Mechanisms underlying synapse-specific clustering of GABAA receptors


Prof A. M. Thomson, as above.


A principle that arises from a body of previous work is that each presynaptic terminal recognises its postsynaptic partner and that each postsynaptic site recognises the origin of the synaptic bouton innervating it. In response, the presynaptic terminal sequesters the proteins whose interactions result in the dynamic transmitter release pattern and chemical modulation appropriate for that connection. In parallel, the postsynaptic site sequesters, inserts or captures the receptors and postsynaptic density proteins appropriate for that type of synapse. The focus of this review is the selective clustering of GABAA receptors (GABAAR) at synapses made by each class of inhibitory interneurone. This provides a system in which the mechanisms underlying transynaptic recognition can be explored. There are many synaptic proteins, often with several isoforms created by post-translational modifications. Complex cascades of interactions between these proteins, on either side of the synaptic cleft, are essential for normal function, normal transmitter release and postsynaptic responsiveness. Interactions between presynaptic and postsynaptic proteins that have binding domains in the synaptic cleft are proposed here to result in a local cleft structure that captures and stabilises only the appropriate subtype of GABAARs, allowing others to drift away from that synapse, either to be captured by another synapse, or internalised.


At one time, central inhibitory GABAergic synapses were considered to be broadly similar, both in their properties and in the role(s) they play in circuit behaviour. This was despite the overwhelming anatomical evidence for the many different subtypes of GABAergic inhibitory interneurones that are found in the brain (e.g. Ramón y Cajal, 1909, 1911; Peters & Jones, 1984; Monyer & Markram, 2004; Butt et al., 2005; Klausberger & Somogyi, 2008). We now know a lot more about inhibitory circuitry and can begin to predict some of the many different roles that inhibitory connections play. In neocortex alone there are > 20 different types of inhibitory connection in each of five layers (2–6). While it has been clear for a long time that gross alterations in inhibitory gain control result in coma on the one hand and seizures on the other, more recently, seemingly small or subtle changes in only one part of the inhibitory system have been linked to perhaps less life-threatening, but nevertheless debilitating and tragic, neurological and psychiatric disease. Similarly, pharmacological or genetic manipulation of single GABAA receptor (GABAAR) subunits produces rather less dramatic, but nevertheless profound, alterations in mood and behaviour (Möhler, 2006a,b; Möhler et al., 2002; Rudolph & Möhler, 2006, for reviews). One important consequence of this selectivity is that, in many cases, changing the activity of a given GABAAR subtype, either by gene mutation, or by pharmacological manipulation, modifies the outputs of a given class of presynaptic GABAergic neurone, rather than modifying the inputs to a given class of postsynaptic neurone. These manipulations can, therefore, indicate the role(s) that different classes of interneurones play.

Synaptic GABAARs are typically pentomers containing a γ2 subunit in addition to two β- and two α-subunits. A β-subunit (β1, β2 or β3) is almost obligatory for plasma membrane insertion. Multimers without β-subunits are often destroyed before leaving the endoplasmic reticulum (ER). β-Subunits may also play a role in subcellular compartment-targeting of receptors: to axon, soma or dendrites. The α-subunits appear to convey an even finer level of specificity, a striking example being the inputs provided by two major subclasses of basket cells to the soma of the same pyramidal cell in neocortex and hippocampus. The GABAARs innervated by parvalbumin (PV)-containing basket cells include α1-, β2/3- and γ2-subunits, while α2,β2/3,γ2-GABAARs are innervated by cholecystokinin (CCK) basket cells (Fig. 1; also Pawelzik et al., 1999; Thomson et al., 2000; Ali & Thomson, 2008). Benzodiazepines acting only at α1-GABAARs produce sedative and anticonvulsant effects, but generate anxiolysis only when they act at α2/3-GABAARs (Möhler et al., 2002), indicating the functional and potential therapeutic relevance of this differentiation.

Figure 1.

