Regulation of mitochondrial dynamics in astrocytes: mechanisms, consequences, and unknowns

Abstract

Astrocytes are the major glial cell in the central nervous system. These polarized cells possess numerous processes that ensheath the vasculature and contact synapses. Astrocytes play important roles in synaptic signaling, neurotransmitter synthesis and recycling, control of nutrient uptake, and local blood flow. Many of these processes depend on local metabolism and/or energy utilization. While astrocytes respond to increases in neuronal activity and metabolic demand by upregulating glycolysis and glycogenolysis, astrocytes also possess significant capacity for oxidative (mitochondrial) metabolism.

Mitochondria mediate energy supply and metabolism, cellular survival, ionic homeostasis, and proliferation. These organelles are dynamic structures undergoing extensive fission and fusion, directed movement along cytoskeletal tracts, and degradation. While many of the mechanisms underlying the dynamics of these organelles and their physiologic roles have been characterized in neurons and other cells, the roles that mitochondrial dynamics play in glial physiology is less well understood. Recent work from several laboratories has demonstrated that mitochondria are present within the fine processes of astrocytes, that their movement is regulated, and that they contribute to local Ca²⁺ signaling within the astrocyte. They likely play a role in local ATP production and metabolism, particularly that of glutamate. Here will review these and other findings describing the mechanism by which mitochondrial dynamics are regulated in astrocytes, how mitochondrial dynamics might influence astrocyte and brain metabolism, and draw parallels to mitochondrial dynamics in neurons. Additionally, we present new analyses of the size, distribution, and dynamics of mitochondria in astrocytes in vivo using 2-photon microscopy.

Introduction

Mitochondria are essential organelles that play multifaceted roles in eukaryotic cells. They provide energy, coordinate local metabolism, regulate Ca²⁺ signals, and integrate survival and death cues (reviewed in Jacobson and Duchen, 2004; Rizzuto et al., 2012). Mitochondria are dynamic organelles; they move, divide from and fuse with one another, and are removed if damaged. In neurons, mitochondria are transported, often over long distance, to sites of elevated [Ca²⁺] and ATP utilization to match energetic demand with supply and provide local Ca²⁺ buffering capacity (for reviews see Hollenbeck and Saxton, 2005; Cai and Sheng, 2009; MacAskill and Kittler, 2010; Schwarz, 2013). Mitochondria may be part of large reticular networks or exist as discrete structures of varying size or shape. Even here, constant remodeling of shape and content takes place facilitating the distribution of mitochondria and homogenization of their content. In addition, mechanisms exist to remove damaged or otherwise dysfunctional mitochondria from the mitochondrial network (for reviews see Chen and Chan, 2009; Youle and Narendra, 2011; Maday et al., 2014; Mishra and Chan, 2014). The totality of these behaviors establishes the steady state distribution of mitochondria (and energy/ Ca²⁺ buffering) within the cell. Much of what we know regarding the mechanisms underlying these behaviors were gleaned from studies conducted in primary neuronal cultures. By contrast less is known regarding the roles and regulation of mitochondrial dynamics in glia, particularly astrocytes.

Astrocytes, thought to be the predominant type of glial cell in the brain, are involved in a wide range of CNS functions, including control of blood flow, glucose metabolism, glutamate clearance, ionic homeostasis (particularly K⁺), synaptic development, and neuronal plasticity (for reviews see Araque et al., 1999; Attwell et al., 2010; Clarke and Barres, 2013; Dienel, 2013; Haydon and Nedergaard, 2014; Weber and Barros, 2015). The wide range of functions ascribed to astrocytes is facilitated anatomically by processes that contact synapses and blood vessels. While primary cultures of astrocytes are largely flat and retain few cellular processes (Shao and McCarthy, 1994; Derouiche and Frotscher, 2001), astrocytes in the CNS, particularly the so called protoplasmic astrocytes, possess numerous, highly branched processes, lamellipodia, and veil-like structures. The smallest of these processes, also known as the peripheral astrocytic processes (PAPs), may be 20–200 nm in diameter and possess very little cytoplasm (Kosaka and Hama, 1986; Ogata and Kosaka, 2002; for review see Benjamin Kacerovsky and Murai, 2015). This gives astrocytes a very large surface area to volume ratio. In the cortex, these PAPs often are intimately associated with synapses (Kosaka and Hama, 1986; Ventura and Harris, 1999; Witcher et al., 2010). In fact, a single astrocyte in the rodent brain may contact ~100K synapses and may contact in excess of 2 million synapses in the human brain (Oberheim et al., 2009).

Based in part on their small size, it was assumed that the fine astrocyte processes were too small to accommodate mitochondria (whose diameter may exceed 1 µm in other cell types). However, new studies, and re-examination of older studies, have revealed the presence of mitochondria within the fine astrocytic processes both in situ and in vivo (Mugnaini, 1964; Aoki et al., 1987; Lovatt et al., 2007; Oberheim et al., 2009; Mathiisen et al., 2010; Genda et al., 2011; Jackson et al., 2014b; Derouiche et al., 2015; Stephen et al., 2015; Agarwal et al., 2017; for review, see Stephen et al., 2014; Benjamin Kacerovsky and Murai, 2015; Robinson and Jackson, 2016) and demonstrated that astrocytes exhibit a high rate of oxidative metabolism (Hertz et al., 2007). To better understand how mitochondrial dynamics and distribution were regulated in adult animals in vivo, we generated DNA constructs to drive expression of fluorescent proteins (mitochondrially-targeted enhanced green fluorescent protein (mitoEGFP) and a membrane-targeted tandem-dimer tomato (Lck-TdTomato) fluorescent protein) under control of a minimal glial fibrillary acid protein promoter (GFA_ABC1D; Lee et al., 2008). These were packaged into adeno-associated virus 2 (AAV2) particles that express the coat-protein from AAV5 (AAV2/5) which displays a tropism for astrocytes (Duale et al., 2005). A similar strategy, combining viral tropism with an astrocyte-selective promoter, has been used previously to confine expression of transgenes to astrocytes in vivo (Ortinski et al., 2010; Shigetomi et al., 2013). Viral particles were injected at a total dose of 3×10⁹ genome copies per animal into the superficial layers of somatosensory cortex of adult mice of either sex. The distribution and trafficking of mitochondria in astrocyte processes in vivo were examined two weeks post-injection. As has been observed in organotypic cultures (Benediktsson et al., 2005) or in vivo (Shigetomi et al., 2013; Benjamin Kacerovsky and Murai, 2015), membrane-targeted TdTomato fluorescence revealed the complex morphology of the astrocyte with many fine processes and branchlets (Fig 1A, A’). EGFP-labelled mitochondria were readily visualized throughout the processes of the cortical astrocytes using confocal microscopy of fixed brain tissue (Fig 1B, B’ or overlay Fig 1C, C’) and in vivo using two-photon microscopy (Fig 1D, E), including distal processes and vascular endfeet (see also Mathiisen et al., 2010; Motori et al., 2013; Benjamin Kacerovsky and Murai, 2015). Interestingly, this approach reveals dense labeling of mitochondria in the astrocyte processes along vessel walls. Mitochondria within the astrocyte processes in vivo possessed a similar, but slightly smaller distribution of mitochondrial lengths to that previously observed in hippocampal organotypic cultures (Fig 1F. median length=1.4 vs 1.8 µm, respectively; Jackson et al., 2014b). These contribute to the many converging lines of evidence that mitochondria are present within astrocytic processes in situ and in vivo.

Figure 1. Visualization of mitochondrial distribution in vivo.

DNA constructs were generated to drive expression of a membrane-targeted TdTomato fluorescent protein (Lck-TdTomato) and a mitochondrially-targeted enhanced green fluorescent protein (mitoEGFP) under control of a minimal GFAP promoter (GFA_ABC1D). These were packaged into AAV2 particles containing the coat-protein from AAV5 (AAV2/5; UPenn viral vector core). Viral particles were injected into the superficial cortical layers of adult mice of either sex. (A–C) Representative images (maximal z-projections) from astrocytes transduced with Lck-TdTomato (A, C; magenta) and mitoEGFP (B, C; green). Images A’-C’ are cross-sections from z-projection images above (A–C) that depict a dense mitochondrial network throughout the astrocyte. Individual mitochondria can be seen in the smaller distal processes and are particularly concentrated near vascular end-feet (arrows). (D–E) Representative maximal z-projection images of the vasculature (visualized by injection of dextran conjugated AlexaFluor633 hydrazide into the tail vein; magenta) and mitochondria (mitoEGFP; green) in an anesthetized (α-chloralose) mouse visualized through an acute cortical window via a two-photon microscope, representative of 4 animals). (F) Distribution of mitochondrial lengths in vivo (magenta; new data) compared with mitochondrial lengths in organotypic cultures (green; data from Jackson et al., 2014b).

Recently, we and others have examined aspects of mitochondrial dynamics within astrocyte processes in both acute slices, organotypic cultures, and in vivo (Motori et al., 2013; Jackson et al., 2014b; Jackson and Robinson, 2015; Stephen et al., 2015; O’Donnell et al., 2016). Here we review these studies and try to link them to the broader context of what is known about mitochondrial dynamics. In particular, we will compare the dynamics of mitochondria within astrocytes with the more extensively characterized neuronal population. We try to point out what is yet unknown about the dynamics of mitochondria within astrocytes and to highlight cellular functions in astrocytes that might be impacted by changes in mitochondrial distribution and dynamics.

