Turning garbage into gold: Autophagy in synaptic function

Abstract

Trillions of synapses in the human brain enable thought and behavior. Synaptic connections must be established and maintained, while retaining dynamic flexibility to respond to experiences. These processes require active remodeling of the synapse to control the composition and integrity of proteins and organelles. Macroautophagy (hereafter, autophagy) provides a mechanism to edit and prune the synaptic proteome. Canonically, autophagy has been viewed as a homeostatic process, which eliminates aged and damaged proteins to maintain neuronal survival. However, accumulating evidence suggests that autophagy also degrades specific cargoes in response to neuronal activity to impact neuronal transmission, excitability, and synaptic plasticity. Here, we will discuss the diverse roles, regulation, and mechanisms of neuronal autophagy in synaptic function and contributions from glial autophagy in these processes.

Introduction: Regulating the synaptic proteome

Maintaining the landscape of the synaptic proteome presents unique challenges. Synapses can be positioned far from the soma where protein synthesis and degradation are concentrated. Neurons fire action potentials that can reach up to rates of 100 Hz, which renders the synaptic proteome vulnerable to damage. Moreover, shaping neuronal circuits in response to experiences requires dynamic changes in the synaptic proteome to support structural and functional remodeling. Thus, neurons must galvanize a transcriptional and translational response along with selective protein degradation. Here, we highlight recent literature implicating autophagy, a lysosomal degradation pathway, as a key regulator of synapse formation, maintenance, and plasticity.

Autophagy is an evolutionarily conserved degradative pathway in which components of the cytoplasm are sequestered within a growing double membrane autophagosome and delivered to the lysosome for degradation (Fig. 1) [1–3]. In this way, autophagy can tightly regulate the integrity and composition of the synaptic proteome. Overwhelming evidence from various neuronal models in vitro and in vivo demonstrates that autophagy levels are coupled with synaptic activity. But what is the function of autophagy at synapses? Over a half century ago, autophagosomes were first reported in axons in landmark electron micrograph studies by Dr. Mary Bunge [4]. Since those studies, the field has advanced to reveal the diversity of roles for autophagy at the synapse that depend on the stage of development and age, neuronal state (e.g., activity and cellular stress), neuronal subtype, and neuronal compartment (e.g., presynaptic vs. postsynaptic). In this review, we will discuss this diversity of roles, regulation, and mechanisms of autophagy at the synapse (Fig. 2A–C). Since glia also play key roles in synaptic development and function, we will also discuss the contributions of glial autophagy to synaptic function.

Figure 1. The autophagy pathway.

Autophagy is initiated with the nucleation of a double-membrane structure called the isolation membrane. This membrane can be sourced from various organelles but primarily from the ER. Expansion of the isolation membrane allows for the sequestration of cytoplasmic cargo. The edges of the isolation membrane fuse, establishing a sealed autophagosome. The autophagosome matures into a degradative organelle after fusion with late endosomes and/or lysosomes. The hydrolases present in the lysosome allow degradation of the contents of the autophagosome. Autophagy-related proteins (e.g. ATG proteins) mentioned throughout this review are highlighted at different steps of the autophagy pathway. LC3 is a ubiquitin-like protein that is important for autophagosome biogenesis and cargo recruitment, and is a key marker for autophagosome organelles [86]. A comprehensive description of the machinery involved in autophagosome formation and fusion with lysosomes is reviewed in [1–3].

Figure 2. Emerging themes in synaptic autophagy.

(A) Proteomics of autophagic organelles isolated from mouse brain indicate that mitochondrial and synaptic proteins are the most abundant cargoes [31–33] but their abundance changes with age [33]. The endoplasmic reticulum (ER) constitutes a small fraction of cargo at baseline, but prominently accumulates when autophagy is disrupted [19,21]. (B) The turnover of AMPA receptors or scaffolding protein PSD-95 through autophagy serves to regulate the excitability and plasticity of synapses [34,50–52]. Autophagy also regulates levels of the regulatory subunits of protein kinase A (PKA), broadly influencing the phosphoproteome of the postsynaptic density [15,19]. (C) Autophagy indirectly influences synaptic vesicle release (Ci, [21]) and regulates the turnover of synaptic vesicle proteins (Cii, [19,25,30]). Scaffolding (Cii, [25]) and endocytic proteins (Ciii, [65]; Civ, [8,59]) enriched at the presynaptic terminal control the local induction of autophagy.

Autophagy in synaptic development

Growing axon terminals.

A remarkable feat during neurodevelopment is the precise wiring of trillions of synapses, in which growing axons must traverse and identify correct postsynaptic partners. Potential targets are surveyed by filopodia forming at the leading edge of a growing axon terminal and incorrect interactions are destabilized. The mechanisms that restrict these incorrect interactions remain a key area of interest. A growing theme in the literature is that autophagy regulates axon growth and the stability of these transient synaptic interactions, reviewed in [5]. Live-cell imaging of developing Drosophila melanogaster photoreceptors reveals that autophagosomes form at the base of axon filopodia before retraction/collapse [6]. Autophagy-deficient photoreceptors show excess synaptic filopodia with increased stability, resulting in the formation of supernumerary synapses throughout the optic lobe [6]. Thus, autophagy can play a restrictive role in regulating the dynamics of a developing circuit.

