Mitochondria-associated membranes as hubs for neurodegeneration

Although most membrane lipid synthesis in eukaryotic cells occurs in the ER, several key metabolic steps and lipid modifications are performed in other organelles, and thus, require transport of lipids. How hydrophobic lipids are distributed within the aqueous environment of the cell towards other organelles remains largely elusive, but recently the role of non-vesicular transport at membrane contact sites has gained a lot of interest [61, 109]. Close membrane apposition at these sites appears to allow efficient shuttling of lipids. In addition, several lipid metabolic enzymes are exclusively localized at certain MCSs, restricting the generation of these lipid species to the site where they are needed. MAMs were originally identified as a fraction highly enriched in phosphatidylserine (PS) synthase-1 and -2 (Pss1 and Pss2) and transfer of phospholipids between the ER and mitochondria was the first function ascribed to ER-mitochondria contacts [109]. Phosphatidylserine, newly synthesized in the MAM by the PS synthases, is subsequently transferred to the closely apposed mitochondrion (Fig. 3). In the inner mitochondrial membrane (IMM), decarboxylation of PS results in the production of phosphatidylethanolamine (PE). The importance of ER-mitochondria MCSs is underlined by the fact that the transfer of PS between both organelles is the rate-limiting step in PE synthesis from PS [112]. In a crude mitochondrial fraction derived from rat liver, newly synthesized PE can then be transferred back to the MAM where it is converted into phosphatidylcholine (PC) by the enzyme PE-N-methyltransferase (PEMT) [110]. Since PEMT activity is restricted to hepatocytes, it is not clear how essential MAMs are for PC synthesis in other tissues.

Protein identification approaches revealed that not only Pss1 and Pss2 are enriched at the MAMs, but also several other lipid synthesizing enzymes involved in cholesterol and sphingolipid biosynthesis (Fig. 3) [39, 45, 109]. This fits well with thin layer chromatography studies showing that MAMs comprise a unique lipid profile, with highly elevated levels of ceramides and cholesterol compared to the bulk ER [45]. While it is yet unclear whether the MAMs participate in trafficking of these lipids to mitochondria, this particular lipid raft-like composition plays a determining factor in the recruitment of several well-established MAM components. Both presenilin 1 and 2 (PS1 and PS2), members of the γ-secretase complex are targeted to the MAMs because of their lipid raft composition [106]. Similarly, the chaperone Sigma non-opioid intracellular receptor 1 (Sigma1R), associated with AD, ALS and FTD is strictly dependent on the presence of cholesterol and ceramide for targeting to the MAMs [45] and for performing its essential functions in calcium regulation (see Table 1; Fig. 4).

Already in the 19th century, the role of calcium as second messenger was established through seminal experiments on perfusion of isolated hearts. Since then, a lot of research has been devoted to understand how increases in calcium levels (so-called oscillations) are generated, decoded and translated in signaling events. Two major sources count for the rise in cytosolic calcium: the extracellular medium with a [Ca²⁺] of ~1 mM and the ER with a [Ca²⁺] >100 μM [87]. Mitochondria play a crucial role in buffering increases in cytosolic calcium, and a rise in cytosolic [Ca²⁺]_cyt is always closely followed by a rise in [Ca²⁺]_mito [89]. Interestingly, upon ER calcium release, mitochondrial levels can rapidly increase 10–100-fold without noticeable rises of cytosolic calcium levels [87]. In contrast to the outer mitochondrial membrane (OMM), which is permeable to calcium ions, influx of calcium into the mitochondrial matrix—driven by the electrical gradient across the IMM—occurs through the recently identified mitochondrial calcium uniporter (MCU) complex. This complex consists of the MCU protein and its regulators (MICU1, MICU2, MCUb, MCUR1, and EMRE) [72]. Paradoxically, the low calcium affinity of the MCU appears incompatible with the observed rapid calcium uptake in mitochondria at low cytosolic calcium levels. This discrepancy was solved by the observation that mitochondria are in close contact with the IP3Rs and ryanodine sensitive channels (RyRs) mediating ER store release [88]. This occurs at the MAMs and leads to the transient formation of “hotspots”, perimitochondrial microdomains in which calcium levels far exceed those in the bulk of the cytosol, enabling rapid import across the IMM [27, 40, 88].

