Mitophagy pathways in health and disease

Macroautophagy, hereafter referred to as “autophagy,” is an evolutionarily conserved pathway involving the engulfment of cytosolic contents by a lipid membrane for recycling of nutrients or removal of harmful aggregates, microbes, and organelles (He and Klionsky, 2009). Mitophagy is one form of macroautophagy that involves selectively targeting and engulfing mitochondria for removal through lysosomal degradation (Rodriguez-Enriquez et al., 2006). Activation of this pathway is a result of mitochondria being damaged beyond the capabilities of other quality control methods or in instances in which the cell needs to get rid of mitochondria for metabolic or developmental purposes (Palikaras et al., 2018). As is the case in other forms of selective autophagy, mitophagy involves some form of “eat me” signal on the surface of the mitochondria designated for clearance (Palikaras et al., 2018).

The efficient functioning of mitochondria is essential for their diverse roles in the cell, including but not limited to ATP synthesis, lipid and heme biosynthesis, calcium buffering, and innate immune surveillance (Friedman and Nunnari, 2014). During instances of mitochondrial damage, mitophagy removes malfunctioning mitochondria to maintain the population at an optimal state (Palikaras et al., 2018). Not only are damaged mitochondria deficient at making ATP and other biosynthetic products, but, as a result, they release greater levels of reactive oxygen species (ROS; Murphy, 2009). This turns into a feedback signal because mitochondria themselves are sensitive to the oxidizing damage of ROS to proteins and DNA in addition to downstream activation of the NLRP3 inflammasome (Heid et al., 2013). Accumulation of defective mitochondria also leads to cell death through the release of prodeath molecules and accumulation of mutations in the mitochondrial DNA (mtDNA; Youle, 2019). The removal of mitochondria is balanced through the regulated biogenesis of new mitochondria (Jornayvaz and Shulman, 2010). Uncontrolled mitophagy would disrupt homeostasis because there would not be enough remaining organelle, in addition to overwhelming lysosomes. Because the careful and timely removal of mitochondria appears crucial for cell survival, cells have evolved numerous and often overlapping pathways to ensure that mitophagy can occur in a balanced way in response to a wide array of stimuli and triggers (Palikaras et al., 2018).

Although this review focuses on mitophagy and its role in mitochondrial quality control, various mitochondrial quality control programs exist and function independently from mitophagy (Youle, 2019). The type of mitochondrial quality control activated depends on the category and level of stress or damage. The most basic of these involves regulation of mitochondrial dynamics through fission and fusion to locally sequester portions of damaged mitochondria (Ni et al., 2015). Before a segment of mitochondria can be fully engulfed in the growing mitophagophore, it must be asymmetrically divided from the remaining mitochondrial reticulum. This ensures that the damaged regions will be isolated in a way that minimizes the lost contents. The mitochondrial proteins Fis1 and MFF recruit and form complexes with DRP1, allowing it to encircle the region of mitochondria to be split off (Friedman and Nunnari, 2014). In addition, the mitochondrial unfolded protein response relieves the burden of misfolded and damaged proteins by increasing the expression and presence of chaperones such as heat shock protein 22 (Hsp22), Hsp60, and Hsp70 and the proteases Lon and ClpP within mitochondria (Moehle et al., 2019). Mitochondrial proteins can also be translocated to the cytosol for proteasomal clearance (Bragoszewski et al., 2015). Finally, other pathways related to mitophagy involve mitochondria-derived vesicles, piecemeal mitophagy, and mitochondria being taken up directly by lysosomes (microautophagy; Kiššová et al., 2007; Neuspiel et al., 2008; Soubannier et al., 2012; McLelland et al., 2014; Le Guerroué et al., 2017; Pickles et al., 2018). Recently, transmitophagy, the transfer of mitochondria from cell to cell for degradation, has been explored, especially in the central nervous system, reflecting the progress the field has made in understanding mitochondrial clearance from within a cell to a system of cells (Davis et al., 2014; Morales et al., 2020).

We start with the question of whether mitophagy begins with the same machinery as the well-characterized autophagy initiation pathways (Fig. 1). There is coordinated colocalization of the ULK1 complex to ATG9 vesicles along the ER, and both are important for mitophagosome initiation (Itakura et al., 2012). The ULK1 complex transmits upstream signals of stress from mammalian target of rapamycin and AMP-activated protein kinase (AMPK) to promote autophagy initiation in times of need, and it is itself regulated by phosphorylation, whereas the ATG9 vesicles offer another source of lipid membrane that will be included in the new autophagosome (Kim et al., 2011; Yamamoto et al., 2012). These areas can then recruit VPS34, the class III phosphatidylinositol 3–kinase (PI3P) complex that produces PI3P (Burman and Ktistakis, 2010). Omegasomes are DFCP1-positive ER segments where PI3P is concentrated and are often regarded as focal points for phagophore synthesis and precursors to the isolation membranes that facilitate autophagy (Axe et al., 2008; Nanao et al., 2015). This PI3P also attracts WIPI1 and WIPI2, which recruit ATG16 and the associated ATG machinery needed to covalently modify ATG8s with phosphatidylethanolamine (Polson et al., 2010). The ATG8 family of proteins associate with phagophores through conjugation to phosphatidylethanolamine and are understood as the tether between phagophore and cargo (Kabeya et al., 2004; Pankiv et al., 2007). Once this machinery is in place, the mitophagophore continues to elongate and mature through the addition of lipids until it can be closed by the endosomal sorting complex required for transport machinery (Zhen et al., 2020).

