Homeostatic synaptic plasticity as a metaplasticity mechanism – a molecular and cellular perspective

Abstract

The molecular mechanisms underlying various types of synaptic plasticity are historically regarded as separate processes involved in independent cellular events. However, recent progress in our molecular understanding of Hebbian and homeostatic synaptic plasticity supports the observation that these two types of plasticity share common cellular events, and are often altered together in neurological diseases. Here, we discuss the emerging concept of homeostatic synaptic plasticity as a metaplasticity mechanism with a focus on cellular signaling processes that enables a direct interaction between Hebbian and homeostatic plasticity. We also identify distinct and shared molecular players involved in these cellular processes that may be explored experimentally in future studies to test the hypothesis that homeostatic synaptic plasticity serves as a metaplasticity mechanism to integrate changes in neuronal activity and support optimal Hebbian learning.

Introduction

One of the defining features of the nervous systems in both invertebrate and vertebrate animals is that they are plastic – changes in the activity and connectivity of the various circuits within the nervous system enable learning, encode memory, and drive behavior. Hebbian and non-Hebbian types of synaptic plasticity have been described as two major mechanisms driving synaptic connectivity changes as a result of synaptic activity experience. Hebbian plasticity, referred to here as input-specific synaptic modifications in the forms of long-term potentiation (LTP) and long-term depression (LTD), is thought to underlie associative learning through bidirectional modification of synaptic strength [1]. The direction of modification is determined by the levels of the postsynaptic responses relative to a “modification threshold” (θ_m), which itself is subject to modification (the sliding threshold model or BCM model) [2,3]. Hebbian plasticity is thought to be self-reinforcing with a tendency to run away if left unchecked (e.g. LTP leads to synaptic strengthening and more correlated pre- and post-synaptic activity, which facilitates additional LTP). The non-Hebbian (sometimes also called anti-Hebbian) plasticity was brought into the picture as a “corrective” mechanism to prevent runaway Hebbian plasticity.

Significant progress has been made in the past two decades toward a molecular understanding of the mechanisms underlying Hebbian plasticity. It becomes clear that although synaptic strengthening (LTP) and weakening (LTD) push synaptic weight changes toward opposite directions by primarily up- or down-regulating presynaptic release probability and moving postsynaptic neurotransmitter receptors in or out of the synaptic membranes, the molecular signaling pathways leading to these changes are quite distinct – instead of sliding on the scale of activation of the same signaling pathway (one end of the scale being LTP and the other LTD), different molecular players are recruited to support opposite changes at synapses. What are the possible mechanisms that mediate the crosstalk between these distinct LTP- and LTD-driving signaling mechanisms in such a manner that they can be engaged in a coordinated fashion to achieve the “sliding” of the θ_m (i.e. metaplasticity)?

Homeostatic synaptic plasticity, as a major form of non-Hebbian plasticity, has been studied historically in a slightly different context. It is thought to stabilize neural networks through negative feedback-based modifications, thus countering the self-reinforcing nature of Hebbian plasticity [4]. Although Hebbian and homeostatic plasticity are believed to achieve distinct purposes (associative learning versus network stability), the biological parameters they modify are often the same. These parameters include, but are not limited to, neuronal excitability [5,6], synaptic strength [7–9], and – on a longer time scale – changes in the number of synaptic contacts [10,11]. A recent review on a similar topic did an excellent job covering in vivo evidence supporting the interactions between synaptic homeostatic and Hebbian mechanisms [12]. We have previously proposed the idea that local homeostatic synaptic plasticity, which potentially maintains activity stability in a neuronal subcompartment (e.g. a segment of dendrite or even single synapses) instead of the entire neuron [13–17], may function as a type of metaplasticity to modulate Hebbian plasticity [18]. In this article, we will first focus on the regulation of postsynaptic AMPA receptor (AMPAR) abundance as an example of a major converging point between Hebbian and homeostatic plasticity. We will discuss the cellular processes and the molecular players involved in each process to identify potential cellular ‘nodes’ through which homeostatic synaptic plasticity may act to impact Hebbian plasticity. In addition, we discuss possibilities that homeostatic plasticity deficit may contribute to impaired cognitive functions through changes that drive synaptic excitation/inhibition (E/I) imbalance in animal models of neuropsychiatric disorders

