that can compete with α-KG such as 2-HG, fumarate and succinate as well as ROS are expected to inhibit their function as well. Early studies using cancer cells with mutations in FH and SDH provided evidence of changes in histone methylation due to competitive inhibition of KDMs by high levels of fumarate and succinate [13]. Later, loss of RISP in hematopoietic stem cells (HSCs) was proposed to lead to histone hypermethylation through accumulation of 2-HG [62]. These changes were observed in several residues, including those associated with promoters (H3K4me3), enhancers (H3K27me3, H3K4me1) heterochromatic regions (H3K9me) and active genes (H3K79me) [13,63]. Such lack of specificity is in line with histone demethylase inhibition driving the effects. Interestingly, fumarate and ROS have been shown to affect histone methylation through means beyond direct inhibition of KDMs. For example, Sullivan and colleagues demonstrated that FH mutations led to accumulation of fumarate that was able to bind glutathione, generating a fumarate-GSH derivative. This derivative, by being a substrate of glutathione reductase, consumed NADPH increasing cellular ROS. Thus, the authors hypothesized that the increased oxidative stress resulting from fumarate-GSH amplified the histone phenotype. To test this, they showed in vitro that fumarate could impair activity of jumonji D2A, a histone demethylase. They also found that exogenous H₂O₂ at the nanomolar level inhibited its activity, an effect that was synergized when a subinhibitory dose of fumarate was added contemporaneously. Thus, they concluded that, mechanistically, FH mutation contributed to histone hypermethylation both through fumarate and ROS [64]. Nevertheless, experiments demonstrating that antioxidants could partially rescue the histone phenotype in FH mutants were not performed but would have strengthen the conclusions.

Interestingly, the opposite effects on the histone methylation landscape were observed in another study in which ROS was increased, overall highlighting the complex relationship between ROS and the epigenome. Using a conditional deletion of MnSOD in HSCs in the mouse, Case and colleagues (2013) found significant increases in the levels of mitochondrial ${\text{O}}_2^{.-}$, which was accompanied by systemic iron deregulation due to ferrochelatase deficiency and severe anemia in the animals. These phenotypes were associated with abnormal regulation of hematopoietic transcription factors and silencing of globins, in addition to altered expression of iron-responsive genes. Because the switch from fetal to adult globins is known to be regulated by histone modifications, it was hypothesized that oxidative stress driven by loss of MnSOD impacted the histone landscape. Using Western blots, the authors found that histone methylation and acetylation levels were altered in MnSOD deleted HSCs, but it was rather surprising to find that H3K4, H3K27 and H3K9 were hypomethylated. Such phenotype could not be attributed to ROS inhibiting KDMs. The authors then proposed that decreases in the cellular levels of SAM in MnSOD deleted cells could underlie the histone phenotype [65]. Whether SAM levels are decreased upon MnSOD deletion and, if so, whether they impact histone methylation levels still needs to be determined. Also, whether the changes in the abundance of histone modifications drove the altered gene expression program remains unknown. Additional experiments using chromatin immunoprecipitation (ChIP)-based assays would be useful in determining the state of histone methylation at the promoter coordinates of the genes aberrantly expressed upon deletion of MnSOD in HSCs. While the Western blot analysis pointed to global histone hypomethylation, it is possible that at the locus of those globin genes histones could still be hypermethylated. This is because the impact of SAM concentrations can be lysine specific since they depend on the K_m of histone methyltransferases (HMTs) for SAM [66]. Also, it is our experience that data obtained by Westerns show a general trend that may reveal to be more heterogenous at the locus level when using more sensitive techniques such as ChIP-seq [67].