The next question involves whether mitochondria are targeted to existing phagophores within the cytosol, like those created following starvation, or whether mitochondrial stress stimulates the de novo synthesis of the autophagosome around or even on the surface of the mitochondria that will be removed. Recent studies support the latter model, and from an energetic point of view, local synthesis would limit the size of the phagophore to one that is as small as possible (Lazarou et al., 2015; Vargas et al., 2019; Zachari et al., 2019). Depolarized mitochondria become systematically labeled with phosphorylated ubiquitin through the actions of PINK1 and Parkin, respectively (Vives-Bauza et al., 2010; Matsuda et al., 2010; Narendra et al., 2010). PINK1 is a mitochondrial-localized serine/threonine protein kinase that phosphorylates ubiquitin (Koyano et al., 2014). Parkin is an E3 ubiquitin ligase that is found in the cytosol until PINK1 phosphorylation directs its recruitment and activation on the surface of damaged mitochondria (Kazlauskaite et al., 2014). In this new location, Parkin ubiquitinates outer mitochondrial membrane (OMM) proteins, providing the substrate for additional PINK1-induced phosphorylation (Matsuda et al., 2010; Ordureau et al., 2014). Ubiquitin acts as a signal to cargo receptors such as OPTN and NDP52, which contain LC3/GABARAP-interacting region (LIR) motifs that allow the subsequent recruitment of ATG8 to the mitochondria (Lazarou et al., 2015). Phosphorylation of OPTN and NDP52 by the kinase TBK1 controls their localization to ubiquitinated mitochondrial proteins (Lazarou et al., 2015; Heo et al., 2015). Recent studies have shown that NDP52 can bring the ULK1 complex to the mitochondrial surface through direct binding to FIP200 to initiate mitophagy (Lazarou et al., 2015; Vargas et al., 2019). Additional cargo receptors such as p62, NBR1, and TAX1BP1 that are known for roles in other selective autophagy processes appear to play a more modest and dispensable role in mitophagy (Shi et al., 2015; Lazarou et al., 2015).

Mitophagosome synthesis has historically been perceived as the linear sequence described above. Using this understanding, the initiation of mitophagy would be LC3 independent because recruitment of NDP52 to mitochondria through ubiquitin binding is enough to initiate a phagophore at the surface of the mitochondria even in the absence of all ATG8s (Nguyen et al., 2016). However, mitophagosomes that form in the absence of LC3 are smaller and do not mature fully, suggesting that the LC3 family is important for elongation and maturation (Padman et al., 2019). This study later confirmed that there is a feedforward cycle whereby LC3 on the mitochondria can drive further recruitment of NDP52 and OPTN rather than strictly the opposite direction. Discovery of this feedback illustrates one of the major trends in mitophagy: that the pathway involves positive feedback signals and amplification loops instead of a linear sequence of effects (Heo et al., 2015). This also provides one explanation for why redundancy in certain steps of the mitophagy pathway may actually be a good thing; these are likely steps that benefit from multiple overlapping amplification loops so that when one protein is absent, the whole pathway can still operate. Interestingly, a recent study reported that OPTN and ATG13 interact with mitochondria in an oscillatory nature, supporting the cyclical amplification that mitochondrial targeting and initiation involves (Zachari et al., 2019).

The functional importance of LC3 localization remains a prominent focus of the field. Reports suggest that LC3 is still targeted to mitochondria in the absence of FIP200 or ATG9A, raising the possibility that LC3 may be brought to mitochondria without the mitophagosome and serve as a marker for damaged mitochondria such as ubiquitin (Itakura et al., 2012). Reports also suggest that LC3 and other ATG8s can facilitate recruitment of initiation machinery such as the ULK1 complex themselves by serving as a scaffold (Alemu et al., 2012). Interestingly, the need for NDP52 and LC3 can be completely bypassed by artificially targeting ULK1 to the mitochondria (Vargas et al., 2019). However, although components may be dispensable under certain artificial setups, extrapolating from these results is challenging at best.

Experiments involving the overexpression of Parkin in combination with a toxin that severely damages a large proportion of the mitochondria have led to a detailed understanding of mitophagy that may not be generalizable to mitophagy following unrelated treatments. In agreement with this, a study investigating the drug ivermectin found that mitophagy was ubiquitin dependent but Parkin independent (Zachari et al., 2019). Following ivermectin treatment, TBK1 acts independently of OPTN, and FIP200 acts independently of ULK1 (Zachari et al., 2019). By studying different experimental setups, groups can uncover notably different pathways for mediating mitophagy. In addition, receptor-mediated mitophagy pathways occur independently from mitochondrial surface ubiquitination, which raises questions involving how the autophagy initiation machinery would be targeted to the mitochondria and what components would be shared with the ubiquitin pathways (Schweers et al., 2007; Zhang et al., 2008; Liu et al., 2012; Murakawa et al., 2015; Bhujabal et al., 2017). NDP52 and OPTN, which are known to detect ubiquitin, may not be involved at all or until a later step, possibly being recruited to mitochondria by LC3 that is bound to the OMM mitophagy receptors. Redundancy and overlap are evolutionarily beneficial because mitochondrial quality control is essential, so the cell must have ways to fine-tune a path, depending on the current conditions.

Clearance of mitochondria is not a self-destruction program that mitochondria complete autonomously in isolation; rather, it depends on additional organelles with the ER as the key player (Böckler and Westermann, 2014; Zachari et al., 2019). The ER provides most lipids for the autophagosome, with notable contributions from the Golgi, plasma membrane, and the mitochondrion itself (Ravikumar et al., 2010; Tooze and Yoshimori, 2010; Böckler and Westermann, 2014). ER–mitochondria contacts are referred to as “mitochondria-associated membranes,” and these sites are of significance for mitophagy for several reasons (Hamasaki et al., 2013). Upon stimulation of autophagy, DFCP1 translocates to mitochondria-associated membranes and is a marker for omegasomes (Axe et al., 2008; Itakura et al., 2012). Interestingly, PINK1-BECN1 localizes to omegasome sites as well, supporting the importance of the sites in mitophagy initiation (Gelmetti et al., 2017). Efficient calcium transfer and balance between the two organelles is needed for energy production and mitophagy initiation (MacVicar et al., 2015; Marchi et al., 2018). Last, crosstalk between mitochondrial and ER stresses has been proposed to induce Parkin expression, which can help trigger the PINK1/Parkin pathway for mitophagy, whereas certain mitochondrial stressors can also activate protein kinase R–like ER kinase (Bouman et al., 2011; Fessler et al., 2020; Guo et al., 2020).