How is autophagy locally induced in the growing axon terminal? Recent work implicates the BAR-domain protein Endophilin-B (EndoB) [7]. EndoB accumulates in the growing axon terminals of photoreceptors in larval D. melanogaster [7]. EndoB promotes membrane deformation in vitro and may be involved in autophagosome closure in vivo [7]. Loss of EndoB function drastically impairs autophagosome biogenesis within developing synapses [7]. Similar to flies deficient for autophagy genes, loss of EndoB in the D. melanogaster visual system results in the formation of supernumerary synapses throughout the optic lobe during development [7]. Another member of the Endophilin family, Endophilin-A (EndoA), can also help coordinate this local initiation of autophagy [8] (discussed below). Thus, proteins enriched within growing axons may coordinate local cues with autophagic activity to help establish proper wiring.

Evidence in mammalian neurons.

The importance of autophagy in refining synaptic connections is also evident in mammals. Loss of autophagy can have deleterious effects on the formation and guidance of axon tracts throughout the brain and spinal cord [9–11]. Moreover, loss of ATG7, a key protein in autophagosome biogenesis [12], causes neurodevelopmental defects in humans [13]. In the developing nervous system, initial connections are established and mature circuits are refined throughout adolescence into adulthood. A deficit in synaptic pruning during neurodevelopment is a prominent pathology in patients with Autism Spectrum Disorder (ASD). In ASD patients, pyramidal neurons in layer V of the temporal lobe exhibited elevated spine densities, which correlated with mTOR hyperactivity and dampened autophagy [14]. Moreover, a mouse model of ASD with hyperactive mTOR also exhibited reduced pruning of dendritic spines in cortical pyramidal neurons that correlated with suppressed levels of autophagy [14]. Restoration of autophagy levels in the ASD mouse model with pharmacological inhibition of mTOR increased spine pruning and attenuated ASD-like social behaviors in mice, suggesting roles for autophagy in spine pruning [14]. Moreover, a conditional knockout of Atg7 in pyramidal neurons was sufficient to cause defects in spine pruning [14]. Thus, autophagy is critical during neurodevelopment in regulating spine pruning and axon outgrowth to ensure proper connectivity of the nervous system.

Autophagy in synaptic function

Neuronal excitability.

Autophagy can regulate the excitability of neurons. Deletion of Atg5, a gene encoding a protein required for autophagosome formation [12], in murine excitatory neurons elicits elevated neuronal network activity and startle-evoked seizures [15] (reviewed in [16]). This elevated activity stems from an accumulation of regulatory subunits (R1α/β) of protein kinase A (PKA), which are key substrates for degradation by neuronal autophagy [15] (Fig. 2B). Prior studies established that the R1α subunit of PKA is targeted to autophagosomes via the autophagy receptor AKAP11 [17]. In the absence of autophagy, reduced PKA activity leads to hypophosphorylation of PKA substrates enriched in the postsynaptic density (PSD), and enlargement of the PSD. Consequently, AMPARs containing GluR1 subunits accumulate at synapses leading to increased AMPAR currents and baseline excitability in primary hippocampal-cortical neurons [15]. Thus, autophagy can control neuronal excitability via regulation of PKA signaling. Since R1β is enriched in dendrites [18], these results highlight a role for autophagy in controlling neuronal excitability in a compartment-specific manner [15].

A corroborative study in human iNeurons and mouse brains with neurons deficient for ATG7 or ATG14 shows an accumulation of the regulatory subunits (R1α/β) of PKA and AKAP11 [19]. Autophagy-deficient neurons exhibited hypophosphorylation of PKA substrates and a reduction in cFOS-mediated neuronal activity in vitro and in vivo [19]. Thus, several studies implicate autophagy in regulating PKA signaling and neuronal activity [15,19].

Interestingly, the role of autophagy in neuronal excitability can differ between neuronal subtypes. Deletion of Atg7 in spiny projection neurons (SPNs) in the mouse striatum, in either the direct (dSPNs) or indirect (iSPNs) pathways reduced motor performance [20]. However, cell-type-specific phenotypes were observed. dSPNs lacking autophagy exhibited a reduction in dendrite length, dendritic complexity, and dendritic spines [20]. Consequently, dSPNs exhibited a reduction in the frequency and amplitude of mEPSCs but not mIPSCs [20]. These phenotypes were not observed in iSPNs. Rather, iSPNs lacking autophagy exhibited elevated intrinsic excitability [20]. Mechanisms underlying this effect include a reduction in currents from Kir2, an inwardly rectifying potassium channel that repolarizes the neuron [20]. Thus, autophagy-deficient iSPNs are hyper-responsive to depolarizing inputs and exhibit a greater tendency to fire action potentials. Under baseline conditions, Kir2 is trafficked to autophagosomes for degradation [20]. But in the absence of autophagy, Kir2 channels accumulate on the postsynaptic membrane and exhibit reduced function due to increased acetylation [20]. In total, these studies underscore distinct and complex roles for autophagy in synaptic function that are specific to neuron subtypes.