ER-mitochondria MCSs are able to establish such microdomains required for efficient calcium transfer through (1) the close apposition of both membranes, (2) the enrichment of the calcium release channels IP3Rs and RyRs at MAMs and (3) the connection of IP3Rs to the mitochondrial VDAC1 by the chaperone Grp75 (Fig. 4). Knockdown of Grp75 abolishes functional coupling between VDAC1 and IP3R, resulting in diminished calcium transfer [100]. On the other hand, bringing the ER closer to mitochondria through an artificial linker stimulates calcium transfer [26]. Numerous studies have correlated altered ER-mitochondrial connectivity to alterations in calcium transfer between both organelles, through overexpression or knockdown of proposed tethers, including MFN2 and VAPB-PTPIP51, or regulators thereof [28, 29, 34, 97] (Fig. 4). Several modulators of IP3R channel activity are present at or recruited to MAMs under various conditions. For example, various ER-resident chaperones and oxidoreductases can affect the exchange of calcium between the ER and mitochondria [96]. Calnexin (CNX) modulates local calcium levels by acting as a calcium buffer and through regulating the activity of IP3R and the sarco/endoplasmic reticulum calcium ATPase (Serca), the main ATPase pumping cytosolic calcium into the ER [52, 68, 90]. CNX targeting to the MAM is dependent on palmitoylation [68], and on the cytosolic sorting protein PACS2 [79] and Rab32 [16] two trafficking molecules that both regulate the composition of the MAMs (Fig. 4). The ER-resident Sigma1R is released from Bip upon ER calcium release, and binds to IP3Rs at the MAMs, in this way preventing their proteasomal degradation and sustaining prolonged Ca²⁺ uptake by mitochondria [46] (Fig. 4). Especially in neurons, both Sigma1R and PACS2 play a crucial role in mediating calcium homeostasis and loss of Sigma1R results in neuromuscular defects [47]. This is underscored by the fact that alterations in PACS2 and Sigma1R levels are associated with the pathogenesis of AD, ALS and FTD (see Table 1; Boxes 1, 2). In addition, presenilin mutations in AD are associated with an increased IP3R calcium release (see Table 1; Box 1). Finally, in Huntington’s disease ER-mitochondrial calcium transfer appears affected as well. Mutant huntingtin, but not the wild type protein, was found to interact with IP3R and facilitate calcium transfer between the ER and mitochondria [12, 83].

Calcium transfer to mitochondria occurs under both physiological and stress conditions. A moderate mitochondrial matrix calcium increase stimulates ATP synthesis through calcium dependency of metabolic enzymes in the Krebs cycle [87]. Functional coupling of the ER and mitochondria therefore stimulates aerobic metabolism, in response to a demand of ATP-requiring processes in the cytosol as well as in the ER [96, 111]. Hence, loss of contact has adverse effects on ER homeostasis, protein folding and energy metabolism [14], and will affect cell survival and proliferation, especially in cells with high-energy demands such as neurons. Mitochondrial calcium overload on the other hand is a potent inducer of the mitochondrial permeability transition pore and can thereby lead to OMM rupture and the escape of pro-apoptotic factors such as cytochrome c into the cytosol [87]. In this respect, regulation of ER-mitochondrial calcium transfer at the MAMs is essential in the switch between cell survival and death during various stress responses. In the context of Huntington’s disease for example, the binding of mutant huntingtin to IP3R may result in a heightened susceptibility of mitochondria to activate the permeability transition pore and the subsequent release of pro-apoptotic factors [23]. Intriguingly, proteins that modulate calcium transfer between mitochondria and the ER are affected in many neurodegenerative diseases and could be tightly linked to a loss in neuronal survival [94] (see Table 1).