Synaptic retinoic acid (RA) signaling, homeostatic synaptic plasticity and their interaction with Hebbian plasticity

The discovery of RA’s involvement in homeostatic synaptic plasticity was a bit serendipitous. Acute RA treatment in cultured hippocampal neurons increases excitatory synaptic transmission with a mechanism that does not require its genomic action [19]. The search for RA synthesis mechanism revealed that it is suppressed by intracellular Ca²⁺ and thus de-repression occurs during prolonged blockade of synaptic activity [19,20]. Putting the two findings together allowed us to uncover the central role of RA and its action on the RA receptor RARα [21] in synaptic silencing-induced homeostatic upregulation of excitatory synaptic strength (Figure 1). RA signaling also modulates inhibitory synapses in the opposite direction [22], thus acting as a master organizer to coordinate synaptic E/I balance in the context of homeostatic synaptic plasticity.

Figure 1. Molecular pathways involved in RA-dependent homeostatic synaptic plasticity and its interaction with Hebbian plasticity.

Left: Molecular pathways involved in RA-dependent homeostatic synaptic plasticity (HSP). A reduction in postsynaptic Ca²⁺ levels resulted from synaptic inactivity triggers RA synthesis, which disinhibits local protein synthesis and promotes synaptic insertion of AMPARs. CaN: calcineurin; RALDH: retinal dehydrogenase. Right: Exocytosis of AMPAR-containing vesicles into synaptic membranes during LTP and HSP is mediated by partially overlapping SNARE components. Postsynaptic deletion of Q-SNARE syntaxin-4 (Stx-4), which is uniquely required for HSP, prevents the impairment of LTP following RA treatment by blocking RA-induced HSP.

Aside from RA-dependent homeostatic synaptic plasticity, synaptic strength can also be modified homeostatically by RA-independent mechanisms. For example, Homeostatic synaptic plasticity induced by prolonged blockade of neuronal firing alone (e.g. tetrodotoxin (TTX) treatment to block voltage-gated sodium channels) does not require RA or local protein synthesis and therefore is not affected by RA synthesis blockers [20,23]. Synaptic activity blockade with co-treatment of glutamate receptor antagonists and TTX reduces dendritic Ca²⁺ concentration further below a critical level, triggers RA synthesis, and induces rapid compensation of synaptic strength by engaging local protein synthesis [13,19,20,24,25] (Figure 1). Thus, an important distinction between RA-dependent and RA-independent homeostatic plasticity is that local protein synthesis activated by RA in neuronal dendrites allows homeostatic modulation of synaptic strength in potentially discrete subcellular compartments. As we discuss below, these different forms of homeostatic plasticity have differential impacts on subsequent Hebbian plasticity.

In the case of RA-dependent homeostatic plasticity, although synaptic RA signaling is not directly involved in Hebbian plasticity, altering synaptic strength with RA suppresses subsequent LTP expression [26] (Figure 1). RA-induced impairment of LTP can be rescued by protein synthesis inhibitors or RARα deletion, indicating RARα signaling and local protein synthesis play significant roles in the interaction between LTP and RA-dependent homeostatic plasticity [26]. Interestingly, chronic TTX treatment, which induces homeostatic upregulation of mEPSC amplitude in an RA-independent manner, has the opposite effect on LTP – it promotes greater LTP [27]. This is because in addition to upregulating strength of existing synapses, TTX treatment also increases the relative proportion of silent synapses by significantly increasing the formation of a number of new synapses that are functionally silent (e.g. AMPAR-lacking). LTP induction in TTX-treated slices leads to AMPAR insertion into and functional activation of these silent synapses, thus generating greater LTP than control slices [27]. A similar study using a longer (3–4 days) TTX treatment conducted in slightly older rat organotypic hippocampal slices concluded that the effect of circuit inactivity on silent synapse formation and subsequent LTP induction is age-dependent, and that TTX treatment reduces fidelity of presynaptic release and thus compromises LTP [28]. Further study is needed to resolve the exact mechanisms underlying the opposite outcomes from these seemingly similar treatments. However, it is worth noting that although RA-treatment and TTX-treatment both induce homeostatic upregulation of the strength of existing excitatory synapses, their impact on subsequent LTP induction is diverse due to their differential effect on silent synapse formation and/or presynaptic release, revealing the complex nature of the interaction between Hebbian and homeostatic plasticity.