Mitochondrial Trafficking/Transport

Mitochondria within neurons maintained in culture are actively transported in axons and dendrites (Morris and Hollenbeck, 1995; Saxton and Hollenbeck, 2012). Experiments have consistently shown that 20–40% of mitochondria are mobile within neuronal processes (axons and dendrites; Overly et al., 1996; Chen et al., 2007;for reviews see Cai and Sheng, 2009; MacAskill and Kittler, 2010; Saxton and Hollenbeck, 2012; Schwarz, 2013). The percentage of mobile mitochondria decreases with age or time in culture and varies between brain regions (Lewis et al., 2016; Smit-Rigter et al., 2016). Recently, several groups have examined factors regulating the distribution and transport of mitochondria within the processes of astrocytes in situ and in vivo (Genda et al., 2011; Bauer et al., 2012; Jackson et al., 2014b; Jackson and Robinson, 2015; Stephen et al., 2015; O’Donnell et al., 2016). While many of the features of mitochondrial movement are similar in astrocytes and neurons, there are differences.

Direct comparisons of mitochondrial movement in organotypic hippocampal cultures shows that a greater fraction of mitochondria is mobile in neuronal dendrites than in astrocytic processes (defined at displacement > 2 µm/15min. We found that ~40% of mitochondria in dendrites were mobile over this 15-minute imaging window vs ~15–20% in astrocyte processes (Jackson et al., 2014b). Using a similar preparation, Stephen et al observed greater, but qualitatively similar results (~54% mobile in neurons vs ~31% in astrocytes; Stephen et al., 2015). In order to determine if similar mobility is observed in vivo, we used two-photon imaging to monitor the movement of mitochondria in transduced cortical astrocytes (AAV2/5-GFA_ABC1D-mitoEGFP) in adult mice (Fig 1D, E). Mice were anesthetized (α-chloralose) and the EGFPmito fluorescence visualized through an acute cranial window. As we previously observed in organotypic cultures (Jackson et al., 2014b), the percentage of mobile mitochondria in vivo (displacement > 0.67 µm/5min) was low (9.7 ± 2.5 %; n=4 mice). This observation is consistent with the apparent stability of mitochondrially-dependent [Ca²⁺] microdomains in astrocyte processes in vivo that was recently reported (Agarwal et al., 2017). Whether this reflects a difference in the preparation (organotypic cultures vs in vivo) or a difference in the developmental age of the astrocytes (neonatal vs adult) is an open question. However, others have observed that the mitochondrial movement decreases in neurons in vivo (Lewis et al., 2016; Smit-Rigter et al., 2016). Using 2-photon in vivo imaging of mTurquise2-labeled mitochondria, Smit-Rigter and colleagues report that ~1% of mitochondria are mobile in layer 2/3 pyramidal neuron axons (V1) in either adult or neonatal mice. Similarly, Lewis et al report <5% of mitochondria are mobile in layer 2/3 pyramidal axon collaterals in vivo. While they found that the percentage of mobile mitochondria decreased coincident with axon maturation in vitro, the percentage of mobile mitochondria did not increase from P10-adulthood in vivo.

The movement of mitochondria in neurons is characterized by brief pauses, accompanied by changes in speed and direction (Chang et al., 2006; Macaskill et al., 2009b). We found that the movement of mitochondria in astrocytes is similarly characterized by brief pauses, as well as changes in speed and direction (Jackson et al., 2014b). However, the features of this movement differ between astrocytes and neurons. Stephen et. al. found that in astrocytic processes, mitochondria exhibited a more oscillatory pattern of motion compared to the directed motion observed in neurons. Differences within the rates at which mitochondria move are apparent between astrocytes and neurons as well. We found that the instantaneous velocity in astrocytes (0.15 µm/s anterograde and 0.2 µm/s retrograde) was significantly slower than in neurons (0.55 µm/s anterograde and 0.65 µm/s retrograde; Jackson et al., 2014b). Stephen et. al. (2015) observed a similar difference in the rates of mitochondrial movement. What might explain these differences? In other cells, differences in the rates of movement, destination, and distance traveled have been ascribed to differences in motor apparatus, adaptor proteins, and cytoskeletal elements. In neurons, for example, long range movements is often attributed to kinesin and dynein motors moving along microtubules, while short range movements depend on mysosin-directed movement along the actin microfilaments (Nangaku et al., 1994; Morris and Hollenbeck, 1995; Tanaka et al., 1998; Pilling et al., 2006; Tanaka et al., 2011; reviewed in MacAskill and Kittler, 2010; Schwarz, 2013). In axons of neurons in culture, mitochondria move along both microtubules and actin microfilaments albeit with different velocities (Morris and Hollenbeck, 1995). Even within the kinesin superfamily, different motor proteins move at velocities that range from 0.02 to 1.5 µm/second and vary with respect to processivity (see Martin et al., 1999). Additionally, different adaptor proteins (TRAK1 and TRAK2) associate with different machineries to transport mitochondria to axons or dendrites, with TRAK1 associating with kinesin-1 and dynactin and TRAK2 associating with dynein-dynactin to mediate axonal or dendritic targeting, respectively (van Spronsen et al., 2013). The rates of mitochondrial movement in astrocytes (~0.15 µm/sec), match the rates of mitochondrial movement along actin microfilaments (Morris and Hollenbeck, 1995). Based on these differences, we would speculate that the transport of mitochondria in astrocytes may depend on different motor or adaptor proteins (Jackson et al., 2014b; Stephen et al., 2014; Stephen et al., 2015) than are utilized by neurons (please see subsection on motor and adaptor proteins).

Recently, mitochondria have been observed in the primary process of oligodendrocytes and their myelin sheaths (Rinholm et al., 2016). There are fascinating differences between these mitochondria and those found in astrocytes and neurons. Mitochondria within the oligodendrocyte primary process or myelin sheath are generally smaller (lengths of ~1.2 and 0.8 µm, respectively) than in astrocytes or neurons (~1.4–3 µm). This is particularly interesting as mitochondrial oxidative capacity is thought to vary inversely with size (Bertoni-Freddari et al., 2003; Bertoni-Freddari et al., 2005; Tondera et al., 2009). Supporting the notion of reduced oxidative capacity, Rinholm and colleagues report that putative mitochondria present in the myelin sheath are relatively devoid of cristae, another factor that correlates with respiratory potential (Palade, 1953; Mannella et al., 2001; Rinholm et al., 2016; reviewed in Zick et al., 2009; Mannella et al., 2013). Oligodendrocyte resident mitochondria are significantly less mobile than in either astrocytes or neurons (~12% and ~8% in primary processes or sheaths, respectively) and move slower (<0.1 µm/s; Rinholm et al., 2016). There are very striking differences in the regulation of mitochondrial movement between oligodendrocytes and either astrocytes or neurons. Whereas mitochondrial movement is arrested in both neurons and astrocytes by glutamate and Ca²⁺, the movement of mitochondria in oligodendrocyte processes is enhanced (by ~76%) by application of glutamate. Conversely, removal of Ca²⁺ abolished mitochondrial movement in oligodendrocytes. The authors suggest that these differences may serve an important function in oligodendrocytes; activity dependent dispersal of mitochondria may preserve lactate production for export to axons (Rinholm et al., 2016). The differences in mitochondrial mobility and its regulation further strengthen the notion that different cells may utilize different strategies to position mitochondria.

Role of activity in regulating mitochondrial movement

Given the energetic demands of continuously moving mitochondria, one might ask “what is the purpose of moving mitochondria?”. The coordinated movement of mitochondria along cellular processes probably serves several roles. First, the movement of mitochondria facilitates the distribution of mitochondria into cellular processes. While mitochondrial biogenesis (which involves nuclear transcription as well as mtDNA replication and fission) can occur locally in neuronal processes (Amiri and Hollenbeck, 2008), most mitochondria are thought to be generated in the cell body and subsequently transported in an anterograde direction into the processes (Davis and Clayton, 1996; for review, see Schwarz, 2013). Second, the retrograde movement of mitochondria out of cellular processes probably facilitates the degradation of depolarized, damaged, or aged mitochondria (Frederick and Shaw, 2007). Third, mitochondrial transport facilitates fusion and fission processes that allow the exchange of proteins and genetic material between discreet mitochondria (Misko et al., 2010). Fourth, mitochondrial transport allows for the relatively rapid redistribution of mitochondria, at least in culture, to sites of elevated activity (for reviews, see MacAskill et al., 2010; Schwarz, 2013) such as active synapses or nodes of Ranvier in a neuron or the perisynaptic astrocytic processes where they can provide local energy and regulate Ca²⁺ signaling. Finally, relatively recent evidence suggests a role for mitochondrial trafficking in regulating the transfer of mitochondria between adjacent cells, termed horizontal transfer (Ahmad et al., 2014; Berridge et al., 2016b).