The function of both the ER and mitochondria strongly depends on a highly regulated balance in membrane dynamics. Mitochondrial dynamics include their transport along cytoskeletal tracks throughout the cell, in addition to regulated fission and fusion. Combined, these dynamics ensure optimal mitochondrial metabolism and, in neurons, the correct distribution of mitochondria to dendrites and synapses. Especially in neurons with long axons this poses a challenge, which is highlighted by the many diseases caused by defects in the proteins involved in mitochondrial dynamics [19] (see Boxes 2‐4; Table 1). Mitochondrial fission is known to require the assembly of dynamin-related protein 1 (Drp1) in a helix around the OMM, followed by constriction of this ring (Fig. 5). However, it has long been unclear what signaling events precede this process and mark the site of Drp1 assembly. In 2011, Friedman et al. showed in Cos7 cells that focal accumulation of Drp1 and subsequent fission occurs at or near sites of contact with the ER, and the same holds true for the yeast Drp1 ortholog Dnm1 [37]. Mitochondrial constriction occurs in the vicinity of the ER even in the absence of Drp1 and the mitochondrial fission factor (Mff). These observations suggested that ER tubules mark the sites of mitochondrial fission and induce initial constriction, followed by the recruitment of Mff and finally Drp1 helix assembly that mediates the division [37]. Indeed, recent studies have described that actin polymerization, induced at the MCSs by an ER-bound isoform of Inverted Formin 2 (INF2), is capable and required to induce the initial constriction required for mitochondrial fission (Fig. 5) [58]. Spire1C, a newly described mitochondrial isoform of the actin-nucleating Spire proteins, interacts with INF2 to specifically nucleate actin filaments only at sites where the ER and mitochondrial membranes are closely apposed. Loss of this interaction results in long tubular mitochondria reminiscent of defective fission [70]. Mitochondrial constriction prior to Drp1 assembly depends on this actin polymerization step, however, additional force generated by Myosin II may also contribute to this process [30, 57]. How mutations in INF2, causing Charcot–Marie–Tooth disease type 2D and glomerulopathy [15] affect this process remains to be determined. Finally, it was shown in a very elegant way that Drp1 activity at the MAMs is strictly regulated by phosphorylation in a process that involves Rab32 and syntaxin 17 (STX17) [4] (Fig. 5). This process seems tightly linked to the regulation of calcium homeostasis and the initiation of autophagy (see further). In addition, SUMOylation of Drp1 was reported to affect its function at the MAM, stabilizing the ER-mitochondria contact site and thus promoting calcium crosstalk and cytochrome c release [86].

In contrast to the role of Gem1p in ER-associated mitochondrial division in yeast, no such role has been described for the mammalian Gem1 orthologs Miro1 and Miro2. In mammalian cells, both Miro proteins 1 and 2 are known as essential regulators of mitochondrial motility [36, 69, 92]. These atypical Rho GTPases are anchored in the OMM and mediate anterograde and retrograde transport by connecting the mitochondria to microtubule (MT)-bound kinesin and dynein through the trafficking kinesin (TRAK1 and TRAK2) adaptor proteins (Fig. 5) [36, 69]. Neuron-specific loss of Miro1 in mice results in a neuronal disease phenotype that most closely resembles that of spastic paraplegia patients [80] (see Box 4; Table 1).

Interestingly, several observations point to a role of MAMs in Miro1/2-mediated cellular transport of mitochondria. First, as mentioned, the yeast ortholog of Miro1/2 is a regulatory subunit of the ERMES complex and Miro1 localizes at ER-mitochondria contact sites in mammalian cells [56]. Second, at sites of close contact, the ER and mitochondria move along acetylated MTs while staying attached, suggesting coordinated transport [38]. Third, in addition to coordinating fusion and tethering to the ER of the OMM, MFN2 is required for axonal transport and both MFN1 and MFN2 interact with Miro1/2 [75]. Fourth, mitochondrial transport is regulated by calcium through the EF-hand motifs in Miro1/2, which function as a calcium sensor. Calcium binding of Miro1/2 induces a conformational change that disconnects the mitochondria from the MT track it moves along, thus halting its transport [20, 69, 92]. Due to the relatively low affinity of Miro1/2 to calcium, calcium-dependent halting of mitochondrial transport is most likely to occur at calcium hotspots, such as the ER-mitochondria MCSs [20, 92]. The dependence on high calcium regions indicates a mechanism whereby the proximity with the ER can determine the redistribution of mitochondria. This provides the cell with a powerful system of targeting mitochondria based on local energy requirements. Particularly in neurons, this might be a relevant factor in determining the distribution of mitochondria to the dendrites, synapses and the nodes of Ranvier that is currently insufficiently understood. That such a system is essential for neuronal survival is underscored by the many reports of mitochondrial transport defects in neurodegenerative disorders [19]. Much research has focused on the connection between mitochondria and MTs in this context. It is becoming increasingly clear, however, that the connection with the ER is also a major determinant for axonal transport of mitochondria (see Boxes 2–4).