In trying to better understand the potential impact of ROS in histone methylation, we used Western blots to get a snapshot of the global landscape in different models of altered mitochondrial ROS in which we find changes in DNA methylation. For instance, we monitored histone methylation in the livers of the control or rotenone-exposed animals at PND22, 6, 12 and 18-month old animals [59]. Generally, no differences were observed between the two cohorts, but decreased levels of H3K4me3 and H3K27me3 were observed only in the rotenone perinatally-exposed animals at 12 months of age (Fig. 1A). Curiously, 12 months was the only time point in adult animals when signs of oxidative stress were also observed [59], which would be consistent with the observations from Ref. [65]. However, whether ROS are involved remains to be addressed. We also looked at the histone methylation landscape in a cell culture model of complex III deficiency based on a 4 base pair deletion on cytochrome b (cytb). Even though this mutation was previously shown to increase cellular ROS [68], we did not observe altered histone methylation by immunoblots (Fig. 1B). Together with the studies from Ref. [64] and Case and co-workers (2013), these data suggest that the impact of modulation of mitochondrial function and ROS to histone methylation is more complex than initially anticipated. Because each of the above described mitochondrial interventions were accompanied by distinct metabolic rewiring, it is likely that the physiological consequences of those mitochondrial changes also play a role in the outcomes for the epigenetic landscape. It is noteworthy that mitochondria dysfunction plays a major role in oxidative stress, a condition in which several oxidative histone changes, including carbonylation, have been identified [69,70]. Whether mitochondrial dysfunction and the resulting state of oxidative stress may drive those histone phenotypes deserves attention.

Fig. 1. Effects of rotenone and comple III mutation on histone methylation.

(A) Histones were extracted through acid purification from the livers of offspring of animals exposed to control or 10 ppm rotenone-containing diet. Western blots were performed for different histone methylation marks at post-natal day 22 (PND22), 6-, 12- and 18-month old animals. Data are the average ± SD of N = 6/group. (B) Histone were purified from 143 B osteosarcoma cells wild type or bearing a 4 base pair deletion on the mitochondrial cytochrome b gene that impairs complex III. Methylation marks were probed by Western blots on 3 independent cell cultures (marked as R1-R3). Total histones were used as loading control., quantification of the changes are shown on the graphs. WT = wild-type cells, cytb = cytochrome b mutants.

Most recently, deletion of the mitochondrial calcium uniporter (MCU) in mouse embryonic fibroblasts was shown to impact histone methylation by affecting cellular metabolism [71]. The authors were interested in understanding the underlying mechanisms associated with fibroblast to myofibroblast differentiation, a universal response to injury that in the heart is associated with fibrosis. As sustained increases in cytosolic calcium had been shown to promote fibroblast to myofibroblast differentiation and given the critical role of mitochondria in buffering cytosolic calcium, the authors hypothesized that this differentiation program would be influenced by loss of MCU. They found that adenoviral-mediated MCU deletion in mouse fibroblasts per se was sufficient to promote differentiation; this was further promoted upon exposure to a pro-fibrotic stimulus such as TGFβ. They then probed the effects of MCU deletion on cellular metabolism given calcium stimulates the TCA cycle dehydrogenases and metabolic rewiring is a requirement for differentiation programs [72]. By using quantitative metabolomics, the authors found that loss of MCU increased glycolysis and glutamine metabolism; it also increased the levels of a few TCA cycle intermediates including citrate and α-KG. To connect the changes in metabolism with the gene expression program, they analyzed epigenetic marks that could be impacted by α-KG including histone methylation. The authors focused on loss of H3K27me2 in MCU deleted cells, which was decreased at baseline and upon 12 h of TGFβ stimulation. In control cells, decreased H3K27me2 levels were only found after 48 h TGFβ exposure. They went on to show that H3K27me2 levels were decreased at the promoters of genes involved in differentiation into pro-fibrotic myofibroblast using ChIP-PCR. Thus, it was concluded that loss of MCU led to metabolic rewiring involving increased glycolysis and glutaminolysis. The latter provided the cells with enhanced levels of α-KG that promoted histone demethylase activity and decreased H3K27me2 levels in the genes involved in the differentiation program [71].