Arguably one of the most important features of mitochondrial movement, particularly in neurons, is its directed distribution and the consequent accumulation of mitochondria at sites of elevated metabolic load and [Ca²⁺]. In neurons, mitochondria actively accumulate in areas of high-energy demand and Ca²⁺ flux including synaptic terminals, nodes of Ranvier and growth cones (Li et al., 2004; Misgeld et al., 2007; Ohno et al., 2011; Bertholet et al., 2013). Blocking mitochondrial redistribution to these areas, either by manipulating mitochondrial transport, decreasing mitochondrial anchoring, or blocking fission (necessary for mitochondrial entry to spines) inhibits synapse formation, axonal/dendritic branching, and maintenance of these structures (Li et al., 2004; Guo et al., 2005; Verstreken et al., 2005; Ishihara et al., 2009; Courchet et al., 2013; Lopez-Domenech et al., 2016). In fact, blocking mitochondrial trafficking (Miro1 knock-out) leads to progressive neurodegeneration in peripheral nerves (Nguyen et al., 2014). Thus, it seems clear that mitochondrial distribution in neurons is important to neuronal maturation and the maintenance of neuronal structures. While it is unknown whether mitochondrial distribution controls the maturation and elaboration of astrocyte or oligodendrocyte processes, it is an intriguing possibility.

The accumulation of mitochondria at sites of elevated activity presumes a signal arresting their movement and retaining them. In cultured neurons, mitochondrial movement is inversely dependent upon neuronal activity. Decreasing neuronal activity (with tetrodotoxin or omega-conotoxin GVIA) increases mitochondrial movement, while conversely increasing neuronal activity (KCl depolarization or activation of NMDA receptors) decreases movement (Rintoul et al., 2003; Li et al., 2004; Mironov, 2006). In addition to increases in local [Ca²⁺], increases in [ADP] subsequent to increased ATP consumption have also been implicated in the arrest of mitochondria at synapses (Mironov, 2006, 2007).

Much of astrocytic activity is tuned to neuronal activity, and in particular excitatory (glutamatergic) activity (Araque et al., 1999; Halassa et al., 2007; Araque and Navarrete, 2010). Using transfected organotypic cultures of rat hippocampus, we found that the percentage of mobile mitochondria in astrocytes was low (~15%). Treatment of the slices with tetrodotoxin, a Na⁺ channel antagonist that blocks action potential generation, increased the percentage of mobile mitochondria to ~45% (Jackson et al., 2014b). Similarly, simultaneous blockade of AMPA, NMDAR, and GABAR receptors (DNQX, APV, bicuculline, respectively) increased the percentage of mobile mitochondria. Recently, work from Joseph Kittler’s lab, confirmed the observation that inhibiting neuronal activity increases mitochondrial mobility in astrocytes (Stephen et al., 2015). As a greater proportion of mitochondria were mobile in their preparations, they were also able to demonstrate that treatment with glutamate or 4-AP (increasing neuronal activity) decreased the percentage of mobile mitochondria (Stephen et al., 2015). Both the Robinson and Kittler laboratories have implicated glutamate in the mechanism of mitochondrial arrest (Jackson et al., 2014b; Stephen et al., 2015), however we differ in some of the mechanistic details of this arrest.

Perhaps one of the most important roles that astrocytes play is in the clearance of glutamate following excitatory neurotransmission. Glutamate, the principal excitatory neurotransmitter in the forebrain, is almost entirely cleared into astrocytes via GLAST and GLT1 (EAAT1 and EAAT2, respectively; Rothstein et al., 1994; Tanaka et al., 1997; for review see Danbolt, 2001). Glutamate uptake is steeply Na⁺ dependent. The glutamate transporters catalyze the movement of glutamate (and aspartate) against its concentration gradient along with 3 Na⁺ ions and a H⁺, and the counter-transport of a K⁺ ion (Zerangue and Kavanaugh, 1996). This large increase in Na⁺ influx activates the Na⁺/K⁺-ATPase (utilizing energy) and causes local reversal of the Na⁺/Ca²⁺-exchanger (NCX), leading to increase in [Ca²⁺] (Pellerin and Magistretti, 1996; Kirischuk et al., 1997; Pellerin and Magistretti, 1997; Rojas et al., 2013). We, and subsequently others, have observed co-compartmentalization of both GLT1 and GLAST with mitochondrial proteins and the Na⁺/K⁺-ATPase in astrocyte processes (Genda et al., 2011; Bauer et al., 2012; Jackson et al., 2014a; Shan et al., 2014). Based on this, and the large depolarization that accompanies glutamate (Na⁺) influx, we hypothesized that mitochondrial movement might be regulated by glutamate uptake. We found that inhibition of glutamate uptake (with TFB-TBOA) or the Na⁺/K⁺-ATPase (with ouabain) increased the percentage of mobile mitochondria in astrocyte processes. Conversely, application of D-aspartate decreased the percentage of mobile mitochondria in the presence of TTX (Jackson et al., 2014b). Further, two structurally dissimilar inhibitors of reversed-mode Na+/Ca²⁺-exchange both increased the percentage of mobile mitochondria. These results demonstrate that activation of glutamate transport is necessary and sufficient to stop mitochondria moving and suggest that increases in [Ca²⁺] might underlie their arrest.

By contrast, Stephen et al (2015) implicate activation of NMDARs on astrocytes in the arrest of mitochondrial movement. In hippocampal slice cultures, they found that glutamate decreased the percentage of mobile mitochondria. This decrease was not blocked or only partially blocked by inhibiting AMPA receptors (with 2,3-Dioxo-6-nitro-1,2,3,4-tetrahydrobenzo[f]quinoxaline-7-sulfonamide; NBQX) or mGluRs (with α-Methyl-4-carboxyphenylglycine; MCPG), respectively. However, the decrease in mobility was blocked by the NMDAR antagonist D-APV or by removing extracellular Ca²⁺. They also found that MK801 was able to block glutamate-induced mitochondrial shortening in slices or in pure astrocyte cultures (Stephen et al., 2015). While both groups broadly agree that glutamate is decreasing mitochondrial movement, and that this proceeds through increases in [Ca²⁺], we differ in mechanisms by which glutamate mediates this [Ca²⁺] increase. These differences could reflect differences in culture preparation. Alternatively, perhaps transporter mediated-depolarization facilitates opening of NMDA receptors on astrocytes. We would suggest that a possible resolution to this would involve knocking out NMDARs specifically in astrocytes.

The Ca²⁺-dependent arrest of mitochondria is thought to facilitate the activity-dependent positioning of mitochondria and underlie the observed steady-state accumulation of mitochondria at synapses and other active regions (Macaskill et al., 2009b; Wang and Schwarz, 2009). Like in neurons, mitochondria in astrocytes are not uniformly distributed. Previously, we observed that mitochondria co-localize with clusters of glutamate transporters in the processes of transfected astrocytes; a phenomenon that occurred more than would be predicted by chance (Genda et al., 2011). This accumulation depends on neuronal activity. We examined the spatial relationship of individual mitochondria with glutamate transporter puncta and synapses in the presence and absence (+TTX) of neuronal activity. We found that ~80% of mitochondria were within 1 µm of an endogenous glutamate transporter puncta (and VGLUT1 positive pre-synaptic terminal). In the absence of neuronal activity (+TTX) the distance between mitochondria and glutamate transporters (and synapses) increased (Jackson et al., 2014b). In co-cultures of hippocampal neurons and astrocytes, glutamate application or electrical field stimulation decreased the mean distance between astrocytic mitochondria and VGLUT+ puncta, as did application of 4-AP in hippocampal slices (Stephen et al., 2015). Thus, multiple lines of evidence from multiple labs suggest that neuronal activity increases the apposition of astrocytic mitochondria with synapses.

Role of motor proteins and adaptors

The movement and distribution of mitochondria to active cellular compartments, whether into the synapse in neurons or perisynaptic processes of astrocytes, is likely determined by their engagement with the cellular transport machinery (reviewed in MacAskill and Kittler, 2010; Saxton and Hollenbeck, 2012; Schwarz, 2013). While, neurons can utilize both the actin and microtubule cytoskeletons for the directed transport of mitochondria, long-range transport proceeds mainly along the microtubule cytoskeleton (reviewed in Hollenbeck and Saxton, 2005; MacAskill et al., 2009a; Schwarz, 2013; Maday et al., 2014; see Fig 2 for model). Microtubules are polymers comprised of α and β-tubulin dimers joined end-to-end to form a polarized structure with β (+) and α (−) ends (Watanabe et al., 2005). In axons, microtubules display uniform polarity with the β (+) end oriented distally toward the axon terminal while they display a mixed orientation in dendrites (Baas et al., 1988). Most plus-end mediated transport is mediated by kinesin superfamily motors, while minus-end mediated transport is via cytoplasmic dynein (Pilling et al., 2006; Hirokawa et al., 2009). Thus, in axons, most anterograde movement is via kinesin motors while retrograde movement is via dynein. In dendrites, because of mixed microtubule polarity, both kinesin and dynein motors can operate in either the retrograde or anterograde direction. The orientation of the microtubule cytoskeleton in astrocytes is not known. However, mitochondria move at similar rates and in similar proportions in both the anterograde and retrograde directions (Jackson et al., 2014b; Stephen et al., 2015), perhaps implying a mixed orientation, as in dendrites.