ATG9A vesicles from the trans-Golgi network are targeted to the autophagosome formation site on damaged mitochondria independently from the ULK1 complex and are required for further recruitment of multiple ATG family members and subsequent expansion of the autophagophore (Itakura et al., 2012). Additional components that help facilitate mitophagy are endosomal Rab-GTPase family members (Yamano et al., 2014). Through ubiquitin detection, RABGEF1 engages Rab5 and Rab7A, which in turn usher ATG9 vesicles to damaged mitochondria (Yamano et al., 2018). TBK1 phosphorylation of RAB7A is important for this role in bringing ATG9 vesicles to ubiquitin-labeled mitochondria (Heo et al., 2018). TBC1D15/17 is a Rab-GTPase–activating protein that controls autophagosome biogenesis by limiting phagophore growth around the cargo by releasing Rab7A from the mitochondrial surface, ensuring precise engulfment (Yamano et al., 2014). In addition to these roles of Rabs in canonical mitophagy, the endosomal signaling pathway has been described as a program of mitochondrial clearance that is dependent on multiple Rabs while being independent of major components of the autophagy machinery, such as the ULK1 complex (Hammerling et al., 2017). In this pathway, depolarized mitochondria are still ubiquitinated by Parkin, and BECN1 is recruited and activates Rab5, which, in combination with the endosomal sorting complex required for transport machinery, engulfs mitochondria in endosomes (Hammerling et al., 2017). These endosomes later mature into Rab7 endosomes toward the end of the pathway, before lysosomal degradation (Hammerling et al., 2017).

While canonical autophagy is induced through lipid conjugation of ATG8s thanks to the concerted actions of ATGs such as ATG5 and ATG7, noncanonical autophagy and mitophagy occur independently of ATG5, ATG7, and ATG8s (Codogno et al., 2012; Saito et al., 2019). Instead, following energetic stresses such as starvation or ischemia, AMPK, ULK1, BECN1, and the trans-Golgi network work in combination with Rab9 to bring late endosome membranes to the mitochondria (Saito et al., 2019). Notably, this method of clearance is Parkin and ubiquitin independent. It is apparent that the cell has multiple different ways of inducing both canonical and noncanonical mitochondrial clearance. This review focuses on the better-characterized pathways that employ canonical autophagy machinery.

At the final step of each of these mitophagy pathways, the mitophagosome fuses with the lysosome, where the acid hydrolase enzymes can recycle the building blocks of the cell for new purposes (Wong et al., 2019). Mitophagy efficiency can be impacted by lysosomal dysfunction as well, and if the lysosomes are not functioning properly, a buildup of toxic organelles and aggregates can accumulate, which would further affect mitochondrial health and clearance (Plotegher and Duchen, 2017). Similarly, mitochondrial health has been known to affect lysosomal function, so maintaining healthy crosstalk remains another important system in the cell (Wong et al., 2019; Deus et al., 2020).

Because it is advantageous for mitochondria to be removed in different circumstances, it is expected that multiple triggers can induce mitophagy. The most investigated stimulus for inducing mitochondrial clearance is depolarization, and several mitophagy inducers accomplish this by targeting various steps of the electron transport chain and oxidative phosphorylation (OXPHOS; Palikaras et al., 2018). Mitochondria that are depolarized are not able to efficiently generate ATP, so these are removed in order for new mitochondria to be synthesized (Narendra et al., 2008). Depolarized mitochondria have difficulty importing nuclear-encoded mitochondrial proteins, which PINK1 uses to relocalize to the surface of mitochondria as a signal for removal (Narendra et al., 2010). Some of the most common toxins are carbonyl cyanide 3-chlorophenylhydrazone and carbonyl cyanide 4-(trifluoromethoxy)phenylhydrazone, which both act as protonophores and disrupt the proton gradient across the inner mitochondrial membrane (IMM; Georgakopoulos et al., 2017). Aside from the protonophores, the combination of oligomycin and antimycin A is also commonly used. Oligomycin is an antibiotic that blocks the ATP synthase proton channel, and antimycin A is an inhibitor of the respiratory chain complex III (Georgakopoulos et al., 2017). Last, valinomycin is a potassium ionophore that also uncouples ATP generation by affecting the H⁺ gradient (Georgakopoulos et al., 2017). Mitophagy following these treatments heavily relies on the PINK1–Parkin system (Georgakopoulos et al., 2017). Interestingly, depolarizing mitochondria alone may not be enough on the basis of a recent study arguing that some second signal, such as acidification of the cytosol, is needed, and these depolarizing treatments also increase acidity of the cytosol (Berezhnov et al., 2016). There are additional chemical triggers that are gaining interest in the field. As mentioned earlier, the antiparasitic drug ivermectin induces a ubiquitin-dependent, Parkin-independent mode of mitophagy, meaning other E3 ubiquitin ligases are involved (Zachari et al., 2019). The exact mechanistic details of this pathway remain to be resolved. Finally, both urolithin A and actinonin have been shown to induce mitophagy through novel mechanisms (Ryu et al., 2016; Fang et al., 2019b). Despite the focus on these triggers, recent work has begun questioning whether the severe mitochondrial damage these treatments induce is relevant to the quality and quantity of stimuli that may occur within the cell or organism physiologically (McWilliams et al., 2018b; Lee et al., 2018).

Hypoxia serves as a physiologically relevant trigger for mitophagy, because in low-oxygen conditions, the mitochondria cannot complete OXPHOS (Zhang et al., 2008). Hypoxic environments stabilize and activate the master regulator transcription factor hypoxia-inducible factor-1α (HIF1α), which drives the transcriptional upregulation of two mitophagy receptors, BNIP3 and BNIP3L (NIX), whereas a third mitophagy receptor, FUNDC1, becomes activated by dephosphorylation (Sowter et al., 2001; Zhang et al., 2008; Liu et al., 2012). Mitophagy following hypoxia is proposed to be independent of PINK1 stability (Wei et al., 2015). AMPK is a pivotal sensor of cellular metabolism and energetics that closely monitors the levels of ATP and ADP/AMP (Herzig and Shaw, 2018). Hypoxia also activates AMPK through the resulting oxidative stress (Mungai et al., 2011). Activated AMPK promotes mitophagy by activating ULK1 and the downstream initiation machinery (Kim et al., 2011; Laker et al., 2017). Another physiological stimulus closely related to hypoxia is iron chelation (Allen et al., 2013). Mitochondrial respiration and energy production rely on iron-sulfur clusters, and iron chelation also leads to HIF1α stabilization (Allen et al., 2013). Iron chelation by deferiprone drives the removal of mitochondria without depolarization and is independent from the PINK1/Parkin pathway (Allen et al., 2013).

