Mitochondria as intracellular signaling platforms in health and disease

For those of us who survived the emotional trauma known as junior high school in America, the strategy for adolescent success appeared to be the unique ability to both blend in and stand out. While we will leave it to social scientists to describe the psychological ramifications of such approaches, we would argue that there is an important hidden biological lesson here as well. Mitochondria, present in hundreds to thousands of seemingly identical copies per cell, certainly can blend in. In doing so, they provide the cell with its bioenergetics requirements, supplying the chemical energy required to power essentially all cellular processes. Yet, to view these structures as indistinguishable and amorphous, organelles would be shortsighted. Endowed with a unique evolutionary history, mitochondria have retained a distinct set of lipid components, as well as their own genome, that under specific stress conditions, enables these organelles to also stand out. This singularity makes mitochondria uniquely situated to act as a signaling platform. Here, we review the evidence for how mitochondria participate in a wide range of cell fate decisions that exploit the unique qualities and properties of this organelle. In particular, we will focus on how specific unique mitochondrial lipids can modulate signaling events, how mitochondria participate in innate immune signaling, and how mitochondria can signal back to the nucleus to alter both transcription and the epigenome. These topics cover the multifaceted stress responses used by mitochondria, from signaling reactions within mitochondria, to the surface of mitochondria, to those pathways involving the nucleus and the extracellular environment. Finally, we will describe the initial foray into leveraging these observations to develop new classes of therapies that may have the potential to treat a wide array of diseases, especially age-related diseases that are closely linked to mitochondrial dysfunction.

Involved in all aspects of cell biology and physiology, lipids are the major components of all cellular membranes, delineating the boundary between cells and subcellular organelles and supporting intercellular and intracellular communication. The most abundant lipid in mammalian cells is phosphatidylcholine (PC), whereas the amounts of other lipids such as phosphatidylethanolamine (PE), phosphatidylserine (PS), phosphatidylinositol (PI), phosphatidic acid (PA), cholesterol, and sphingolipids vary depending on the type of cell and subcellular organelle. As an α-proteobacteria-derived organelle, mitochondria exhibit distinct lipid composition in both their inner and outer membranes, which establishes a biochemical basis for mitochondrial signaling. Mitochondrial lipid synthesis and transport have been the subject of several excellent recent reviews (Horvath and Daum, 2013; Tatsuta and Langer, 2017; Tatsuta et al., 2014), and as such, here, we will focus instead on mitochondrial signaling and metabolism regulated by two unique lipids, cardiolipin and PE (Fig. 1).

Cardiolipin comprises nearly 20% of the inner mitochondrial membrane (IMM) and a much smaller fraction (up to 3%) of the outer mitochondrial membrane (OMM), but it is not found in any other subcellular organelle (de Kroon et al., 1997; Gebert et al., 2009). Structurally, cardiolipin is composed of a dimeric phosphatidylglycerol lipid, with two PA molecules connected with a glycerol backbone. This cone-shaped structure makes cardiolipin ideal for the highly curved IMM, the structure of which is dramatically disrupted when cardiolipin is depleted (Dudek, 2017). Cardiolipin is also well known for its integration into all respiratory chain complexes in the IMM and several translocase of outer membrane protein complexes (Dudek, 2017; Schlame and Greenberg, 2017; Xu et al., 2006). In humans, cardiolipin is synthesized at the IMM by cardiolipin synthase, followed by acyl chain remodeling leading to the formation of mature cardiolipin with polyunsaturated fatty acyl chains (Schlame and Greenberg, 2017; Tatsuta et al., 2014). Perturbing cardiolipin synthesis or remodeling causes mitochondrial dysfunction including defective oxidative phosphorylation and severe oxidative stress, as seen in Barth syndrome, a devastating X-linked pediatric disease (Schlame and Greenberg, 2017). In mice, deletion of cardiolipin synthase is embryonic lethal, whereas neuronal specific depletion causes disrupted mitochondrial crista structure and respiration, altered mitochondrial calcium handling, and neurodegeneration (Kasahara et al., 2020), indicating an essential role for cardiolipin in mitochondrial structure and function.