This work highlights yet a potential different means through which mitochondrial functional changes can affect histone methylation, namely, by providing enough α-KG that can be used to support histone demethylase activity. However, a few outstanding questions remain. For example, KDM activity was not assayed in WT and MCU null cells. Also, the authors focused on loss of H3K27me2 but essentially the same effects were detected for the promoter mark H3K4me3 and for H3K9me3 levels, which is normally enriched in heterochromatic regions. While these broader effects are more in line with whole cell changes in α-KG levels, they raise fundamental questions about the specific role of H3K27me2. In addition to impacting the TCA cycle, loss of MCU is expected to influence extra-mitochondrial calcium dynamics and ROS signaling [73]. Interestingly, previous work showed that increased intracellular calcium led to histone hyperacetylation, including on H3K27, in cardiomyocytes [74]. Given that methylation and acetylation of H3K27 are mutually exclusive events [75], it would be interesting to probe the status of the histone acetylation landscape upon loss of MCU. This seems particularly relevant in view of the fact that citrate, an acetyl-CoA precursor, was significantly increased in MCU null cells [71]. Thus, more work is required to determine the full extent to which modulation of mitochondrial calcium dynamics affects the epigenome.

4. Mitochondria and histone acetylation

Unlike methylation reactions, the relationship between mitochondrial function and histone acetylation seems straightforward since the primary source of acetyl-CoA in cells is the TCA cycle. Generally, citrate generated in the mitochondria is exported to the cytoplasm, where ATP citrate lyase (ACL) converts is back into acetyl-CoA that is used for several cellular processes including histone acetylation [9]. In the absence of ACL, despite normal TCA cycle activity, cells utilize acetate as a source of acetyl-CoA since the cytosolic conversion of citrate is inhibited [16]. Accordingly, the histone acetylation landscape is maintained despite loss of ACL [16,76] Interestingly, in the absence of a functional mitochondria cells do not compensate by using acetate, leading to genome-wide changes in the histone acetylation landscape, involving several residues in H3 and H4, that reflect the low steady state levels of mitochondrial acetyl-CoA/citrate production [35,67]. These latter findings challenge the notion that the mitochondrial pool of acetyl-CoA does not impact the nucleus given it is compartmentalized (Sivanand et al., 2017). While it is unlikely that the mitochondrial-derived pool of acetyl-CoA is the one utilized to acetylate histones, lack of a functional organelle influences the cytosolic generation of acetyl-CoA by limiting the amount of citrate for ACL. Using a series of assays in cell culture models of mtDNA depletion, we provided unequivocal evidence that supports this conclusion [67]. Nevertheless, why compensatory means to upregulate cytosolic acetyl-CoA production under conditions in which mitochondrial function is impaired are not deployed as when ACL is inhibited remains unclear.

The first study that probed the effects of loss of mitochondrial function to histone acetylation used the DNPOLG cells. Using both Western blots and mass spectrometry, significant decreases in the levels of acetylated H3K9, 14, 18, 23, 27, 56, H4K5 and 8 in the nucleus were found. These changes were accompanied by decreased levels of TCA cycle intermediates, including citrate, and were completely reversed by ectopic expression of NDI1/AOX, which fully resumed TCA activity and metabolite levels [35]. Subsequently, using this same cell culture model and employing ChIP-seq and following specifically the status of the promoter mark H3K9ac, we found that this lysine got progressively hypoacetylated as mtDNA became depleted and mitochondrial dysfunction prominent. Nevertheless, the earliest detectable nuclear changes in histone acetylation were observed prior to any signs of measurable mitochondrial dysfunction, underscoring the sensitivity of the histone acetylation landscape to subtle modulation of mitochondrial function [67]. RNA-seq analysis revealed that, unlike observed in the absence of ACL, no compensatory means to increase cytosolic acetyl-CoA production were activated as mtDNA became depleted. Instead, cells rewired metabolism to spare acetyl-CoA consumption by diminishing, for instance, lipid synthesis while increasing catabolism of branch chain amino acids whose end product is acetyl-CoA [28]. Changes in the promoter H3K9ac mark occurred preferentially in the promoters of genes differentially expressed. Most importantly, both the epigenome and transcriptomic changes were fully reversed in mtDNA-depleted cells expressing NDI1/AOX, suggesting a directly cause-effect relationship between alterations in TCA cycle-derived acetyl-CoA and nuclear histone acetylation [67].