In some of the studies discussed below probing molecular mechanisms of homeostatic synaptic plasticity downstream of RA synthesis, RA treatment was used as a proxy for homeostatic synaptic plasticity induction, much similar to the use of DHPG to induce LTD in studies probing mechanisms of mGluR-dependent LTD. In this review, we will cover cellular events and molecular mechanisms involved in regulating excitatory synaptic strength in, but not limited to, RA-dependent homeostatic synaptic plasticity.

AMPA receptor trafficking – the final common pathway for excitatory postsynaptic modification

Similar to Hebbian plasticity, homeostatic synaptic plasticity is expressed post-synaptically as modulation of synaptic strength by up- or down-regulation of AMPA receptor (AMPAR) abundance in the postsynaptic density (PSD). Regulation of AMPAR-containing trafficking vesicles can occur at multiple stages of exocytosis and endocytosis.

Within the exocytosis process, the SNARE complex molecules for AMPAR-vesicle fusion has been studied in both LTP and RA-dependent homeostatic synaptic plasticity, and was shown that the SNARE components required for both types of plasticity are partially overlapping. Both processes require the R-SNARE synaptobrevin-2 (Syb-2) and the Q-SNARE SNAP-47 [26,29]. However, the dependence on complexin and the other Q-SNARE syntaxins are different: LTP requires complexin and syntaxin-3 [29,30], while RA-induced AMPA receptor exocytosis requires syntaxin-4 and does not involve complexin [26]. Importantly, direct acute treatment with RA increases excitatory synaptic strength and blocks subsequent induction of LTP. The RA blockade of LTP can be reversed by syntaxin-4 knockdown and prevention of RA-induced AMPA receptor insertion [26], indicating that the step of AMPA receptor exocytosis may act as a critical interaction point between some forms of Hebbian and homeostatic synaptic plasticity (Figure 1).

Another key step in regulated exocytosis is Ca²⁺-trigger vesicle fusion (e.g. Ca²⁺-dependent neurotransmitter release). Both RA- and LTP-mediated AMPAR exocytosis requires NMDA receptor (NMDAR) activation (we will come back to the specific point of NMDAR-dependence later), suggesting that a Ca²⁺-triggered fusion process may be involved. Indeed, postsynaptic synaptotagmin-1 and synaptotagmin-7 act as redundant Ca²⁺-sensors for activity-regulated AMPAR exocytosis during LTP [31]. Curiously, RA-dependent AMPAR insertion is intact in synaptotagmin-1/7 double knockout neurons, indicating a yet-to-be-identified Ca²⁺-sensor is involved for AMPAR exocytosis during homeostatic plasticity [31].

Down-regulation of excitatory synaptic strength in Hebbian (i.e. LTD) and homeostatic plasticity both involve endocytosis of AMPARs. The immediate early gene Arc/Arg3.1, upregulated by elevated synaptic activity [32], seems to be a central node in AMPAR endocytosis in both processes by recruiting clathrin-dependent endocytosis machinery to AMPARs and mediating their removal from the synapse [33–35]. Interestingly, aside from modulating AMPAR endocytosis, Arc also localizes to nucleus and suppresses GluA1 transcription through expression of promyelocytic leukemia nuclear bodies [36], thus reinforcing the changes at synapse to weaken synaptic strength during homeostatic down scaling. Additionally, it was proposed that clathrin-independent constitutive endocytosis of AMPARs may be involved in homeostatic downscaling, which requires small GTPase Rac1 and F-actin [37].

In addition to components directly related to vesicular trafficking, posttranslational modification of AMPA receptors, in particular phosphorylation of AMPARs, also influence the trafficking pathways involved in synaptic plasticity. AMPAR synaptic targeting and channel properties are largely affected by phosphorylation of its C-terminal sequences [38]. Among the most studied, phosphorylation of the two serine residues in the C-terminal sequence of GluA1 (S831 and S845) appear to govern the conductance and trafficking of AMPARs in and out of synaptic membranes during LTP and LTD [39]. PKA-mediated phosphorylation of GluA1 S845 has been shown to promote plasma membrane insertion of GluA1 and synaptic retention, thereby facilitating LTP [40–43], whereas dephosphorylation of S845 by calcineurin (CaN) and other phosphatases has been correlated with AMPAR endocytosis and LTD [39,42,44]. Additionally, it has been suggested that regulation of GluA1 S845 phosphorylation by PKA and CaN is involved in AMPAR trafficking during bidirectional homeostatic synaptic plasticity in cortical neurons [45,46], but not in hippocampal neurons [47]. More recent evidence using GluA1 knockin mice lacking the two phosphorylation sites [48] further support the notion that the involvement of phosphorylation-dependent AMPAR trafficking in homeostatic synaptic plasticity may not be as universal but is brain region/neuronal type-specific [24].