Cardiolipin functions as a reactive oxygen species (ROS) scavenger that protects cells from oxidative stress via cardiolipin oxidation and degradation (Fig. 1). ROS in animal cells is largely produced by the respiratory chain complexes which are assembled in the cardiolipin-rich IMM. The proximity of cardiolipin to the source of ROS and the presence of polyunsaturated fatty acyl chains in cardiolipin makes it an ideal sensor and scavenger of oxygen radicals. Oxidized cardiolipin is toxic (Paradies et al., 2001, 2002) and needs to be quickly degraded by enzymes including phospholipase A2γ (PLA2γ; Liu et al., 2017; Tyurina et al., 2014) and 17-β-hydroxysteroid dehydrogenase 10 (HSD10 [also known as amyloid β-peptide-binding alcohol dehydrogenase]; Boynton and Shimkets, 2015). PLA2γ, which can directly hydrolyze cardiolipin, appears to be the major enzyme for the degradation of oxidized cardiolipin, since either genetic depletion or pharmacological inhibition of PLA2γ results in robust accumulation of oxidized cardiolipin upon mitochondrial stress (Liu et al., 2017; Tyurina et al., 2014). HSD10 uses a different mechanism by in fact further oxidizing cardiolipin, causing subsequent spontaneous breakdown of the oxidized product into diacylglycerol, dihydroxyzcetone, and orthophosphate (Boynton and Shimkets, 2015). Thus, to ensure efficient removal of oxidized cardiolipin, the protein levels and enzymatic activity of PLA2γ and HSD10 must be tightly regulated. Indeed, increased levels of HSD10 have been observed in patients with neurological conditions, including multiple sclerosis and Alzheimer’s disease (Krištofiková et al., 2009). In that regard, amyloid β peptide, which is mechanistically linked to Alzheimer’s disease, directly binds and inhibits the enzymatic activity of HSD10 (Yan et al., 1997; He et al., 1998). Moreover, in a rotenone-induced rat model of Parkinson's disease, inhibiting PLA2γ activity causes increased oxidative stress, more mitochondrial lipid oxidation, augmented apoptosis, and a subsequent exacerbation of Parkinson’s disease symptoms (Chao et al., 2018). From a signaling viewpoint, metabolism of oxidized cardiolipin by PLA2γ and HSD10 generates a large group of second messengers, including oxidized fatty acid lipids and diacylglycerol (Boynton and Shimkets, 2015; Liu et al., 2017; Tyurina et al., 2014), which may be involved in mediating the cellular response to mitochondrial stress.

Cardiolipin externalization from the IMM to the OMM is another important lipid reorganization event that often occurs in response to various conditions involving moderate levels of mitochondrial damage (Fig. 1). In healthy cells, cardiolipin is maintained at very low levels on the OMM and almost exclusively accumulates within the inner membrane (∼96.5 mol%; Kagan et al., 2016). In contrast, with mitochondrial damage or membrane depolarization, cardiolipin moves to the OMM. Three proteins have been reported to transport cardiolipin from the IMM to the OMM: phospholipid scramblase-3 (Liu et al., 2003), mitochondrial creatine kinase, and nucleoside diphosphate kinase D (Epand et al., 2007; Kagan et al., 2016; Schlattner et al., 2013). Although it is not clear whether these proteins work sequentially or in parallel, individual depletion of either phospholipid scramblase-3 or nucleoside diphosphate kinase D substantially reduces cardiolipin relocalization to the OMM (Chu et al., 2013; Kagan et al., 2016; Liu et al., 2008; Schlattner et al., 2013). Direct lipid transport through membrane contact sites between subcellular organelles have been described (Cockcroft and Raghu, 2018; Scorrano et al., 2019; Wu et al., 2018), but it is not clear exactly how these three cardiolipin transporters act, and whether they localize to IMM–OMM contact sites (Horvath et al., 2015).

Externalized, oxidized cardiolipin at the OMM establishes a signaling platform to orchestrate cellular response to mitochondrial damage that can include apoptosis or mitophagy (Fig. 1). In mitochondrial-dependent apoptosis, the recruitment and activation of the apoptotic initiator protease caspase-8, its substrate BH3-interacting domain death agonist (BID), and the downstream effector BCL-2–associated X (BAX) are all dependent on the presence of cardiolipin in the OMM (Gonzalvez et al., 2008; Kuwana et al., 2002; Lutter et al., 2000). The interactions among truncated BID (tBID), BAX, and cardiolipin drive the formation of supramolecular complexes containing BAX oligomers on the OMM surface that mediate membrane permeabilization and cytochrome c release (Kuwana et al., 2002; Lai et al., 2019; Lutter et al., 2000). The oxidation of cardiolipin reduces its interaction with cytochrome c, which further accelerates cytochrome c release to initiate apoptosis. In mitophagy-promoting conditions, increased cardiolipin at the OMM functions as an “eat-me” signal, recognized by the autophagy marker protein microtubule-associated-protein-1 light chain 3 (LC3) allowing for autophagic capture of the damaged mitochondria (Chu et al., 2013). As such, interfering with cardiolipin synthesis or maturation or its IMM-to-OMM transport potently affects mitophagic and apoptotic signaling (Chu et al., 2013; Garcia Fernandez et al., 2000; Hsu et al., 2015; Kagan et al., 2016).

Cardiolipin has been shown to interact with a litany of proteins linked to both protein aggregation and neurodegenerative diseases. Proteins such as α-synuclein, amyloid β, and Tau, known to form aggregates in neurodegenerative diseases, have all been reported to bind to cardiolipin-containing membranes (Ahyayauch et al., 2012; Camilleri et al., 2013, 2020; Nakamura et al., 2011; Robotta et al., 2014). Such binding is believed to reduce cardiolipin levels and/or interfere with cardiolipin function in mitochondrial respiration and/or ROS sensing. However, cardiolipin may play a role in resolving these protein aggregates, including amyloid β (Ordóñez-Gutiérrez et al., 2015) and α-synuclein fibrils (Ryan et al., 2018). Thus, further work is needed to more precisely define the physiological relationship between cardiolipin and protein aggregates in neurodegeneration.