We then questioned whether the lack of activation of alternative means to generate cytosolic acetyl-CoA in DNPOLG cells was because of the acute nature of mitochondrial dysfunction. To address whether cells needed more time to adapt, we used ethidium bromide-derived rho⁰ cells that by virtue of having been generated decades ago were in a chronic state of mitochondrial dysfunction. However, rho⁰ cells showed decreased steady state levels of cellular acetyl-CoA, and H3K9 and 27 were hypoacetylated. As observed in DNPOLG, no compensatory means to increase cytosolic generation of acetyl-CoA was found. Consistent with the lower steady-state levels of acetyl-CoA, histone acetyltransferase (HAT) activity was diminished in rho⁰ but could be reversed by pharmacological means to boost the TCA cycle. For example, dichloroacetate (DCA), which increases pyruvate metabolism within the mitochondria, or dimethyl α-KG exposure, which provides α-KG as another entry point of the TCA [77], increased acetyl-CoA levels and HAT activity in rho⁰ but had no effects in the rho⁺ controls. Also, pharmacological inhibition of ACL decreased acetyl-CoA and histone acetylation in rho⁺ but not rho⁰ cells, demonstrating that acetyl-CoA levels were rate-limiting for histone acetylation in the latter. ACL activity was comparable between rho⁰ and control cells, ruling out the possibility that ACL function was inhibited in conditions of chronic loss of mtDNA. Finally, we showed that exposure of rho⁰ cells to the cell permeable dimethyl α-KG reversed site-specific H3K9ac and H3K27ac levels and gene expression in the promoters of approximately 500 differentially expressed genes in rho⁰ relative to rho⁺ [67]. Collectively, these results were the first to show cause-effect relationships between mitochondrial production of acetyl-CoA, maintenance of histone acetylation and gene expression regulation. They also highlighted the malleability of the epigenome in its response to metabolic changes. Finally, they demonstrated that even long-term epigenetic and gene expression remodeling associated with chronic mitochondrial dysfunction were amenable to pharmacological intervention.

Following this work, other studies also highlighted the sensitivity of the histone acetylation landscape to mitochondrial changes. For instance, increased levels of heteroplasmy, a condition in which normal and mutant mtDNA populate the same cell leading to mitochondrial dysfunction, was associated with decreases in acetyl-CoA and histone hypoacetylation in H3 and H4 lysine residues. The authors focused on changes in H4K acetylation and demonstrated that only when heteroplasmy levels were above 70% the abundance of acetylated lysines on H4 decreased [78]. This contrasts with our data on mtDNA-depleted cells where we detected changes in H3K acetylation prior to overt mitochondrial dysfunction [67]. However, while H3K residues can be methylated or acetylated, the H4 tail residues studied, namely H4K5, K8, K12 and K16 have only been shown to be acetylated. It is possible that maintenance of steady state H4 acetylation levels predominates over those in H3 residues in conditions when acetyl-CoA is limiting, a concept we previously raised [67]. Finally, using metabolic tracing the authors demonstrated that about 70% of glucose-derived acetyl-CoA was used to acetylate H4, but this pathway was essentially gone when cells carried 100% mutant mtDNA. They also showed that glutamine can supply acetyl-CoA for histone acetylation; nevertheless, the glutamine-derived contribution to H4 acetylation is rather small (~8%) under normal conditions. Taken these findings together, the authors concluded that glucose-derived mitochondrial acetyl-CoA is the primary substrate for histone acetylation, strengthening the key role of mitochondria in maintaining the acetylated epigenetic landscape.