 The subtype of GABAAR inserted at an inhibitory synapse is determined by the class of presynaptic inhibitory interneurone. The cartoon in the middle indicates the synaptic connections made and received by four classes of GABAergic cortical interneurones (and a pyramidal cell). The very different dynamic properties of the excitatory inputs that two classes (bitufted and basket cells) receive from pyramidal cells are illustrated on the left. In the centre and right, CCK-containing basket cells (yellow) and chandelier cells (green) innervate clusters of α2/3-GABAARs. PV-containing basket cells activate α1-GABAARs. Bitufted interneurones activate α5-containing GABAARs. (Bottom) At each class of GABAergic synapse, one subtype of GABAAR dominates.

Specificity and selectivity in cortical circuits

With recent technical and theoretical advances, the complexity of the synaptic machinery and the exquisite selectivity with which the array of pre- and postsynaptic properties and mechanisms are targeted and inserted is beginning to be unravelled. What at first appeared to be almost unfathomable diversity and complexity is becoming more accessible as the ‘rules’ that apply to circuit construction are elucidated. Dual intracellular recordings with dye-labelling have been very informative here. They have documented the vast range of properties displayed when different classes of synaptic connections are studied in detail and compared, each class displaying its own unique combination of properties (e.g. Thomson & Lamy, 2007 for review; see also Fig. 1). Synaptic connections are not made randomly with just any neuronal element that happens to be near to a particular axon. They involve only certain classes of target neurones and, moreover, specific subcellular compartments of those cells (Somogyi & Klausberger, 2005; Klausberger & Somogyi, 2008; for review of interneuronal axon targets; Thomson & Lamy, 2007, for review of pyramidal targets).

Distribution of γ and β subunits

These selective innervation patterns probably account for some of the GABAAR subtype segregation apparent in pharmacological and immunocytochemical studies, as each class of interneurone targets only certain subcompartments of its postsynaptic principal cell partners. In polarized epithelial cells, GABAARs containing the β1-subunit are sorted to the apical membrane (Perez-Velazquez & Angelides, 1993) and β2/3-subunits to the basolateral membrane (Connolly et al., 1996a). Something similar appears to be happening in hippocampal pyramidal cells. Synapses supplied by basket cells (both PV- and CCK-containing) that innervate the soma and proximal dendrites of pyramidal cells were enhanced by low concentrations of Etomidate, an anaesthetic whose potency is greater at β2/3-subunit-containing GABAARs than at β1-subunit-containing GABAARs, but independent of the α-subunit included (Hill-Venning et al., 1997). When inputs to pyramidal cell dendrites supplied by bistratified cells were tested, however, they were very much less sensitive to Etomidate (H. Pawelzik, unpublished data). It therefore appears that proximal GABAergic synapses on pyramidal cells are supplied with β2/3-subunit-containing GABAARs, while at least some dendritic synapses contain β1-subunit-containing GABAARs.

Most synaptic GABAARs in cortical regions contain a γ2L-subunit; indeed, a γ2-subunit appears obligatory for synaptic receptors in cortical pyramidal cells (Essrich et al., 1998; Schweizer et al., 2003). The predominant class of extrasynaptic receptors contain a δ-subunit instead. In addition to the γ2-subunit most GABAARs are thought to contain two β-subunits and two α-subunits. Summarising simplistically, therefore, somatic synaptic GABAARs contain a γ2- and two β2/3-subunits, while at least some dendritic synaptic receptors contain a γ2- and two β1-subunits.

The α-subunits appear to convey a finer level of specificity

The α-subunits appear to be selectively sequestered or clustered at synapses innervated by specific subtypes of interneurones. For example, the synapses supplied by fast-spiking PV-immunopositive basket cells, in neocortex (Ali & Thomson, 2008) and hippocampus (Pawelzik et al., 1999; Thomson et al., 2000), are extremely sensitive to the α1-selective benzodiazepine site ligand Zolpidem. They are insensitive to zinc and to IAα5 (an α5-subunit-selective partial inverse agonist: Chambers et al., 2004; Street et al., 2004) and are partially blocked by the broad spectrum inverse agonist flumazenil. This benzodiazepine type 1 (BZ1) pharmacological profile indicates mediation by α1βγ2 receptors (Fig. 1).