Sphingolipids, usually maintained at very low levels in mitochondria compared with other organelles (the Golgi complex, endolysosomes, and the plasma membrane), have been increasingly implicated in age-dependent mitochondrial dysfunction. Ceramide, which is at the center of sphingolipid metabolism, accumulates in response to almost all types of cellular stress, including chemotherapeutic stress, ionizing and ultraviolet radiation, serum starvation, injury, and infection (Ogretmen, 2018). Like cardiolipin, ceramide can also induce apoptosis by triggering BAX-dependent mitochondrial outer membrane permeabilization (Birbes et al., 2001; Ganesan et al., 2010). Mechanistically, this may involve the direct targeting of voltage-dependent anion channel 2 (VDAC2) by ceramide (Dadsena et al., 2019). In addition, during the initial phase of mitochondrial-mediated apoptosis, ceramide is further processed into two lipid messengers, sphingosine-1-phosphate and hexadecenal, which specifically activate Bcl-2 homologous antagonist killer (BAK) and BAX, respectively, for mitochondrial membrane permeabilization (Chipuk et al., 2012). Moreover, ceramide may also contribute to mitophagy, as the lethal autophagy induced by a natural bioactive sphingolipid, N-stearoyl-D-erythro-sphingosine (C18-ceramide), is dependent on the direct interaction between C18-ceramide and LC3 (Sentelle et al., 2012). As might be expected, crosstalk between cardiolipin and ceramide exists. For instance, cardiolipin has been shown to inhibit ceramide synthesis and promote its cleavage by ceramidase (El Bawab et al., 2001; Kim et al., 2016; Okino et al., 2003). On the other hand, manipulation of ceramide modulates in vivo cardiolipin levels (Babenko and Storozhenko, 2017), which is consistent with observations of an age-related accumulation of ceramide and a corresponding decline in cardiolipin (Helmy et al., 2003; Monette et al., 2011; Petrosillo et al., 2008; Sen et al., 2007). In addition, alterations in these sphingolipid-dependent pathways have been increasingly linked to conditions such as Alzheimer’s disease (Pera et al., 2017).

The IMM also contains the highest molar ratio of PE compared with other subcellular membranes, which is consistent with its bacterial origin, as PE is quite abundant in many bacteria (Kaval and Garsin, 2018; López-Lara and Geiger, 2017). This is mainly achieved by direct synthesis of PE from PS at the IMM (Fig. 1). Most phospholipids, including PC, PS, and PI, are synthesized in the ER and then transported to other organelles (Tatsuta and Langer, 2017). PS synthesized in the ER is transported to the mitochondrial outer membrane at ER–mitochondrial contact sites. PS is then further transferred from the mitochondrial outer to the inner membrane after which it is converted by PS decarboxylase (PISD) into PE (Schuiki and Daum, 2009). PISD is an enzyme conserved from bacteria to humans that is critical for PE synthesis at the IMM, malfunction of which causes mitochondrial fragmentation and embryonic lethality in mice (Hartmann et al., 2016; Steenbergen et al., 2005), indicating that direct PE synthesis at the IMM cannot be compensated by other pathways.

PE synthesis at the IMM is dynamically controlled to balance cell growth and mitochondrial biogenesis. One key protein regulated by PISD-produced PE at the IMM is YME1L, a transmembrane ATP-dependent protease essential for mitochondrial proteostasis and dynamics, apoptotic resistance, and cell proliferation (Anand et al., 2014; Stiburek et al., 2012). Normal levels of PE suppresses the proteolytic activity of YME1L (Fig. 1). However, stresses such as hypoxia or nutrient deprivation induce a lipid signaling cascade upon inactivation of mechanistic target of rapamycin complex 1 (mTORC1) to reduce PE levels at the IMM, leading to activation of YME1L-mediated proteolysis of mitochondrial proteins and subsequent inhibition of mitochondrial biogenesis (MacVicar et al., 2019). This involves activation of LIPIN1, a substrate of mTORC1 that is phosphorylated and hence inhibited under basal conditions (e.g., available oxygen and nutrients). Inactivation of mTORC1 leads to dephosphorylation and activation of LIPIN1 that reduces the level of PA, resulting in a decline in the subsequent synthesis of PS, a substate of PE (MacVicar et al., 2019). Consistent with a key role for PE in suppressing YME1L, blocking PS transfer from the mitochondrial outer to inner membrane (hence reducing PE levels) activates YME1L (MacVicar et al., 2019). Thus, signals that modulate lipid transport and/or synthesis pathways can rewire mitochondrial metabolism in response to physiological changes. Interestingly, the proteolytic degradation of YME1L, as well as another mitochondrial protease, OMA1, is regulated by various cellular insults that perturb mitochondrial functions (Baker et al., 2014; Rainbolt et al., 2016). It would therefore be important to investigate whether this phenomenon is also modulated by changes in mitochondrial lipids.