Macroautophagy is the process whereby damaged proteins and organelles are cleared from the cell by sequestering them in a double membrane-bound vesicle termed the autophagosome. Subsequent delivery to the lysosome allows proteasomal breakdown and recycling of the substrates. Debate is still ongoing with respect to the origin of the phagophore and the identity of membrane donor sources needed for its expansion to form the mature autophagosome [62]. Axe et al. identified the omegasome, a phosphatidylinositol-3-phosphate-enriched membrane structure attached to the ER, as a membrane source for phagophore development [8]. In addition, the mitochondria, harboring many autophagy-related (ATG) proteins and regulators thereof, have been proposed as an origin for phagophore formation, based on the transfer of a fluorescently labeled mitochondrial marker to the phagosome [43]. Intriguingly, several ATG proteins have been found to accumulate specifically at ER-mitochondria MCSs in conditions of starvation, and more and more evidence is accumulating that MAMs might be the actual site of autophagosome formation, unifying both earlier models [43]. Hailey et al. first speculated on the requirement of lipid transfer from the ER to mitochondria during autophagosome biogenesis. In mouse cells lacking MFN2, autophagy induction was disturbed. The authors contributed this to a decreased lipid transfer towards mitochondria as a consequence of ER-mitochondrial uncoupling and concluded that lipids transferred from the ER accumulate in the OMM, from where they are trafficked to the expanding phagophore [43] (Fig. 6). A more direct involvement of MAMs in autophagosome biogenesis was proposed by Hamasaki and coworkers [44]. Upon starvation of mammalian cells, Vps15, Vps34, Atg14L, Beclin1—early markers of phagophore formation—and the omegasome marker double FYVE-containing protein 1 (DFCP1) accumulate in the MAM fraction (Fig. 6) [44]. This might be initiated by early translocation of STX17 [4]. In starvation conditions, STX17 translocates in an mTOR/ULK dependent manner to a domain of ER-mitochondrial contact resistant to digitonin, where it recruits ATG14L [4]. ATG14L then interacts with the phosphoinositide 3-kinase complex consisting of Beclin1, p150 and Vps34, resulting in local phosphatidylinositol-triphosphate production. This alters the local lipid composition of the MAM membrane, which leads to the recruitment of WD-repeat domain phosphoinositide interacting (WIPI) proteins, the effector proteins and the mediators of autophagosome biogenesis. It has been proposed that this could be sufficient to induce a deformation of the membrane [62]. Consistent with a role for the MAMs in autophagosome formation, knockdown of MFN2 or PACS2 abolished ATG14L puncta formation and downstream LC3 lipidation in starved cells, and caused defective STX17 localization at the MAM [44].

In addition, during the selective autophagy of mitochondria, known as mitophagy, ER-mitochondria contact sites appear to constitute a platform promoting mitophagosome formation. In yeast, efficient mitophagy depends on coupling of mitochondria and the ER through ERMES, while bulk autophagy can apparently operate in the absence of this tether [13]. ERMES colocalizes with sites of mitophagosome biogenesis and affects the formation of the phagophore that engulfs the mitochondria destined for degradation [13, 71].

In mammalian cells, mitochondria destined for degradation recruit the E3 ubiquitin ligase Parkin to the OMM through PTEN-induced putative kinase 1 (PINK1) kinase activity. Ubiquitination of Parkin substrates subsequently recruits the autophagy machinery [31]. ER-mitochondria MCSs are implied in this process in several ways. First, focal accumulation of DFCP1 along mitochondria labeled for degradation with exogenously expressed Parkin suggests a role for the ER/MAM as a membrane source for the mitophagosome [115]. Second, Drp1-mediated fission, at least in part mediated by contacts with the ER [37], is suggested to promote mitophagy by breaking off pieces of mitochondria promoting their engulfment by the mitophagosome [105]. Lastly, depolarization of the IMM through calcium (over)loading could be involved in PINK1 and Parkin translocation to the OMM, activating mitophagy [31, 53].

While the exact mechanisms governing autophagy induction and progression are not entirely clear, several lines of evidence point to an essential role of the ER-mitochondria interface during this process. The presence at the MAMs of several important players in metabolism, such as mTORC2 [11], allows to speculate that ER-mitochondria MCSs could represent a site of crosstalk between sensors of the cells’ energetic needs and the early players in autophagy. Additionally, through interorganelle communication, dysfunctional mitochondria may be sensed at the MAMs, followed by mitophagy induction and targeted removal of these unhealthy mitochondria. Future research is needed to study these mechanisms and how they are disrupted in disease. Dysregulation of neuronal energy metabolism and mitochondrial homeostasis are important factors contributing to neurodegeneration [94]. This appears relevant in particular for PD (see Box 3), however, autophagy is a process whose dysregulation is widely implied in neurodegenerative disorders [73].