Endoplasmic reticulum–mitochondria signaling in neurons and neurodegenerative diseases

Mitochondria have a major role in the energy metabolism of eukaryotic cells (McBride et al., 2006). However, the long-held dogma that mitochondria are only the ‘powerhouse of the cells’ has now been firmly abandoned owing to extensive work showing that mitochondria also play an essential role in several other cellular functions, including regulation of Ca²⁺ homeostasis, apoptosis, cell growth and differentiation, and reactive oxidative species (ROS) production among others (McBride et al., 2006). In order to spatially and temporally orchestrate different cellular mechanisms, mitochondria rely on the signals from other organelles and in particular the endoplasmic reticulum (ER). Indeed, ER and mitochondria form close contacts through specialized regions of ER called mitochondria-associated membranes (MAMs) (Csordas et al., 2006). MAMs are dynamic structures composed of membrane segments from both the outer mitochondrial membrane (OMM) and the ER, as well various proteins (Box 1) (Csordas et al., 2006). The mechanism by which MAMs are recruited to the mitochondrial surface involve ER–mitochondrial tethering proteins that physically link the organelles (Csordas et al., 2006; Rowland and Voeltz, 2012). Alterations to ER–mitochondria tethering and MAMs have been strongly linked to the pathogenesis of major neurodegenerative diseases, including Alzheimer's disease (AD), Parkinson's disease (PD) and amyotrophic lateral sclerosis with related frontotemporal dementia (ALS/FTD) (Lau et al., 2018; Paillusson et al., 2016). Interestingly, several functions that are regulated by ER–mitochondria signaling are disrupted in neurodegenerative diseases, including bioenergetics and mitochondrial ATP production, Ca²⁺ homeostasis, axonal transport, lipid metabolism, inflammation, autophagy, ER stress responses and the unfolded protein response (UPR) (Paillusson et al., 2016). However, it is yet to be elucidated whether disruption of ER–mitochondria signaling is a primary trigger of neurodegeneration and an early step in a cascade of pathogenic mechanisms involved in neuronal death, or a repercussion of different cellular insults that subsequently heighten the process of neurodegeneration.

Neurons in particular are dependent on functional ER–mitochondria signaling, not only due to their high energy demands, but also because ER–mitochondria interactions are involved in axonal transport and neurotransmission (Gomez-Suaga et al., 2019; Hirabayashi et al., 2017; Paillusson et al., 2016). Thus, in this Review, we will focus on the role of ER–mitochondria signaling in neurons and discuss how tethering and regulatory proteins that shape communications between the two organelles contribute to key neuronal functions. Finally, we will describe the evidence for disruption in the ER–mitochondria axis in several neurodegenerative diseases.

ER–mitochondria contacts and signaling are important since they impact upon so many different cellular functions (Csordas et al., 2018; Paillusson et al., 2016). However, recent evidence has uncovered ER–mitochondria signaling roles specifically within neurons. The use of electron microscopy to investigate ER–mitochondria contacts in brain has revealed that these contacts are extensive within neurons, being present in cell bodies, axons, dendrites and synaptic regions (Hirabayashi et al., 2017; Leal et al., 2018; Wu et al., 2017b).

One of the primary functions of ER–mitochondria signaling is the regulation of intracellular Ca²⁺ homeostasis (Hajnoczky et al., 2000). The main pathway for Ca²⁺ transfer from ER stores to mitochondria involves release of Ca²⁺ by inositol 1,4,5-trisphosphate receptors (IP3Rs) and import into mitochondria via the OMM voltage-dependent anion channel (VDAC) proteins (VDAC1–VDAC3) and the inner mitochondrial membrane (IMM) protein mitochondrial Ca²⁺ uniporter (MCU) (Rizzuto et al., 1998) (Fig. 1). Ca²⁺ is required for mitochondrial ATP production because dehydrogenases in the TCA cycle are Ca²⁺ dependent (Rowland and Voeltz, 2012; van Vliet et al., 2014). Ca²⁺ plays a vital role in sparking synaptic transmission as concentrations of presynaptic Ca²⁺ control neurotransmitter release, while changes in dendritic Ca²⁺ levels regulate synaptic activity (Devine and Kittler, 2018; Segal, 2005; Sjostrom et al., 2008). Moreover, release of Ca²⁺ from intracellular stores orchestrates neurogenesis by activating signaling pathways that are involved in proliferation, differentiation and migration (Toth et al., 2016).

