Unraveling mechanisms of homeostatic synaptic plasticity

SUMMARY

Homeostatic synaptic plasticity is a negative feedback mechanism neurons use to offset excessive excitation or inhibition by adjusting their synaptic strengths. Recent findings reveal a complex web of signaling processes involved in this compensatory form of synaptic strength regulation, and in contrast to the popular view of homeostatic plasticity as a slow, global phenomenon, neurons may also rapidly tune the efficacy of individual synapses on demand. Here we review our current understanding of cellular and molecular mechanisms of homeostatic synaptic plasticity.

INTRODUCTION

The primary function of a neuron is to receive, integrate and transmit information as an electrical or chemical signal to other neurons in the brain. In response to extrinsic stimuli, neurons can change and adapt the strength of their connections, or synapses. The most widely studied form of such activity-dependent adaptation of synaptic strength is Hebbian plasticity, which includes long-term potentiation (LTP) and its counterpart, long-term depression (LTD) (Collingridge et al., 2004; Feldman, 2009; Malenka and Bear, 2004). In Hebbian plasticity, synaptic changes are associative, rapidly induced and input specific. Because these hallmark features facilitate reinforcement of synaptic connections that are active with a given set of stimuli or ‘experience’, Hebbian plasticity has been extensively studied as a cellular basis for learning and memory (Neves et al., 2008; Sjöström et al., 2008). Nevertheless, Hebbian plasticity is a positive feedback process; for example, upon inducing LTP, synapses are more excitable and the same connections have a reduced threshold for undergoing further LTP with a propensity for runaway excitation. In order to prevent neural networks from reaching such extremes, a homeostatic negative feedback regulation that could constrain activity levels would be highly desirable for maintaining network stability, and such an idea has been supported by network models of learning (Turrigiano, 2008).

Experimental evidence for adaptive compensatory mechanisms suggestive of homeostasis in the central and the peripheral nervous systems have been first reported decades ago (Cannon, 1939; Sharpless, 1964). However, it is only in recent years that homeostatic mechanisms of neuronal circuit adaptations have been subjected to close scrutiny (reviewed in Burrone and Murthy, 2003; Davis, 2006; Davis and Bezprozvanny, 2001; Marder and Goaillard, 2006; Pérez-Otaño and Ehlers, 2005; Rabinowitch and Segev, 2008; Rich and Wenner, 2007; Shah and Crair, 2008; Thiagarajan et al., 2007; Turrigiano, 1999, 2008; Yu and Goda, 2009). The findings to date point to two major targets to achieve homeostasis: intrinsic excitability and synaptic efficacy.

This review will focus on synaptic mechanisms of homeostatic adaptations, primarily at mammalian synapses, mostly drawing on recent developments in this rapidly growing field. Collectively, investigations into the cellular properties and the underlying molecular mechanisms are beginning to unfold a complicated picture in which synapses implement homeostatic adaptations through a variety of cellular processes. The mechanisms appear to differ depending on the developmental stage, the cell type, and the mode of activity manipulation that elicits synaptic homeostasis. Moreover, these differences are further confounded by variables that are introduced by the experimental systems used. In an attempt to simplify the problem, we have divided the review into three main sections. The first part addresses the physiological relevance of homeostatic synaptic plasticity by focusing on studies carried out in preparations that retain the native neural connectivity. In the second section we consider the cellular mechanisms of expression of homeostatic plasticity in the pre- and the postsynaptic neurons. The third part examines the signaling pathways that neurons use to execute homeostatic synaptic adaptations. We conclude the review by reflecting on the overall current state of knowledge and the major issues that remain to be tackled in the future. We apologize to authors whose work could not be cited directly owing to space limitations.

PHYSIOLOGICAL RELEVANCE OF HOMEOSTATIC SYANPTIC PLASTICITY

In the intact brain, neurons are precisely organized into structured circuits that communicate between each other to perform physiological brain functions. Whereas studies in dissociated neuronal cultures have provided important insights into the cellular and molecular properties of homeostatic synaptic plasticity, a lack of network architecture typical of the intact brain may have obscured some aspects, particularly of mechanisms that rely on precise patterns of synaptic connections. The use of intact in vivo models and organotypic slice cultures that partly preserves the in vivo connectivity and its properties (De Simoni et al., 2003), provides a complementary approach to further refine findings from dissociated cell culture work. Organotypic slice preparations maintain the ease of pharmacological activity manipulations and single cell molecular interventions which are typical of neuronal cultures that facilitate studies of detailed mechanisms (e.g. Aptowicz et al., 2004; Bartley et al., 2008; Deeg, 2009; Kim and Tsien, 2008). Furthermore, in vivo models permit for directly testing the physiological relevance of mechanisms of homeostatic synaptic plasticity identified in studies in vitro (e.g. Kaneko et al., 2008; Maffei et al., 2006; Maffei and Turrigiano, 2008; Mrsic-Flogel et al., 2007). Here we focus on properties and mechanisms of homeostatic synaptic plasticity pertinent to physiological brain function by highlighting recent findings from preparations that better preserve the native pattern of synaptic connections. In addition, we further consider the pros and cons of native preparations and dissociated neuronal cultures as experimental models for studying homeostatic synaptic adaptations.

Spatial and developmental regulation

Homeostatic synaptic plasticity has been studied in vivo in the visual cortex, where experience-dependent scaling of glutamatergic synaptic responses is subject to spatial and developmental regulation (Desai et al., 2002; Goel et al., 2006; Goel and Lee, 2007). In this in vivo model, network activity is altered by intraocular injection of tetrodotoxin (TTX) to block action potentials or by manipulating sensory inputs, either by exposing to or depriving animals from light, although there are subtle differences in the mechanisms of homeostatic compensation between those induced by TTX and light deprivation (Maffei and Turrigiano, 2008). In young animals, dark rearing increased glutamatergic quantal size in visual cortical layers 4 and 2/3, which was correlated with an increased abundance of AMPA receptors; re-exposure of animals to light reversed these changes (Desai et al., 2002; Goel et al., 2006). Importantly, homeostatic changes in layer 4 were observed only if activity was blocked in early developmental stages before the end of the critical period for ocular dominance plasticity. In contrast, layer 2/3 showed no such restriction, and dark rearing of adult animals produced an increase in AMPA miniature excitatory postsynaptic current (mEPSC) amplitudes (Desai et al., 2002; Goel and Lee, 2007). Interestingly, the mechanism of homeostatic increase in mEPSCs amplitude could be different between adult and juvenile animals as only the juvenile animals showed a multiplicative scaling that affected all synapses uniformly (Goel and Lee, 2007). The underlying basis for the layer-specific, developmental differences in the homeostatic scaling mechanisms in the visual cortex remains to be delineated.

The spatial specificity of homeostatic synaptic plasticity in the adult cortex is also highlighted in the whisker-to-barrel pathway, although synaptic changes appear to manifest primarily at the level of synapse number (Knott et al., 2002; Quairiaux et al., 2007). In this system, each whisker is connected to layer 4 in the cortical barrel, which integrates signals generated by whisker movement and transduces it to layers 2/3 in the same barrel column. Persistent whisker stimulation produced a compensatory increase in the number of inhibitory synapses in layer 4 neurons; this in turn, decreased the spontaneous firing rate of layer 4 neurons and consequently that of layer 2/3 neurons. The effect of enhanced stimulation was confined to the barrel receiving inputs from the stimulated whisker and did not extend to adjacent barrel columns associated to non-stimulated whiskers.