Synaptic scaffold proteins – modulators of AMPAR trafficking

It is probably not surprising that many of the synaptic scaffold proteins are involved in synaptic plasticity as their roles in trafficking and synaptic stabilization of AMPARs are well established. PSD-95 and PSD-93 are membrane‐associated guanylate kinase (MAGUK) family proteins that have been shown to be involved in Hebbian and homeostatic synaptic plasticity [49–52]. MAGUK family proteins interact with many transmembrane proteins, including the transmembrane AMPAR regulatory proteins (TARPs) that are considered AMPAR auxiliary subunit instrumental for AMPAR surface and synaptic targeting [53,54]. TARPs are also known to participate in both Hebbian and homeostatic synaptic plasticity [55–58].

Given the significant roles of AMPAR C-terminal phosphorylation in its surface trafficking, scaffold proteins that anchor relevant kinases (e.g. PKA) and phosphatases (e.g. calcineurin) were also studied for their roles in synaptic plasticity. Indeed, non-MAGUK scaffold proteins such as PICK1, GRIP and AKAP150 contribute to Hebbian plasticity regulation [59–61] and homeostatic synaptic plasticity [62–65]. It is worth noting that in addition to posttranslational modification of AMPARs, kinases and phosphatases may also be recruited to dendrites and postsynaptic density to regulate other critical steps of homeostatic synaptic plasticity [24,66].

The Homer protein family is another class of postsynaptic density scaffold protein that supports postsynaptic structure and mediates postsynaptic signaling. Homer-1a, a short variant of homer-1 encoded by an immediate-early gene, was first identified as a mGluR-binding protein [67]. Regulation of mGluR1/5 trafficking by Homer1a not only affects mGluR activation, but also has been shown to be involved in Hebbian plasticity [68,69] and homeostatic plasticity [70] in in vitro studies. A recent study further explored the Homer-1a-mGluR1/5 interaction in an in vivo model of homeostatic synaptic plasticity, namely excitatory synapse weakening during sleep. In this context, synaptic targeting of Homer-1a, which removes AMPARs from postsynaptic density by activating mGluR5 and downstream signaling pathways, is coupled to the state of arousal via wake- and sleep-promoting neuromodulators [71]. Thus, the same Homer1a/mGluR-mediated signaling cascade may be used in both Hebbian and homeostatic synaptic plasticity to regulate synaptic AMPAR removal.

Metaplasticity – a mechanism that integrates homeostatic and Hebbian plasticity in health and disease?

As summarized above, various cellular processes activated during homeostatic synaptic plasticity adjust the state of the synapses (e.g. modification of synaptic AMPAR abundance) in response to activity experience. Multiple signaling pathways are altered in the process of achieving this new status quo, including the activation states of various kinases and phosphatases, the phosphorylation states of AMPARs, and the availability of synaptic slots through addition or removal of synaptic scaffolds, etc. Changes in one or a combination of these signaling pathways may impose a limitation/new constrains onto subsequent Hebbian plasticity (Figure 2). For example, it is conceivable that homeostatic upregulation of the strength of existing synapses through AMPAR insertion will constrain the ability of the affected synapses to undergo LTP but facilitate their ability for LTD within a certain time window – a process that may contribute mechanistically to the sliding threshold model. Meanwhile, the phosphorylation states of AMPARs, the availability of synaptic slots for AMPAR anchoring, and the abundance of silent synapses could also change as a result of homeostatic signaling pathway activation, adding additional complexity to the sliding threshold mechanisms. More importantly, numerous in vitro and in vivo studies show that homeostatic modifications occur at both excitatory and inhibitory synapses (reviewed in [72,73]). Synaptic inhibition is known to modulate Hebbian plasticity through regulating integration of dendritic excitatory inputs both temporally and spatially [74–79]. Homeostatic modulation of synaptic E/I balance in local dendrites may push the θ_m of a particular excitatory synapse in either directions depending on the state the synapse. In the case of increased E/I after homeostatic plasticity, for example, LTP may be facilitated at unsaturated synapses due to reduced local inhibition but may be constrained if synaptic excitation is already saturated. Thus, homeostatic modification of synaptic E/I ratio may be another metaplasticity mechanism through which sliding threshold for Hebbian plasticity may be achieved at individual synapses.