While mitochondrial cardiolipin and PE have been extensively studied, other potential lipid signaling pathways are not yet well characterized. Although the mitochondrial outer membrane is very weakly charged, dynamic turnover of certain negatively charged phosphoinositides has been reported. For example, a low level of phosphatidylinositol 4,5-bisphosphate [PI(4,5)P₂] can be detected on mitochondria (Watt et al., 2002). Removal or masking this mitochondrial pool of PI(4,5)P₂ leads to fragmentation and autophagic targeting of mitochondria (Rosivatz and Woscholski, 2011). However, the downstream effectors for these and related lipid signals are largely undefined.

In summary, the double-membrane structure of mitochondria and its unique lipid composition establish an unusual membrane environment that are essential for maintaining mitochondrial activity and homeostasis. In particular, mitochondrial lipid signaling, mediated through cardiolipin and PE, provides an immediate capacity to both sense and respond to mitochondrial stress. When this stress goes beyond the handling capacity of local lipid signaling, more mitochondrial damage is induced. This often leads to the activation of the next level of response, both on the surface mitochondria and in the cytosol, as we discuss below.

Mitochondria are linked to various innate immune signaling pathways, with the best characterized being the cytosolic RNA-sensing pathway for which mitochondria function as an essential platform (Fig. 2). While endolysosomal microbial RNAs are sensed by Toll-like receptors (specifically Toll-like receptors 3/7/8), those in the cytosol are directly captured by cytosolic RNA sensors known as retinoic acid–inducible gene I (RIG-I)-like receptors (RLRs; Tan et al., 2018). These include RIG-I (Yoneyama et al., 2004), melanoma differentiation-associated gene 5 (MDA5; Kang et al., 2002), and laboratory of genetics and physiology 2 (Saito et al., 2007), the last of which has no known intrinsic signaling capacity. Once bound to RNA ligands, both RIG-I and MDA5 form oligomers that directly bind and activate mitochondrial antiviral-signaling protein (MAVS [also known as IPS-1, VISA, and CARDIF]), a signaling scaffold with a C-terminal transmembrane domain anchored on the OMM (Kawai et al., 2005; Meylan et al., 2005; Seth et al., 2005; Xu et al., 2005). Activated MAVS forms aggregates that recruit signaling molecules (e.g., inhibitor of nuclear factor-κB kinase [IKK] and TANK-binding kinase [TBK1]), leading to the activation and nuclear translocation of NF-kB and IRF3/7, initiating transcriptional up-regulation of proinflammatory cytokines and type I IFNs (Liu et al., 2013). One key feature of MAVS signaling are its feedforward properties, whereby a small amount of MAVS aggregates can induce the formation of large MAVS aggregates by directly recruiting and activating other MAVS molecules on the mitochondria (Cai et al., 2014; Hou et al., 2011; Xu et al., 2015). The anchorage of MAVS on the mitochondrial surface is critical for its function, as removal of the MAVS transmembrane domain by a hepatitis C virus–encoded protease abrogates its signaling capacity (Li et al., 2005a; Li et al., 2005b; Meylan et al., 2005). Evidence suggests that MAVS may also localize to peroxisomes and mitochondrial-associated membranes (Dixit et al., 2010; Horner et al., 2011).

There is extensive communication between MAVS signaling and mitochondrial metabolism (Fig. 2). It is well established that MAVS is C-terminally anchored on the OMM and that the formation of large MAVS aggregates causes substantial changes to mitochondrial morphology (Cai et al., 2014; Hou et al., 2011; Xu et al., 2015). Even endogenous levels of MAVS aggregates are sufficient to drive apoptosis, which is dependent on the MAVS transmembrane domain, likely linked to MAVS-driven mitochondrial fragmentation (Hwang et al., 2019). Driving such morphological changes are MAVS-dependent mitochondrial ROS generation, mitochondrial membrane hyperpolarization, inhibition of spare respiratory capacity, and modulation of ATP synthesis (see below; Buskiewicz et al., 2016; Lei et al., 2009).

An interesting link between MAVS signaling and glucose metabolism has recently been established (Fig. 2). The first step of glucose metabolism involves the phosphorylation of glucose into glucose-6 phosphate by hexokinase. Although glycolysis occurs in the cytosol, hexokinase 2 (HK2), the major enzyme that initiates glycolysis, requires mitochondrial localization for its glycolytic function (DeWaal et al., 2018; Roberts and Miyamoto, 2015; Wolf et al., 2016). The mitochondrial localization and activity of HK2 is dependent on the physical interactions of HK2 with both MAVS and VDAC situated on the OMM (Roberts and Miyamoto, 2015; Wolf et al., 2016; Zhang et al., 2019). Upon RLR activation, RIG-I binding to MAVS releases HK2 from mitochondria and thus inactivates glycolysis at its initial step (Zhang et al., 2019). On the other hand, up-regulation of glycolysis in turn suppresses MAVS signaling. Lactate, a key metabolite of anaerobic glycolysis, is a direct suppressor of RLR signaling, since it directly binds to the transmembrane domain of MAVS and prevents the formation of MAVS aggregates on the mitochondrial surface (Zhang et al., 2019).