Figure 2. Cartoon model depicting mitochondrial trafficking in astrocytes.

Neuronal activity increases the concentration of glutamate (and other transmitters) at the synaptic cleft. Glutamate is cleared by a family of Na⁺-dependent glutamate transporters, in particular, GLAST and GLT-1 (EAAT1 and EAAT2), that reside on the perisynaptic astrocyte membranes. Activation of glutamate uptake activates Na⁺/K⁺-ATPase isoforms and leads to a reversal (Na⁺ out, Ca²⁺ in) of the Na⁺/Ca²⁺ exchanger (NCX) in the astrocyte processes, thus consuming ATP and increasing local [Ca²⁺], respectively. In addition, astrocytes possess receptors for numerous neurotransmitters (glutamate, ATP, etc) that couple to increases in [Ca²⁺]_I increases. Increases in Ca²⁺ result in the arrest of mitochondria within astrocyte processes via a Ca²⁺-dependent disengagement of Miro proteins from the motor apparatus. The identity of motor proteins mediating mitochondrial movement in astrocytes is unknown.

What motor proteins are responsible for the movement of mitochondria in astrocytes? The kinesin family of motors is relatively large, including at least 45 members (Hirokawa et al., 2009). Several kinesin family members have been implicated in the transport of mitochondria. In neurons, most plus-end mediated movement is mediated by Kif5 (kinesin-1) superfamily members (MacAskill et al., 2010). Of the three Kif5 superfamily members, Kif5a and Kif5c are principally neuronal, while Kif5b is ubiquitously expressed (Nangaku et al., 1994; Tanaka et al., 1998; Kanai et al., 2000; Zhang et al., 2014). Deletion of Kif5b leads to the perinuclear accumulation of mitochondria (Tanaka et al., 1998) in cultured extraembryonic cells, whilst expression of a KIF5B mutant lacking the motor domain blocks transit of mitochondria into neurites in Neuro2 cells (Tanaka et al., 2011). In addition to Kif5, other kinesin family members have been associated with mitochondrial transport. Kif1B (kinesin-3) localizes to mitochondria in neurons and purified protein can transport mitochondria along microtubules in vitro. Kinesin-like protein 6 (Klp6) mutants inhibit anterograde transport of mitochondria in differentiated neuro 2a. By contrast, only one dynein type motor, cytoplasmic dynein, has been implicated as the primary retrograde mitochondrial motor (Pilling et al., 2006). It is, as yet, unknown which motor protein(s) mediates mitochondrial movement in astrocytes.

The movement of mitochondria along the microtubule cytoskeleton depends on several adaptor proteins that link mitochondria to the motor proteins (Glater et al., 2006; for review see MacAskill and Kittler, 2010; Sheng and Cai, 2012; Schwarz, 2013). Amongst these are the mitochondrial Rho proteins (Miros; also known as RhoTs) and the trafficking kinesin binding proteins (TRAKs, known as Milton in drosophila). Mutations in Milton lead to loss of mitochondria from photoreceptor terminals and blindness in drosophila (Stowers et al., 2002). The Milton/TRAK proteins bind kinesin motors (KIF5) and Miro protein (Glater et al., 2006; Smith et al., 2006; Macaskill et al., 2009b). The Miro proteins (Miro1 and Miro2) possess an n-terminal outer mitochondrial membrane anchoring motif, two EF-hand Ca²⁺ binding domains, and 2 Rho-GTPase domains (Fransson et al., 2003; Fransson et al., 2006). In neurons and other cells the Ca²⁺-dependent arrest and accumulation of mitochondria in response to elevated [Ca²⁺] is thought to be largely dependent on the Miro proteins (Saotome et al., 2008; Macaskill et al., 2009b; Wang and Schwarz, 2009). Mutations in the Ca²⁺ binding domain of Miro proteins block the Ca²⁺-dependent arrest of mitochondrial movement. In neurons, both sets of proteins are important for correct trafficking of mitochondria to processes (Glater et al., 2006; MacAskill et al., 2009a; Brickley and Stephenson, 2011; van Spronsen et al., 2013).

Similar roles for the TRAK and Miro proteins in astrocytes have recently started to be appreciated. In primary cultures of astrocytes, the TRAK proteins are not expressed. Upon co-culture with neurons, astrocytes increase expression of TRAK2, allowing transit of mitochondria into astrocyte process (Ugbode et al., 2014). Both Miro1 and Miro2 are expressed in astrocytes and localize to mitochondria within the processes of astrocytes (Zhang et al., 2014; Jackson and Robinson, 2015; Stephen et al., 2015; reviewed in Stephen et al., 2014). As in neurons, the Miros appear to mediate the Ca²⁺ sensitive arrest of mitochondria within the processes of astrocytes in response to neuronal activity and glutamate (Jackson et al., 2014b; Jackson and Robinson, 2015; Stephen et al., 2015). Exogenous expression of wild-type Miro proteins results in an increase in the percentage of mobile mitochondria (Jackson and Robinson, 2015; Stephen et al., 2015), suggesting that Miro availability might limit mobility in astrocytes. Expression of Ca²⁺-insensitive Miro1 or Miro2 mutants increase the percentage of mobile mitochondria in astrocyte processes and blocks the arrest of mitochondria in response to increased Ca²⁺ signaling (Jackson and Robinson, 2015; Stephen et al., 2015).

In addition to regulating mitochondrial movement, Miro proteins may play additional roles in regulating mitochondrial function and signaling. Mitochondria closely approach endoplasmic reticulum (ER) Ca²⁺ release channels and form physical structures with the ER in multiple cell types (Rizzuto et al., 1993; Rizzuto et al., 1994; Rizzuto et al., 1998; Csordas et al., 1999; Csordas et al., 2006). Close apposition of the mitochondria and ER facilitates Ca²⁺ transfer between these organelles (Csordas et al., 2006). Miro proteins may be involved in the formation of tethers linking the ER with mitochondria. Gem1, a Miro homologue found in yeast, was identified as a regulator of the endoplasmic reticulum-mitochondrial encounter structure (ERMES; Kornmann et al., 2011). This complex, which includes Fis1 and Mfn2, forms a physical link between the ER and mitochondria and may facilitate lipid exchange between the organelles. The existence of mitochondrial/ER tethers has not been explored in astrocytes, however there is evidence that the ER exists within the astrocyte processes (Spacek, 1982; Benjamin Kacerovsky and Murai, 2015).

Like in neurons, the majority of mitochondria in astrocytes are, in fact, immobile (Morris and Hollenbeck, 1993, 1995; Miller and Sheetz, 2004; Kang et al., 2008; Jackson et al., 2014b; Stephen et al., 2015). We found that up to 90% of mitochondria were immobile over a 15 minute viewing window. Even in the absence of neuronal activity, (TTX) the percentage of mobile mitochondria was not 100% (Jackson et al., 2014b; Stephen et al., 2015). This suggests either that other processes regulate mitochondrial mobility in astrocytes or that the mitochondria might be physically tethered to domains within astrocytes. Such physical tethering/anchoring proteins have been identified in neurons. One of these, syntaphilin, a microtubule binding protein, has been suggested to mediated the permanent capture/immobilization of mitochondria within axons (Kang et al., 2008). However, syntaphilin mRNA is absent from astrocytes (Zhang et al., 2014). The existence of different proteins regulating this immobile pool of mitochondria in astrocytes is an interesting area for future exploration.

In addition to the microtubule cytoskeleton, the actin cytoskeleton and myosin motors contribute to the distribution of cellular cargo. In contrast to microtubule-based transport, the myosin family of proteins are the only known actin-based motor (reviewed in Hartman and Spudich, 2012). While the motor/head domain of the myosin proteins is largely conserved, many of these proteins possess variable tail domains that enable specificity with respect to the binding of different cargos or adaptor proteins (reviewed in Boldogh and Pon, 2006; Akhmanova and Hammer, 2010).

While most long-range mitochondrial movement in neurons depends on microtubule-based motors, there is evidence, sometimes conflicting, of mitochondrial interactions with the actin cytoskeleton and myosin motors (Bradley and Satir, 1979; Morris and Hollenbeck, 1993; Ligon and Steward, 2000; Saxton and Hollenbeck, 2012). In neurons, it has been hypothesized that mitochondrial movement might proceed along actin filaments where microtubules might be sparse, such as at active growth cones (Bradley and Satir, 1979). In cultures of hippocampal neurons, Ligon and Steward found that mitochondrial movement was arrested by treatment with cytochalasin D, which aggregates actin filaments, although latrunculin (an actin polymerizing agent) had no effect (Ligon and Steward, 2000). In cultures of dorsal root ganglion neurons, mitochondria accumulate near to nerve growth factor (NGF) beads. This accumulation is blocked by Latrunculin B without blocking overt mitochondrial transport (Chada and Hollenbeck, 2004); suggesting a role for the actin cytoskeleton in anchoring/retaining mitochondria.