One category of mitophagy that occurs during differentiation of RBCs and retinal ganglion cells is unique and distinct because it does not cause damage and is not a stressor (Schweers et al., 2007; Sandoval et al., 2008; Esteban-Martínez et al., 2017). In both types of cells, there is dramatic removal of mitochondria through NIX-dependent mitophagy, and deficiencies in mitophagy at these developmental milestones result in diseases such as anemia (Schweers et al., 2007; Esteban-Martínez et al., 2017). Another notably distinct trigger for mitophagy is Listeria infection through the toxin listeriolysin O (Zhang et al., 2019). Following infection, mitophagy promotes Listeria propagation through the reduction of ROS, which occurs through the Nod-like receptor NLRX1 independently from the PINK1/Parkin pathway, but many details of the mechanism remain to be explored further (Zhang et al., 2019). Another emerging mitophagy trigger is high glucose (HG) levels (Devi et al., 2017; Alcántar-Fernández et al., 2019). Although glucose is normally thought to prevent nonselective autophagy, levels at the high extreme can trigger removal of mitochondria. TBK1 inhibitor blocks this pathway, suggesting involvement of TBK1 substrates such as OPTN and NDP52 (Devi et al., 2019). TXNIP is involved in triggering mitophagy in HG conditions through increasing oxidative stress and AMPK activation rather than being involved in the targeting of mitochondria (Devi et al., 2017, 2019). Oxidative stress is a necessary signal because N-acetylcysteine can prevent HG-induced mitophagy (Devi et al., 2019). An alternative hypothesis is that HG supplementation shifts a cell’s energy production away from OXPHOS to glycolytic metabolism, and therefore large mitochondrial content is not required (Doménech et al., 2015; Esteban-Martínez et al., 2017). Interestingly, mitochondrial dysfunction following most of the treatments listed above will invariably increase oxidative stress, suggesting that ROS could be the common stimulus for certain components of the pathway, while also amplifying mitochondrial dysfunction and the degradative signals (Zhang et al., 2008; Devi et al., 2019). Accumulation of misfolded proteins in the mitochondria due to the mitochondrial Hsp90 inhibitor Gamitrinib-triphenylphosphonium or overexpression of a mutant form of ornithine transcarbamylase also leads to mitophagy (Burman et al., 2017; Fiesel et al., 2017). Because these triggers induced PINK1-dependent mitophagy without membrane depolarization, it is likely that the accumulation of misfolded proteins alone was sufficient to prevent proper import of PINK1 and possibly other mitochondrial proteins (Burman et al., 2017; Fiesel et al., 2017).

The last mitophagy trigger we discuss is NAD⁺. Recently, NAD⁺ regulation has become a physiological target that is gaining interest in increasing mitochondrial function, biogenesis, and clearance (Jang et al., 2012; Fang et al., 2016). NAD⁺ supplementation induces mitophagy mainly through Sirtuin-dependent pathways (Aman et al., 2020). Sirtuins are a class of signaling proteins that regulate many aspects of cellular metabolism, including mitochondrial biogenesis and mitophagy, and rely on NAD⁺ levels to function (Aman et al., 2020). SIRT1 is a deacetylase that can modify multiple ATGs that are needed for LC3 lipidation while also affecting multiple targets downstream of AMPK, such as ULK1, PGC1α, FOXO1, and FOXO3a (Cantó et al., 2009; Fang et al., 2019a). Aside from SIRT1, SIRT2 can modify ATG5 acetylation, whereas SIRT3 affects the FUNDC1 mitophagy receptor pathway (Liu et al., 2017; Fang, 2019).

Aside from exogenous molecules or triggers to induce mitophagy, low levels of background mitophagy maintain the mitochondrial population at optimum efficiency (McWilliams et al., 2018b; Lee et al., 2018). This basal, physiological mitophagy likely happens with a local signal when only a small percentage of mitochondria are being turned over to maintain the health of the whole, rather than engaging in global depolarization. However, the cells must still receive some form of signal for clearance of these mitochondria, possibly a localized signature of oxidative stress or altered metabolite levels. Interestingly, although PINK1 and Parkin are important for basal turnover of mitochondrial proteins, PINK1- and Parkin-knockout flies and mice do not display any defect in basal mitophagy, which is especially high in metabolically active tissue such as the heart and muscle (Vincow et al., 2013; McWilliams et al., 2018b; Lee et al., 2018). On the contrary, the adenine nucleotide translocator (ANT) complex is involved in basal mitophagy, with genetic removal of ANT leading to accumulation of damaged mitochondria (Hoshino et al., 2019). This mechanism involves the ability of ANT to restrict mitochondrial protein import through TIM23 by directly interacting with TIM44 (Hoshino et al., 2019). Cells lacking ANT are unable to stabilize PINK1 following depolarization, but because PINK1 is not needed for basal mitophagy, this suggests that the retention of mitochondrial proteins in the cytosol may be a common mechanism multiple proteins use to induce mitophagy (Hoshino et al., 2019). Aside from this, other autophagy components are important for basal clearance, such as ATG5, as well as the mitophagy receptor BNIP3, whose deficiency leads to accumulation of defective mitochondria in mammary tumor cells and the liver (Tal et al., 2009; Chourasia et al., 2015b; Glick et al., 2012).