Neurotransmission.

Neuronal autophagy can impact presynaptic neurotransmission by controlling the abundance of the endoplasmic reticulum (ER) in the axon [21] (Fig. 2Ci). Deletion of Atg5 in excitatory neurons in the cortex and hippocampus enhances excitatory neurotransmission by increasing synaptic vesicle release probability [21]. Loss of ATG5 resulted in an accumulation of ER, but only in axons and not in the soma or dendrites [21], revealing compartment-specific functions for autophagy in presynaptic terminals. Human iNeurons and mouse brains with neurons deficient for ATG7 or ATG14 also demonstrated an enrichment of ER and ER-phagy receptors [19]. Other compartments that were enriched include presynaptic proteins such as synaptic vesicle proteins, mitochondria, Golgi, and endosomes [19], but the largest effects were on the ER. Electron microscopy confirmed an expansion of tubular ER in axons [19]. Increased ER abundance from a loss of ATG5 led to elevated release of calcium from ER stores and exocytosis of synaptic vesicles [21]. Thus, autophagic degradation of axonal ER regulates axonal calcium homeostasis, thereby preventing aberrant neurotransmission.

Further, autophagy can negatively regulate presynaptic neurotransmission by controlling the pool of synaptic vesicles [22]. Pharmacological stimulation of autophagy in striatal slices reduces the abundance of synaptic vesicles within the presynaptic terminal [22]. Conversely, knockout of Atg7 in dopaminergic neurons increases evoked release of dopamine [22]. Along these lines, Gu et al., also find that presynaptic autophagy limits the size of synaptic vesicle pools and synaptic vesicle release probability [23]. Immunostain analysis revealed the presence of newly endocytosed synaptic vesicle proteins with a population of LC3-labeled autophagosomes [23]. These results raise the possibility that synaptic vesicles, or their components, may be substrates for autophagy as a potential mechanism to regulate neurotransmission, but this possibility needs further exploration.

The mechanisms by which autophagy may degrade synaptic vesicle components are starting to be revealed. For example, deletion of Bassoon, a presynaptic scaffolding protein and negative regulator of autophagy [24], reduces the synaptic vesicle pool size by increasing the turnover rate of synaptic vesicles by autophagy [25] (Fig. 2Cii). Indeed, the decrease in synaptic vesicle density with loss of Bassoon can be reversed by inhibiting autophagy [25]. Bassoon inhibits members of the E3 ubiquitin ligase family [26], enzymes that catalyze the final step in the ubiquitination pathway and can target cargo for autophagic degradation [27]. Analysis of synaptosomes from Bassoon knockout neurons revealed an increase in ubiquitinated synaptic vesicle-associated proteins compared to controls [25]. Further, the increase in autophagosome number and colocalization with synaptic vesicle proteins in Bassoon knockout neurons can be reversed by knockdown of the E3 ligase Parkin [25]. Thus, Bassoon negatively regulates the autophagic clearance of synaptic vesicle proteins by controlling Parkin-dependent ubiquitination [25]. In the absence of Bassoon, elevated ubiquitination of synaptic vesicle proteins leads to enhanced clearance by autophagy [25]. Additional links between synaptic vesicles and autophagosomes involve the small GTPase Rab26 and its Guanine Nucleotide Exchange Factor Plekhg5 [28,29]. However, other studies did not find alterations in synaptic vesicle number [19] or bulk levels of synaptic vesicle proteins [15,21] in autophagy-deficient neurons, warranting further investigation.

Evidence suggests that autophagy can selectively degrade synaptic vesicle proteins rather than a synaptic vesicle en masse [30]. Indeed, focally-induced damage to specific synaptic vesicle proteins using light-activated generation of ROS activates autophagy within 5–10 minutes [30]. Only the damaged synaptic vesicle protein, and not other synaptic vesicle proteins, were targeted to autophagosomes, suggesting a selective clearance of the damaged protein rather than the entire synaptic vesicle [30]. Lastly, ROS-induced damage to synaptic vesicle proteins only affects synaptic function in the absence of autophagy [30], suggesting that autophagy maintains the integrity of the synaptic vesicle pool.

Cargoes for neuronal autophagy and their links to development and neuronal state.

A key approach that has expanded our understanding for how autophagy may affect synaptic function comes from proteomic analysis of cargoes of autophagic vacuoles (AVs) isolated from the mouse brain (Fig. 2A). Key substrates include mitochondrial components, presynaptic and postsynaptic proteins, ER, and selective autophagy receptors for aggrephagy, mitophagy and ER-phagy [31–34]. Mitochondrial proteins and synaptic proteins are the most abundant and prominent cargoes of autophagy in the brain [31–33]. Also reported is the constitutive turnover of aggregates, ER, and Golgi [33]. Efficient clearance of damaged mitochondria by autophagy protects synaptic function as reduced clearance of synaptic mitochondria is linked to cognitive impairments [35–37].