In contrast, the neighbouring synapses supplied by CCK (cholecystokinin)-immunopositive basket cells are much less sensitive to Zolpidem, but are also insensitive to zinc and IAα5. This pharmacological profile is typical of BZ2 receptors. These CCK basket cells therefore act through α2/3-subunit-containing GABAARs on pyramidal cells (later confirmed with immunocytochemistry at the ultrastructural level: Nyiri et al., 2001). It is unlikely that many of these receptors include α1- as well as α2/3-subunits, as when α1 and α2/α3 are included in the same receptor, α1-benzodiazepine site (BZ1) pharmacology dominates (e.g. Araujo et al., 1996). It is, however, possible that some of the receptors displaying α1 pharmacology at PV basket cell synapses also contain an α2 or α3 subunit. We still do not know whether synaptic receptors can contain two different α-subints, or indeed two different β-subunits. It is, however, clear that the inputs provided by two major subclasses of basket cells to the soma of the same pyramidal cell are mediated by GABAARs containing different α-subunits and displaying different pharmacology, though both contain a γ2- and two β2/3-subunits. Chandelier, or axo-axonic, cells also innervate synapses rich in α2-subunits (Nusser et al., 1996).

Finally, some dendritic GABAergic synapses, those supplied by CA1 bistratifed cells (Pawelzik et al., 1999; Thomson et al., 2000) and those supplied by bitufted, dendrite-preferring cortical interneurones (including somatostatin-immunopositive Martinotti cells: Ali & Thomson, 2008) have a BZ3 pharmacological profile indicating that they are mediated by α5-subunit-containing GABAARs. These synapses are insensitive to Zolpidem, but enhanced by Diazepam and partially blocked by zinc, IAα5 and flumazenil. These receptors may include an α1-subunit, as in this combination α5-benzodiazepine site (BZ3) pharmacology dominates. The postsynaptic location of α5-subunits has been confirmed immunocytochemically at the ultrastructural level (Serwanski et al., 2006; also Fig. 1).

The existence of synaptic α5-subunit-containing GABAARs has been controversial, but the evidence in favour of a predominately extrasynaptic site for these receptors is largely circumstantial. The failure of single-electrode experiments to find evidence for synaptic α5-subunit-containing GABAARs is, perhaps, not surprising. Somatic recordings are dominated by the large fast inhibitory postsynaptic currents (IPSCs) and inhibitory postsynaptic potentials generated by basket and chandelier cells. The smaller dendritic α5-GABAAR-mediated events are slowed in time course as they transfer to the soma and are difficult to identify from somatic recordings of spontaneous and miniature synaptic events. Their existence only becomes apparent in paired intracellular recordings. Other evidence for a largely extrasynaptic site for these GABAARs comes from recordings of ‘tonic’ inhibition, i.e. recordings of a conductance that persists in the absence of action potential-elicited GABA release. However, many of these studies involve the use of GABA uptake inhibitors. The final piece of evidence is that α5-subunits are not typically colocalised with gephyrin, a postsynaptic scaffold protein originally thought to be present at, if not an essential component of, all GABAergic synapses. However, while gephyrin is now considered important if not essential for clustering α2/3-GABAARs (Tretter et al., 2008), it is not necessary for the clustering of α5-GABAARs in retina or spinal cord (Kneussel et al., 2001). The failure to find evidence for a synaptic location or role for a receptor subtype is not evidence for the absence of such a location or role. Some years ago, the weight of opinion was against a normal physiological synaptic role for NMDA (N-methyl-d-aspartate) receptors, until, that is, the appropriate experiment was performed (Thomson et al., 1985).