While mitochondria play protective roles in innate immune signaling, abnormal activation of this system can nevertheless cause undesired inflammation and even cell death. For example, the integrity of the mitochondrial genome and RNAs must be tightly controlled. In situations where membrane damage leads to the leakage of mitochondrial nucleic acids, immune responses and/or cell death can be triggered (Fig. 2). Due to bidirectional transcription, mitochondrial RNAs form extensive double-stranded RNAs that are efficiently digested by the mitochondrial RNA helicase SUV3 and the 3′–5′ RNA exonuclease PNPT1 (Aloni and Attardi, 1971; Dhir et al., 2018; Young and Attardi, 1975). Hypomorphic mutations of PNPT1 result in robust accumulation and cytosolic leakage of mitochondrial double-stranded RNAs that trigger MDA5- and MAVS-dependent inflammation (Dhir et al., 2018). Additionally, mitochondrial double-stranded RNA processed by RNase L in p53-deficient cells might also trigger MAVS signaling and transcription of proinflammatory factors (Wiatrek et al., 2019). Likewise, leakage of mitochondrial DNA (mtDNA) activates the cytosolic DNA sensor cyclic GMP-AMP synthase (cGAS), leading to the further activation of stimulator of IFN genes (STING) and type I IFN production (Kim et al., 2019; West et al., 2015). Such mtDNA-stimulated innate immune signaling appears to facilitate neurodegeneration in mouse models with mitochondrial stress but defective mitophagy (Sliter et al., 2018). It would be important to further elucidate the role of inflammation mediated by mitochondrial-derived nucleic acid in other disease states.

Various RNA-independent mechanisms activating MAVS-mediated transcriptional up-regulation of proinflammatory cytokines have been described. One such mechanism involves the communication between oxidative stress and MAVS (Fig. 2). Oxidative stress is a common condition in infectious and inflammatory diseases, which can be sensed by MAVS to trigger RNA- and RLR-independent MAVS oligomerization and type I IFN production (Buskiewicz et al., 2016). Moreover, antioxidants can suppress ROS-induced MAVS oligomerization and IFN production (Buskiewicz et al., 2016), which may potentially benefit patients with autoimmune or inflammatory diseases. The fact that MAVS aggregates are resistant to detergent but sensitive to reducing agents is consistent with MAVS being an ROS sensor (Cai et al., 2014; Hou et al., 2011). In agreement with a possible role for ROS-driven disulfide bond formation in MAVS oligomerization, the MAVS-C79F variant does not oligomerize when exposed to ROS (Buskiewicz et al., 2016). Indeed, systemic lupus erythematosus patients carrying the MAVS-C79F variant show reduced plasma levels of MAVS oligomers, reduced type I IFN production, and milder disease symptoms (Buskiewicz et al., 2016; Pothlichet et al., 2011). Of note, while MAVS senses both cellular and mitochondrial ROS, MAVS aggregation can in turn promotes the generation of mitochondrial ROS and appears to be essential for infection-induced mitochondrial hyperpolarization, decrease in ATP synthesis, and drop in the spare respiratory capacity (Buskiewicz et al., 2016; Lei et al., 2009). It is still elusive how the prion-like aggregation of MAVS and the feed-forward cycle between ROS and MAVS aggregation are ultimately terminated and how mitochondrial function is restored in healthy patients after the clearance of an infection. Further mechanistic investigation of these critical deactivation steps could yield novel therapeutic strategies for autoimmune and inflammatory diseases.

Another unconventional mechanism that activates mitochondrial MAVS involves abnormal sphingolipid metabolism (Fig. 2). Sphingolipids are well known for their roles in promoting inflammation in both physiology and diseases. Expression changes of sphingolipid metabolizing enzymes have been documented upon induction of neuronal inflammation (Morita et al., 2020). In experimental autoimmune encephalomyelitis, a murine model of multiple sclerosis, the sphingolipid lactosylceramide promotes inflammation and subsequent neurodegeneration (Mayo et al., 2014). Recently, lactosylceramide has been found to trigger an interaction between MAVS and cytosolic phospholipase A2 (cPLA2) that activates MAVS aggregation and type I IFN production in astrocytes independently of RNA ligands and RLRs (Chao et al., 2019). Lactosylceramide directly binds to and activates cPLA2, a calcium-dependent phospholipase that cleaves phospholipids, releasing inflammatory lipid messengers (Chao et al., 2019; Leslie, 2015; Nakamura et al., 2013). Once activated, cPLA2 is recruited to mitochondria, where it binds and activates MAVS and changes mitochondrial metabolism by multiple mechanisms (Chao et al., 2019). First, cPLA2 binding directly triggers MAVS aggregation, leading to transcriptional production of pro-inflammatory cytokines and type I IFN. Second, cPLA2 displaces HK2 from MAVS, which causes reduced HK2 activity in glycolysis (see above) and a subsequent decline in lactate production. Third, cPLA2 activation by lactosylceramide increases the expression of cPLA2 itself and the generation of ROS, both of which, in turn, further fuel MAVS aggregation (Chao et al., 2019). Given the reported associations of MAVS signaling, cPLA2, and sphingolipids with various neurodegeneration diseases, interventions targeting the cPLA2–MAVS signaling axis might prove beneficial.