Connection-specific regulation

Understanding how different synapses within a same network adapt to long-term alterations in activity poses a problem of a higher degree of complexity. Whereas the original studies on modulating network activity in dissociated cultures have revealed uniform multiplicative synaptic scaling (e.g. Turrigiano et al., 1998), recent work in vivo and in organotypic slices indicates that synaptic scaling is not always uniform and suggests that changes in network activity do not affect all synaptic inputs onto a given neuron equally (e.g. Cingolani and Goda, 2009; Echegoyen et al., 2007; Goel and Lee, 2007). Moreover, hippocampal and cortical networks display differential regulation of homeostatic synaptic plasticity for specific connections (Bartley et al., 2008; Kim and Tsien, 2008). For example, in hippocampal slice cultures, TTX-treatment strengthened CA3-to-CA1 synapses while inputs onto CA3 cells were regulated differentially depending on their origin: recurrent CA3-to-CA3 inputs were weakened whereas mossy fiber-to-CA3 connections became stronger (Kim and Tsien, 2008).

In the neocortical inhibitory circuits, inputs from two subtypes of inhibitory neurons, parvalbumin-positive (Parv) and somatostatin-positive (Som) neurons, were shown to adapt differentially to chronic action potential blockade (Bartley et al., 2008). In TTX-treated neocortical slice cultures, the inhibitory drive of Parv neuronal inputs onto excitatory neurons was decreased, at least in part due to reduced synapse number, whereas that of Som neurons remained unchanged. In contrast, when short-term plasticity was examined, Som but not Parv inputs showed enhanced depression following TTX. Notably, Parv neurons preferentially synapse onto the soma and proximal dendrites of excitatory neurons whereas Som neurons target distal dendrites (e.g. Somogyi and Klausberger, 2005). The differential tuning of two types of inhibitory connections, therefore, supports the idea that the receiving dendrite could homeostatically regulate synaptic strengths in a subcompartment-specific manner (see below: e.g. Branco et al., 2008; Sutton et al., 2006). Such local homeostatic adjustments of synaptic strengths might be advantageous for dendritic integration as it allows for a flexible control of the excitability of each dendritic branch independently of one another (Häusser and Mel, 2003; Polsky et al., 2004; Rabinowitch and Segev, 2006, 2008).

Excitatory-inhibitory balance

In concert with up-regulation of excitatory inputs, modulation of inhibitory synaptic transmission is crucial for restraining network activity and avoiding potential epileptogenic states (Treiman, 2001). For instance, chronic TTX delivery in the intact hippocampus produces an increase in the amplitude and frequency of spontaneous inhibitory currents in addition to enhanced excitation (Echegoyen et al., 2007). Moreover, in the CA3 region of hippocampal slice cultures, upon activity blockade with glutamate receptor antagonists, the level of the GABA-synthetic enzyme glutamate decarboxylase isoform GAD65, and GABA_A receptor α1 subunits was maintained rather being reduced (Buckby et al., 2006). These observations are in sharp contrast with findings from dissociated cortical cultures in which activity suppression scales down synaptic GABA currents (Hartman et al., 2006; Kilman et al., 2002). Inhibitory connections in native circuits may thus be wired to provide an additional layer of negative feedback control that operates in conjunction with homeostatic up regulation of excitatory inputs (Karmarkar and Buonomano, 2006). In support of such a proposal, in organotypic hippocampal slices, changes in excitation and inhibition elicited by chronic modulation of network activity are dissociable and are expressed in temporally distinct order, with changes in excitation occurring prior to those of inhibition (Karmarkar and Buonomano, 2006).

An additional complexity to the regulation of excitatory-inhibitory balance is suggested by the differential developmental dependence of excitatory and inhibitory inputs to homeostatic modulation (Echegoyen et al., 2007; Karmarkar and Buonomano, 2006). In the intact hippocampus, TTX treatment increased mEPSC amplitude only in juvenile animals and mEPSC frequency in both adult and juvenile animals. As for inhibitory inputs, TTX increased mIPSC amplitude in both adult and juvenile animals while mIPSC frequency was increased only in adult animals (Echegoyen et al., 2007). Curiously, the age-dependency of quantal responses has been demonstrated in dissociated cultures (e.g. Burrone et al., 2002; Han and Stevens, 2009; Hartman et al., 2006; Wierenga et al., 2006), albeit not necessarily in the same direction. For instance, in dissociated hippocampal cultures, TTX treatment reduced mIPSC amplitude in both mature and young cultures whereas it reduced mIPSC frequency significantly in young cultures only (Hartman et al., 2006).

The observed developmental dependence of homeostatic synaptic changes are not surprising given the higher rate of synapse remodeling in developing networks and GABA_A responses that are depolarizing in young neurons. Regardless, many other factors that shape synaptic activity are likely to contribute to the heterogeneity of excitatory and inhibitory homeostatic responses, such as the developmental changes in the spatial distribution of ion channels (Lai and Jan, 2006) and associated changes in intrinsic excitability (Debanne et al., 2003; Marder and Goaillard, 2006; Zhang and Linden, 2003), as well as the vast number of distinct GABAergic cell types with specific functions in influencing circuit behavior (Klausberger and Somogyi, 2008).

Hebbian vs. homeostatic synaptic plasticity

Hebbian forms of plasticity, such as LTP, rapidly modify the efficacy of individual synapses associatively in an input-specific manner, and they are thought to represent the cellular mechanisms for storing memories (Collingridge et al., 2004; Feldman, 2009; Malenka and Bear, 2004; Neves et al., 2008; Sjöström et al., 2008). LTP represents a positive feedback mechanism, and once induced, the induction threshold for further rounds of potentiation is successively reduced such that runaway excitation could potentially occur. Neurons have however developed homeostatic mechanisms that sense and prevent saturated synapses from undergoing further potentiation (Roth-Alpermann et al., 2006; Seeburg and Sheng, 2008) and maintain the stability of network activity within a set range that could help preserve stored information (Davis, 2006; Turrigiano, 2008). In a prevailing view, the strength of all synapses received by a neuron is scaled up or down by a multiplicative factor, which may be crucial for maintaining the relative differences in the strengths between synapses (Turrigiano, 2008). However, in addition to the findings from organotypic cultures and in vivo studies discussed above, several studies in dissociated cultures have demonstrated non-uniform scaling of synaptic strengths in response to chronic activity modulation. Homeostatic synaptic changes were found to be more potent at larger synapses than smaller synapses (Minerbi et al., 2009; Thiagarajan et al., 2005), and also, within a dendritic tree, the changes could be restricted to small subsets of synapses along a dendritic branch (Branco et al., 2008; Hou et al., 2008; Ju et al., 2004; Sutton et al., 2006; see above).