Recently, several myosin motors have been identified that associate with mitochondria and influence mitochondrial dynamics in other cell types. Myo 19 is anchored to the outer-mitochondrial membrane through its c-terminus. Its overexpression resulted in a nearly 2-fold increase in mitochondrial mobility in A549 cells (an epithelial cell line), while a dominant-negative tail domain mutant decreased movement (Quintero et al., 2009). Myo19 also facilitates the localization of mitochondria to filopodial extensions in U2OS cells (Shneyer et al., 2016). RNAi-mediated knockdown of Myo19 decreased filopodial size and number. Moreover, mitochondria were absent from the remaining filopodia, suggesting that myo19 might regulate filopodial extension and mitochondrial transit to those filopodia. By contrast, RNAi-mediated knockdown of myosin IV and V increased both the percentage of mobile mitochondria and their velocity in cultured drosophila neurons (Pathak et al., 2010), suggesting that myosin motors might oppose mitochondrial movement in neurons. In astrocytes, we found that application of cytochalasin D or vinblastine decreased the percentage of mobile mitochondria, suggesting that both the actin and microtubule cytoskeletons, respectively, contribute to the movement of mitochondria in astrocytic processes (Jackson et al., 2014b). Further, while only suggestive, the rates of mitochondrial movement in astrocytes (~0.15 – 0.2 µm/sec), match the rates of mitochondrial movement along actin microfilaments (Morris and Hollenbeck, 1995). The identity of the motor proteins, whether microtubule or actin resident, that mediate the movement of mitochondria in astrocyte processes, and their relative contributions to that movement remains unknown.

In addition to actin and microtubule networks, intermediate filament (IF) proteins constitute part of the cytoskeleton. Unlike the actin or microtubule cytoskeletons, there are no known motors that operate along intermediate filaments. Astrocytes express three types of intermediate filament proteins (vimentin, nestin, and glial fibrillary acid protein (GFAP); Pekny and Pekna, 2004). Vimentin and GFAP, in particular, increase dramatically during reactive astrocytosis (Eng et al., 1987; Chen et al., 1993; Janeczko, 1993). Intermediate filaments associate with mitochondria in a number of cell types, including muscle and neurons (Mose-Larsen et al., 1982; Summerhayes et al., 1983; Linden et al., 1989; Leterrier et al., 1994; Tang et al., 2008). In fact, vimentin directly bind to mitochondria (Nekrasova et al., 2011). Further, the association of vimentin with mitochondria regulates its movement and function. Disruption of vimentin IFs increases mitochondrial movement and alters the distribution of mitochondria in cell lines and fibroblasts (Nekrasova et al., 2011). shRNA mediated knockdown of vimentin decreases mitochondrial membrane potential (Tang et al., 2008; Chernoivanenko et al., 2015). It is unclear whether changes in vimentin or GFAP expression in astrocytes might alter mitochondrial dynamics or function in astrocytes, however Motori et al (2013) found that pro-inflammatory stimuli in astrocyte cultures (LPS + IFNγ) or in vivo (cortical stab wound) elicited transient changes in mitochondrial dynamics favoring fission over fusion (discussed in the next section).

Fission/Fusion

While we often conceptualize mitochondria as discrete organelles, the reality is more complicated. Mitochondria in most cells exist as a complex continuum between a reticular network of connected mitochondria and discrete organelles. The shape and interconnectedness of this network varies between cells, within cells, and is dynamic. The mitochondrial network is characterized by the near constant division and condensation of individual mitochondria. These opposing activities, termed fission and fusion, determine the size and shape of the mitochondrial network in eukaryotic cells and, along with movement, determine its distribution. These structural rearrangements also facilitate the exchange of mitochondrial components, including mtDNA, lipids, and proteins (Parone et al., 2008; Ban-Ishihara et al., 2013; Mishra and Chan, 2014; Ishihara et al., 2015; Bertholet et al., 2016), and contribute to homogenization of the mitochondrial pool.

Mitochondria are membrane encapsulated organelles that possess both inner and outer membranes. Mitochondrial fusion, in which two or more mitochondria condense into one larger mitochondrion, involves the merger of both the inner and outer mitochondrial membranes of the parent mitochondria. Mitochondrial fusion is controlled by the mitofusin proteins (Mfn1, and Mfn2) and optic atrophy 1 (OPA1; Santel and Fuller, 2001; for review see Chen and Chan, 2009; Liesa et al., 2009). Fission, by contrast, describes the division of a single mitochondria into two organelles. Mitochondrial fission is controlled by dynamin-like protein 1, (DRP1) and fission 1 protein (Fis1; Chang and Blackstone, 2010). The balance between mitochondrial fission and fusion events determines the shape, connectivity, distribution, and density of mitochondria within neuronal, and presumably, astrocytic processes. This balance between fission/fusion is controlled both at the level of expression of these proteins and by post-translational modification (e.g. phosphorylation, ubiquitination, etc; Cribbs and Strack, 2007; Ziviani et al., 2010; for review see Flippo and Strack, 2017).

Mitochondrial fission and fusion are critical for proper development of the nervous system. Neural cell-specific ablation of the mitochondrial fission factor Drp1 is embryonic lethal in mice (Ishihara et al., 2009). Primary cultures of forebrain neurons derived from Drp1 knockout mice possess fewer neurites and display defective synapse formation (Ishihara et al., 2009). Mutations in the genes encoding the fusion-related proteins MFN2 and OPA1 are associated with neurodegenerative diseases. Mutations in MFN2 lead to Charcot-Marie-Tooth subtype 2A (Zuchner et al., 2004), while mutations in OPA1 lead to autosomal optic atrophy (Alexander et al., 2000; Delettre et al., 2000; Delettre et al., 2002). Mitochondrial fragmentation and perturbations in mitochondrial fission and fusion have been observed in other neurodegenerative disorders, including amyotrophic lateral sclerosis (ALS), Huntington’s and Parkinson’s disease, however the precise role that these events play in the etiology of these disorders is yet unclear (Calkins et al., 2011; Shirendeb et al., 2011; DuBoff et al., 2012; Shirendeb et al., 2012; for reviews see Reddy et al., 2011; Anne Stetler et al., 2013; Hroudova et al., 2014; Burte et al., 2015; Flippo and Strack, 2017).

In neuronal somata, mitochondria often resemble a reticular network, while mitochondria, particularly within the axons, exist as discrete organelles. The distribution of mitochondria within astrocytes appears to mirror that of neurons, with cell body resident mitochondria resembling the reticular network seen in neuronal soma (Li et al., 2004; Popov et al., 2005; Jackson et al., 2014b), whilst mitochondria in processes are discrete, vermiform structures ranging from 0.2–6 µm in length (Genda et al., 2011; Motori et al., 2013; Jackson et al., 2014b; Derouiche et al., 2015; Stephen et al., 2015; O’Donnell et al., 2016). Recently, we, and others, demonstrated mitochondria distributed throughout the astrocytic arborization, even into the small distal processes where the volume of the mitochondria occupies almost the entire process (Genda et al., 2011; Motori et al., 2013; Jackson et al., 2014b; Stephen et al., 2015; see also Fig 1). Indeed, the existence of a population of diminutive mitochondria (0.2 µm in diameter) in the very distal tips of the peripheral astrocyte processes was described in astrocyte cultures (Derouiche et al., 2015). The small size and discrete nature of these mitochondria would suggest a prominent role for Drp1-mediated mitochondrial fission in astrocyte processes. There is a precedent for this. In neurons, activity drives mitochondrial fission and subsequent recruitment of very small mitochondria (<1 µm) to spines and filopodia (Li et al., 2004). Drp1 inhibition reduces the number of mitochondria that are present in dendrites (Li et al., 2004) and synaptic terminals (Verstreken et al., 2005). Thus, Drp1 mediated mitochondrial fission facilitates the entry of mitochondria into the distal processes of neurons, perhaps it serves a similar role in astrocytes.

Using viral expression of a mitochondrially-targeted photoconvertible fluorescent protein, mito-Dendra, in organotypic hippocampal cultures, Stephen et al (2015) showed evidence for mitochondrial fusion on a constitutive basis in the astrocyte processes. While, no direct comparison of the fission/fusion dynamics between astrocytes and neurons has been conducted, the slower movement and greater percentage of immobile mitochondria in astrocyte processes (relative to neurons), as well as the often-tortuous geometry of these processes, might suggest that mitochondrial fusion (and exchange of mitochondrial content) might occur less frequently than in neurons. Limited fusion-mediated exchange of mitochondrial content would be expected to manifest as heterogeneity of individual mitochondria within astrocyte processes. Supporting this notion, α-ketoglutarate dehydrogenase is heterogeneously distributed in mitochondria within individual astrocytes in primary culture (Waagepetersen et al., 2006).

As with trafficking, the dynamics and morphology of mitochondria in astrocytes appears to be regulated by neuronal activity (glutamate). Bath application of glutamate decreased mitochondrial fusion in astrocytes in organotypic cultures. Whether this was a primary effect on fusion or secondary to decreases in mitochondrial transport is unclear (Stephen et al., 2015). In addition to decreasing fusion, increasing neuronal activity (4-AP) or glutamate decreased the length of mitochondria from ~3 µm to ~1µm (Stephen et al., 2015). This is similar to what was observed in primary cultures of hippocampal neurons in response to glutamate (Rintoul et al., 2003).