Remarkably high concentrations of mitochondria in synaptic terminals were reported over 60 years ago (Palay, 1956), and their presence is now known to be linked to the high levels of ATP required for synaptic activity (Laughlin et al., 1998). In addition, the ER is also present in synaptic regions, where ER–mitochondria contacts are now known to form (Gomez-Suaga et al., 2019; Hirabayashi et al., 2017; Mironov and Symonchuk, 2006). Synaptic activity is also dependent on mitochondrial fission, that is, depletion of the key mitochondrial fission protein dynamin-related protein 1 (Drp1; also known as DNM1L) leads to impairment in synaptic transmission and memory functions (Oettinghaus et al., 2016) (Fig. 1). As Drp1 has a potential role in stabilizing the interaction between ER and mitochondria, its loss and subsequent disruption of synaptic transmission could be directly linked to impaired signaling between the two organelles (Oettinghaus et al., 2016; Prudent et al., 2015).

Pathological conditions such as axonal injury has been shown to trigger recruitment of ER and mitochondria to the damaged site (Han et al., 2016; Rao et al., 2016; Zhou et al., 2016). Interestingly, an increase in ER–mitochondria tethering and subsequent Ca²⁺ transfer to mitochondria promotes regrowth of injured axons (Lee et al., 2019) thus pointing to ER–mitochondria signaling as a potential therapeutic target in neurodegenerative diseases (Fig. 1).

Finally, ER–mitochondria signaling is also involved in the regulation of autophagy and mitophagy (selective degradation of damaged mitochondria) (Son et al., 2012). Moreover, a core component of many neurodegenerative diseases is autophagic dysfunction (Hara et al., 2006; Nixon, 2013). However, the role of ER–mitochondria contacts in autophagy is somewhat controversial. Although it has been initially shown that ER–mitochondria contacts are sites for autophagosome formation (Hamasaki et al., 2013), subsequent work has reported that loss of ER–mitochondria contacts stimulates autophagy (Gomez-Suaga et al., 2017; Rimessi et al., 2020; Wu et al., 2018) (Fig. 1), possibly due to the different methods used to induce autophagy.

Many years after electron microscopy revealed the occurrence of close interactions between ER and mitochondria (Robertson, 1960; Shore and Tata, 1977), the first tether protein complex was discovered in yeast and termed the ER–mitochondria encounter structure (ERMES) (Kornmann and Walter, 2010). PDZD8 is a mammalian homolog of one subunit of the ERMES complex, but its mitochondrial binding partner is unknown (Fig. 1). Although it is not clear how PDZD8 acts as a tether, it has been shown to promote ER–mitochondria contacts and induces Ca²⁺ delivery to mitochondria at the synapse in cortical neurons (Hirabayashi et al., 2017). Several other mammalian proteins have also been identified as ER–mitochondrial tethers.

The interaction between IP3R, VDAC and the glucose-regulated protein 75 (GRP75; also known as HSPA9) was one of the first proposed tethering complexes (Rizzuto et al., 1998) (Fig. 1). This complex has a functional role in directing Ca²⁺ transfer between ER and mitochondria and subsequent ATP production (Bartok et al., 2019; D'Eletto et al., 2018; Rapizzi et al., 2002; Szabadkai et al., 2006a). However, a complete knockout of all three isoforms of IP3R does not modify ER–mitochondria contacts, suggesting that the IP3R–GRP75–VDAC complex associates at contact sites rather than being a bona fide tether (Csordas et al., 2006).

Homotypic and heterotypic interactions between mitochondrial mitofusin 1 (MFN1) and/or mitofusin 2 (MFN2) and ER-located MFN2 have also been proposed as a tethering complex (de Brito and Scorrano, 2008, 2009). Although MFN2 appears to regulate MAM dynamics, its role as genuine tether is still debated, as a loss of MFN2 expression has been reported to both decrease and increase mitochondrial contacts (Cosson et al., 2012; de Brito and Scorrano, 2008; Filadi et al., 2015; Leal et al., 2016; Naon et al., 2016; Wang et al., 2015).

Several studies have suggested that the interaction between the integral ER protein vesicle-associated membrane protein-associated protein B (VAPB) and the OMM component protein tyrosine phosphatase interacting protein 51 (PTPIP51; also known as RMDN3) acts as a major linker between ER and mitochondria (Cieri et al., 2018a; De Vos et al., 2012; Rimessi et al., 2020; Stoica et al., 2014; Yeo et al., 2021) (Fig. 1). More recently, an interaction between VAPB and PTPIP51 was found within neuronal synapses, where it contributes to synaptic activity (Gomez-Suaga et al., 2019). Here, we primarily focus on the three major ER–mitochondria tether complexes mentioned above, and further proteins identified as MAM tethers are listed in Table 1.

ER–mitochondria contacts are now known to change in response to physiological stimuli and are increased by induction of synaptic activity (Gomez-Suaga et al., 2019). The mechanism underlying these changes is not fully understood yet, but a number of proteins are known to regulate ER–mitochondria contacts, either to promote or disrupt them. These can act in close vicinity to the contacts such as Sigma-1 receptor (SigmaR1) or can operate more distally by initiating a cascade of signaling events that eventually results in a modulation of contacts, as is the case for ALS-linked proteins TAR DNA-binding protein 43 (TDP-43; also known as TARDBP) and fused in sarcoma (FUS) (Hayashi and Su, 2007; Stoica et al., 2014, 2016).