How might neurons discriminate between non-uniform homeostatic adaptations and input-specific changes when these two opposing forms of plasticity are expressed in overlapping domains? An appealing mechanism by which local homeostatic changes could at least be implemented without losing the relative differences in synaptic weights created by input-specific changes was proposed by Rabinowitch and Segev (2006, 2008). In their model, when a given synapse undergoes LTP, then the strengths of synapses adjacent to the potentiated synapse are compensated homeostatically by weakening. This local interplay of Hebbian and homeostatic changes between neighboring synapses would help maintain the relative differences in synaptic strengths while keeping the overall activity in a dendritic branch constant. In agreement with such a proposal, previous studies have reported of heterosynaptic depression that accompanies LTP induction in a population of neurons in the hippocampus (Abraham and Goddard, 1983; Lynch et al., 1977; Scanziani et al., 1996) and the amygdala (Royer and Pare, 2003). However, in these studies, the precise relationship between neighboring synapses along a local dendritic region has not been directly examined. While this model awaits experimental testing, the local expression of LTP and homeostatic synaptic plasticity could be coordinated by secreted molecules such as BDNF (see below), that are shared between mechanisms of these two forms of plasticity. Furthermore, following LTP induction at a given synapse, local elevation in intracellular Ca²⁺ together with signaling proteins that escape from activated spines, such as the small GTPase Ras (Harvey et al., 2008), could help delimit homeostatic plasticity to adjacent synapses by interacting with molecules/signaling events which are specific for homeostatic changes, involving, for example, TNFα (see below; Stellwagen and Malenka, 2006) and the GluA2 cytoplasmic tail (see below; Gainey et al., 2009).

In vivo circuits vs. dissociated neuronal cultures

Notably, studies in vivo and organotypic cultures are revealing properties of homeostatic synaptic plasticity that are not necessarily shared by findings in dissociated cultures, such as the differences in developmental dependence of inhibitory quantal responses and the prevalence of non-multiplicative synaptic scaling described above. While the divergent findings may be disconcerting at the outset, given that basic properties of the mechanisms of neurotransmitter release and reception are conserved at synapses formed in vitro compared to synapses formed in vivo, one of the main reasons contributing to the differential homeostatic synaptic responses could be the organization of synaptic connections. For example, in dissociated cultures, the relative proportion of excitatory and inhibitory neurons is variable, let alone the different types of glutamatergic and GABAergic neurons present. Their abundance and survival are subject to a variety of experimental factors, such as the developmental stage of the animal when the tissue is harvested and growth conditions in culture (Banker and Goslin, 1998). Moreover, microcircuits formed in dissociated cultures, including feedback and feedforward inhibition, are likely to be defined by chance positioning of neurons and extension of neuronal processes to form synaptic contacts. Also, when comparing hippocampal dissociated cultures – a popular primary neuronal culture model – to intact hippocampal circuits, dissociated cultures do not encounter alterations of network activity arising from adult neurogenesis in the dentate gyrus (Li and Pleasure, 2010). Given such major differences in synaptic network organization between the two systems, it is expected that neurons in vitro experience activity patterns that are distinct from those in vivo. Consequently, activity manipulation in the two systems, even if they derive from the same brain region, can recruit divergent mechanisms to achieve homeostasis.

Beyond the obvious differences in synaptic connectivity between in vivo and in vitro systems, including a lack of external driving inputs in vitro, there could be other differences in molecular factors which shape network activity, such as the expression pattern of ion channels, their subunit composition (e.g. Marder and Goaillard, 2006) and their subcellular distribution (e.g. Lai and Jan, 2006), that give rise to differential homeostatic responses. Moreover, there is a noted difference in basal synaptic activity between in vivo and in vitro preparations. For example, mEPSCs are much more frequent in dissociated hippocampal cultures compared to organotypic or acute hippocampal slices (e.g. 10-fold difference in mEPSC frequency between dissociated and organotypic hippocampal cultures: Cingolani and Goda, 2009) despite a similar or a lower synapse density in dissociated cultures compared to the brain (Boyer et al., 1998). While the nature of the difference in release machinery between these systems remains to be understood, basal dendritic activity would be considerably different between neurons in dissociated hippocampal cultures compared to native hippocampal circuits that see much fewer quantal excitatory events.

Altogether, when studying homeostatic synaptic plasticity, the choice of the preparation is crucial with respect to the specific problem being addressed. If on the one hand, the question of interest involves cellular and molecular mechanisms of homeostatic synaptic plasticity at individual synapses, then dissociated cultures would be an optimal starting point. This is because dissociated cultures form relatively simple circuits, and the exact synaptic connectivity could be easily mapped out amongst groups of neurons. This contrasts to intact systems in which identifying presynaptic neurons that form specific connections onto a dendrite of postsynaptic neurons poses a big challenge due to dense packing of neurons and synapses. Also, in neuronal cultures, reagents can permeate the preparation readily by virtue of being grown on a flat surface, and such a property makes cultured neurons highly amenable to molecular and pharmacological perturbations that are essential for dissecting detailed mechanisms. If on the other hand, the problem of homeostatic synaptic plasticity mechanisms concerns a higher degree of interaction between synaptic connections and circuits, for instance, the way in which neurons sense the need for homeostatically adapting their excitability in the first place, or for identifying the biological functions of homeostatic synaptic plasticity, such studies would be physiologically relevant if they are based on preparations that preserve the native synaptic connectivity. Recent technological advances in molecular manipulation of identified neuronal population in the brain are providing means to bridge the gap between in vitro and in vivo models. One could perform experiments in the intact brain similarly to those performed in vitro, albeit for a synapse population, by targeted gene expression in the mouse brain or delivering recombinant viral vectors in specific brain regions, to control and measure neuronal activity using optogenetic probes (O’Connor et al., 2009; Scanziani and Häusser, 2009) while interfering with function of molecules implicated in homeostatic synaptic plasticity, for instance, by conditional loss of protein expression.

CELLULAR MECHANISMS OF EXPRESSION OF HOMEOSTATIC SYNAPTIC PLASTICITY

Over the past decade, homeostatic synaptic plasticity at mammalian synapses has been typically studied in rodent dissociated primary neuronal cultures, a popular model system in which activity can be readily manipulated. However, increasingly, the expression of homeostatic synaptic plasticity has been further investigated in more intact preparations such as slice cultures and in vivo as discussed above. Collectively, these studies have revealed that synaptic strengths are compensated by altering presynaptic neurotransmitter release and controlling the abundance of postsynaptic receptors. The emerging view highlights the involvement of multiple parallel mechanisms with distinct changes in synaptic properties being elicited for different experimental conditions such as the method used to manipulate activity, the preparation and the source of neurons, and their age (e.g. Moulder et al., 2006b; Murthy et al., 2001; O’Brien et al., 1998; Sutton et al., 2006; Turrigiano et al., 1998). We next address the mode of expression of homeostatic synaptic plasticity in the pre and the postsynaptic neurons in turn (Figure 1).

Figure 1. Basic scheme of homeostatic synaptic plasticity at an excitatory synapse.

a. Basal conditions: Synaptic transmission is mediated via the liberation of neurotransmitters from the presynaptic terminal and subsequent activation of receptors on the postsynaptic cell. The efficacies of neurotransmitter release and reception are major determinants of pre- and postsynaptic strengths, respectively. b,c. Neurons offset imposed changes in network activity by adapting their pre and postsynaptic strengths. (b) Reduced activity is offset presynaptically by enhancing the recycling of vesicles, the number of docked vesicles, and the release probability. Postsynaptically, additional neurotransmitter receptors are incorporated at the synapse by a mechanism involving lateral diffusion from extrasynaptic sites and exocytosis from intracellular pools. (c) To compensate an increased network activity, presynaptic neurons decrease their release probability while postsynaptic cells reduce the number of postsynaptic receptors by endocytosis or by lateral diffusion from synaptic to extrasynaptic sites. Depending on the developmental stage and on experimental conditions, pre- and postsynaptic changes can occur concurrently or separately. Glial cells (shown in grey) can also contribute to the changes in synaptic strength, for example, by secreting soluble factors that signal through cell surface receptors. See text for details.