Thus, MAVS is an integral part of the cellular stress response, which is supported by the extensive communications between MAVS signaling and cellular metabolism. While the MAVS pathway has likely evolved to protect cells from pathogens and damage, undesired, chronic activation of this RNA-sensing pathway, together with the cGAS–STING DNA-sensing pathway, is emerging as underlying contributor fueling inflammation-driven diseases.

Besides lipid signaling within mitochondrial membrane and innate immune signaling on the mitochondrial surface and in the cytosol, mitochondria also send out stress signals via interorganelle communication, adding another layer of potential regulation. Crosstalk between the mitochondria and nucleus is well defined in the regulation of cellular bioenergetics and stress responses. This includes anterograde signaling from the nucleus to the mitochondria, as most mitochondrial proteins are encoded by the nuclear genome, translated in the cytosol, and imported into mitochondria. On the other hand, metabolic, proteostatic or oxidative stress within mitochondria can be communicated in a retrograde fashion back to the nucleus (Fig. 3).

One such retrograde pathway is controlled by the lipid signaling we have already discussed. When mitochondrial function is compromised, cells activate a transcriptional response known as the mitochondrial unfolded protein response (UPR; Shpilka and Haynes, 2018). In yeasts, general cellular membrane stress caused by global lipid disequilibrium can be resolved by transcriptional activation of the mitochondrial UPR, which in turn maintains cellular membrane morphology (Thibault et al., 2012). Likewise, in worms, mitochondrial protein-folding stress can be decoded into local lipid signaling, which is further transmitted to the nucleus to transcriptionally turn on both the mitochondrial UPR and the cytosolic heat shock response. Specifically, disruption of mitochondrial proteostasis triggers an up-regulation of cardiolipin and down-regulation of ceramide to initiate a nuclear transcription program (Kim et al., 2016). Both cardiolipin accumulation and ceramide reduction are necessary and sufficient to trigger this program, which protects worm cells from cytosolic proteotoxicity when mitochondrial proteostasis is impaired (Kim et al., 2016).

An increasing number of proteins encoded by the nuclear genome have been shown to reside in, or on, mitochondria under basal conditions, whereas metabolic or other cellular stresses stimulate their translocation into the nucleus to activate transcriptional stress responses. For example, a well-conserved oxidative stress response factor, nuclear factor erythroid-2–related factor 2 (NRF2, known as SKN-1 in Caenorhabditis elegans) is sequestered by interacting with Kelch-like ECH associated protein 1 (KEAP1) and the mitochondrial serine/threonine protein phosphatase PGAM5 on the surface of the OMM (An and Blackwell, 2003; Lo and Hannink, 2008). Oxidative stress disrupts this protein complex and triggers NRF2 translocation to the nucleus, where it activates the transcription of a large number of genes, including a set of antioxidant proteins, to help restore redox homeostasis (Lo and Hannink, 2008; Sekhar et al., 2002). Other nuclear-encoded mitochondrial proteins that translocate into the nuclei upon physiological or exogenous stresses have been recently reviewed (Monaghan and Whitmarsh, 2015). Interestingly, a 16-amino-acid peptide called MOTS-c, encoded by the mitochondrial genome of mammalian cells, also translocates from cytoplasmic structures to the nucleus upon metabolic stresses accompanied by ROS production (Kim et al., 2018). Activation of 5′-AMP–activated protein kinase (AMPK) downstream of ROS is required for MOTS-c nuclear translocation (Kim et al., 2018). In the nucleus, MOTS-c in turn initiates the transcription of a set of stress response genes that may protect against conditions characterized by metabolic dysfunction (Lee et al., 2015).

In additional to regulating translocation of transcription factors, mitochondria can also transmit stress signals that result in epigenetic modifications, such as histone acetylation and histone and DNA methylation. A mild increase in mitochondrial ROS, especially at the early stage of life, has been shown to benefit multiple organisms from yeast to humans. In C. elegans, early exposure to oxidants results in epigenetic change in the adult organism that correlates with increased stress resistance and a longer lifespan (Bazopoulou et al., 2019). Although in many cases it remains unclear as to how mitochondrial stress is transmitted to the epigenome, some downstream mitochondrial ROS effectors have been identified in yeasts and worms with conserved homologues in mammals. For instance, in murine cells, the heterogeneous ribonucleoprotein A2 (hnRNPA2) acetylates lysine 8 of histone H4 at mitochondrial stress–responsive promoters in an acetyl-coenzyme A–dependent manner to activate gene transcription (Guha et al., 2016). In yeasts, two serine/threonine protein kinases, Tel1p and Rad53p, homologues of mammalian ataxia telangiectasia mutated (ATM) and Chk2 (Schroeder et al., 2013), promote yeast chronological lifespan in response to mitochondrial ROS by inactivating the histone H3K36 demethylase Rph1p, leading to methylation and transcriptional suppression of subtelomeric regions (Schroeder et al., 2013). The highly conserved histone lysine demethylases JMJD-1.2/PHF8 and JMJD-3.1/JMJD3 also play key roles in mitochondrial stress–induced lifespan extension in different species (Merkwirth et al., 2016). Another histone methyl transferase, MET-2, is involved in this process in worms by regulating chromatin reorganization as part of mitochondrial stress response in the nucleus, which is regulated by a nuclear cofactor LIN-65 that translocates from cytosol to the nucleus upon mitochondrial stress (Tian et al., 2016). Such chromatin reorganization causes gene silencing of many loci but often allows for transcriptional activation of stress response genes at other chromosome locations (Tian et al., 2016). Mitochondrial stress also triggers global elevation of N6-methyldeoxyadenine in stress responsive genes, which can be inherited to modulate stress response signaling across generations of worms (Ma et al., 2019). Thus, in general, mitochondrial stress induces epigenetic reprograming of gene expression in order to promote stress adaptation.