SigmaR1 is predominantly localized at MAMs where it acts as an IP3R chaperone to facilitate the transfer of Ca²⁺ from ER to mitochondria and support the synthesis of ATP (Hayashi and Su, 2007; Mavlyutov et al., 2010). Moreover, SigmaR1 regulates dendritic spine formation in hippocampal neurons via activation of Rac1, a small Rho GTPase involved in regulation of neuronal development (Tsai et al., 2009a; Watabe-Uchida et al., 2006) (Fig. 1). Ca²⁺ transfer from ER to mitochondria at MAMs is also facilitated by Tespa-1, transglutaminase type 2 (TG-2; also known as TGM2) and the deglycase DJ-1 (also known as PARK7), which physically interact with GRP75 to stabilize the IP3R–GRP75–VDAC complex (D'Eletto et al., 2018; Liu et al., 2019; Matsuzaki et al., 2013) (Fig. 1). In addition, the E3 ubiquitin ligase MITOL (also known as March5 and MARCHF5) promotes ER–mitochondria interaction by stimulating MFN2 oligomerization; accordingly, its loss in neurons leads to a disruption of ER–mitochondria contacts, oxidative stress and neuroinflammation (Nagashima et al., 2019; Sugiura et al., 2013).

Maintenance of mitochondrial function is of key importance for neurons, not only due to their high energetic requirements, but also as a protection against the generation of neurotoxic ROS (Angelova and Abramov, 2018). The mitochondrial E3 ubiquitin ligase Mul1 protects mitochondrial morphology, restrains mitophagy and promotes ER–mitochondria coupling in neurons by targeting MFN2 for degradation, thus acting as a positive regulator of ER–mitochondria contacts (Puri et al., 2019) (Fig. 1). In addition, Mul1 promotes SUMOylation of Drp1 to stabilize its oligomers on the mitochondrial membrane, which subsequently promotes ER–mitochondria interactions and regulates the release of pro-apoptotic factor cytochrome c (Prudent et al., 2015). Moreover, Drp1 plays a crucial role in mitochondrial fission, which is particularly important in neurons for the maintenance of normal synaptic transmission and terminal branching during the final step of axon development (Lewis et al., 2018; Oettinghaus et al., 2016) (Fig. 1). Loss of the OMM proteins Miro1 and Miro2 leads to disruption of ER–mitochondria interactions and mitochondrial cristae morphology, although the precise mechanism is still unclear (Modi et al., 2019). Moreover, Miro1 forms a complex with disrupted-in-schizophrenia 1 (DISC1) to promote anterograde mitochondrial transport in neuronal axons, which is known to have a crucial role in development of neurons and neurotransmission (Ogawa et al., 2013). DISC1 also orchestrates Ca²⁺ transfer from ER to mitochondria and disrupts ER–mitochondria contacts by binding to IP3R, thus precluding binding of its ligand, while its loss results in Ca²⁺ overload in mitochondria and the generation of oxidative stress in neurons (Park et al., 2017).

The serine/threonine-protein kinase Akt1 localizes to MAMs and has been linked to neurodegeneration (Betz et al., 2013; Giorgi et al., 2010). Akt1 has been shown to play a role in the maintenance of ER–mitochondria contacts, which may be linked to its ability to phosphorylate IP3Rs (Betz et al., 2013). The full list of regulators of ER–mitochondria interactions and their functions are listed in Table 2.

Given that either mutations or disruption in function of the abovementioned tethers and regulators have been reported in several neurodegenerative diseases, in the following sections we will focus on highlighting the role of ER–mitochondria signaling in common neurodegenerative diseases.

AD is the most common form of dementia, currently affecting more than 55 million people worldwide (World Alzheimer Report 2021; https://www.alzint.org/u/World-Alzheimer-Report-2021.pdf). Its key pathologic hallmarks are plaques composed of extracellular amyloid β (Aβ) peptides and intracellular neurofibrillary tangles formed of highly phosphorylated aggregates of the Tau protein, a microtubule-associated protein found at greater levels in the brains of AD patients compared to controls (Goedert and Spillantini, 2006; Khatoon et al., 1994; Masters et al., 2015). Aβ is generated by cleavage of amyloid precursor protein (APP) by two proteases known as β-secretase and γ-secretase. Presenilin 1 (PS1) and presenilin 2 (PS2) (also known as PSEN1 and PSEN2, respectively) are active subunits of the γ-secretase complex, and mutations in genes encoding for presenilin proteins and APP cause some familial forms of AD (Goedert and Spillantini, 2006; Schellenberg and Montine, 2012).