Presynaptic expression of homeostatic synaptic plasticity

Homeostatic regulation of synaptic transmission involving changes in presynaptic activity has been extensively studied at the Drosophila neuromuscular junction (NMJ). Earlier reports demonstrated an involvement of a retrograde signaling process from muscles that altered neurotransmitter release from presynaptic motor neurons. Compromising muscle excitability either by the loss of postsynaptic glutamate receptors DGluRIIA or by overexpressing a constitutively active protein kinase A (PKA) that phenocopied the loss of DGluRIIA, concomitantly increased neurotransmitter release from the motor neuron. Importantly, this effect counterbalanced the decreased postsynaptic efficacy and served to maintain muscle excitation to nerve stimulation (Davis et al., 1998; Petersen et al., 1997; also see Davis and Goodman, 1998).

Subsequently, it was shown also in cultured rodent neurons that synaptic efficacy could be adjusted by modifying presynaptic function in a direction compensatory to changes in network activity. For instance, in cultured hippocampal neurons, chronically suppressing neuronal activity by TTX or glutamate receptor antagonists enlarged active zones and increased the number of docked vesicles (Moulder et al., 2006a; Murthy et al., 2001). Such changes implied an increase in the number of readily releasable vesicles, and in turn, neurotransmitter release probability (Murthy et al., 2001). Chronic activity suppression also increased the frequency of mEPSCs and enhanced vesicle recycling activity as detected by styryl dyes or synaptotagmin luminal domain antibody uptake assays (Bacci et al., 2001; Burrone et al., 2002; Han and Stevens, 2009; Moulder et al., 2006a; Thiagarajan et al., 2005). In contrast to activity suppression and in line with the nature of homeostatic synaptic plasticity as a negative feedback process, persistent elevation of network activity in neuronal cultures decreased presynaptic release at excitatory synapses: the number of readily releasable synaptic vesicles and the probability of neurotransmitter release were both diminished (Branco et al., 2008; Moulder et al., 2004, 2006a). Presynaptic expression of homeostatic changes is not limited to targeting the synaptic vesicle cycle and the efficacy of neurotransmitter release. Altered expression of the glutamate transporter v-Glut1 could also accompany the bi-directional homeostatic adaptations in presynaptic efficacy to change the glutamate content in synaptic vesicles (e.g. De Gois et al., 2005; Wilson et al., 2005).

In studies in cultured neurons described above, neuronal activity was modified globally by bath applying pharmacological reagents. Thus, while the experiments did not discriminate the behavior of individual synapses, they suggested that at least all synapses could undergo presynaptic changes in response to altered network activity. A recent study, however, used local electrical stimulation in combination with styryl dyes to visualize vesicle turnover at single synapses to show that release probability could be homeostatically adapted in a subset of synapses via spatially restricted mechanisms (Branco et al., 2008). Moreover, as reported for homeostatic adaptations at the Drosophila NMJ, postsynaptic receptor activity was required for eliciting the compensatory change in presynaptic release. These findings suggest that mammalian central synapses also use retrograde signaling for homeostatic modulation of presynaptic release and crucially, that such regulation can occur locally.

Postsynaptic expression of homeostatic synaptic plasticity

Following from original studies demonstrating the scaling of AMPA mEPSC amplitudes upon chronic activity modulation (O’Brien et al., 1998; Turrigiano et al., 1998), much effort has been made towards delineating the postsynaptic mechanisms of homeostatic synaptic plasticity, particularly at excitatory synapses (Turrigiano, 2008). Fast excitatory synaptic transmission is mediated by AMPA receptors consisting of heterotetrameric combinations of GluA1-A4 subunits (Hollmann and Heinemann, 1994). Changes in postsynaptic strength that occur during homeostatic plasticity result from alterations in the composition and the abundance of synaptic AMPA receptors, which are regulated by multiple processes: exo/endocytic membrane traffic, lateral diffusion of cell surface receptors between extrasynaptic and synaptic sites, and by the interactions of receptor subunits with associated proteins such as TARPs and cytoplasmic scaffold proteins (Bredt and Nicoll, 2003; Collingridge et al., 2004; Elias and Nicoll, 2007; Newpher and Ehlers, 2008; Nicoll et al., 2006; Santos et al., 2009; Sheng and Hoogenraad, 2007; Shepherd and Huganir, 2007; Triller and Choquet, 2008). Therefore, postsynaptic expression of homeostatic plasticity could rely on a variety of mechanisms that mediate activity-dependent delivery and stabilization of synaptic AMPA receptors of distinct composition (e.g. Ehlers et al., 2007; Harms et al., 2005; also see Makino and Malinow, 2009).

The accumulation of postsynaptic AMPA receptors that occurs upon chronic activity deprivation at least partly involves membrane insertion of newly synthesized AMPA receptors (Ju et al., 2004; Sutton et al., 2006). Interestingly, rather than being synthesized and delivered from the soma, AMPA receptors are locally translated in the dendrites, as new receptor subunits still accumulate when dendrites are physically isolated from the soma (Ju et al., 2004). Local translation of AMPA receptors for synaptic scaling is further supported by the finding that newly synthesized receptors are inserted to membranes within defined dendritic regions deprived of activity using microperfusion (Sutton et al., 2006; but see Ibata et al., 2008). Nevertheless, to what extent the expression of homeostatic synaptic plasticity is spatially confined, and where and how the increase involves AMPA receptors of specific subunit composition, which is thought to determine the trafficking mode of AMPA receptors (Bredt and Nicoll, 2003; Shepherd and Huganir, 2007), remain to be clarified. Global regulation of AMPA receptors requiring somatic protein synthesis could contribute to mechanisms of homeostatic synaptic scaling under some conditions (Turrigiano, 2008).

No general agreement has been found to date regarding the preferential homeostatic insertion of GluA1 and/or GluA2-containing AMPA receptors at synapses of activity deprived neurons. Studies have reported of postsynaptic recruitment of GluA1 and not GluA2 (Ju et al., 2004; Sutton et al., 2006; Thiagarajan et al., 2005) or of both GluA1 and GluA2 subunits in response to inactivity (O’Brien et al., 1998; Sutton et al., 2006; Wierenga et al., 2005). The apparent differential accumulation of postsynaptic GluA1 or GluA2 receptors under different experimental conditions could represent the existence of multiple types of synaptic scaling. In another possibility, such differences could imply a two-step process of synaptic AMPA receptor accumulation as has been suggested for LTP (Plant et al., 2006), in which homomeric GluA1 subunits are inserted to the synapses first, followed by a gradual incorporation of GluA2-containing AMPA receptors. In this case, for some conditions (e.g. Ju et al., 2004; Thiagarajan et al., 2005), synaptic scaling would be suggested to proceed as far as the first step of GluA1 insertion. In agreement with such a proposal of sequential insertion of AMPA receptors, combined treatment of hippocampal neurons with TTX and the NMDA receptor antagonist APV, triggered a rapid and local translation of GluA1. Functional incorporation of GluA1 into synapses could be detected within 1 h of APV application but not at later times. Concurrently, additional GluA2 subunits were recruited to synapses over a period requiring several hours (Sutton et al., 2006; also see Hou et al., 2008). Note that following synaptic insertion of GluA1, later insertion of GluA2 might be required to counteract the increased Ca²⁺ permeability contributed by the GluA2-lacking AMPA receptors with the effect of reverting synapses back into the basal state, similarly to the regulated insertion of GluA2-containing AMPA receptors by Ca²⁺-influx through Ca²⁺-permeable AMPA receptors in cerebellar stellate cells (Liu and Cull-Candy, 2000). Interestingly, synaptic scaling of mEPSC amplitudes induced by TTX (and also a local form of synaptic scaling induced by genetically silencing a small number of presynaptic inputs) required GluA2-lacking AMPA receptors, as synaptic scaling was blocked by pharmacological inhibition of GluA2-lacking AMPA receptors (Hou et al., 2008). Altogether, synaptic insertion of GluA1 could provide a requisite signal for synaptic delivery of GluA2 for synaptic scaling. In such a case, what controls the synaptic delivery of GluA1? Interestingly, the finding of a rapid induction of GluA1 synthesis upon APV application highlights an inhibitory role for basal NMDA receptor-mediated Ca²⁺signaling in suppressing local GluA1 synthesis under basal conditions (Sutton et al., 2006). It remains to be clarified if the release from the negative feedback signals downstream to NMDA mEPSCs alone is sufficient for synaptic scaling of GluA1 and GluA2 or if additional signals are involved and how such changes are implemented.