Interestingly, a comparison of the autophagosome degradome of adolescent, adult, and aged mice found that mitophagy is enriched in adolescent mice and synaptic protein turnover is enriched in adult and aged mice [33] (Fig. 2A). Thus, there is a shift toward degradation of synaptic proteins by autophagy with age [33]. Moreover, autophagic organelles isolated from the brain exhibit cell-type-specific cargoes that uniquely change in response to cellular stress [32]. For example, the most dramatic changes in the brain AV proteome in response to nutrient deprivation were in proteins expressed in glia and endothelial cells rather than in neurons [32]. These results corroborate findings that starvation elicits a more robust activation of autophagy in astrocytes versus neurons in vitro [38]. In sum, proteomic analyses have revealed a diversity of autophagic cargoes in cell types of the brain, which are dynamically regulated with age and cellular stress.

Profiling the content of purified AVs has revealed differences from profiling proteins that accumulate in total brain or neuronal lysates in the absence of autophagy. Indeed, ER-related proteins were represented in relatively low abundance in purified AVs [31–33] but exhibited the greatest accumulation in ATG5 [21], ATG7 and ATG14-deficient neurons [19]. This enhanced accumulation of ER in autophagy deficient models may reflect the cumulative effects of blocking degradative flux or a potential backlog of membrane that would have been donated to newly forming autophagosomes. The lower abundance of ER in AV preparations from the mouse brain could also be due to an enrichment of cargoes from non-neuronal cells. An exciting next step will be sorting out AV preparations by brain cell types to understand the cell-type-specific cargoes of autophagy.

Another distinction is that analysis of protein levels in total lysates in several autophagy deficient models have reported no changes in bulk levels of synaptic vesicle proteins [15,21] or synaptic-associated proteins [21], but proteomics of purified AVs reveals high synaptic content [31–33]. Grosso Jasutkar et al. [39] find that the effects of autophagy deficiency may be masked in total homogenates versus isolated organelle populations. In fact, acute deactivation of autophagy in the adult mouse initially resulted in an accumulation of synaptic proteins in whole brain homogenates, but by later timepoints of autophagy deactivation, this accumulation was reduced [39]. By contrast, the synaptosomal fraction maintained a significant accumulation of synaptic proteins at later time points of autophagy deactivation. The reduction of synaptic protein levels in total brain homogenates at later time points was attributed to a compensatory activation of proteasomal activity that might affect protein levels in bulk but not at the organelle level. Thus, the findings from Grosso Jasutkar et al. [39] emphasize that the effects of autophagy deficiency manifest differently depending on compartment and time.

Autophagy in learning and memory

Autophagy plays a key role in hippocampal-dependent learning and memory [40–44]. For example, autophagy is upregulated in the hippocampus of mice in response to tasks that stimulate novel memories [40]. Acute reduction of autophagy in hippocampal neurons did not impact learning, but reduced memory performance in two different associative learning tasks (e.g., contextual fear conditioning and novel object recognition) [40]. Memory performance depended on the upregulation of autophagy specifically during the learning phase of the task [40]. Thus, stimulation-induced autophagy in the hippocampus is required for memory formation.

The learning phase of another associative learning task (e.g., inhibitory avoidance; IA) upregulates autophagy-lysosomal proteins and autophagic flux in the hippocampus [41]. IA-induced autophagy depends on de novo synthesis of autophagy proteins and not changes in their mRNA levels [41]. Interestingly, injection of pharmacological inhibitors of autophagy, or siRNAs to knockdown key autophagy genes, into the mouse hippocampus had no effect on IA learning or short-term memory, but impaired the formation of long-term memory [41]. These results build upon prior literature showing that spatial learning also increases autophagy in the hippocampus [42], and autophagy is important for the formation of long-term spatial memories [42,43]. Thus, an emerging theme is that autophagy induced by learning is required for the formation of long-term memories. Interestingly, stimulating hippocampal autophagy can improve spatial memory [42] and reverse age-related memory deficits [40]. Activating autophagy in the dorsal hippocampus can also mitigate cognitive impairments in Angelman syndrome [45]. Indeed, tuning autophagy levels in the hippocampus can impact cognitive function with potential for translational benefits.

Grosso Jasutkar et al. [39] also find that acute disruption of autophagy in the adult mouse leads to deficits in memory-dependent tasks. Loss of autophagy impaired cognition but not gross locomotor skills [39]. Changes in cognition were not associated with hippocampal cell death or synapse loss, but rather alterations in levels of synaptic proteins [39]. These results suggest that loss of autophagy may not affect all neuronal circuits equally.