Astrocytes respond to a wide range of noxious stimuli or damage by a set of changes in morphology and expression patterns that are collectively termed “reactivity”. In the case of prolonged or severe injury this reactivity can be accompanied by the proliferation of astrocytes (gliosis; reviewed in Sofroniew, 2009). These charges vary according to the type of stimulus and can be beneficial or harmful (Bush et al., 1999; Menet et al., 2003; Okada et al., 2006; Zamanian et al., 2012). Changes in metabolic pathways have been observed in the response of astrocytes to inflammation and tissue injury (Hamby et al., 2012; Zamanian et al., 2012). In addition, changes in mitochondrial morphology and mitochondrial dynamics have been observed post injury. In organotypic hippocampal cultures oxygen-glucose deprivation (OGD) leads to decreased mitochondrial length and increased mitochondria number (due to either increased fission or decreased fusion) beginning several hours following the insult. This is followed by the loss of ~50% of mitochondria from the astrocyte processes (O’Donnell et al., 2016). Similarly, cortical stab wound or inflammatory stimuli (LPS + IFNγ) result in large scale remodeling of mitochondrial networks in astrocytes in vivo and in situ, respectively, favoring fission over fusion. These effects are mediated by phosphorylated Drp1 and result in decreases in respiratory capacity (Motori et al., 2013). Along a similar line, blocking reactive astrocytosis, by disruption of astrocytic STAT3, decreases mitochondrial function and increases mitochondrial oxidative stress (Sarafian et al., 2010). Conversely, others have found that increases in inflammation stimulate mitochondrial oxidative metabolism and mitogenesis in astrocytes (Jiang and Cadenas, 2014). It is clear, however, that the large scale structural and phenotypic remodeling of astrocytes to injury and inflammation is associated with similar changes in the mitochondrial network.

Remodeling of the mitochondrial network plays a role in the response of astrocytes to injury and inflammation. Following pilocarpine induced status epilepticus (SE), Ko et al (2017) observed apoptosis of astrocytes in the molecular layer of the dentate gyrus and clasmatodendrotic astrocytes in area CA1 (characterized by round edamotous cell bodies, short processes, loss of distal processes, nuclear dissolution, and LAMP-1⁺ lysozomes) of rat hippocampus. This astrocytic damage was accompanied by regionally-distinct mitochondrial remodeling, with mitochondrial length decreasing in dentate gyrus but increasing in area CA1 (Ko et al., 2016). Pharmacologic inhibition of mitochondrial fission (Mdvi-1) attenuates, while enhancing fission (WY14643) increases, the cell death seen in the dentate gyrus post-SE. Increasing mitochondrial fission increased astrocytic autophagy in area CA1. In related work, SE-evoked increases in astrocytic death and reactive astrocytosis were attenuated by inhibitors of cyclin-dependent kinase 5 (CDK5) acting upstream of Drp1-s616 phosphorylation (Hyun et al., 2017).

Similarly, changes in the fission/fusion dynamics and the astrocytic mitochondrial network are observed in several neurodegenerative diseases. Drp1, the human homologue of DLP1, is decreased in astrocytes of Parkinson’s disease patients (Hoekstra et al., 2015). SiRNA-mediated knockdown of Drp1 leads to a hyperfused mitochondrial morphology, altered Ca²⁺ signaling and decreases glutamate (aspartate) uptake (by ~30%). This loss of glutamate uptake is associated with an impaired ability of astrocytes to protect dopaminergic neurons in culture against the addition of excess glutamate (Hoekstra et al., 2015).

The mechanisms regulating mitochondrial transport, mitogenesis, and fission/fusion do not operate in isolation. There are considerable interactions between the fission/fusion apparatus and various transport-mediating proteins. Dynein and dynactin recruit the fission related protein Drp1 to the outer mitochondrial membrane (Varadi et al., 2004). The fusion protein Mfn2 physically interacts with Miro proteins (Misko et al., 2010) and knock-out of MFN2 or expression of disease-causing variants of Mfn2 decrease axonal mitochondrial transport (Baloh et al., 2007; Misko et al., 2010) in neurons. In addition, mutations disrupt axonal mitochondrial positioning and promote axon degeneration (Misko et al., 2012). Conversely, proteins that are conventionally associated with mitochondrial transport can influence mitochondrial fusion/fission and shape. Expression of Miro1/2 isoforms harboring null mutations in their GTPase or Ca²⁺-binding domains results in increases in mitochondrial length (Fransson et al., 2006; Saotome et al., 2008). Similarly, expression of Ca²⁺-insensitive mutants of Miro1 or Miro2 modestly increase the length of mitochondria in astrocytes (Jackson et al., 2014b; Stephen et al., 2015). Myosin V increases length of axonal mitochondria (Pathak et al., 2010). The coordination of mitochondrial trafficking and fission/fusion dynamics likely contributes to appropriate distribution of mitochondria within neuronal and astrocytic processes.

Quality control/mitophagy

Depolarized or damaged mitochondria are degraded through a specialized, cargo specific form of macroautophagy termed mitophagy (Lemasters, 2005; reviewed in Youle and Narendra, 2011; Wong and Holzbaur, 2015). Mitophagy is often preceded by the arrest of mitochondrial movement and mitochondrial fission (Twig et al., 2008), possibly to isolate individual dysfunctional mitochondria from the larger network and to make mitochondrial fragments small enough to be engulfed by the autophagosome apparatus.

PTEN-induced putative kinase protein (Pink1) and the cytosolic ubiquitin E3 ligase, Parkin (PARK2) play important roles in the initialization of this pathway. Mitochondrial damage or collapse of the mitochondrial matrix potential (Δψm) results in the accumulation of Pink1 (due to loss of inner-mitochondrial membrane localized protease presenilin-associated rhomboid-like protein (PARL) -mediated Pink1 degradation) and the subsequent recruitment of parkin from the cytosol to the outer mitochondrial membrane (Jin et al., 2010; Deas et al., 2011; Greene et al., 2012; Meissner et al., 2015; Reviewed in Wong and Holzbaur, 2015). Mutations in the genes encoding parkin and Pink1 have been linked to forms of familial Parkinson’s disease (Kitada et al., 1998; Valente et al., 2004b; Valente et al., 2004a . Parkin catalyzes the ubiquitination of various substrates on the outer mitochondrial membrane, including Drp1 (Wang et al., 2011), VDAC (Geisler et al., 2010), Miro (Weihofen et al., 2009; Liu et al., 2012), and Mfn1/2 (Poole et al., 2010; Ziviani et al., 2010). This is followed by autophagosome formation and the engulfment and degradation of damaged mitochondria (reviewed in Narendra et al., 2012; Wong and Holzbaur, 2015).

In neurons, autophagasomes are continuously formed in the distal tips of neurites and are transported retrogradely to the cell body (Maday and Holzbaur, 2012; Maday et al., 2012; Ashrafi et al., 2014). A small percentage (~10%) of these autophagosomes contain mitochondrial fragments (Maday et al., 2012), suggesting a constitutive, but low level turnover of mitochondria in neuronal processes. Similarly, basal autophagosome formation (and mitophagy) appears to be low in astrocyte processes (Motori et al., 2013; O’Donnell et al., 2016). In astrocyte processes in vivo, Motori et al observed that ~1% of GFP+ mitochondria co-localized with LC3BII (a marker of autophagosomes). Cortical stab wound resulted in an increase in autophagosome formation, increased mitochondrial fragmentation, and resulted in a 3-fold increase in the percentage mitochondria that co-localize with the autophagosome marker (Motori et al., 2013). Using organotypic hippocampal cultures, we found that oxygen-glucose deprivation (OGD) increased the colocalization of mitochondria with the autophagosome marker LC3B and caused a delayed loss of mitochondria (~50%) from astrocytic processes between 8–24 hrs post injury (O’Donnell et al., 2016). These results are consistent with observation in neurons demonstrating (mito)autophagosome generation in distal axons in response to damaged mitochondria (Ashrafi et al., 2014). It remains to be seen whether autophagosome maturation is local or whether these organelles are transported to the soma (retrograde) for degradation in astrocytes.

In neurons and other cells, removal of damaged or dysfunctional mitochondria is thought to protect the cell from oxidative damage (Loeb et al., 2005). In astrocytes, blocking of mitophagy is associated with an inability to re-establish the mitochondrial network and increased cell death. Blocking autophagosome induction (either by knockout of ATG7 or expression of a dominant-negative ATG4B mutant), altered mitochondrial fission/fusion balance, increased ROS formation, and increased inflammation-mediated astrocyte cell death (Motori et al., 2013). Genetic ablation of Pink1 decreases mitochondrial function (decreased mitochondrial mass, Δψm, ATP production, and increased ROS generation) and attenuates astrocyte proliferation in primary cultures of astrocytes (Choi et al., 2013). Given the role of astrocytes in coordinating neuronal metabolism, survival, and synaptic maintenance, it will be interesting to determine the extent to which mitochondrial quality control in astrocytes contributes to neuronal death and disease progression in neurodegenerative diseases.