Another link between mitochondria and histone acetylation came from studies in hematopoietic stem/progenitor cells (HSPCs) in mice where levels of TFAM (transcription factor A, mitochondrial) were reduced [63]. The development of HSPCs into lineage-committed cells has been widely studied as a paradigm for gene regulatory events governing cellular differentiation. Chief among them is the specification of erythroid lineage, which involve coordinated programs of proliferation, maturation and metabolism. Using transcriptomics and proteomics approaches, the authors identified progressive increases in mtDNA, mitochondrial mass, inner membrane potential and intracellular ATP during erythroid specification. Accordingly, they also identified high levels of TFAM and prohibitin 2 (PHB2). While TFAM is known to regulate mtDNA content, PHB2 is a scaffold in the inner mitochondrial membrane that indirectly impacts mitochondrial genome abundance. To better understand the role of mitochondria in erythroid specification, the authors then depleted cells from TFAM or PHB2, which led to impaired erythroid differentiation, proliferation and increased apoptosis. Transcriptomics profiles revealed that either TFAM or PHB2 loss impaired the silencing of HSPC genes, the expression of genes required for the above-described mitochondrial changes and erythropoiesis. These effects were then recapitulated in TFAM erythroid-specific KO animals, which also showed severe anemia. Thus, it was concluded that TFAM is indispensable for erythropoiesis in vivo [63].

Analysis of the histone landscape revealed significant increases in the levels of acetylated histones in TFAM-deficient erythroid cells, with H3K27ac levels shown by ChIP-seq to be enriched at the promoters of HSPC-specific genes consistent with their increased expression. This hyperacetylated phenotype was unexpected because loss of TFAM is associated with mtDNA depletion, and different studies have connected loss of mtDNA - either partially or completely - to histone hypoacetylation [35,67]. Additionally, we had shown that mouse embryonic fibroblasts derived from TFAM heterozygote mice had 50% of mtDNA and decreased HAT activity, which we presumed reflected loss in total levels of acetyl-CoA [67]. The authors then performed metabolomics analysis in the K562 erythroid cell line depleted of TFAM by shRNA. While they confirmed decreased cellular levels of acetyl-CoA, as presumed by us, they also identified high levels of β-hydroxybutyrate - a ketogenic metabolite that is known to inhibit histone deacetylases (HDACs). Indeed, HDAC activity was significantly inhibited in TFAM depleted cells. Moreover, exposure of WT K562 cells to β-hydroxybutyrate recapitulated the phenotypes observed upon TFAM depletion both in terms of epigenetics and altered erythroid differentiation. Hence, it was concluded that loss of TFAM impaired mitochondria function and metabolism, resulting in increased β-hydroxybutyrate levels and HDAC inhibition. This in turn led to histone hyperacetylation and the inability to silence HSPC genes, resulting in dysregulated erythroid gene expression and differentiation [63].

Finally, similar impairment in differentiation program was found in embryonic stem (ESCs) cells depleted for the mitochondrial carrier homolog 2 (Mtch2). While the exact function of MTCH2 in mitochondria remains elusive, its loss in ESCs was found to lead to mtDNA depletion, histone hyperacetylation and impaired naïve to prime interconversion - a key transition from the pluripotent to the differentiated state. No changes in acetyl-CoA levels were observed between WT and MTCH2 deleted cells, which were essentially confirmed upon carbon tracing experiments following the fate of glucose or glutamine in both cell types [79]. Neither HAT or HDAC activity were probed, but given no changes in acetyl-CoA levels, our data on the relationship between acetyl-CoA and HAT activity [67] and data from Liu and colleagues (2017), a prediction would be that the histone hyperacetylation phenotype resulted from HDAC inhibition. Additional experiments are required to test this possibility. Despite the remaining questions, the data obtained using ESCs depleted of MTCH2 add to the idea that mitochondrial function is important for the execution of proper differentiation programs, at least partly, by influencing the epigenetic landscape.