Synaptic plasticity: Autophagy in LTP and LTD

What is the molecular basis for how autophagy can influence learning and memory? The principal mechanism underlying learning and memory is synaptic plasticity, an activity-dependent process that modifies the strength or efficacy of synapses in response to experience [46,47]. Persistent memories, in particular, are thought to be generated by long-term potentiation (LTP) of glutamatergic synapses, reviewed in [48]. LTP has been widely investigated in the hippocampus and is triggered by high frequency stimulation leading to sustained increases in synaptic excitability. Information processing in the brain also requires flexibility: some synapses are strengthened, whereas others are weakened. Lasting decreases in synaptic efficacy is referred to as long-term depression (LTD), and is experimentally triggered by low frequency stimulation of neurons [48]. LTD has been proposed to prevent the saturation of synapses and maintain global excitation levels by counteracting LTP. On the molecular level, both LTP and LTD require changes in the synaptic proteome, particularly the density of glutamate receptors (e.g. AMPARs) on the postsynaptic membrane [48]. LTP elicits insertion of AMPARs to the postsynaptic membrane whereas LTD elicits removal of AMPARs from the postsynaptic membrane to increase or decrease responsiveness to glutamate, respectively. Beyond AMPARs, synaptic plasticity requires massive remodeling of the number and organization of synaptic proteins. Recent studies in the autophagy field converge on a common theme: autophagy is critical for synaptic plasticity and can influence LTP and LTD.

Long-term potentiation.

Chemical induction of LTP (cLTP) increases autophagy in primary hippocampal neurons and this activation of autophagy was required for cLTP-induced formation of novel dendritic spines in vitro [40]. Acute inhibition of autophagy in the mouse hippocampus also blocked activity-dependent formation of dendritic spines in vivo [40]. Interestingly, acute inhibition of autophagy affected dendritic spine density only in the context of cLTP, and not in baseline conditions [40]. Moreover, autophagy inhibition in hippocampal slices did not affect baseline glutamatergic neurotransmission, but attenuated LTP induced by high frequency electrical stimulation [40]. These results suggest that autophagy induction facilitates activity-dependent changes at the synapse. Similarly, hippocampal slices isolated from mice lacking Wdr45/WIPI4, a gene encoding a PI3P-binding protein involved in autophagosome formation, also exhibit attenuated LTP [43]. Thus, autophagy is important for structural and functional synaptic plasticity and promoting LTP in hippocampal neurons to support memory performance.

In specific circumstances, autophagy may suppress the induction of LTP. In fact, mice undergoing fasting show an activation of autophagy in the hypothalamus but suppression of autophagy in the forebrain, the latter effect mediated by local BDNF signaling [49]. This BDNF-mediated suppression of autophagy promotes cell survival in vivo and LTP in hippocampal slices ex vivo [49]. Thus, fasting does not elicit a global activation of autophagy throughout the brain. Rather, autophagy levels may be tuned based on neuronal subtype to enable BDNF-induced synaptic plasticity and behavioral adaptations to survive periods of fasting.

Long-term depression.

Autophagy is also important for LTD [34,50–52]. Chemical induction of LTD (cLTD) increased autophagy in dendrites, and autophagy promoted the degradation of AMPARs [50]. LTD-induced downregulation of AMPARs may involve the fusion of newly endocytosed AMPARs with autophagosomes for delivery to lysosomes (Fig. 2B). Indeed, cLTD rapidly generates autophagosomes locally within dendrites in primary hippocampal neurons and hippocampal slices, within 15 minutes [34]. Cargoes enriched in hippocampal AVs upon cLTD included postsynaptic components such as AMPAR subunits, scaffolding proteins that localize receptors to the postsynaptic density (e.g., PSD-95, PICK1, SAP97), actin-associated proteins, and kinases involved in synaptic plasticity (e.g., CAMK2a) [34] (Fig. 2B). Importantly, acute inhibition of autophagy in hippocampal neurons prevented the induction of LTD by low frequency stimulation, and the degradation of AMPARs and PSD-95 that would be elicited by LTD [34]. However, baseline synaptic transmission was unaffected [34]. Moreover, loss of autophagy in only postsynaptic pyramidal neurons in the CA1 area was sufficient to inhibit LTD [34].

Autophagy may promote LTD by degrading PSD-95 to increase lateral diffusion of AMPARs [51]. PSD-95 is a key scaffolding protein of the excitatory post-synaptic density and forms nanoscale clusters that stabilize AMPARs at synapses. NMDAR-dependent LTD reduces the number of PSD-95 molecules per cluster and increases AMPAR mobility [51]. Mechanisms underlying this effect involve phosphorylation of PSD-95 at T19 by GSKβ that targets PSD-95 to autophagosomes for degradation, thereby reducing the PSD-95 available to trap surface AMPARs [51]. Interestingly, T19 phosphorylation is inhibited by cLTP [53]. In silico simulations reveal that the increase in lateral AMPAR mobility enables depressed synapses to respond to high frequency stimulation [51]. Curiously, parallel studies suggest that autophagy is upregulated during development to decrease the inducibility of LTD in adult CA1 neurons [44].

How is the regulation of autophagy distinguished in LTD versus LTP? Pan et al. also find that autophagy is induced during cLTD and is specifically required for the long-lasting degradation of PSD-95 and surface AMPAR subunits in cLTD [52]. This elevation in autophagy and downregulation of AMPAR subunits in cLTD critically depend on CREB-mediated transcription of autophagy genes, but only in late-phases of cLTD [52]. Both cLTD and cLTP induce nuclear translocation of the CREB cofactor CRTC1 [52]. But this effect is stronger in cLTD and distinguished from cLTP by dephosphorylation of CRTC1 at S151 [52]. Dephosphorylation of CRTC1 increases CRTC1 association with CREB and CRTC1-CREB binding to cis-elements of autophagy genes [52]. Thus, CRCT1-CREB is a critical regulator of transcription-dependent autophagy in response to LTD and not LTP [52]. In total, alterations in these pathways may provide a mechanistic basis for the learning and memory deficits reported in autophagy-deficient models.