Transcellular/ intercellular transfer of mitochondria

Until recently, the conventional view of mitochondrial inheritance was that mitochondria are transmitted during cell division, with newly generated mitochondria split between the mother and daughter cells. However, recent studies have demonstrated the existence of other mechanisms that allow the transfer of mitochondria between cells in the absence of cell division. This process, termed horizontal transfer, can be broadly placed into two categories, those favoring degradation of mitochondria and those that allow incorporation of mitochondrial DNA and proteins into the existing mitochondrial network of the acceptor cell (for reviews see Las and Shirihai, 2014; Berridge et al., 2016b; Torralba et al., 2016).

The first report of horizontal mitochondrial transfer involved the rescue of mitochondria-depleted cells by transfer of healthy mitochondria from human stem cells (Spees et al., 2006). A549 cells were exposed to ethidium bromide to mutate and deplete mtDNA, rendering them incapable of aerobic respiration and growth unless grown in a permissive medium containing uridine and pyruvate. Co-culture with non-haematopoetic progenitor cells from human bone marrow (hMSCs) resulted in a rescue of the A549 cells. Genetic analysis demonstrated that rescued cells contained mtDNA from donor cells that correlated with increase in mitochondrial activity (decreased production of extracellular lactate, decreased levels of reactive oxygen species, increased [ATP], and increased oxygen consumption). Since this initial report, multiple groups have demonstrated that the horizontal transfer of mitochondria can occur in a number of cell types both in culture and in vivo (Islam et al., 2012; Pasquier et al., 2013; Ahmad et al., 2014; Tan et al., 2015; Hayakawa et al., 2016; reviewed in Berridge et al., 2016b; Torralba et al., 2016). A common feature involved in the horizontal transfer of mitochondria is cell stress and loss of mitochondrial function in the acceptor cell (Torralba et al., 2016).

Several mechanisms have been proposed to underlie the transfer of mitochondria between cells, including extracellular vesicles (exosomes and microvesicles), membrane evulsions (during transcellular mitophagy) and tunneling nanotubes (see Torralba et al., 2016). Extracellular vesicles have been suggested to facilitate the transfer of proteins (cytosolic and membrane-bound), lipids, and RNA, between cells (reviewed in Budnik et al., 2016). Exosomes (30–100 nm microvesicles) containing mtDNA have been isolated from astrocyte or glioblastoma-conditioned medium (Guescini et al., 2010). In addition, several groups have isolated larger microvesicles (0.3–8 µm) that contain intact mitochondria (Falchi et al., 2013; Hayakawa et al., 2016). In addition, mitochondria can be transferred along bridging structures termed tunneling nanotubes (Rustom et al., 2004). These are 50–100 nM long structures containing actin and microtubules that are assembled between cells in response to cell stress. Interestingly, transfer of mitochondria along tunneling nanotubes requires Miro1. Knockdown of Miro1 decreases the transfer of mitochondria from mesenchymal stem cells to injured (rotenone-treated) epithelial cells, while overexpression increases mitochondrial transfer and facilitates functional rescue (Ahmad et al., 2014; reviewed in Las and Shirihai, 2014 ).

Recently, Hayakawa et al proposed that mitochondria derived from astrocytes might serve to rescue neurons whose mitochondria are damaged, such as occurs following stroke. As others have observed (Falchi et al., 2013), they described microvesicles (300–1100 nm diameter) containing functional mitochondria and displaying surface expression of β1-integrin in medium from primary cultures of astrocytes. In primary cultures of cortical neurons, they observed loss of [ATP] and neuronal viability following oxygen-glucose deprivation. This was rescued by addition of astrocyte-conditioned medium containing mitochondrial-laden microvesicles, but was not mimicked by addition of ATP-containing liposomes or microvesicle-depleted medium. Using confocal microscopy, they demonstrated that “treatment” of OGD injured neurons with astrocyte results in an accumulation of astrocytic mitochondria (labelled with mitoTracker Red CMXRos) in neuronal soma and axons in culture. Some of these transferred mitochondria colocalize with neuronally born mitochondria (labelled with CellLight Mitochondria-GFP), suggesting fusion and incorporation of astrocyte-derived mitochondria into the pre-existing neuronal mitochondrial network (Hayakawa et al., 2016; although see Berridge et al., 2016a for caveats). While the mechanisms involved in mitochondrial release and subsequent neuronal internalization are not yet clear, this involves cyclic ADPribose-mediated Ca²⁺ increases in astrocytes and integrin signaling (Hayakawa et al., 2016). This mechanism has been proposed to provide healthy mitochondria to neurons following stroke aid in their functional recovery.

Several lines of evidence suggest that astrocytes might serve as acceptors for mitochondria destined for degradation. Davis et al recently identified a new mechanism that might mediate the degradation of axonal mitochondria from retinal ganglion cells in astrocytes, which they termed transmitophagy. Using serial block-face scanning electron microscopy, they identified axonal protrusions filled with mitochondria that appose astrocytic processes. These protrusions are pinched off from the axons to form membrane enclosed evulsions that are surrounded by axonal cytoplasm. These axonally-derived evulsions containing mitochondria appear to be degraded via fusion with the lysosome in astrocytes (Davis et al., 2014).

How widespread is this? Davis and colleagues suggest that the degradation of retinal ganglion cell mitochondria in the optic nerve head astrocytes might exceed that seen within the retinal ganglion cell soma (Davis et al., 2014). In addition, they report the existence of morphologically similar structures (mitochondrial filled evulsions) within superficial layers of cortex, although with a significantly decreased frequency. The astrocyte-mediated phagocytosis of membranous evulsions though is reminiscent of astrocyte-mediated phagocytosis and removal of synapses identified in the retinogeniculate system (Chung et al., 2013). As noted by Davis et al, it is possible that the molecular machinery for both activities is conserved.

Role of mitochondria in astrocytic processes

Like neurons, astrocyte processes are decorated with receptors and transporters for numerous neurotransmitters, allowing astrocytes to sense and regulate multiple aspects of brain function. Unlike neurons, astrocytes are not electrically active (i.e. their activation does not lead to the generation of action potentials). Instead, the activation of astrocytic neurotransmitter receptors and transporters is frequently coupled, either directly, or indirectly, to increases in [Ca²⁺]_i. There are multiple sources for these Ca²⁺ increases, including IP₃ receptor-mediated Ca²⁺ release from the endoplasmic reticulum, release from mitochondria, and influx across the plasma through a plethora of ion channels and exchangers. The attribution of Ca²⁺ signal to source has, however, proven contentious (for recent discussion and reviews see Rusakov et al., 2014; Volterra et al., 2014; Bazargani and Attwell, 2016). Increases in intracellular [Ca²⁺] are implicated in numerous astrocyte functions, including regulation of local blood flow (Petzold et al., 2008; Schummers et al., 2008; Petzold and Murthy, 2011; Otsu et al., 2015, reviewed in MacVicar and Newman, 2015), release of transmitter (Nedergaard, 1994; Parpura et al., 1994; Marchaland et al., 2008; Navarrete et al., 2013), and regulation of synaptic plasticity (Di Castro et al., 2011; Min and Nevian, 2012).

Calcium signals in astrocytes are highly varied, ranging from small, local, uncoordinated elevations in [Ca²⁺] (Nett et al., 2002; Shigetomi et al., 2010a; Di Castro et al., 2011) to large, somatic [Ca²⁺] increases that are coordinated between many cells, such as occur in response to adrenergic signaling in vivo (Ding et al., 2013). Numerous groups have observed spatially restricted [Ca²⁺]_i increases within the processes of astrocytes (Grosche et al., 1999; Perea and Araque, 2005; Shigetomi et al., 2010b; Di Castro et al., 2011; Jackson and Robinson, 2015; Srinivasan et al., 2015). Using hippocampal organotypic cultures expressing membrane tethered Ca²⁺ indicators, we observed that these Ca²⁺ signaling microdomains frequently co-localized with mitochondria (Jackson and Robinson, 2015; O’Donnell et al., 2016).

Mitochondria play important roles in regulating intracellular [Ca²⁺]. In addition to providing ATP necessary to fuel the various Ca²⁺ pumps that maintain low levels of intracellular [Ca²⁺], mitochondria can also actively import Ca²⁺. Electron transport generates a proton electrochemical gradient (Δψm) across the inner mitochondrial membrane. The membrane potential associated with this gradient (~-180–200 mV) facilitates the import of Ca²⁺ into the mitochondria against its concentration gradient via a Ca²⁺ uniporter (MCU) protein in the inner mitochondrial membrane (Kirichok et al., 2004). Mitochondria can release accumulated Ca²⁺ via the mitochondrial Na⁺/Ca²⁺ exchanger (NCXL) (Werth and Thayer, 1994) or via transient opening of the mitochondrial permeability transition pore (mPTP; Bernardi and Petronilli, 1996). Together, these activities allow mitochondria to serve as both sink and source of Ca²⁺ (see Fig 3 for model).

Figure 3. Cartoon representation depicting proposed roles of mitochondria in astrocyte processes.