Several lines of evidence suggest that MAMs and ER–mitochondria signaling may play a key role in the pathogenesis of AD. First, the levels of proteins tethering or associating with MAMs, including VAPB, PTPIP51, IP3R1, MFN1, MFN2, phosphofurin acidic cluster sorting protein 2 (PACS-2) and SigmaR1, were found to be either increased or decreased in post-mortem AD brain tissue (Hedskog et al., 2013; Lau et al., 2018; Leal et al., 2020; Sepulveda-Falla et al., 2014). Moreover, disruption to ER–mitochondria contacts, including damage to VAPB–PTPIP51 tethers (Lau et al., 2020), is present in human post-mortem AD tissue, even at early disease stages (Hedskog et al., 2013; Lau et al., 2018; Leal et al., 2018; Sepulveda-Falla et al., 2014). Second, MAMs are a platform for major proteins linked to AD pathogenesis and are enriched in both PS1 and PS2, whereas acyl-CoA:cholesterol acyltransferase, an enzyme proposed to be necessary for Aβ production, is predominantly located at MAMs (Area-Gomez et al., 2009; Puglielli et al., 2001; Rusinol et al., 1994). Third, upon cleavage of full-length APP by β-secretase, the intermediate fragment, called C99, is transported to MAMs and processed by γ-secretase to form two peptides, amyloid precursor protein intracellular domain (AICD) and Aβ (Pera et al., 2017). Thus, MAMs are sites of Aβ production (Area-Gomez et al., 2009; Schreiner et al., 2015). Finally, the E4 allele of apolipoprotein E (ApoE4), a major genetic risk factor for AD, promotes ER–mitochondrial communication and MAM function (Tambini et al., 2015) (Fig. 2).

Most of the evidence for impaired ER–mitochondria signaling in AD arises from cellular and transgenic mouse models. However, these studies present conflicting data; whereas some studies have reported that PS2 is important for modulating ER–mitochondria signaling (Rossi et al., 2021; Zampese et al., 2011) and that PS1 is dispensable in this process (Zampese et al., 2011), others have shown that PS1 does affect ER–mitochondria interactions (Area-Gomez et al., 2012; Sepulveda-Falla et al., 2014). Interestingly, AD-linked PS2 mutants diminish mitochondrial Ca²⁺ uptake while increasing ER–mitochondria contact sites (Area-Gomez et al., 2012; Filadi et al., 2016; Kipanyula et al., 2012; Zampese et al., 2011), suggesting increased ER–mitochondria interaction as a cell-intrinsic compensatory mechanism to boost transfer of Ca²⁺ from ER to mitochondria. The effect of Aβ and APP on ER–mitochondria interactions has also resulted in conflicting data, with some studies showing increased contacts (Del Prete et al., 2017; Hedskog et al., 2013; Leal et al., 2020), whereas others have reported decreased contacts (Martino Adami et al., 2019). In addition, expression of disease-associated phosphorylated (Perreault et al., 2009) and truncated Tau (Cieri et al., 2018b) increases the number of ER–mitochondria contact sites. Similar to what is seen for PS2 mutants, truncated Tau has been reported to decrease the basal levels of ER Ca²⁺. Thus, the subsequent increase in ER–mitochondria contact sites is perhaps acting to restore mitochondrial Ca²⁺ homeostasis (Cieri et al., 2018b) (Fig. 2).

It has also been investigated how loss of ER–mitochondria tethering and diminished MAM function affect AD pathogenesis. Indeed, two studies show that loss of MFN2 decreases γ-secretase activity and lowers Aβ production; however, loss of MFN2 has been suggested to both reduce (Area-Gomez et al., 2012) and increase (Leal et al., 2016) ER–mitochondria contacts. From a clinical point of view, modifying ER–mitochondria interactions could be a potential treatment for AD, as a Drosophila model of AD demonstrated that the expression of a synthetic linker, which increases contacts between ER and mitochondria, successfully enhanced locomotion of the flies and expands their lifespan (Garrido-Maraver et al., 2019). However, even though damage to ER–mitochondria contacts is established in AD, data are still conflicting, partly due to the heterogenicity of experimental models and methods used (see Box 2).

PD is the second most frequent neurodegenerative disease, clinically characterized by a motor tetrad, resulting from a loss of dopaminergic neurons in the subtantia nigra pars compacta (Damier et al., 1999). This loss of dopaminergic neurons is mostly associated with the presence of proteinaceous aggregates, named Lewy bodies, in the surviving neurons; these are mostly composed of phosphorylated and aggregated α-synuclein (Kalia and Lang, 2015).