Recently, in visual cortical neurons, GluA2 subunit was suggested to be necessary for synaptic scaling induced by TTX (Gainey et al., 2009). Knock-down of GluA2 by shRNA prevented the synaptic accumulation of AMPA receptors. Moreover, the C-terminal tail of GluA2 was crucial as synaptic scaling could be rescued by exogenous expression of GluA2 resistant to shRNA but not by a chimeric GluA2 in which the C-terminal domain was exchanged with that of GluA1. It would be of interest to examine how general the requirement for the GluA2 C-terminal tail is for different experimental paradigms of synaptic scaling, and whether GluA2 C-terminal-dependent synaptic scaling also requires GluA2-lacking AMPA receptors as reported for hippocampal neurons.

SIGNALING PATHWAYS UNDERLYING HOMEOSTATIC SYNAPTIC PLASTICITY

With increasing research efforts, our knowledge about the cellular properties of homeostatic synaptic plasticity in a number of systems has rapidly advanced. Nevertheless, we still understand very little of the underlying molecular mechanisms of the negative feedback process by which synaptic strength is adjusted. A fragmentary picture is emerging through the identification of molecules whose loss of function interferes with the experimental expression of homeostatic synaptic plasticity. These molecules represent a variety of classes of cellular functions ranging from transcription and translation, trophic signaling, to cell-cell adhesion (Figure 2). Such an array of players, in turn, emphasizes the complex nature of signaling pathways that neurons use to implement the homeostatic synaptic changes. The particular pathway(s) engaged could depend on a specific set of conditions attributable to the experimental system being used or the state of neurons shaped by the past history of synaptic activity. In this section we highlight some of these molecular players.

Figure 2. Summary of the molecular mechanisms underlying homeostatic synaptic plasticity.

Changes in network activity are detected by an unknown mechanism by neurons or glial cells and activate intracellular mechanisms to modify presynaptic release and/or the abundance of functional postsynaptic receptors. This can involve activation of gene expression in neurons and triggering of local dendritic protein synthesis (e.g. Arc/Arg3.1, AMPA receptors) as well as the release of soluble factors such as BDNF or TNFα from neurons and glial cells that then engage additional signaling pathways. See text for details.

Activity-induced gene expression and local protein synthesis

Homeostatic synaptic plasticity is modulated by transcriptional events (e.g. Ibata et al., 2008; Han and Stevens, 2009) and protein synthesis, including local dendritic translation (e.g. Arc/Arg3.1 and GluA1; see below). Persistent changes in synaptic activity trigger signaling cascades that activate transcription (Cohen and Greenberg, 2008), and newly transcribed mRNAs are packaged into granules and transported across the dendritic tree. Subsequent local translation of mRNAs is thought to supply proteins used for strengthening or weakening of synaptic efficacy (Bramham and Wells, 2007; Martin and Zukin, 2006). Note that local protein synthesis may not necessarily incur new gene expression and occur by translating mRNAs that are already present in dendrites (Steward and Schuman, 2003).

Arc/Arg3.1

A series of recent studies have highlighted synaptic functions of Arc/Arg3.1 protein in regulating synaptic plasticity, including homeostatic synaptic plasticity (Tzingounis and Nicoll, 2006). Arc/Arg3.1 is encoded by an immediate-early gene that is rapidly expressed in glutamatergic neurons by strong synaptic stimulation. Following the transport of its mRNA along the dendrite, Arc/Arg3.1 is locally translated at active synapses (Bramham and Wells, 2007; Guzowski et al., 2005), where it forms a complex with dynamin and endophilin to facilitate AMPA receptor endocytosis (Chowdhury et al., 2006). The activity-dependent up-regulation of Arc/Arg3.1 protein levels and the reduction of synaptic AMPA receptor abundance by the Arc/Arg3.1 protein (Chowdhury et al., 2006; Rial Verde et al., 2006; Shepherd et al., 2006) point to its role in homeostatic synaptic scaling (Shepherd et al., 2006). Indeed, cultured neurons from Arc/Arg3.1 knock-out mice showed increased basal synaptic strength, and did not undergo homeostatic scaling up of AMPA receptors; moreover, exogenous expression of Arc/Arg3.1 reduced synaptic strength on its own and interfered with scaling up of AMPA receptors induced by chronic activity block (Shepherd et al., 2006).

Arc/Arg3.1 regulates other forms of synaptic plasticity in addition to homeostatic plasticity, such as NMDA receptor-dependent and mGluR-dependent forms of LTD and late phase LTP (Guzowski et al., 2005; Park et al., 2008; Plath et al., 2006; Rial Verde et al., 2006; Waung et al., 2008). Neurons must therefore tightly regulate the actions of Arc/Arg3.1 in tune with the specific type of synaptic strength changes being engaged, which likely involves the interplay between different intracellular signaling pathways.

eEF2

One mechanism by which synaptic activity could directly control local dendritic protein synthesis is by modifying the phosphorylation state of translational effectors, such as eukaryotic elongation factor-2 (eEF2). Chronic silencing of synaptic inputs increases the dephosphorylated, active form of eEF2, which promotes dendritic protein synthesis; in contrast, spontaneous neurotransmitter release enhances phosphorylation of eEF2 that renders it inactive and suppresses protein synthesis (Sutton et al., 2007; Sutton et al., 2004; but see Park et al., 2008). Thus eEF2 acts as a local sensor that inversely couples synaptic transmission to local protein synthesis. Such translational control would be an attractive mechanism for homeostatic synaptic modulation if, for example, synaptic silencing that activates eEF2 drives translation of target proteins that facilitate excitatory synaptic strength.

Many other effectors molecules, including microRNA, participate in neurotransmitter-coupled pathways to regulate dendritic protein translation (Bramham and Wells, 2007). Further work is needed to decipher whether and how these different pathways interact and contribute to homeostatic synaptic plasticity.