Astrocytic mitochondria apposed to synapses coordinate local metabolism and Ca²⁺ signaling in astrocyte processes. Increases in neuronal activity result in increases in local [Ca²⁺] in astrocytes. Calcium is removed from the cytosol into the mitochondria via the mitochondrial Ca²⁺ uniporter (MCU). Increases in matrix [Ca²⁺] stimulate dehydrogenases in the TCA cycle and the ATP-synthase, while increases in extramitochondrial [Ca²⁺] stimulate the glutamate-aspartate carrier, aralar. Ca²⁺ accumulated by the mitochondria may be released via transient opening of the permeability transition pore or by the mitochondrial Na⁺/ Ca²⁺-exchanger (NCLX). Glutamate can be transferred into the mitochondria via aralar and subsequently converted to α-ketoglutarate (a-KG) for entry into the TCA cycle.

Several lines of evidence suggest a role for mitochondria controlling Ca²⁺ signaling within astrocytes. Dissipating the mitochondrial membrane potential with carbonyl cyanide p-trifluoromethoxy-phenyl-hydrazone (FCCP; a proton uncoupler) slows the rate of decay of [Ca²⁺] transients both in primary cultures of astrocytes and within astrocyte process, where it also increases the amplitude of spontaneous Ca²⁺ transients and increases the distance over which these transients propagate (Boitier et al., 1999; Jackson and Robinson, 2015). Both we, and Joseph Kittler’s group have used Ca²⁺-insensitive Miro mutant proteins to block the Ca²⁺ sensitive positioning of mitochondria within astrocyte processes. Both groups see an increase in Ca²⁺ signaling characterized by increases in the frequency, amplitude, and half-life of spontaneous events. Further, we have found that the ablation (KillerRed-mito) of individual mitochondria leads to similar increases in Ca²⁺ signaling within astrocyte processes. Additionally, loss of mitochondria from astrocyte processes following oxygen-glucose deprivation is associated with large increases in Ca²⁺ signals (O’Donnell et al., 2016). Together, these observations suggest a role for mitochondrial Ca²⁺ uptake in the clearance of physiological [Ca²⁺] signals in astrocytes. This may contribute to the restriction of these signals into microdomains within the astrocyte processes.

Similarly, evidence suggests that under certain circumstances, mitochondria may also serve as sources of Ca²⁺ signaling within the astrocyte. Mitochondria within astrocytes co-localize with domains of elevated [Ca²⁺] within the processes (Jackson and Robinson, 2015; O’Donnell et al., 2016), suggesting that mitochondria might contribute to Ca²⁺ release. In cultures of astrocytes, inhibition of the NCLX decreased exocytotic glutamate release, wound closure, and proliferation (Parnis et al., 2013). Recently, a role for mitochondria has been suggested in generating Ca²⁺ signals within the processes of astrocytes (Agarwal et al., 2017) in vivo. Using two-photon in vivo imaging of awake mice, Agarwal et al determine that a significant proportion (up to 85%) of spontaneous microdomain [Ca²⁺]_i increases correlated with mitochondrial localization. Further, they found that ~35% of these microdomain elevations could be blocked by inhibition of the mitochondrial permeability pore opening (mPTP; using cyclosporine A + rotenone). In addition, they found the application of picrotoxin (to increase neuronal activity) increased efflux of Ca²⁺ from the mitochondria (increased microdomain number and frequency) and that this was blocked by mPTP inhibition. Together, these results strongly suggest a role for mitochondrial Ca²⁺ efflux in the generation of Ca²⁺ signals in astrocyte processes and adds to the diversity of Ca²⁺ signals that are present in astrocytes.

In addition to contributing to the termination and compartmentalization of Ca²⁺ signals within astrocyte processes, the positioning of mitochondria near sources of elevated [Ca²⁺] may also be important in matching ATP supply with demand. Simultaneous measurement of [ATP] and [Ca²⁺] in Hela cells revealed that mitochondrial [Ca²⁺] increases are matched by increases in ATP production (Jouaville et al., 1999). Work from the Foskett lab has shown that loss of ER-to-mitochondrial Ca²⁺ transfer in unstimulated DT40B lymphocytes results in cell starvation even in the presence of nutrients (Cardenas et al., 2010), highlighting the importance of matrix Ca²⁺ for cell survival. Several mechanisms have been proposed to explain the Ca²⁺-dependent increases in respiration. Increases in intra-mitochondrial [Ca²⁺] increase ATP production through activation of the mitochondrial dehydrogenases (FAD-glycerol phosphate dehydrogenase, pyruvate dehydrogenase, NAD-isocitrate dehydrogenase and oxoglutarate dehydrogenase (Denton et al., 1972; Denton, 2009). Increases in matrix [Ca²⁺] also increase the velocity of ATP production via the ATP-synthase at a given mitochondrial membrane potential (Territo et al., 2000; Territo et al., 2001a; Territo et al., 2001b; reviewed in Balaban, 2009) in cardiac mitochondria. More recently, it was found that nanomolar (S_0.5=~300 nm) elevations in extramitochondrial [Ca²⁺] increase activation of the aspartate-glutamate carrier-1(AGC1/Aralar) (Palmieri et al., 2001). This results in an increased transfer of reducing equivalents (NADH) into the mitochondria and has been proposed to increase glutamate-dependent respiration by increasing substrate supply (Safer et al., 1971; Palmieri et al., 2001; Pardo et al., 2006; Contreras et al., 2007; Gellerich et al., 2012; reviewed in Gellerich et al., 2013; Llorente-Folch et al., 2015). Interestingly, glutamate-dependent increases in intracellular [Na⁺] and [Ca²⁺] stimulate the glucose transporter GLUT1 in astrocytes, a mechanism that is mimicked by D-aspartate application and inhibited by TBOA (Loaiza et al., 2003; Porras et al., 2008). Regardless of the site of Ca²⁺ activation (intra- or extra-mitochondrial) active positioning of mitochondria near to sites of elevated [Ca²⁺] would favor the local production of ATP.

As well as providing local ATP and Ca²⁺ buffering, the movement of mitochondria to sites of high activity and glutamate uptake within the astrocyte may be related to the metabolism of glutamate itself (see Fig 3 for model). Most glutamate in the forebrain is cleared into astrocytes via the glial glutamate transporters GLT1 and GLAST (Rothstein et al., 1994; Danbolt, 2001). It is generally assumed that most of this glutamate is converted to glutamine via the astrocyte selective enzyme glutamine synthetase (Martinez-Hernandez et al., 1977; Norenberg and Martinez-Hernandez, 1979). This glutamine is, in turn, recycled to neurons to regenerate glutamate; the glutamate-glutamine cycle. However, a relatively large percentage (~30%) of the glutamate that is taken up by astrocytes is converted to α-ketoglutarate and subsequently metabolized via the TCA cycle on an ongoing basis (Yu et al., 1982; McKenna et al., 1996; for review see McKenna, 2013). In fact, Yu et al found that in primary cultures of astrocytes, ~50% more glutamate was oxidized vs converted to glutamine. In addition, the percentage of glutamate that is oxidized increases as a function of glutamate concentration (McKenna et al., 1996). Thus, the positioning of mitochondria within astrocytes may allow selective fueling of glutamate uptake by catabolism of glutamate itself (for more extensive reviews see (Dienel, 2013; McKenna, 2013; Olsen and Sonnewald, 2015; Robinson and Jackson, 2016).

From the perspective of the astrocyte, the release of glutamine (derived from α-ketoglutarate) can be regarded as a loss of a TCA cycle intermediate. By some estimates, ~60% of TCA cycle intermediates are lost in astrocytes (Hassel et al., 1994; Hassel and Sonnewald, 1995; for review see Hassel, 2000). Anaplerosis is necessary to replace these intermediates. Pyruvate (derived from glucose metabolism) can be carboxylated via the astrocyte- selective, mitochondrial enzyme, pyruvate decarboxylase, to form the TCA cycle intermediate oxaloacetate (Yu et al., 1983; Shank et al., 1985). Thus, one consequence of perisynaptic mitochondrial positioning in astrocytes may be to facilitate the replenishment of TCA cycle intermediates.

Conclusion

Until recently, the existence of mitochondria and mitochondrial metabolism within astrocyte processes was highly contested. It is increasingly clear that, as in neurons, the dynamics of mitochondria in astrocytes is highly regulated and influences multiple aspects of mitochondrial, and presumably, astrocytic functions. Given that astrocyte signaling and metabolism is highly integrated with that of neurons, one possible avenue of investigation should be to determine whether mitochondrial dynamics in astrocytes influences neuronal signaling and function. In addition to the physiologic influence of mitochondria, dysfunction of mitochondria has been implicated in the pathology of both neuro-developmental and neurodegenerative diseases (Chen and Chan, 2009; Liesa et al., 2009). Relatively unexplored is whether mitochondrial dynamics in astrocytes contributes to these conditions. Understanding the differences underlying the mechanisms regulating mitochondrial dynamics between neurons and glia may yield new targets for pharmacologic manipulation of these disorders.

Main Points.

• This review highlights what is known about mitochondrial dynamics in astrocytes, their regulation, and how these dynamics influence local signaling and metabolism. New analyses of mitochondrial distribution and movement in vivo are presented.