It is clear that there are multiple retrograde signaling pathways deriving from the mitochondria. However, at present, these pathways do not always seem to be highly conserved among species. It is also conceivable that these observed species differences reflect the role mitochondrial communication plays in a largely postmitotic organism like C. elegans compared with more dynamic mammalian species. Whatever the explanation, it is fair to say that the molecular and biochemical mechanisms linking mitochondrial stress to the activation or inactivation of transcription factors and epigenetic enzymes are still, for the most part, poorly defined.

Most human diseases are age related. Aging is a complex process involving the impairment of mitochondrial respiratory activity, increase of oxidative stress, compromised stress response, and accumulation of damaged protein and organelles, leading to overall decline of cellular functions and induction of various age-related diseases. Since mitochondrial ROS contribute to much of the cellular damage in aging and disease, the most straightforward therapeutic approach would appear to be modulating mitochondrial ROS levels (Fig. 4). With that said, the history of broadly acting ROS scavengers is not particularly encouraging (Bjelakovic et al., 2007), although the antioxidant edaravone was recently approved for the treatment of amyotrophic lateral sclerosis (Jaiswal, 2019). One promising class of mitochondria-targeted antioxidants are plastoquinone derivatives that can accumulate within mitochondria and, once oxidized, can be reduced by accepting electrons from the mitochondrial respiratory chain (Feniouk and Skulachev, 2018b). One plastoquinone derivative, SkQ1 (a decyltriphenyl phosphonium cation conjugated to a quinone moiety), has been extensively characterized in vitro and in vivo and appears to have significant antioxidant activity at even nanomolar concentrations (Feniouk and Skulachev, 2018a). Long-term treatment with SkQ1 increased lifespan of a range of species, including mammals, and various clinical trials are currently testing clinical efficacy on various age-related diseases. An eyedrop form of SkQ1 (Visomitin) is currently in phase 3 trials for dry eye disease, a condition associated with oxidative stress (Feniouk and Skulachev, 2018a).

Another group of mitochondria-targeted antioxidants are peptides (Fig. 4). In 2004, Szeto and Schiller designed a series of cell-permeable antioxidant tetrapeptides (SS peptides) that specifically accumulate at the IMM (Zhao et al., 2004). Among these peptides, the most promising one, SS-31 (D-Arg-2′6′-dimethylTyr-Lys-Phe-NH2 [also known as MTP-131, elamipretide, or Bendavia]) is believed to directly bind to cardiolipin in the IMM and protect it from peroxidation, leading to restored mitochondrial bioenergetics (Birk et al., 2013). SS-31 has proven effective in protecting mitochondrial functions in different animal models, and clinical trials have also demonstrated promising effects in humans (Szeto, 2014). In an early small trial, 5 d of SS-31 treatment substantially increased exercise performance of patients with primary mitochondrial myopathy (Karaa et al., 2018). In another clinical trial for human atherosclerotic renal artery stenosis that required a revascularization procedure, SS-31 significantly suppressed procedure-associated ischemic renal injury and increased renal blood flow and renal function after the procedure, with overall improved outcome for revascularization (Saad et al., 2017). However, disappointingly, a recent randomized, phase 3 trial with this compound failed to demonstrate efficacy in patients with primary mitochondrial diseases.

When blocking ROS generation may be difficult or problematic, improving the removal of damaged mitochondria is an alternative therapeutic approach. Indeed, accumulation of damaged mitochondria is implicated in various degenerative diseases, and pharmacologically stimulating mitophagy has proven an effective way to prevent cell death (Fig. 4). A chemical screen identified the antibiotic actinonin as a potent activator of mitophagy (Sun et al., 2015). More recently, urolithin A and actinonin have been found to stimulate mitophagy and reverse memory impairment in various animal models of Alzheimer's disease (Fang et al., 2019), suggesting that mitophagy stimulation may be a promising therapeutic intervention for neurodegeneration.