PINK1/Parkin-driven mitophagy is the most characterized pathway, and the mitophagy field has been built on investigation of these two proteins (Park et al., 2006; Narendra et al., 2008, 2010). Recent reviews go into depth and detail on this path, so we will just touch on high-level points (Pickles et al., 2018; Montava-Garriga and Ganley, 2020). Under normal mitochondrial conditions, PINK1 is imported into the mitochondria, where it is exposed to and cleaved by mitochondrial proteases MPP and PARL (Whitworth et al., 2008; Deas et al., 2011a; Greene et al., 2012). Upon mitochondrial depolarization or accumulation of misfolded mitochondrial proteins, PINK1 import is prevented, and, as a result, PINK1 is stabilized on the surface of the mitochondria (Fig. 2; Narendra et al., 2010; Jin and Youle, 2013; Burman et al., 2017; Fiesel et al., 2017). Through its kinase activity, it phosphorylates the ubiquitin attached to several OMM proteins in addition to the cytosolic E3 ligase Parkin (Kane et al., 2014; Koyano et al., 2014; Shiba-Fukushima et al., 2014; Zhuang et al., 2016). Parkin is then recruited to these mitochondria and ubiquitinates multiple surface proteins as well, some of which will serve as a signal for autophagy receptors such as NDP52 and OPTN, whereas others will be targeted by the proteasome that degrades components of the OMM, which is essential for mitophagy (Vives-Bauza et al., 2010; Ziviani et al., 2010; Lazarou et al., 2015; Tanaka et al., 2010; Yoshii et al., 2011). This pathway contains an amplification loop whereby phosphorylation encourages Parkin recruitment and ubiquitination, resulting in more phosphorylation, sequentially building up the signal for degradation on the surface of mitochondria (Heo et al., 2015). As mentioned earlier, Parkin-mediated ubiquitination of mitochondria seems to be important for an endosomal Rab5-dependent path of clearance that operates independently of the canonical autophagy machinery (Hammerling et al., 2017).

Although the importance of this pathway is indisputable for in vitro assays that trigger mitophagy using depolarization, the role of PINK1 and Parkin in vivo has been more difficult to ascertain (McWilliams et al., 2018b; Lee et al., 2018). Mice lacking PINK1 or Parkin do not spontaneously develop a phenotype, and levels of basal mitophagy in metabolically active tissues such as the heart and the brain are not affected by the loss of either of these proteins (McWilliams et al., 2018b). Similarly, platelets from PINK1-deficient mice function normally and do not exhibit any defect in basal mitophagy (Walsh et al., 2018). These knockout animals require another hit such as the mtDNA mutator mouse background that induces mtDNA mutations due to deficiency in proofreading of the mtDNA polymerase POLG, exhaustive exercise, or aging before seeing a phenotype (Sliter et al., 2018; Noda et al., 2020). Because PINK1-Parkin mitophagy is triggered following severe mitochondrial stress, it is reasonable that some of the numerous other mitophagy pathways are important in maintaining basal levels of mitophagy when the stress is milder. Alternatively, due to the high levels of redundancy, perhaps loss of any one mitophagy pathway does not result in an overwhelming phenotype, where the others can compensate to accomplish the low levels of mitophagy needed to maintain balance. Deubiquitinating enzymes such as USP30 and USP35 are constitutively active and dampen the ubiquitin-dependent pathway, allowing the fine-tuning of the ubiquitin signal (Bingol et al., 2014; Wang et al., 2015; Marcassa et al., 2018). It makes sense, then, to have a ubiquitin-independent group of mitophagy receptors that can be activated and override this dampening through other means.

Mitophagy receptors are typically mitochondrial proteins that contain an LIR motif that enables the recruitment of LC3 and the growing mitophagophore to the mitochondria designated for removal (Wang et al., 2019b). In addition to possessing this motif, these receptors are usually embedded in the OMM via a transmembrane domain, except for a few of the less characterized receptors (Hamacher-Brady and Brady, 2016). As mentioned earlier, the mitophagy receptor NIX mediates mitophagy during RBC differentiation, as well as the hypoxia-driven glycolytic switch during metabolic transitions associated with retinal ganglion cell development (Schweers et al., 2007; Sandoval et al., 2008; Esteban-Martínez et al., 2017). NIX activity is enhanced through phosphorylation and dimerization and is transcriptionally regulated by HIF1α stabilization, supporting its role in hypoxia-mediated mitophagy (Sowter et al., 2001; Gustafsson, 2011). Closely related to NIX is BNIP3, which is similarly regulated transcriptionally by HIF1α, and also mediates mitophagy following hypoxia (Sowter et al., 2001; Zhang et al., 2008). The affinity that BNIP3 has for LC3 is regulated by phosphorylation within the LIR motif (Gustafsson, 2011). In addition to hypoxia, BNIP3 contributes to PINK1-Parkin mitophagy through multiple steps, including stabilizing PINK1 on the OMM, aiding in translocation of DRP1, and freeing BECN1 by preventing the interaction with BCL-2 (Zhang et al., 2008, 2016; Lee et al., 2011). The final mitophagy receptor induced by hypoxia is FUNDC1, which, instead of being transcriptionally regulated, is regulated by its own phosphorylation within the LIR motif (Wu et al., 2014). FUNDC1 has also been implicated in depolarization-induced mitophagy by maintaining the ER–mitochondria contact sites through interaction with IP3R2 and is regulated through direct phosphorylation by ULK1 (Wu et al., 2014, 2017). Recently, the critical function of FUNDC1 in exercise-induced mitophagy within skeletal muscle has been discovered and will be discussed in detail later (Fu et al., 2018).

Additional, less studied receptors have also been tied to mitophagy. Bcl2L13 is another OMM protein reported to regulate mitophagy by binding LC3 in addition to mediating mitochondrial fission (Murakawa et al., 2015). FKBP8 is another mitophagy receptor with a binding preference for LC3A that mediates mitophagy as well as fission, independently from Parkin (Bhujabal et al., 2017). Aside from the OMM proteins, some mitophagy receptors exist in other mitochondrial locations. Prohibitin 2 (PHB2) is an IMM protein that is unique not only because of its location but also because the proteasome-driven OMM rupture involved in PINK1-Parkin mitophagy is required for its exposure and activity (Wei et al., 2017). PHB2 regulates PINK1 stability on mitochondria in addition to binding LC3 upon OMM rupture and is required for mitophagy following depolarization (Wei et al., 2017; Yan et al., 2020). Another mitophagy receptor that resides within the mitochondria is NLRX1, a NOD-like receptor that is localized within the mitochondrial matrix and contains an LIR motif (Zhang et al., 2019). Recently, infection with Listeria was shown to promote a PINK1-independent mitophagy program through NLRX1 (Zhang et al., 2019). The direct mechanism remains to be determined, including an explanation of how a mitochondrial matrix protein could be interacting with LC3 from its location. The last mitophagy receptor we will mention is cardiolipin, a unique phospholipid that is present in the IMM (Chu et al., 2013). Upon mitochondrial damage, cardiolipin relocalizes to the OMM, where it can interact with LC3 and is potentially involved in PINK1-Parkin mitophagy (Chu et al., 2013).