α-Synuclein is a key protein in PD physiopathology and point mutations and duplications in its encoding gene, SNCA, are responsible for familial forms of PD (Srinivasan et al., 2021). Four groups have reported that modulating alpha synuclein expression affects ER-mitochondria signaling and its associated functions, such as Ca²⁺ homeostasis (Cali et al., 2012; Paillusson et al., 2017), phospholipid synthesis (Guardia-Laguarta et al., 2014), autophagy (Cali et al., 2012), and ATP production (Faustini et al., 2019; Paillusson et al., 2017). However, there is no clear consensus on the effects of α-synuclein on ER–mitochondria contacts. Initially, overexpression of α-synuclein was found to increase MAM contacts and mitochondrial Ca²⁺ in HeLa cells (Cali et al., 2012, 2019). However, we and another group have subsequently shown that overexpression of α-synuclein results in a decrease in ER–mitochondria interactions and associated functions, and it appears that the familial α-synuclein mutations A53T and A30P have a similar effect (Guardia-Laguarta et al., 2014; Paillusson et al., 2017). One possible reason for this discrepancy might be the relative levels of α-synuclein expressed in the respective models, as well as the methods used (Box 2). Another key point is that α-synuclein is found within the MAM compartment (Guardia-Laguarta et al., 2014; Paillusson et al., 2017), probably due to its high affinity for lipid rafts (Fortin et al., 2004). We have shown that α-synuclein localized at MAMs binds to VAPB and prevents its interaction with PTPIP51, therefore decreasing ER–mitochondria contacts, and demonstrating that α-synuclein can directly break ER–mitochondria contacts (Paillusson et al., 2017) (Fig. 2).

Loss-of-function mutations in PTEN-induced putative kinase 1 (PINK1) and Parkin (an E3 ubiquitin ligase; symbol PRKN), two proteins involved in mitophagy, are responsible for familial forms of PD (Berenguer-Escuder et al., 2019; Cali et al., 2013; Celardo et al., 2016; Gautier et al., 2016; Gelmetti et al., 2017; Parrado-Fernandez et al., 2018). Notably, they both localize to the MAM compartment (Gelmetti et al., 2017; Van Laar et al., 2015), suggesting a role in the regulation of ER–mitochondria signaling (Fig. 2). It has also been found that PINK1- and Parkin-mediated phospho-ubiquitylation of MFN2 complexes promotes their degradation, and subsequent loss of ER–mitochondria contacts, as well as increased mitophagy (Gautier et al., 2016; Gegg et al., 2010; McLelland et al., 2018). Conversely, several groups have reported that decreased Parkin or PINK1 expression leads to a loss of ER–mitochondria contacts that is associated with dysregulation of Ca²⁺ homeostasis (Fig. 2) (Basso et al., 2018; Cali et al., 2013; Parrado-Fernandez et al., 2018), suggesting a loss of ER-mitochondria contact in PD. Mechanistically, ubiquitylation of MFN2 on lysine 416 by Parkin has been shown to be a pre-requisite for promotion of ER–mitochondria interaction (Basso et al., 2018). However, some studies showed the opposite; in primary fibroblasts from PD patients and Drosophila models expressing loss-of-function mutations of either PINK1 or Parkin have an increase in ER–mitochondria contacts (Celardo et al., 2016; Valadas et al., 2018) that is mediated through the activation of the PKR-like ER kinase (PERK; also known as EIF2AK3), one of the key UPR signal activator proteins (Celardo et al., 2016). This increase in contacts was associated with an excessive phosphatidylserine transfer in the MAMs, which is responsible for sleep disorder in flies, a premotor symptom also found in PD (Valadas et al., 2018). Therefore, although it is evident that PINK1 and Parkin have an important role in ER–mitochondria signaling, their contribution is not fully understood.

The deglycase DJ-1 is small protein that scavenges ROS, and has been strongly associated with PD (Repici and Giorgini, 2019). Loss-of-function mutations or deletion of PARK7, the gene encoding DJ-1, are responsible for early onset forms of PD (Repici and Giorgini, 2019). Although DJ-1 is a cytoplasmic protein, it is also found at ER–mitochondria contact sites, where it promotes contacts (Ottolini et al., 2013; Parrado-Fernandez et al., 2018) (Fig. 2). Moreover, DJ-1 has recently been described to interact with and stabilize the IP3R–GRP75–VDAC complex, therefore increasing ER–mitochondria tethering and facilitating Ca²⁺ exchange between the two organelles to sustain mitochondria needs in Ca²⁺ (Liu et al., 2019); this suggests a positive effect on ER–mitochondria contacts (Fig. 2).