How does an induction of autophagy contribute to both LTP and LTD? Pan et al. finds that the CREB cofactor CRTC1 helps distinguish LTD-specific induction of autophagy [52]. Continuing to identify stimulus-specific regulators of the transcription, translation or nucleation of autophagy proteins will be crucial to parse out these different roles. Moreover, how is this stimulus-specific autophagy induction coupled to cargo selection? Compans et al. show that LTD is distinguished from LTP in cargo selection based on specific phosphorylation events [51]. Cargoes specific to LTP remain poorly understood. Future work is needed to further identify how different models of synaptic plasticity may coordinate upstream inducers of autophagy with downstream cargo selection.

Autophagy can impact brain-wide plasticity.

In D. melanogaster, there is an age-dependent decline in the presynaptic plasticity of olfactory neurons and olfactory memory formation [54]. Specifically, aging is associated with a brain-wide enlargement of the presynaptic active zone and enhanced synaptic vesicle release (termed, ‘metaplasticity’); a state which is incompatible with memory formation [54]. Attenuation of autophagy in young flies accelerates presynaptic metaplasticity and impairs olfactory memory formation [55]. Strikingly, knockdown of Atg5 or Atg9 specifically within the mushroom body (i.e., a higher integration region important for memory formation) was sufficient to drive these brain-wide changes in presynaptic plasticity [55]. Thus, the autophagic status of specific subtypes of neurons can regulate the plasticity of the D. melanogaster brain at large.

How is autophagy coupled to synaptic activity?

Overwhelming evidence shows that autophagy is engaged by neural activity to tune the molecular composition of the synapse and regulate synaptic function and learning and memory. But what are the molecular underpinnings that decode synaptic activity to alter levels of autophagy? Proteins enriched in presynaptic terminals regulate the local formation of autophagosomes. Bassoon, a large scaffolding protein that is localized to the presynaptic active zone [56], can negatively regulate presynaptic autophagy by sequestering ATG5 [24]. Synaptojanin, a lipid phosphatase important in synaptic vesicle trafficking, promotes autophagosome formation in presynaptic terminals [57]. Mutations in the Synaptojanin SAC1 lipid phosphatase domain arrest autophagosome biogenesis likely due to dysregulation of the lipid composition of nascent autophagosomal membranes [57]. Endophilin-A (EndoA), a BAR-domain protein that is required for synaptic vesicle endocytosis [58], also promotes autophagosome formation in presynaptic terminals [59]. EndoA is important for autophagy activation in response to both synaptic activity or metabolic stress [59]. Phosphorylation of EndoA at S75 by LRRK2 regulates autophagy in presynaptic terminals by promoting curved membranes for the docking of autophagic factors including Atg3 [59]. Phosphomimic EndoA S75D constitutively induces autophagy at baseline whereas phosphodead EndoA S75A blocks starvation-induced autophagy [59]. Calcium influx is a critical activator for presynaptic autophagy, and EndoA acts as a molecular sensor to connect these events [8] (Fig. 2Ciii). Indeed, the flexibility of EndoA, mediated by the D265 residue, is important for responsiveness to calcium [8]. Mutations in EndoA that increase rigidity (D265A) block autophagy induced by calcium influx [8]. By contrast, mutations that enhance flexibility in EndoA (D265R) constitutively activate autophagy at baseline [8]. Interestingly, the D265 residue uncouples autophagy activated by synaptic activity from autophagy activated by metabolic stress [8]. In fact, the flexible EndoA D265R activates autophagy even in the presence of the phosphodead S75A mutation [8]. Furthermore, the rigid EndoA D265A that blocks calcium-induced autophagy does not block starvation-induced autophagy [8]. In this way, presynaptic autophagy is orchestrated by distinct stimuli.

Hernandez-Diaz et al. [60] have further expanded our understanding of the mechanisms that distinguish autophagy activated in presynaptic terminals by neuronal activity versus metabolic stress. Specifically, they identify Synaptogyrin, a presynaptic protein found in synaptic vesicles, as a negative regulator of presynaptic autophagy induced by neuronal activity but not by starvation [60] (Fig. 2Ciii). By comparison, loss of EndoB, a positive regulator of autophagy, blocks autophagy induced by starvation but not activity [7,60]. Further, overexpression of Synaptogyrin doesn’t block induction of autophagy caused by overexpression of EndoB, and loss of EndoB doesn’t block induction of autophagy caused by loss of Synaptogyrin [60]. Thus, Synaptogyrin and EndoB function in different autophagy pathways. Taken together, stimuli for synaptic autophagy are decoded differently in presynaptic terminals, and neuronal activity and metabolic stress regulate synaptic autophagy via distinct molecular mechanisms. Excitingly, these differences in regulation also influence cargo selection [34,51]. Hernandez-Diaz et al. find synaptic vesicle components in autophagosomes generated by loss of Synaptogyrin, but not in autophagosomes generated by nutrient deprivation [60]. These data support an emerging theme that cargo selection is stimulus-dependent.