Secreted molecules

Secreted molecules play a role in shaping homeostatic adaptation of synaptic strength. Here we discuss examples representing three different types of molecules: a neurotrophin mediating neuron-to-neuron communication, a cytokine derived from glia that acts on neurons, and a membrane permeant small molecule. Involvement of secreted molecules that diffuse across extracellular space may suggest of a global mode of synapse modulation, such as that required for multiplicative synaptic scaling. Nevertheless, local regulation could also be achieved via spatially targeted secretion of molecules coupled with their interaction with the extracellular matrix proteins (Hynes, 2009). Thus, the actions of secreted molecules are not necessarily limited to coordinating synaptic plasticity broadly amongst a large population of synapses but they may also regulate local forms of homeostatic feedback signaling at individual or a small number of neighboring synapses.

BDNF

Brain-derived neurotrophic factor (BDNF) is amongst one of the first molecules identified to play a role in homeostatic synaptic plasticity. Its precursor, pro-BDNF is synthesized by excitatory and inhibitory neurons, and it is processed and stored intracellularly as BDNF (Matsumoto et al., 2008) or processed to BDNF extracellularly by plasmin (Lu et al., 2008). In hippocampal neurons, BDNF is released by a Ca²⁺-dependent mechanism, most effectively following a patterned electrical stimulation (Balkowiec and Katz, 2002), and its release can be modulated by synaptotagmin-IV (Dean et al., 2009). Once released, BDNF binds to TrkB receptors to initiate signaling cascades important for controlling synaptic plasticity (Carvalho et al., 2008; Minichiello, 2009).

In TTX-treated cortical cultures, exogenous application of BDNF altered quantal size differentially at excitatory synapses formed onto pyramidal neurons and those formed onto interneurons: synaptic scaling of AMPA currents induced by TTX alone was prevented at pyramidal-pyramidal synapses whereas it was enhanced at pyramidal-interneuron synapses (Rutherford et al., 1998). Furthermore, BDNF treatment increased the firing rate of interneurons but not of pyramidal neurons, while co-application of BDNF and TTX attenuated the increase in firing rate observed with TTX alone in both pyramidal neurons and interneurons (Rutherford et al., 1998; Desai et al., 1999). In addition, when GABAergic synapses were examined in cultured hippocampal neurons, co-application of BDNF with TTX blocked the reduction in mIPSC amplitude that occurred with TTX alone (Swanwick et al., 2006).

Collectively, these findings suggest that BDNF can negatively control homeostatic up-regulation of dendritic excitability by preferentially promoting inhibitory synaptic activity. In addition, BDNF can also influence synaptic strength changes by modulating local dendritic protein synthesis, and BDNF itself can be locally translated by activity (e.g. Aakalu et al., 2001; Bramham and Wells, 2007; Lu et al., 2008). Further work is needed to clarify the multifaceted actions of BDNF in homeostatic synaptic plasticity. In particular, a better understanding of the spatio-temporal control of BDNF expression and its release (e.g. Dean et al., 2009) and the mechanisms by which BDNF-TrkB signaling exerts its cell-type specific effects on synaptic strength changes are warranted.

TNFα

Tumor necrosis factor-α (TNFα) is a cytokine which is required for synaptic scaling of excitatory and inhibitory synapses in vivo and in vitro (Kaneko et al., 2008; Stellwagen and Malenka, 2006), and its role in synapse modulation highlights a crossover of immune system/inflammatory response mechanisms to the CNS (Di Filippo et al., 2008; Goddard et al., 2007). Previous work in hippocampal neurons had shown that TNFα promoted the delivery of GluA2-lacking AMPA receptors to the cell surface and the removal of GABA_A receptors from the cell surface (Beattie et al., 2002; Stellwagen et al., 2005). Interestingly, a follow-up study found that chronic activity suppression triggered slow secretion of TNFα from glial cells, and the build-up of TNFα increased postsynaptic AMPA receptor levels in a homeostatic manner through a mechanism requiring TNFα receptors (Stellwagen and Malenka, 2006).

The glial origin for TNFα in controlling synaptic scaling was supported by the following observations. First, wild type neurons did not show synaptic scaling when co-cultured with glial cells from TNFα knockout mice. Second, whereas neurons from TNFα knockout mice did not show synaptic scaling when cultured on glial cells from TNFα knockout mice, culturing on wild type glial cells was sufficient to restore synaptic scaling in TNFα knock-out neurons. At excitatory synapses, the up-regulation of synaptic AMPA receptors by TNFα may involve β3 integrin signaling pathway (Cingolani et al., 2008) whereas the TNFα-dependent signaling mechanism for down-regulating GABA_A receptors at inhibitory synapses is not clear.

Importantly, TNFα is selectively involved in homeostatic synaptic scaling induced by activity deprivation, and it is not required for synaptic scaling induced by enhanced activity (Stellwagen and Malenka, 2006). Moreover, TNFα is not required for Hebbian forms of synaptic plasticity (Stellwagen and Malenka, 2006). Therefore, although homeostatic synaptic plasticity is bi-directional, the underlying mechanisms do not necessarily control receptor abundance in both directions and also be engaged in Hebbian changes. That TNFα KO mice show normal LTP and LTD but are impaired for a component of experience-dependent visual cortical plasticity in vivo underscores an important physiological function for TNFα-dependent form of homeostatic synaptic scaling in the intact brain (Kaneko et al., 2008; Stellwagen and Malenka, 2006). Furthermore, the role of glia-derived TNFα in synaptic scaling emphasizes the critical contribution of glial cells in shaping synaptic efficacy as part of the neuron-neuron-glia “tripartite” synapse (Perea et al., 2009).

Retinoic acid

Retinoic acid (RA), also called vitamin A, is a recent addition to the list of diffusible molecules involved in homeostatic synaptic plasticity. Although RA is primarily known for regulating gene expression during development, it also plays important roles in the adult brain, including LTP and LTD (Lane and Bailey, 2005).

In a recent study using dissociated and organotypic hippocampal cultures, synaptic scaling induced by blocking neuronal activity with TTX and APV, but not with TTX alone, was shown to accompany increased RA synthesis (Aoto et al., 2008). Notably, applying RA by itself rapidly scaled up AMPA receptors, and this occluded synaptic scaling induced by TTX and APV. This form of synaptic scaling promoted local GluA1 synthesis through signaling via the RA receptor, RARα, which was dendritically localized (Aoto et al., 2008; Maghsoodi et al., 2008). The requirement for concurrent NMDA receptor block with TTX for scaling up of GluA1 is consistent with a previously suggested role for basal synaptic NMDA receptor activity in suppressing local translation of GluA1 (Sutton et al., 2006). Moreover, this finding demonstrates a role for RA signaling in regulating local GluA1 synthesis. It would be of interest to further delineate the molecular mechanism by which NMDA receptor-dependent signaling controls RA production.

RA signaling is involved in various forms of synaptic plasticity, and therefore different members of the RA receptor family might mediate particular forms of synaptic plasticity, such as RARα for synaptic scaling (Aoto et al., 2008; Maghsoodi et al., 2008) and RXRγ for LTD (Chiang et al., 1998). Furthermore, regulation of numerous neuronal genes by retinoid signaling suggests that RA could affect synaptic plasticity in a multifaceted manner (Lane and Bailey, 2005).