Nucleic acid–stimulated innate immune signaling is emerging as a key mechanism for multiple age-related diseases. Downstream of mitochondrial oxidative stress and mitochondrial damage, RNA- and/or DNA-stimulated inflammation has been shown to be an essential contributing factor for tissue damage in various conditions (Dhir et al., 2018; Maekawa et al., 2019; Sliter et al., 2018). This usually involves genetic mutations or cellular damage that causes abnormal exposure of nucleic acids to innate immune sensors or gain-of-functions mutations directly activating nucleic acid sensors such as MDA5. Given the age-dependent increase of oxidative stress that could cross-activate MAVS signaling (Buskiewicz et al., 2016) and the age-dependent increase in circulating mtDNA (Pinti et al., 2014), inhibitors specifically targeting the MAVS pathway or the cGAS–STING pathway (Haag et al., 2018; Hall et al., 2017; Lama et al., 2019; Vincent et al., 2017) could be considered for inflammation-driven diseases (Fig. 4). This concept has been recently expanded on by the demonstration of mtDNA escape from the mitochondria via oxidative stress–induced macropores formed by VDAC oligomers on the OMM (Kim et al., 2019). Small molecules that inhibit VDAC oligomerization (Ben-Hail et al., 2016) appear to provide a mitochondrial-centric strategy to lessen the inflammatory symptoms seen in an animal model of lupus (Kim et al., 2019).

The ideal therapeutics for aging and age-related diseases may be a drug that can function through multiple mechanisms. Metformin, the most commonly used medication for type 2 diabetes, is one such drug that has drawn increasingly more attention due to its antiaging effects in many models (Barzilai et al., 2016). Metformin reduces insulin levels, mTORC1 signaling, ROS generation, DNA damage, and inflammation, whereas it appears to activate AMPK, autophagy, and removal of senescent cells (Barzilai et al., 2016). These effects may all be a result of the effects of metformin as a mild mitochondrial inhibitor. The Metformin in Longevity Study (MILES) was launched in 2014 with 14 patients and demonstrated the potential effects of metformin in modulating metabolic and nonmetabolic pathways linked to aging (Kulkarni et al., 2018). A much larger trial, Targeting Aging with Metformin (TAME), has been planned in order to test the effects of metformin on various age-related diseases (Barzilai et al., 2016). The mitochondrial genome-encoded retrograde hormone MOTS-c appears to act similarly to metformin by activating AMPK and increasing methionine turnover (Lee et al., 2015). Mice injected with this 16-amino-acid peptide show increased sensitivity to insulin and resistance to diet-induced obesity (Lee et al., 2015). Interestingly, the levels of MOTS-c decrease with age, suggesting that replenishing MOTS-c levels might have therapeutic potential in certain age-related diseases.

Blending in and standing out, mitochondria are uniquely positioned to function as signaling hubs. Though expressed in hundreds of copies per cell, their unique composition and genome make them a complicated mix of the familiar and the distinct. Here, we have described how these properties can affect cell fate decisions, such as initiating cell death pathways, triggering innate immunity, or altering the epigenetic code. Many questions remain. These include a better understanding of how various other critical properties of mitochondrial function (e.g., fusion/fission and membrane potential) impact signaling. In that regard, it would be important to understand how age-dependent impairment of mitochondrial quality control alters, for instance, MAVS aggregation or epigenetic regulation. While we have focused on retrograde communication, namely the communication from mitochondria to the nucleus, other signaling connections are currently less understood. The tight physical connection between ER and mitochondria likely means these organelles can cross-regulate their biology, yet details are currently woefully incomplete. More distant connections might also exist (e.g., mitochondrial–lysosomal interactions). Of note, individual bacteria communicate with each other through mechanisms such as quorum sensing, and it would not be surprising if analogous mechanisms allowed the multiple mitochondria within a given cell to somehow talk to each other. Finally, it remains likely that mitochondria can signal beyond the cell. We have touched briefly on mitochondrial-derived peptides such as MOTS-c, which circulate and are readily detectable in human serum. The notion of other circulating “mitokines” is attractive and represents a way to potentially coordinate organismal adaptation to mitochondrial stress. In simpler organisms, such as C. elegans, there is ample evidence for such pathways (Durieux et al., 2011; Zhang et al., 2018). In mammals, mitochondrial dysfunction in one tissue can trigger increased expression of circulating factors, such as FGF21, which may help to mitigate the initial stress (Kim et al., 2013). Interestingly, transgenic overexpression of FGF21 in mice appears to extend lifespan (Zhang et al., 2012). However, these animals exhibited supraphysiological levels of FGF21 (5–10 times), and in fact, in humans, levels of FGF21 actually appear to increase as we age (Conte et al., 2019). As such, questions remain as to whether FGF21 and related factors (e.g., GDF15) are sensitive or specific biomarkers of mitochondrial function and whether they play a protective or detrimental role.

While key questions remain unanswered, even with this very rudimentary understanding and significant knowledge gaps, the clinical applications of these insights are accelerating. Unlike the lessons of junior high school, these applications will likely have enduring consequences, as successful manipulation of mitochondrial signaling has the potential to impact a wide range of inflammatory and age-related conditions.

We thank members of the Finkel laboratory for helpful comments.

This work was supported by National Institutes of Health grants P30AG024827, 1R01HL142663-01 and 1R01HL142589-01A1.

The authors declare no competing financial interests.