When compared with the PINK1/Parkin pathway that has been studied for years, these mitophagy receptor pathways appear relatively new and underexplored. Apart from clarifying the location, the trigger for inducing mitophagy and the possession of an LIR motif, these pathways lack mechanistic exploration, and questions remain. Is this mitophagy independent of NDP52, OPTN, and TBK1, which are indispensable for ubiquitin-driven mitophagy, or is there some crosstalk and ability to recruit these components as well, possibly through the LC3-based targeting of NDP52 and OPTN because the ubiquitin label may not be present on the surface (Padman et al., 2019)? Ubiquitin has been reported to drive mitochondrial localization of ULK1 and LC3 (Lazarou et al., 2015; Padman et al., 2019). The state of being Parkin independent does not necessitate that ubiquitin is absent, because other E3 ligases have been reported, such as Gp78, ARIH1, CIAP1/2, and TRAF2 (Fu et al., 2013; Villa et al., 2017; Zachari et al., 2019). Mitophagy receptor pathways contain a ubiquitin-independent “eat me” signal in the form of a receptor that can bind LC3, but there is a possibility of ubiquitin-dependent recruitment of other components for autophagosome synthesis.

With the new paradigm of mitophagosomes forming on mitochondria, another question involves how ULK1 may be recruited without ubiquitin bringing NDP52 to mitochondria. AMPK phosphorylation and activation of ULK1 may lead to ULK1 localization on mitochondria independently of NDP52 following certain stimuli (Tian et al., 2015). Mitophagy receptors such as FUNDC1 and Bcl2L13 can bind and localize ULK1 to the mitochondria directly (Wu et al., 2014; Murakawa et al., 2019). In addition, the LC3 on the surface of mitochondria through interaction with the mitophagy receptors is the ideal replacement label for damaged mitochondria instead of ubiquitin. It can further drive NDP52/OPTN recruitment along with other ATG members (Padman et al., 2019). As mentioned earlier, the ATG8 family has also been proposed to serve as a scaffold for mobilizing the ULK1 complex, with many of the family members having the ability to bind ULK1, ULK2, FIP200, and ATG13 (Alemu et al., 2012). This would allow signal amplification and positive feedback because more LC3 labeling would allow more phagophore synthesis and additional LC3. These ideas are currently speculation, and there is a need for detailed mechanistic investigations into these mitophagy receptor pathways and triggers.

We end this discussion of how mitophagy is occurring by asking if these are truly separate and isolated pathways or if it is wrong to think of PINK1/Parkin-driven versus independent pathways. It is possible that many mitophagy receptors may be involved in the PINK1/Parkin pathway, because they are OMM proteins with the ability to bind LC3, so they could be involved in LC3 amplification steps due to their presence alone. Maybe PINK1-Parkin and depolarization-induced mitophagy uses these mitophagy receptors, whereas the sheer number and redundancy of them means that not all are necessary.

In the beginning of the present review, we touched on why mitophagy is important for any cell that contains mitochondria. When moving from this idea of a nondescript eukaryotic cell to differentiated, specific cells that make up our body’s organs, certain tissues become focal points for discussion. These include muscle cells and neurons because of their specific functions, metabolism, and energy requirements (McWilliams et al., 2018b). These cells make up organs with high energy consumption and vulnerability, and slight perturbations in homeostasis can lead to pronounced effects (Baker et al., 2010; van der Kooij et al., 2018; Muchlinski et al., 2018). Although we focus on these two cell types, mitophagy is important in all organs where mitochondria play important roles. Multiple recent reviews explore the role of mitophagy in cardiovascular, liver, metabolic, immune, and inflammatory diseases and cancer (Chourasia et al., 2015a; Zhao et al., 2018; Ke, 2020; Liu et al., 2020; Morciano et al., 2020; Xu et al., 2020).

The heart is a highly energetic organ that generates ATP mainly through OXPHOS (Kolwicz et al., 2013). Cardiomyocytes are known to consume large amounts of fatty acids through β-oxidation to maintain contraction and blood circulation (Kolwicz et al., 2013). Interestingly, mitochondria within cardiomyocytes are more fragmented than in other cells, and turnover of mitochondrial proteins is less rapid than cytosolic proteins (Saito et al., 2019). There is a notable age-related reduction in mitophagy that is believed to contribute to the fibrotic nature of aged cardiac tissue through accumulated oxidative stress and misfolded proteins (Liang and Gustafsson, 2020). Following ischemia reperfusion in cardiomyocytes, both Parkin- and FUNDC1-mediated mitophagy are protective (Fig. 3; Kubli et al., 2013; Zhang et al., 2017). Aside from canonical mitophagy, the Rab9 alternative mitophagy pathway also plays a role in cardiomyocytes, an example of multiple redundant pathways functioning in parallel (Saito et al., 2019). An alternative trigger for mitophagy in the heart involves acute exercise, with LC3 lipidation increasing in cardiomyocytes after exercise (Ogura et al., 2011). Cardiac mitophagy under these conditions is likely beneficial for endurance capacity through optimized oxygen use during the stressful condition.

The other tissue overtly activated during exercise-mediated stress is skeletal muscle (Fig. 3; Laker et al., 2017; Fu et al., 2018). FUNDC1-driven mitophagy in skeletal muscle greatly affected the endurance capacity of mice (Fu et al., 2018). This study further showed that skeletal muscle–localized mitophagy defects affect whole-body metabolism, with substantial effects on adipose tissue and fat metabolism (Fu et al., 2018). BNIP3 expression is enhanced in exercised muscle, suggesting that it has a role in mitophagy (Lira et al., 2013). In agreement with the importance of autophagy machinery in exercise, mice deficient in exercise-induced autophagy throughout the entire body, named BCL2 AAA mice, as well as whole-body BECN1 heterozygotes had a reduced endurance capacity (He et al., 2012). Exercise-induced mitophagy in skeletal muscle was shown to be ULK1 dependent, placing ULK1 activation downstream of AMPK (Laker et al., 2017). Interestingly, skeletal muscle–specific AMPK–dominant negative mice did not display defective endurance capacity, and mice with a ULK1 deficiency in skeletal muscle similarly did not have reduced endurance capacity (Laker et al., 2017). Similarly, skeletal muscle–specific knockout of ATG7 before exercise did not affect the endurance capacity of mice (Lo Verso et al., 2014). Taking the skeletal muscle–specific knockout data together with the whole-body data, the mice that consistently show a reduced endurance capacity lack mitophagy in more organs than just the skeletal muscle, which suggests that the cardiac muscle would be the most reasonable location of local mitophagy importance.