Figure 2. A schematic diagram depicting the relationship between Hebbian and homeostatic plasticity.

It is traditionally believed that network activity drifts as a result of Hebbian plasticity, which drives homeostatic synaptic plasticity (①). Here we propose the possibility that an animal’s behavioral experience may directly lead to homeostatic plasticity at both excitatory and inhibitory synapses that results in a shift in synaptic E/I balance and modification of the BCM curve. In this context, homeostatic plasticity plays the role of a metaplasticity mechanism, which in turn affects Hebbian plasticity (②).

Given the accumulating knowledge on molecular players specifically involved in homeostatic plasticity, direct testing of the hypothesis that homeostatic plasticity acts as a metaplasticity mechanism for the sliding threshold model is becoming possible and will likely be a major future direction for the synaptic plasticity field. Currently, observations from studies investigating the pathophysiology of neurological diseases provided indirect support for this hypothesis. Among these observations, a disrupted synaptic E/I balance appears to be a common theme in many neurological disorders, including but not limited to autism spectrum disorders (ASDs), schizophrenia, epilepsy and neurodegenerative disorders such as Alzheimer’s disease and Huntington’s disease [80–84]. In extreme cases, E/I imbalance leads to severe network instability, which is consistent with the fact that many neurological disorders have a comorbidity of epileptic activity in subsets of patients [85]. By contrast, intellectual disability is a much more common symptom affecting most patients with ASDs, Schizophrenia and degenerative disorders. How may impaired homeostatic plasticity and defective E/I balance contribute to cognitive dysfunctions?

Homeostatic synaptic plasticity has been relatively more extensively studied in ASDs probably because as a form of neurodevelopmental disorders, ASDs provide an attractive model system for studying long-term functional consequences of defective synaptic plasticity and circuit remodeling without apparent synapse loss or neurodegeneration. For example, mutations in Mecp2 gene, which encodes the transcriptional regulator methyl-CpG-binding protein 2 (MeCP2), leads to Rett syndrome and shows high co-morbidity with ASDs [86]. Mecp2 deletion not only leads to defective Hebbian plasticity and learning [87], but also causes impaired excitatory synaptic up-scaling in visual cortex [88] and down-scaling in hippocampus [89]. Fragile X syndrome (FXS) is another neuropsychiatric disorder characterized by developmental problems including intellectual disability, deficits in communication and social interaction, and in some cases seizures. FXS, in most cases, is caused by silencing of the Fmr1 gene and a complete loss of expression of its protein product FMRP, an RNA-binding protein known to regulate protein synthesis of a subset of neuronal transcripts [90–92]. Hyperactivity of neural network [93] and altered Hebbian plasticity [94] have been described in FXS model mice. Moreover, a complete absence of RA-dependent homeostatic synaptic plasticity at both excitatory and inhibitory synapses have been reported in both FXS mouse [22,23] and human neurons differentiated from FXS patients [95].

Defective learning and memory in various disease models are often attributed to impaired Hebbian plasticity. Altered homeostatic synaptic plasticity found in these disease models is usually considered synaptic phenotypes independent of impaired Hebbian plasticity due to the multifaceted functions of the mutated genes. However, we would like to posit here that these disease phenotypes may be more connected than expected. Most in vivo homeostatic synaptic plasticity studies have been carried out in sensory cortices, which have the advantage of being easily accessible for activity perturbation via sensory modality-specific input manipulations [96–98]. The demonstration of homeostatic plasticity in various in vivo systems not only allows validation of molecular contributors to homeostatic mechanisms in intact circuits, but more importantly, permits further exploration of functional significance of homeostatic plasticity in intact circuits in vivo. In other words, if the signaling pathway known to be specific for homeostatic synaptic plasticity is disrupted, what would be the impact on subsequent Hebbian plasticity and behavior?