That specific GABAARs are located at specific synapses should, perhaps, not be surprising when it is remembered that glutamate receptors are largely found apposed to glutamatergic boutons and GABA receptors apposed to GABAergic terminals. This is merely a finer level of detail. Moreover, presynaptic receptors can also be selectively inserted. Both the glutamatergic inputs from CA1 pyramidal cells (Shigemoto et al., 1996, 1997) and the GABAergic inputs from other interneurones (Somogyi et al., 2003) onto the OLM (oriens lacunosum moleculare) interneurones in CA1 stratum oriens are highly enriched in presynaptic mGluR7a receptors (metabotropic glutamate receptor type 7a), but these receptors are absent from the synaptic inputs onto other classes of interneurones or pyramidal cells in the same region. That is, the boutons supplied by a single axon will either express or not express a particular presynaptic receptor, depending on the type of neurone that is present postsynaptically.

How do the pre- and postsynaptic elements recognise each other?

Specifically, how do postsynaptic neurones cluster just one type of GABAAR at each type of inhibitory synapse when, as is the case with CA1 pyramidal cells, they contain up to 10 different GABAAR subunits (α1, α2, α3, α4, α5, β1, β2, β3, γ2, δ). Small, highly mobile molecules that readily diffuse across the synaptic cleft are commonly considered to be good candidates for pre/postsynaptic recognition signals and, indeed, there is substantial supportive evidence for their importance here. However, something more robust and structural may be needed to stabilise and maintain the pre- and postsynaptic machinery. That many synaptic properties are prescribed and extremely stable, at least in the healthy adult, is demonstrated by the narrow range of properties, such as quantal amplitude and release probability, exhibited by the synapses of a single class, even though these parameters are widely disparate across classes (e.g. Brémaud et al., 2007).

An increasingly popular hypothesis, therefore, is that membrane-spanning proteins derived from both pre- and postsynaptic elements mediate trans-synaptic recognition, by interacting across the synaptic cleft (see Fig. 2). There would certainly appear to be enough diversity within the synaptic adhesion molecule protein families to explain the functional diversity apparent in many neuronal circuits, if we only knew how the choices are made. Disruption of these interactions is beginning to be linked to neurological disease: mutations in neurexins and neuroligins have been implicated in autism spectrum disorders (Tabuchi et al., 2007; for review Pardo & Eberhart, 2007; Buxbaum, 2009), chromosomal exon deletions that affect neurexin 1 appear to increase the risk of schizophrenia (Rujescu & Collier, 2009; Kirov et al., 2009; for review), while increased cleavage and shedding of soluble NCAM (neural cell adhesion molecule) is apparent in schizophrenic brains (e.g. Vawter et al., 2001). It is many years since electron microscopists demonstrated that the synaptic cleft was far from an empty space devoid of structure; it is time to explore the function of that structure.

Figure 2.

 This cartoon describes a sequence of events that may contribute to the specific clustering of different subtypes of GABAARs (in red and blue) at different synapses during synapse formation and maturation. The top panel represents an axon terminal and part of a potential postsynaptic neurone before synapse formation. Pre- and postsynaptic cleft-spanning proteins (e.g. neurexins and neuroligins) are indicated by black figures. Postsynaptic density constituents are represented by grey ovals and presynaptic release machinery constituents by yellow shapes. As a synapse begins to form, interactions between cleft-spanning proteins occur, promoting the alignment of pre- and postsynaptic synapse constituents with the synaptic membranes (middle panel), and ‘appropriate’ receptors are stabilized at the site of the future postsynaptic density. Following the congregation of presynaptic release machinery components and vesicles and postsynaptic receptors, postsynaptic scaffold proteins are recruited (bottom panel).

Synaptic cleft-spanning proteins

Several families of proteins derived from either the presynaptic or the postsynaptic neurone that span all or part of the synaptic cleft have been identified. Some of those that might mediate interactions between presynaptic GABAergic terminals and postsynaptic neurones are sumarised here.