After three consecutive days of exhaustive exercise, PINK1/Parkin-dependent mitophagy is induced, preventing systemic inflammation and the resulting motor deficits seen in neurodegenerative disease (Sliter et al., 2018). Parkin was also reported to drive mitophagy following a much shorter, acute period of exercise; however, these data conflict with those of another study that did not see PINK1 stabilization following exercise (Chen et al., 2018; Drake et al., 2019). None of the studies noted any endurance capacity defect in animals lacking PINK1-Parkin. Because loss of FUNDC1-driven mitophagy in skeletal muscles was enough to decrease the endurance capacity of mice, perhaps this suggests the difference in experimental setup including an exercise regimen, or perhaps the method of quantifying endurance capacity is responsible for these differing results. A separate explanation is that the altered fatty acid metabolism seen in FUNDC1-deficient mice is independent of its mitophagy defect because it is not recapitulated when other proteins essential to mitophagy are absent in the muscle.

The deep molecular understanding of mitophagy we have stems from the thorough investigation into PINK1 and Parkin, which are both major recessive risk factors for developing early onset Parkinson’s disease (PD; Pickrell and Youle, 2015). Neurons are highly specialized cells built to generate and propagate action potentials and are dependent on mitochondria for several functions (Kann and Kovács, 2007). The unique structure of the neuron creates an environment where not only does the mitochondrial pool have to be healthy, but it also must be properly transported down the axon to quite distant sites where ATP production and calcium buffering are its two most important functions (Kann and Kovács, 2007). An aged nervous system coupled with a decline in mitophagy leads to accumulation of bad mitochondria and is a hallmark of neurodegeneration (Fivenson et al., 2017).

The second most common neurodegenerative disease, PD, is a motor neuron disease arising from the progressive loss of dopaminergic neurons in the substantia nigra pars compacta and characterized by bradykinesia, tremors, rigidity, and postural instability (Poewe et al., 2017). Genetic risk factors of PINK1, Parkin, DJ-1, and others led to a proposed disease mechanism of mitochondrial dysfunction (Poewe et al., 2017; Hsieh et al., 2016). Mitophagy is defective in PD tissue and in models that recapitulate the disease (Deas et al., 2011b; Hsieh et al., 2016; Gao et al., 2017). However, a question remains whether defective mitophagy is the cause of PD or if the cause is mitochondrial dysfunction more generally or even a completely other defective process that results in mitochondrial dysfunction (Chen et al., 2019). A recent study showed that the neurotoxin oxidopamine induces mitophagy in various neurons but not in dopaminergic neurons, suggesting that dopaminergic neurons may not induce mitophagy as readily as others, which can explain why they are specifically vulnerable to underlying stress (Katayama et al., 2020). Inducing PINK1-Parkin mitophagy removes the damaged mitochondria, but this does not mean that defective clearance of the mitochondria was the cause of the disease; rather, it means that the buildup of toxic mitochondria leads to cell death (Chen et al., 2019). Interestingly, mice lacking PINK1 or Parkin do not develop spontaneous PD (Perez and Palmiter, 2005; McWilliams et al., 2018b). On the contrary, rats lacking PINK1 accumulate PD neurodegeneration and behavioral deficiencies, although rats lacking Parkin did not show any abnormalities, which supports a Parkin- and mitophagy-independent role of PINK1 driving the phenotype (Dave et al., 2014). In Parkin-knockin mice harboring a Ser65Ala (S65A) mutation, phosphorylation-based activation of Parkin by PINK1 is ablated. Although later developing motor deficiencies, these mice do not show neurodegeneration or defective mitophagy (McWilliams et al., 2018a). Last, the importance of Parkin was also demonstrated in mice aged 110 wk, in which Parkin-knockout mice showed motor deficiencies, neuron loss, and mitochondrial abnormalities, suggesting that the second hit of aging is needed before a PD phenotype is noticeable (Noda et al., 2020). Having learned more about the various functions of PINK1 and Parkin, we move from a model with defective mitophagy as the sole cause of PD to a more complex understanding in which multiple mitochondrial functions are disturbed in a neuronal cell type that is less efficient at clearing out mitochondria. However, disrupted mitochondrial function remains a commonality in PD, and various dangerous processes stem from it.

Mitochondrial dyshomeostasis can stem from multiple processes, including decreased mitochondrial biogenesis, misfolded protein stress, deficient OXPHOS, α-synuclein (α-syn) aggregation and toxicity, ROS accumulation, and iron dyshomeostasis (Chen et al., 2019). Although these have the potential to impact mitochondrial function, they are not caused exclusively by defective mitophagy. PINK1 and Parkin may affect these other processes because they have been implicated in the mitochondrial unfolded protein response (UPR^mt), clearance of the toxic protein PARIS whose buildup negatively regulates PGC1α levels, mitochondrial dynamics, transport, and biogenesis (Jin and Youle, 2013; Lee et al., 2017; Yu et al., 2011). Alternatively, deficiency in PINK1 leads to an exacerbated immune response to mitochondrial antigens and was recently shown to lead to PD symptoms following bacterial infection (Matheoud et al., 2016, 2019). This new direction suggests an underlying role for PINK1 and Parkin in modifying the immune system that may explain the genetic susceptibility seen in mutation-bearing patients and offers an interesting avenue for future research. Although Lewy bodies made up of aggregated α-syn remain the prominent pathology associated with PD, α-syn not only slows mitophagy but also impacts mitochondrial health in additional ways (Vicario et al., 2018; Shaltouki et al., 2018; Wang et al., 2019a). These parallel processes eventually overwhelm the quality control system and are seen as defective mitophagy (Chen et al., 2019). In summary, there is a circle of events underlying PD, and mitochondrial dysfunction is part of that. Abrogated mitophagy causes mitochondrial dysfunction, making it a contributor to PD, but it is not the sole cause of PD; therefore, it may not be a sufficient treatment.