Mutations in the leucine-rich repeat kinase 2 (LRRK2) gene are the most common cause of familial PD and found in 2% of late-onset sporadic cases (Ferreira and Massano, 2017). Recently, LRRK2 knockout has been shown to decrease ER–mitochondria contacts in mouse embryonic fibroblasts (Toyofuku et al., 2020). This loss of contact is associated with a decrease in mitochondrial Ca²⁺ uptake and oxidative phosphorylation. Interestingly, the LRRK2 G2019S mutation, which is characterized by a constitutively active kinase domain, decreases ER–mitochondria contacts (Fig. 2), whereas the artificial kinase-dead LRRK2 D1994A mutation increases these contacts, suggesting that the LRRK2 kinase domain can regulate ER–mitochondria associations (Toyofuku et al., 2020). This work also reported that, similar to the action of Parkin or PINK1, LRRK2 negatively regulates PERK activity through the activation of E3 ubiquitin ligases, such as MARCH5, Mul1 or Parkin itself (Toyofuku et al., 2020), highlighting a potential common pathophysiological mechanism in PD.

The small Rho GTPase Miro1, which is involved in mitochondrial transport, has also recently been implicated in PD pathology. Four loss-of-function mutations in RHOT1, the gene encoding Miro1, have been recently found in PD patients (Berenguer-Escuder et al., 2019; Grossmann et al., 2019; Saeed, 2018), and fibroblasts from these patients exhibit a loss of ER–mitochondria contacts associated with a decrease in energetic metabolism and increase in LC3-dependent autophagy (Berenguer-Escuder et al., 2019, 2020; Grossmann et al., 2019; Modi et al., 2019) (Fig. 2). Interestingly, overexpression of PINK1 and Parkin promotes Miro1 degradation in rat neurons (Wang et al., 2011), whereas loss-of-function mutations in PINK1 lead to increased levels of mitochondrial Ca²⁺ in flies through Miro activation (Lee et al., 2018), therefore providing further support for their role in reducing ER-mitochondria associations (Fig. 2).

ALS and FTD are two related and incurable neurodegenerative diseases, and their features can include pathological protein inclusions rich in TDP-43, FUS, superoxide dismutase-1 (SOD1) and dipeptide repeat proteins derived from the C9ORF72 gene in affected neurons (Arai et al., 2009; Ash et al., 2013; Deng et al., 2010; Mori et al., 2013; Neumann et al., 2006). Mitochondrial dysfunction is a common feature of ALS, and targeting mitochondrial dysfunction, in particular early in the disease process, is a potential therapeutic target, as demonstrated by increased survival in animal models of ALS (Mehta et al., 2019). Disruptions to mitochondrial function and Ca²⁺ regulation have been linked to mutations in the major causative genes of ALS, including TARDBP, FUS, SOD1 and the C9ORF72 repeat expansion, despite these not encoding mitochondrial or ER proteins (Smith et al., 2019). Mutations in genes encoding the MAM proteins VAPB and SigmaR1 are also known to cause ALS (Al-Saif et al., 2011; Nishimura et al., 2004).

Induced pluripotent stem cell (iPSC)-derived motor neurons containing the C9ORF72 hexanucleotide repeat expansion exhibit a reduced mitochondrial membrane potential, increased ER stress and reduced levels of Bcl-2, an anti-apoptotic protein that modulates cellular Ca²⁺ signaling (Dafinca et al., 2016). Levels of GRP78 (also known as BiP), which is associated with the ER stress response, are also increased, and swollen mitochondria were observed (Dafinca et al., 2016). In these cells, Ca²⁺ homeostasis is also disrupted, with a greater Ca²⁺ release on depolarisation, delayed recovery following glutamate stimulation and reduced levels of calbindin, a cytosolic Ca²⁺ buffering protein, resulting in reduced clearance of excess Ca²⁺ (Dafinca et al., 2020). Furthermore, similar to the phenotype of TARDBP-mutant iPSC-derived motor neurons, Ca²⁺ permeable subunits of the glutamatergic receptors AMPA and NMDA are upregulated and mitochondrial Ca²⁺ buffering is inhibited, suggesting that glutamate cytotoxicity, a pathological feature of ALS in motor neurons (Van Den Bosch et al., 2006), is aggravated by impaired mitochondrial Ca²⁺ uptake (Dafinca et al., 2020). In vivo, mitochondrial abnormalities are induced by expression of poly-GR, one of the dipeptide repeat protein inclusions that are a pathological feature of C9ORF72-ALS. In a mouse overexpression model, poly-GR binds to mitochondrial ATP synthase protein ATP5A1 (also known as ATP5F1A) (Choi et al., 2019); this, in turn, increases its ubiquitylation and subsequent degradation, consistent with the reduced ATP5A1 protein levels seen in patient brains (Choi et al., 2019). These mitochondrial defects in the presence of poly-GR were observed before any other cellular defects became evident, supporting the notion that mitochondrial defects are an early pathogenic event in C9ORF72-ALS (Choi et al., 2019).