Cell adhesion molecules

Cell adhesion molecules stabilize synapses and mediate cell-cell or cell-extracellular matrix signaling. In addition to their structural function that is particularly important in synapse formation, recent studies highlight an increasing role for cell adhesion molecules in modulating synaptic efficacy, including homeostatic adaptations. Integrins, as extracellular matrix receptors, could communicate changes in the extracellular matrix, such as those resulting from activity-dependent secretion of signaling proteins, to intracellular signaling pathways or actin scaffolds that control dendritic spines, and in turn, regulate synaptic strength changes. Homophilic or heterophilic adhesion proteins that link the pre and the postsynaptic sides of a synapse could coordinate changes in neurotransmitter release and postsynaptic receptors during homeostatic adaptation. Unlike secreted molecules that in principle diffuse across space, synapse adhesion proteins are membrane anchored to mediate local changes at individual synapses. Note, however, that regulated trafficking of adhesion proteins, for example, activity-dependent endocytosis of N-cadherin (Tai et al., 2007; Yasuda et al., 2007), could also modify sensitivity of individual synapses to synapse adhesion protein-dependent adaptive responses.

Integrins

Integrins are heteromeric transmembrane receptors for extracellular matrix and counterreceptors on adjacent cells that mediate a variety of cell signaling (Hynes, 2002), and their role in homeostatic synaptic plasticity has recently come to light. Several integrin subtypes are expressed in the nervous system where they regulate synapse maturation and function (Chan et al., 2003; Chavis and Westbrook, 2001; Shi and Ethell, 2006).

Recent work in hippocampal neurons showed a specific postsynaptic requirement for β3 integrins in scaling up of synaptic AMPA receptors induced by activity suppression (Cingolani and Goda, 2009; Cingolani et al., 2008). Under basal conditions, β3 integrins acted to stabilize synaptic AMPA receptors as disrupting integrin ligation to the extracellular matrix promoted GluA2 internalization and reduced synaptic AMPA currents, and overexpressing a dominant negative form of β3 integrins in postsynaptic neurons also reduced synaptic AMPA currents. Importantly, the loss of β3 integrin specifically prevented the homeostatic scaling up of mEPSCs with TTX treatment.

How might β3 integrins detect changes in network activity levels to modulate synaptic AMPA receptors? Extracellular proteolysis is involved in LTP and for activity-dependent dendritic spine remodeling (e.g. Wang et al., 2008), a process that suggests of a requisite change in the surrounding space occupied by the extracellular matrix (Frischknecht et al., 2009) to make room for spine volume changes. It is thus tempting to speculate that β3 integrins may sense activity-dependent remodeling of extracellular matrix and coordinate such changes to the tuning of synaptic strength.

N-Cadherin/β-catenin complex

N-cadherin/β-catenin complex represents another class of adhesion proteins involved in homeostatic synaptic plasticity. N-cadherin is a Ca²⁺-dependent homophilic cell adhesion protein with an established role in regulating synapse formation and spine morphology (Mysore et al., 2008; Takeichi and Abe, 2005). It links to the actin cytoskeleton via β- and α-catenins, and it can also associate with synaptic scaffolding proteins via the PDZ binding motif of β-catenin to modulate pre and postsynaptic functions (Bamji et al., 2003, 2006; Murase et al., 2002; Okuda et al., 2007; Tang et al., 1998). In addition, N-cadherin can also interact with AMPA receptor subunits directly via the extracellular domains (Nuriya and Huganir, 2006; Saglietti et al., 2007) to modify synaptic activity (Saglietti et al., 2007). A recent study suggested a role for the N-cadherin/β-catenin complex in the bi-directional regulation of synaptic AMPA receptors during homeostatic synaptic scaling (Okuda et al., 2007). In cultured hippocampal neurons, conditional deletion of β-catenin following synaptogenesis prevented both scaling up and scaling down of mEPSC amplitudes induced by chronic TTX and bicuculline treatments, respectively. The precise mechanism by which the N-cadherin/β-catenin complex regulates activity-dependent synaptic scaling remains to be established.

Ephrin/Eph receptors

The cell adhesion/repulsion proteins, ephrin/Eph receptors, are heterophilic adhesion proteins implicated to play a role in homeostatic synaptic plasticity. A recent study at the Drosophila NMJ reported a requirement for Eph receptor signaling in motor neurons for homeostatic retrograde control of synaptic transmission (Frank et al., 2009). Chronically impairing postsynaptic activity up-regulated presynaptic release by targeting presynaptic Ca_v2.1 channels (Frank et al., 2006, 2009), and this involved a signaling cascade in the motor neuron in which Eph interacted with the Rho-GEF ephexin and the Rho-GTPase Cdc42 (Frank et al., 2009). The nature of putative, muscle derived signals that activate Eph receptor – whether it involves Drosophila ephrin (Tsuda et al., 2008) or other molecules – remains to be determined.

It is not known if ephrin/Eph receptor signaling plays a role in homeostatic synaptic plasticity at mammalian synapses. However, it has been reported to play roles in LTP (Klein, 2009; Lai and Ip, 2009). For example, in hippocampal mossy fiber LTP, facilitation of neurotransmitter release requires trans-synaptic reverse signaling of the postsynaptic EphB receptors to the presynaptic B-ephrins (Armstrong et al., 2006; Contractor et al., 2002). Similarly, ephrin/Eph receptors could also mediate trans-synaptic signaling involved in the expression of homeostatic synaptic plasticity, although this remains to be tested.

LTP at hippocampal CA3-CA1 synapses is modulated by ephrin/Eph receptor signaling involving astrocytes (Filosa et al., 2009). Here, reverse signaling from postsynaptic EphA receptor to A-type ephrins in glial cells was demonstrated to control the levels of glial glutamate transporters; this in turn, was suggested to alter extracellular glutamate concentrations near synapses to affect LTP induction (Filosa et al., 2009). Given that extracellular glutamate levels affect the extent accumulation of secreted TNFα (Stellwagen and Malenka, 2006), it would be of interest to test whether this EphA-ephrin signaling also modulates TNFα-dependent forms of homeostatic synaptic scaling.

Other regulators

In addition to the molecular players discussed in above categories, other proteins representing a variety of cellular functions are likely to contribute to mechanisms of homeostatic synaptic plasticity. Perhaps an analogy could be made to LTP, whose research over the past few decades has implicated the involvement of over 100 molecules (Sanes and Lichtman, 1999; Lisman et al., 2003). The increasing number of participating molecules could represent different types of homeostatic synaptic plasticity that have yet to be clearly defined. In other words, homeostatic synaptic plasticity could refer to a generic process encompassing a variety of plasticity mechanisms. Even for a single neuron, synapses vary in their functional state and molecular composition, and such heterogeneity could also contribute to differential molecular requirements in expressing homeostatic synaptic plasticity. In addition, any molecules that participate in regulatory mechanisms of presynaptic release or postsynaptic receptor abundance are potential candidates, of which there are many. Here we highlight the role of post-translational protein modifications and of a recently identified notable player, dysbindin, in homeostatic synaptic plasticity.

Protein palmitoylation

A recent study in cultured hippocampal neurons highlights an important role for protein palmitoylation in synaptic scaling (Noritake et al., 2009). Previous studies have shown that palmitoylation state of PSD-95, a postsynaptic scaffolding protein, controls the postsynaptic recruitment of PSD-95 and consequently that of AMPA receptors (El-Husseini et al., 2002; Elias and Nicoll, 2006). Interestingly, DHHC2, a dendritically localized member of the family of DHHC-type palmitoylation enzymes, was found to translocate to the postsynaptic density upon TTX-silencing of neurons. Concomitantly, activity silencing produced stoichiometric palmitoylation of PSD-95 and augmentation of postsynaptic PSD-95, GluA1 and GluA2 levels in a DHHC2-dependent manner (Noritake et al., 2009). Therefore, activity-dependent regulation of protein palmitoylation machinery is one mechanism by which neurons homeostatically scale their synaptic AMPA receptors.