Until recently, surprisingly little evidence directly linked mitophagy to the most common neurodegenerative disease, Alzheimer’s disease (AD; Kerr et al., 2017; Chakravorty et al., 2019; Fang et al., 2019b). AD is the major cause of dementia and involves loss of the cholinergic neurons in the brain, with a notable focus on the hippocampal region (Chen and Mobley, 2019). Historically, PD etiology was more focused on defective mitophagy, whereas investigation of AD has focused on accumulation of amyloid-β (Aβ) plaques and phospho-tau neurofibrillary tangles (Kerr et al., 2017). Recent studies have shown that mitophagy is, in fact, affected in AD and, more important, that inducing mitophagy could benefit the pathological and cognitive outcomes (Fang et al., 2019b). Tissues from postmortem patient samples and patient-derived induced pluripotent stem cells display defective mitophagy, contributing to the energetic stress and mitochondrial dysfunction that have been characterized in AD. Models of toxic Aβ and tau were shown to impair mitophagy, and increasing mitophagy helped to reduce the plaque and neurofibrillary tangle burden in Caenorhabditis elegans and mouse models (Fang et al., 2019b). Mitophagy was important not only in neurons but also in microglia that showed improved phagocytic function for clearing out Aβ plaques (Fang et al., 2019b).

AD might also be targeted through modification of NAD⁺ levels, which were low in AD models (Martire et al., 2016; Hou et al., 2018). NAD⁺ supplementation or addition of the precursor nicotinamide mononucleotide was able to induce mitophagy (Fang et al., 2019b). NAD⁺ has also been linked to other pathways of axonal degeneration, which makes it an interesting treatment that could increase mitophagy and abrogate axonal degeneration–related cell death together (Figley and DiAntonio, 2020). Like in PD, AD pathogenesis is likely a cycle and feedback loop whereby Aβ and phosphorylated tau lead to mitochondrial dysfunction, which can further exacerbate the accumulation of protein aggregates, and modifying mitophagy is not the sole cause or solution. Once one part of the cycle is broken through targeted treatment, perhaps the other becomes more vulnerable as well, lending itself to be more amenable to additional therapy.

As with most therapeutics that control a biological process, increasing levels of mitophagy must be carefully controlled because passing an upper limit would induce cell death, so careful modulation rather than constitutive activation would be ideal for this style of treatment. The diseases discussed earlier involve other contributors and steps in a disease amplification cycle which implies that controlling mitophagy may help reduce disease burden but may not be a cure for the underlying cause. When starting to discuss what compounds would be suitable for mitophagy-based treatment, it is obvious that compounds such as carbonyl cyanide 3-chlorophenylhydrazone are toxic and will not be usable due to their widespread off-target effects (Ashrafi et al., 2014). One strategy for a PD treatment involves upregulating the transcriptional coactivator PGC1α to induce mitochondrial biogenesis (Corona and Duchen, 2015). However, PGC1α modifies many different paths, so it will not be exclusively selective for mitochondrial control (Corona and Duchen, 2015). This setback is similar to other upstream components, such as AMPK or the ULK1 complex (Egan et al., 2015; Day et al., 2017).

A promising alternative involves investigating induced pluripotent stem cells from patients who have mutations commonly linked with diseases but are only carriers to see how they are compensating, potentially through other pathways that are being increased to maintain homeostasis (Chang et al., 2020; Penney et al., 2020). Similarly, in patients in whom PINK1/Parkin are mutated and deficient mitophagy is suspected to be a contributor to disease, treatment could involve inducing the other parallel mitophagy receptor pathways. This approach would be used when supplementing or inducing the PINK1/Parkin pathway would not be possible.

Last, by having a deeper understanding of basal mitophagy and its physiological trigger, we will gain insight into a readily tolerable treatment, one that does not irreparably damage the remaining mitochondrial population. Physiologically relevant stimulation through NAD⁺ supplementation has been effective in mouse and C. elegans studies in AD (Fang et al., 2019b). By supplementing with a molecule that is already present in the body, the safety concerns are greatly reduced. Alternatively, by removing the brakes on the mitophagy system, such as the deubiquitinating enzymes, we would also increase levels of basal mitophagy (Bingol et al., 2014; Marcassa et al., 2018). Regardless of the treatment approach, the ideal therapy will be targeted to the dysfunctional organ because affecting the balance of mitophagy in off-target organs that do not have mitochondrial dysfunction will create additional problems.

Investigation of the molecular players involved in mitophagy is a rapidly advancing field, not only due to the layers of complexity and interest in the discovery but also because mitochondrial dysfunction is at the foundation of numerous diseases. Major advancements in any research field are usually associated with paradigm shifts in the way we understand a given pathway, and mitophagy is no different. By renewing our understanding of mitophagosome initiation, what localizes the initiation machinery to a designated mitochondrion, we gain insight into the triggers that the cell uses to remove the organelles. Through a new understanding of LC3 and its crucial role not only in linking cargo to the autophagosome but also in aiding the recruitment of the autophagy initiation machinery as well as ensuring phagosome maturation and elongation, the importance of LIR motifs and targeting LC3 is increased. By understanding the need for redundant mitophagy pathways, we start to question whether they truly operate in isolation or if they contribute to a cyclic amplification process whereby they are all important in their own way. Last, with a new appreciation for basal mitophagy that is likely independent of depolarization and the triggers usually associated with mitophagy, we begin to investigate how mitochondrial homeostasis is maintained at baseline, and we gain new pathways and triggers that could be exploited for their therapeutic potential.

This work was supported by the Canadian Institutes of Health Research (operating grant 000303157).

The authors declare no competing financial interests.

Author contributions: S.A. Killackey, D.J. Philpott, and S.E. Girardin designed the manuscript’s structure and organization. S.A. Killackey wrote the manuscript, which was edited by D.J. Philpott and S.E. Girardin.