In both cell models and spinal cords of transgenic mice that overexpress TDP-43 (either the wild-type or a form with ALS-linked mutations, such as M337V), the interaction between the ER–mitochondrial tethers VAPB and PTPIP51 are disrupted. This is due to the modulation of GSK3β activity by TDP-43, which then disrupts the VAPB–PPTPIP51 interaction (Stoica et al., 2014) (Fig. 2). The same is true of models with overexpression of wild-type or ALS-linked mutants of FUS. Indeed, overexpression of aggregation-prone FUS results in decreased VAPB–PTPIP51 interactions, reduced mitochondrial Ca²⁺ uptake and reduced ATP production (Stoica et al., 2016).

Mutant SOD1 accumulates at the MAMs specifically in neurons, possibly due to aberrant binding of the mutant protein to the mitochondrial membrane, where it inhibits ER–mitochondria tethering, notably before the onset of any symptoms in mutant SOD1^G93A mice (Watabe-Uchida et al., 2006). Mutant SOD1 also results in reduced ATP production, dysregulation of intracellular Ca²⁺ and a loss of MAM proteins, including SigmaR1, IP3R3 and calreticulin, demonstrating a disruption of MAM function and structure (Watanabe et al., 2016).

The MAM-specific factor SigmaR1 is highly expressed in spinal motor neurons and has been shown to have an important role in ALS, as mutations cause ALS, ALS/FTD and juvenile ALS (ALS16) (Al-Saif et al., 2011; Luty et al., 2010; Ullah et al., 2015). SigmaR1 mutations cause ALS owing to the inability of the mutant protein to bind to IP3R3s, which induces a dissociation of the ER from mitochondria, resulting in disrupted Ca²⁺ homeostasis and reduced ATP synthesis (Watanabe et al., 2016). The colocalization of SigmaR1 with calreticulin is also reduced and mutant forms mislocalize away from the MAMs (Gregianin et al., 2016). Furthermore, loss of SigmaR1 results in reduction of MAM proteins involved in regulation of Ca²⁺ homeostasis, such as IP3R3 (Bernard-Marissal et al., 2015; Watanabe et al., 2016). SigmaR1 function at MAMs has also been shown to be important in ALS with SOD1 mutation G85R. SigmaR1 knockout in SOD1^G85R mice accelerated the disease onset, although SigmaR1 deficiency without the SOD1 mutation caused only a modest motor phenotype (Watanabe et al., 2016). Mutant SOD1 accumulation and SigmaR1 deficiency combined results in a disruption of the MAMs and mitochondrial dysfunction (Watanabe et al., 2016). Furthermore, SigmaR1 activation in SOD1-mutant transgenic mice can prevent the disruption of ER–mitochondrial contacts and restore Ca²⁺ homeostasis (Hyrskyluoto et al., 2013; Ono et al., 2014; Watanabe et al., 2016). This further illustrates the importance of MAMs in ALS and the potential of targeting them to delay disease onset, even in the presence of a known ALS mutation.

Mutations in VAPB can also cause ALS, with clinical outcomes ranging from severe ALS with rapid progression, autosomal-dominant ALS8, or late onset spinal muscular atrophy in affected individuals (Borgese et al., 2021; Nishimura et al., 2004). Mutant VAPB forms inclusions in the ER, resulting in alterations to ER structure, which have been suggested to underlie motor neuron pathogenesis in ALS (Fasana et al., 2010; Kuijpers et al., 2013), and the VAPB P56S mutation in particular disrupts axonal mitochondrial transport due to an alteration of Ca²⁺ homeostasis (Morotz et al., 2012). VAPB aggregates in the ER have also been observed in sporadic ALS patients even in the absence of any VAPB mutations and without changes in its gene expression (Cadoni et al., 2020).

It is clear that mitochondrial dysfunction and disruption of Ca²⁺ homeostasis are common and important features in ALS pathology, pointing to MAMs as a potential therapeutic target, especially as MAM defects appear to occur early in disease progression. Moreover, in sporadic ALS, therapeutically targeting mitochondrial function may be particularly valuable as it appears to be a common feature to ALS even when a causative mutation has not been identified.

Huntington's disease (HD) is a genetic autosomal dominant neurodegenerative disorder caused by the abnormal expansion of CAG repeats in the huntingtin (Htt) gene (Wanker, 2000). Such an expansion generates a polyglutamine (polyQ) stretch at the N-terminus of the Htt protein, resulting in its misfolding and aggregation in the cortical and striatal regions (Finkbeiner, 2011; Gutekunst et al., 1999). Additionally, pathogenesis of HD is also accompanied by mitochondrial abnormalities, such as mitochondrial fragmentation, decreased ATP production, loss of mitochondrial membrane potential and disruption of Ca²⁺ signaling (Lim et al., 2008; Milakovic et al., 2006; Naia et al., 2016; Panov et al., 2002; Rockabrand et al., 2007).