Synaptic activity can also modulate palmitoylation state of AMPA receptor subunits that affects their intracellular trafficking and stability (Hayashi et al., 2005; Yang et al., 2009). For example, palmitoylation of GluA1 controls its phosphorylation and activity-dependent surface delivery at extrasynaptic sites, and in turn, regulates the availability of GluA1 for synaptic strengthening (Lin et al., 2009). Palmitoylation can thus affect synaptic strength in many ways, and therefore, it would be important to further delineate how synaptic protein palmitoylation is regulated and how the palmitoylated state of particular proteins affects other determinants of AMPA receptor traffic in mediating specific forms of synaptic plasticity.

Protein phosphorylation

Other signaling molecules, for example, protein kinases acting downstream of Ca²⁺or BDNF, also contribute to the homeostatic modification of synaptic strength. These include TrkB receptor tyrosine kinase (Lu et al., 2008; Minichiello, 2009) and serine threonine protein kinases, such as CaMKII and CAMKIV (Wayman et al., 2008). In hippocampal neurons, CaMKII signaling up-regulates mEPSC amplitude and presynaptic release efficacy upon chronic activity blockade (Thiagarajan et al., 2002; but also see Thiagarajan et al., 2005). In cortical and hippocampal neurons, CaMKIV signaling has also been shown to be required for increasing quantal size by a mechanism involving transcriptional modulation (Ibata et al., 2008; Thiagarajan et al., 2002). In contrast, cyclin-dependent kinase 5 (Cdk5) and polo-like kinase 2 (Plk2) are involved in down-regulating excitatory synaptic strength in hippocampal neurons with elevated activity levels (Seeburg et al., 2008; Seeburg and Sheng, 2008). Upon increased activity, Cdk5 and Plk2-dependent phosphorylation of SPAR, a RapGAP (GTPase-activating protein) that also functions as a postsynaptic scaffold protein, targets SPAR for degradation. This loss of SPAR, a condition that would enhance RapGTPase activity, in turn, reduces synaptic AMPA receptors to diminish synaptic efficacy. Interestingly, Rap1 signaling might act downstream to β3 integrins in controlling synaptic AMPA receptor abundance, where active Rap1 is required for down-regulating quantal size upon blocking integrins (Cingolani et al., 2008). Therefore, the level of Rap activity could serve as an integrator of different synaptic stimuli to exert homeostatic feedback control of synaptic AMPA receptor abundance.

Disease-related gene product

A recent study reported a role for dysbindin, a protein encoded by a major susceptibility gene for schizophrenia (DTNBP1: Straub et al., 2002; Williams et al., 2005) in homeostatic presynaptic adaptation at the Drosophila NMJ (Dickman and Davis, 2009). In an elegant forward genetic screen of more than 250 mutants using electrophysiology to monitor homeostatic synaptic changes, the authors found a mutation in the Drosophila homolog of dysbindin gene that specifically interfered with enhanced presynaptic release induced by blocking muscle excitation. Importantly, the mutant showed normal basal synaptic transmission, and presynaptic expression of wild type dysbindin was sufficient to rescue the homeostasis deficit.

Consistent with such a presynaptic role for Drosophila dysbindin in homeostatic regulation of neurotransmitter release, in vertebrates, dysbindin is found associated with synaptic terminals (Benson et al., 2001) and synaptic vesicles (Talbot et al., 2006; Taneichi-Kuroda et al., 2009), and it may regulate synaptic vesicle biogenesis to affect neurotransmitter release (Chen et al., 2008; Taneichi-Kuroda et al., 2009). Dysbindin is also present at the postsynaptic density although its postsynaptic function is not known (Talbot et al., 2006). Further understanding of how dysbindin contributes to homeostatic synaptic plasticity mechanisms might shed novel insights into how synapse dysfunction contributes to the etiology of schizophrenia.

The list of molecular regulators of homeostatic synaptic plasticity discussed above is by no means exhaustive, but rather, it is likely to represent the tip of the iceberg. The expanding knowledge of participating proteins/molecules with dedicated cellular functions would help in obtaining a clearer picture of the general mechanisms by which neurons implement feedback changes in synaptic efficacy, and moreover, perhaps give us clues about how neurons might monitor and sense the level of network activity in the first place.

CONCLUDING REMARKS

Homeostatic synaptic plasticity is a rapidly burgeoning area of cellular neuroscience research. Here we have reviewed recent advances in cellular properties and molecular mechanisms of this adaptive synaptic process. We have considered pre and postsynaptic mechanisms engaged in changes induced by imposing activity perturbations in simple systems such as primary cultures of dissociated rodent neurons or the genetically amenable Drosophila NMJ, and highlighted molecules with identified function in implementing the homeostatic synaptic changes. The picture that emerges is that of a complex regulatory network involving multiple mechanisms with different temporal and spatial properties. Moreover, studies in vivo and organotypic cultures show that in preparations where the topological relationship of neuronal connections are preserved to drive stereotypic spatiotemporal patterns of synaptic activation, homeostatic synaptic changes do not necessarily occur in the same manner as they do in dissociated cultures, and display even more heterogeneity in response to the same activity manipulation with region and cell-type specific differences. The intact systems emphasize a crucial point, demonstrated in original studies in cultured neurons but often overlooked since, that the effects of network activity modulation on unitary currents must be interpreted ultimately with the effect they have on maintaining the spike output of the neuron, and furthermore, that synaptic changes constitute only a part of the big picture (Davis 2006; Turrigiano 2008). Thus physiological functions of homeostatic synaptic plasticity and a better understanding of how it is engaged in native neural circuits must be considered together with changes in intrinsic membrane excitability and the balance of excitation and inhibition.

Although less physiological in representing the brain, simple systems of dissociated cultured neurons and Drosophila NMJ are highly amenable to molecular characterization, and we expect that they will continue to yield important molecular insights into mechanisms of homeostatic synaptic plasticity. Moreover, in dissociated cultures, the exact synaptic connections can be mapped and activity can be manipulated in identified neurons and at the level of individual synaptic inputs. Thus such a system may also be valuable for addressing questions concerning how homeostatic synaptic plasticity compete or cooperate with different forms of synaptic plasticity at identified synapses.

Notably, the molecular players implicated in homeostatic synaptic plasticity to date are mostly effectors of pre and postsynaptic changes that are induced downstream of activity perturbation. This raises a number of key questions. When and how does a neuron decide whether to engage homeostatic synaptic responses? That is, what is the permissive range of activity and how do neurons detect deviations from it? What are the timescale and the spatial extent over which homeostatic sensors detect and integrate deviations in activity levels and implement changes? Where does the apparent multiplicity of mechanisms involved in homeostatic synaptic plasticity originate from? Is it due to divergent signaling pathways engaged by a key sensor or does it represent multiple detection systems that work in parallel? While much work will be needed to address these questions, synthesis of different approaches, such as identifying the basic rules of homeostatic synaptic plasticity in simple systems and testing of such rules in intact preparations, will help in unraveling of the underlying mechanisms.