A recent study using conditional RARα knockout mice investigated the role of RA-dependent synaptic signaling in whisker-based sensory processing in the barrel cortex [99]. Expression of RARα in layer 5 (L5) pyramidal neurons in the somatosensory cortex was found necessary for normal tactile sensory processing. Transcranial two-photon imaging revealed a significant increase in elimination of more mature-looking dendritic spines on apical dendrites of L5 pyramidal neurons in the absence of RARα. Consistent with RARα’s role in homeostatic plasticity, the enhancement of spine elimination was whisker experience-dependent as whisker trimming rescued the spine elimination phenotype [99]. Although mechanisms underlying whisker-dependent texture encoding remain largely unexplored, it is conceivable that RARα deletion in L5 pyramidal neurons impairs experience-dependent homeostatic synaptic plasticity that fine-tunes the strength of active synapses (i.e. a balance maintained through elimination of immature thin spines and maintenance of mature mushroom-type spines), and negatively impacts sensory information integration from L2/3 neurons to L5 neurons. Future studies are required to investigate how impaired RA signaling and homeostatic plasticity affects whisker experience-dependent Hebbian plasticity, and more broadly speaking, any experience-dependent plasticity, in the context of local circuit wiring and behavioral output.

Although often studied separately, homeostatic synaptic plasticity-inducing sensory manipulations have been shown to affect Hebbian plasticity within the same circuit. Arguably the best examples come from studies in the visual system where prolonged visual deprivation, which is known to induce homeostatic plasticity in the V1 cortical synapses, shifts the modification threshold θ_m of the Hebbian plasticity BCM curve (reviewed in [100]). Most strikingly, in adult animals, prolonged visual deprivation through dark rearing reopens visual cortical critical period and restores ocular dominance plasticity [101]. These modifications of Hebbian plasticity is thought to be achieved through experience-dependent GluN2A/GluN2B ratio shift [102–104]. Dark rearing, a widely used visual deprivation approach to study visual cortical plasticity, is known to upregulate GluN2B expression and shifts GluN2A/GluN2B ratio [105] toward favoring LTP by lowering LTP threshold [103,106]. Increases in miniature excitatory postsynaptic current (mEPSC) amplitude after dark rearing has been characterized as a typical homeostatic upregulation of excitatory synapses [107]. Intriguingly, a recent study demonstrated that mEPSC amplitude is actually potentiated through a Hebbian mechanism mediated by increased GluN2B-containing NMDAR at synapses, not by a synaptic scaling mechanism (i.e. lowering spontaneous firing) although NMDAR-independent synaptic scaling may still be induced by extreme reduction of activity [108]. This seemingly blurry distinction between ‘true’ homeostatic mechanisms and Hebbian mechanisms may reflect our incomplete understanding of biological processes involved in these different forms of plasticity, it also serves as a reminder that Hebbian and homeostatic plasticity are not two separate entities co-exist within the same system, but are intertwined and even coupled to adjust synaptic and network activity for optimal function.

Conclusions

In this short article, we summarized cellular processes and signaling molecules involved in these processes that are shared between Hebbian and homeostatic synaptic plasticity, and outlined synaptic modifications and behavioral outcomes that are likely the consequence of the interaction between these two forms of synaptic plasticity. Behavioral experiences in an animal drive network activity that may lead to Hebbian or homeostatic plasticity, or both. The latter, through its impact on synaptic excitation and inhibition, alters the modification threshold of Hebbian plasticity and its subsequent involvement in future behavior (Figure 2). The synaptic plasticity field is at an exciting time when we are equipped with rapidly growing insight on the molecular basis of plasticity processes, in conjunction with cutting-edge technologies allowing sophisticated genetic and circuit manipulations. Future experiments will be possible to systematically explore how homeostatic synaptic plasticity, as a mechanism of metaplasticity, impacts Hebbian plasticity and cognitive function.

Highlights.

• Hebbian and homeostatic synaptic plasticity converge on shared cellular processes

• Homeostatic plasticity adjusts the state of synapses to impact Hebbian plasticity

• Homeostatic plasticity alters synaptic E/I and drives the Hebbian sliding threshold

• Impaired homeostatic plasticity may be linked to cognitive deficits in ASDs