Mitochondrial Ca²⁺ overload, which leads to an opening of permeability transition pores (PTPs) and subsequent release of pro-apoptotic factor, was initially linked to HD (Bezprozvanny and Hayden, 2004; Choo et al., 2004; Panov et al., 2002). Moreover, polyQ-expanded Htt proteins were found to interact with IP3R, resulting in its increased activation in response to IP3 and subsequent excessive Ca²⁺ transfer to mitochondria (Bezprozvanny and Hayden, 2004). Moreover, mitochondrial Ca²⁺ overload has been suggested to result from increased TG-2 expression, which has been detected in the cortex and striatum of HD patients (Min and Chung, 2018). Although many groups have demonstrated changes in Ca²⁺ levels in HD, only one study thus far has directly linked alterations in ER–mitochondria contacts with HD (Cherubini et al., 2020). Here, primary striatal neurons obtained from R6/1 mice, a standardized HD mouse model, showed Drp1-mediated reduction in ER–mitochondria contacts, characterized by a reduced interaction between IP3R3 and VDAC, and a subsequent decrease in transfer of Ca²⁺ from ER stores to mitochondria (Cherubini et al., 2020). Although the exact role of ER–mitochondria interactions in HD is not fully clear, it is evident that Ca²⁺ delivery to the mitochondria plays an important role in disease pathogenesis.

Charcot–Marie–Tooth (CMT) disease is a motor and sensory peripheral neuropathy characterized by demyelination (damage to the myelin sheath) and/or loss of axonal integrity (Baets et al., 2014). CMT is a hereditary disease and among the causative genes, mutations in MFN2 have been linked to an axonal form of the disease called CMT2 subtype A (Feely et al., 2011; Zuchner et al., 2004). Indeed, several studies have reported that CMT2A-linked mutations in MFN2 disrupt ER–mitochondria contacts (Basso et al., 2018; Bernard-Marissal et al., 2019; Larrea et al., 2019). Accordingly, in vitro studies have shown that the different CMT-causing mutations in MFN2 failed to restore damaged ER–mitochondria contacts in Mfn2^−/− MEFs (Basso et al., 2018; de Brito and Scorrano, 2008). Moreover, silencing of ganglioside-induced differentiation associated protein 1 (GDAP1), whose loss-of-function mutations have also been linked to CMT, causes a decrease in ER-mitochondria contacts and disruption to Ca²⁺ homeostasis in in vitro and in vivo models (Barneo-Munoz et al., 2015; Gonzalez-Sanchez et al., 2017; Pla-Martin et al., 2013). These findings were recently reproduced as motor neurons derived from GDAP1^−/− mice showed a reduction in the number of ER–mitochondria contacts, reduced ATP production and disruption of mitochondrial axonal transport (Civera-Tregon et al., 2021). Finally, the gene encoding for IP3R3 was recently identified as a CMT-causing gene, and patient fibroblasts showed disrupted Ca²⁺ homeostasis (Ronkko et al., 2020). These findings further support the importance of ER–mitochondria signaling in the etiology of CMT disease.

Accumulating evidence shows that defects in ER–mitochondria signaling are a common feature of almost all neurodegenerative disorders. Interestingly, proteins linked to familial forms of neurogenerative disorders, such as α-synuclein or TDP-43, appear to be direct or indirect strong disruptors of MAM dynamics. It is now clear that MAMs are crucial for neuronal functions, and although their complexity is not yet fully understood, recent discoveries have identified key factors involved in MAM structure, regulation and functions. In this regard, the role of the VAPB–PTPIP51 interaction as bona fide MAM tethers is strongly suggested by independent findings. In contrast, the role of MFN2, originally considered as an ER–mitochondria tether, still remains unclear, and it might instead have a role in regulating MAMs. Conversely, SigmaR1 has been described to positively modulate MAM activity such as Ca²⁺ exchange through IP3R activation (Hayashi and Su, 2007), and its activation was demonstrated to mitigate the cellular defects seen in some neurodegenerative diseases such as ALS (Lee et al., 2020; Mancuso et al., 2012), suggesting that modulating ER–mitochondria signaling could be beneficial in neurodegenerative disorders. Nevertheless, future efforts should be directed towards providing further insights into the biological functions of MAMs, especially in terms of their structural and dynamic regulation, as well as identifying new pharmacological or genetic approaches to modulate MAM functions. However, conflicting results in studies of MAM dynamics highlight the need to develop new and appropriate methods for their study (Box 2), but they might also shed light on the different mechanisms that overall damage neuronal functions. Given the role of MAMs in neurodegenerative disorders, promoting MAM function is a promising approach to alleviate patient symptoms or slow down the progression of the disease. In this regard, developing new drug-screening assays is crucial in order to be able to identify candidates, for instance SigmaR1 agonists, that can be safely used in patients.