5. Future perspective

The work described herein demonstrate the different ways that mitochondria can affect epigenetics, with a summary of the current data based on experimental evidence depicted on Fig. 2 and a timeline of the contributions on Fig. 3. While in many cases cause-effect relationships are still missing owed to the complexity of some metabolic pathways or to experimental challenges, the data seem compelling enough to put mitochondria at the crossroads of metabolism and epigenetic maintenance. In addition to expanding the work to interrogate the effects of other mitochondrial biochemical processes on the epigenome and the extent of their dependence on cell metabolism and mitochondrial reliance, more information regarding the signals triggering this communication is still required. Importantly, the field needs to better understand how broad effects in metabolism can lead to lysine-specific histone modifications, and how the genome-wide changes in methylation or acetylation lead to specific gene expression programs. In this context, reporting of the epigenetic effects in all loci that they occur when deep-sequencing based assays are utilized will help the non-experienced researcher understand the breadth of the changes and the inherent complexity of that type of data. Along the same lines, when cross-referencing the epigenetic changes with gene expression, it would be useful to get a complete picture of genes affected rather than information only on the genes of interest, which tend to portray a high degree of specificity that in reality is missing. Lastly, readers should be mindful of the challenges of correlating the physiological relevance of DNA methylation for gene expression. Essentially the entire genome can have altered DNA methylation patterns, and whether they occur on gene bodies, promoters or intergenic regions may differentially impact transcription. Likewise, hypo or hypermethylation may have very different outcomes to gene expression, depending on where they occur. Thus, description of specific parameters adopted can set more realistic expectations to the reader, who must be mindful of the challenges associated with this type of analysis.

Fig. 2. Means of mitochondria-epigenetic crosstalk.

The metabolites through which mitochondria affects epigenetics based on experimental evidence are shown; large arrows depict whether the level of metabolites is increased (green) or decreased (red). Types of mitochondrial interventions are color-coded on the right. The effects on the different epigenetic marks of each intervention are depicted with the same color. Mechanisms proposed, e.g. enzymatic activities involved, are shown within parentheses when data are available; question mark indicates that the mechanisms are unclear or were not tested. N = nucleus, M = mitochondria. For information on all potential ways through which mitochondria might impact epigenetics, please see other reviews such as [84].

Fig. 3. Timeline of contribution from different laboratories to the understanding of the impact of mitochondrial function on epigenetics.

Studies are cited based on chronological order of their publication; earlier studies are depicted closer to the year/bar. Following colors on Fig. 2, studies involving DNA methylation are shown in blue, those reporting changes in histone methylation are shown in purple while the ones showing alterations in levels of histone acetylation are depicted in red.

Defining whether mitochondrial-driven metabolic and epigenetic changes can more broadly alter chromatin architecture or 3D configuration is an area still to be explored. Do genome-wide changes in mitochondrial-driven epigenetic marks influence topologically associating domains (TADs) or chromosome territories? Can the pervasive changes in the epigenome resulting from mitochondrial dysfunction affect transcription in yet novel ways? Interestingly, release of promoter-proximal paused RNA polymerase II into elongation is stimulated by histone deacetylation [80], but whether this can be influenced by mtDNA loss, for example, is unknown. Finally, studies on the relationship between mitochondria and epigenetics have to date focused on the nuclear genome. However, the increasing importance and physiological relevance of RNA epigenetics cannot be understated. Presumably, modulation of mitochondrial function can also alter the epitranscriptome, but such findings have yet to be reported.

From a health perspective, it will be important to contextualize the impact of those mitochondrial-driven epigenetic changes to normal and pathological processes. In addition to understanding whether they play any role in the heterogeneity of tissues affected and severity of mitochondrial disorders, it will be relevant to define the degree that environmental exposures or even “simple” mitochondrial dysfunction associated with more complex diseases such as diabetes, cardiomyopathy or cancer impact epigenetics and health outcomes. Similarly, it will be interesting to understand whether there is a cause-effect relationship between the methylation drift [81] observed in aging and the increasing degree of mitochondrial dysfunction detected over the lifespan of an organism [82]. It also remains unclear whether normal metabolic changes required for important physiological processes such as those observed upon immune activation or stem cell differentiation rely on mitochondrial-driven epigenetic remodeling or occur parallel to the epigenetic changes. Also, the extent to which mitochondria may indeed be the unifying mechanism or a component of DoHaD (Developmental origins of Health and Disease), which posits that early life exposures influence late life health outcomes - and epigenetic alterations may be central for such protracted effects [46,83] - is now an open question. Clearly, the next years will present with many challenges but also exciting opportunities in this emerging field.