Another mechanism to coordinate autophagy with synaptic activity is via ATG9 (Fig. 2Civ). ATG9 is the only transmembrane protein in the machinery that generates autophagosomes. ATG9 functions as a lipid scramblase that translocates phospholipids from the cytoplasmic to the luminal leaflet of the nascent autophagosome membrane (termed isolation membrane) (Fig. 1) [61,62]. ATG9 collaborates with ATG2, a lipid transporter that transfers phospholipids from a membrane source (e.g., the ER) to the isolation membrane for subsequent translocation between leaflets by ATG9 [61,63]. Thus, the concerted efforts of ATG2 and ATG9 help source and distribute lipids in the nascent autophagosome membrane to drive membrane elongation. In neurons, ATG9-positive vesicles are derived from the trans-Golgi network (TGN) and are transported to the distal axon [10,64–67]. Preventing exit of ATG9 from the TGN reduces the formation and early stages of maturation of autophagosomes in the distal axon [10,64,66]. Thus, ATG9-positive vesicles are an important component of autophagosome biogenesis in the axon. Interestingly, ATG-9-positive vesicles in presynaptic terminals undergo cycles of endocytosis and exocytosis that are coupled with synaptic activity, such that an increase in synaptic activity increases ATG-9 cycling [65]. This process is dependent on machinery that mediates synaptic vesicle cycling (e.g. Dynamin, Synaptojanin, and EndoA) [65]. However, ATG9-positive vesicles are distinct from synaptic vesicles [68–70]. Excitingly, ATG-9 exo-endocytosis is coupled with autophagosome formation in the presynaptic terminal. In fact, blocking endocytosis or exocytosis prevents autophagosome formation in response to synaptic activity [65]. Thus, activity-dependent cycling of ATG-9 between vesicles and the plasma membrane may regulate ATG-9 availability to coordinate autophagosome biogenesis with synaptic activity [65]. Interestingly, Hernandez-Diaz et al. report an interaction between Atg9 and Synaptogyrin which may provide another mechanism to coordinate autophagosome formation with the synaptic vesicle cycle [60]. The active zone protein Clarinet genetically interacts with endocytic proteins to regulate ATG-9 sorting during exo-endocytosis at synapses to facilitate activity-dependent autophagosome formation [70]. Due to its large size, Clarinet may facilitate ATG-9 sorting by bridging the exocytic active zone with endocytic periactive zones [70]. In this way, scaffolding proteins of the active zone may coordinate autophagosome biogenesis with synaptic activity.

Availability of ATG9 may also regulate autophagosome formation in dendritic spines. Enrichment of ATG9A in dendritic spines requires Rab11a, a small GTPase that localizes to recycling endosomes [71]. Interactions between ATG9A and Rab11a may facilitate autophagosome formation in a manner coordinated by synaptic activity and local mTOR activity [71]. Thus, Rab11a may control the local pool of ATG9A to nucleate autophagosomes in dendritic spines.

Once autophagosomes are formed, how do they mature into degradative organelles? Moreover, is autophagosome trafficking regulated by synaptic activity? A distinguishing feature between dendritic and axonal autophagy is the scale over which the process is executed, reviewed in [72]. Axonal autophagy involves long-range motility that delivers newly-formed autophagosomes from distal, presynaptic sites to the soma for degradation where lysosomes are concentrated. A recent study in rat optic nerve demonstrates this long-range retrograde motility of autophagic vesicles in vivo [73]. By contrast, dendritic autophagy appears more localized and autophagic organelles undergo short-range, oscillatory dynamics [74]. Interestingly, synaptic activity dampens the motility of autophagic organelles in dendrites and positions them near synapses [74]. This effect on motility is compartment-specific and is not observed in axons, suggesting that a population of autophagosomes formed in presynaptic terminals maintains long-distance transport irrespective of the activity state of the neuron [74]. Strikingly, synaptic activity stimulates the maturation of autophagic vacuoles into degradative compartments in dendrites [74]. In this way, autophagy may be positioned locally at postsynaptic compartments to couple protein degradation with protein synthesis to fuel dynamic changes in the synaptic proteome.

Regulation of glial autophagy to promote synaptic function

Throughout the lifespan of a neuron, glia play critical roles in the proper connectivity of synapses. Astrocytes and microglia influence aspects of synapse formation, plasticity and elimination [75,76]. Oligodendrocytes and oligodendrocyte precursor cells (OPCs) also play a role in synaptic plasticity and synapse elimination [77,78]. What is the role for glial autophagy in facilitating these functions?

Synapse Formation.

Astrocytes secrete synaptogenic cues to influence the formation and maturation of synapses. A subtype of astrocytes in the aging hippocampus exhibits an accumulation of defective autophagic organelles within dystrophic astrocytic branches [79]. Areas proximal to these astrocytes had a reduction in excitatory synapses and dendritic spines [79]. Synaptic pathologies were attributed to impaired secretion of synaptogenic proteins from these dystrophic astrocytes [79]. The mechanisms by which autophagy in astrocyte branches contributes to synaptic support, however, remains unexplored.

Synaptic Plasticity.

The role of glial autophagy in synaptic plasticity and memory formation is largely unknown. However, microglia and astrocytes increase expression of p62 (receptor for selective autophagy) and LAMP1 (marker of late endosomes and lysosomes) during inhibitory avoidance learning [41]. Thus, the autophagic status of glia may be coupled to memory formation. Future work will need to explore if this coupling has functional implications for memory formation. Consistent with this idea, accumulation of p62 occurs in astrocytes in the aging hippocampus [40], which could reflect an impairment of autophagy, as reported by [79]. Stereotaxic injections of TAT-Beclin1, a subunit of the PI3K complex I that generates PI3P important for formation of the isolation membrane [12], within the hippocampus to induce autophagy could reverse age-dependent memory deficits [40]. However, the contributions of glial autophagy to this rescue in cognitive function are unknown and represent an exciting area for future investigation.

Elimination of synapses and axonal debris.

Microglia, astrocytes and OPCs can eliminate synapses through phagocytosis [78,80]. Interestingly, ATG7-deficient microglia exhibit attenuated synapse elimination [80]. Moreover, loss of ATG7 in myeloid cells, progenitors of microglia, led to social behavioral defects and repetitive behaviors, hallmarks of ASD [80]. Thus, autophagy plays a critical role in microglia for synapse and circuit refinement. The role of autophagy in astrocyte or OPC-dependent synapse elimination, however, remains largely unexplored.

LC3-associated phagocytosis facilitates glial degradation of axon debris after injury in Drosophila [81]. Thus, autophagy-related processes in glia promote recovery of the nervous system from injury. This study adds to prior literature implicating autophagy in Schwann cells in digesting myelin from injured nerves [82]. To date, studies exploring autophagy in oligodendrocytes have been limited. However, autophagy in oligodendrocytes is essential for routine turnover of myelin [83]. Knockout of Atg7 specifically within oligodendrocytes increases myelin sheath thickness and leads to motor deficits and neurodegeneration [83]. Thus, specialized cell types of the nervous system use autophagy to maintain proper connectivity and survival.

Conclusions and avenues for future exploration

The field is converging on common themes: Autophagy is coupled to the activity state of the neuron and in turn, autophagy serves critical roles on either side of the synapse to regulate synaptic vesicle release, neuronal excitability, and synaptic plasticity. The impact of depleting autophagy on synaptic function has revealed a diversity of functions for neuronal autophagy that depend on several parameters: (a) the time and duration of autophagy depletion, (b) the neuronal subtype, (c) the compartment targeted, and (d) activity stimulus. A distinguishing feature of autophagy in neurons is that several regulatory molecules that control the local induction of autophagy in presynaptic terminals are enriched at synapses, including scaffolding proteins of the active zone and proteins involved in the synaptic vesicle cycle. In this way, activity-based information can be decoded and coupled with autophagosome formation. Mechanisms regulating autophagosome biogenesis and maturation in post-synaptic compartments, however, are less well understood.

Given the diversity of functions for autophagy at the synapse, how does autophagy decode slight variations in synaptic signaling? Different models of plasticity elicited varying effects on autophagy. Thus, how is autophagy tuned to different types of synaptic plasticity? Most of the studies to date investigated hippocampal-dependent memories, but how is autophagy regulated in other contexts of plasticity in other brain regions? A recent study suggests a role for the secretion of autophagosomes in activity-induced structural plasticity [84]. This result adds to prior work showing activity-dependent lysosomal secretion in the remodeling the extracellular matrix to facilitate the growth and structural plasticity of dendritic spines [85]. Thus, an exciting next step is to explore mechanisms underlying non-canonical pathways for autophagy in functional and structural synaptic plasticity.

How are nuances in these activity-based cues distinguished to ensure that the correct cargoes are selected for degradation? In other words, how are differences in upstream activating signals coordinated with downstream processes in cargo selection? We are only at the inception of understanding how cargo are selected for degradation in response to specific synaptic signals.

The literature also distinguishes functions for basal autophagy versus autophagy induced by synaptic activity. Thus, synaptic autophagy is tightly controlled to perform localized functions at individual synapses. By contrast, basal autophagy may primarily perform important functions to maintain protein and organelle homeostasis to support neuronal health and protect against neurodegeneration. Lastly, the role of glial autophagy in neuronal health and connectivity is largely elusive. Neuron-glial interactions are critical for brain function, thus how autophagy is coordinated between these cell types will be an important area for future studies.

Highlights.

• Local induction of autophagy at synapses is coupled to synaptic activity

• Proteins enriched in the presynaptic active zone control autophagy induction

• Activity-induced autophagy is important for synaptic plasticity and long-term memory

• Glial autophagy may be crucial for synaptic support