Presynaptic neurexins

Neurexins (1α, 2α, 3α, 1β, 2β, 3β) exhibit extensive alternative splicing, from which more than 2000 potential variants can be predicted (Missler & Südhof, 1998; Tabuchi & Südhof, 2002). These splice sites are found particularly within the laminin neurexin sex hormone binding protein (LNS) domains. The six LNS domains in α-neurexin and the one in β-neurexin exhibit Ca2+-dependent binding to the extracellular domains of neuroligins, dystroglycan and neurexophilin, and provide a high-affinity α-latrotoxin (a spider neurotoxin that elicits transmitter release) binding site (Craig & Kang, 2007; for review). By altering Ca2+ binding affinity, splice insertions at these sites alter their interactions with other proteins (Sheckler et al., 2006). In neuronal cultures, neurexin-1β alone (on beads, or expressed in non-neuronal cells) can promote postynaptic differentiation (Graf et al., 2004; Nam & Chen, 2005). All three α-neurexins promote clustering of postsynaptic GABAAR-associated gephyrin and neuroligin 2 (NL2) via the sixth LNS domain, but not of glutamate receptor (GluR)-associated postsynaptic density (PSD)-95, or of NL1, NL3 or NL4. This clustering is negatively modulated by up-stream neurexin sequences (Kang et al., 2008).

A C-terminus binding motif is required for neurexin to leave the endoplasmic reticulum and for targeting to and insertion at synaptic plasma membranes. Neurexin is transported along the axon in vesicles that do not contain active zone precursor proteins, but which carry CASK (Calcium/calmodulin- dependent serine protein kinase), RIM1α (Regulating synaptic membrane exocytosis protein 1α) and calcium channels and possibly other elements of the transmitter release machinery (Fairless et al., 2008). Insertion of neurexin in the presynaptic plasma membrane is clearly important for binding essential components into the presynaptic release machinery. In addition, the interactions of neurexins with neuroligins promote postsynaptic differentiation, presumably because they help to stabilise both proteins and thereby their pre- and postsynaptic binding partners. These interactions and the influence they have are affected by the splice variants present. For example, NL1 lacking an insert in splice site B binds both α and β neurexin and, if overexpressed, has a more powerful effect on synaptic size than on number, unlike the variant with the insert, which affects synapse number more powerfully (Boucard et al., 2005). These studies have led to the suggestion that the combination of neurexin and neuroligin isoforms that is expressed influences a wide range of synaptic properties.

The many binding partners and extensive alternative splicing of neurexins, the conservation of splice insert sequences and positions across species, and the co-expression of several neurexin isoforms in single cells may suggest that they are mediators of synapse specificity and that this specificity is important. How different splice variants may be concentrated at different presynaptic terminals remains to be established, but a mechanism shared with that underlying the specific localisation of certain release machinery components seems likely.

Postsynaptic neuroligins

Neuroligins (NL1, NL2 and NL3) are the postsynaptic neurexin interactors. They exhibit less extensive alternative splicing, which occurs at their single LNS domain and at the AChE (acetylcholine esterase)-homologous regions, but important selectivity nevertheless (Kang et al., 2008). NL2 promotes formation of and is localised to GABAergic synapses (Varoqueaux et al., 2004), while NL1 promotes glutamateric synapse formation. NL3 aggregates at subsets of both glutamatergic and GABAergic synapses, forming complexes with NL1 or NL2 (Budreck & Scheiffele, 2007). Without NL2, GABAAR clusters do form in the plasma membranes of transfected HEK 293T cells co-cultured with neurones. However, the clusters that form are reported to be small, functionally silent and labile, and do not recruit the scaffolding protein gephyrin. Functional synapses with aggregated presynaptic vesicles and with larger, more stable, postsynaptic GABAAR clusters and gephyrin, form when the HEK cells are also transfected with NL2 (Scheiffele et al., 2000; Biederer et al., 2002). The effect of alternative splicing is seen here, since the lack of an insert at the B site, but addition of an insert at the A site promote localisation of NL2 to GABAergic synapses (Graf et al., 2004).

In isolated cultured neurones, surface GABAAR clusters are also small but, after GABAergic axons arrive, larger clusters of receptors form, apposed to the GABAergic boutons. With time, these large clusters become surrounded by regions emptied of the smaller nonsynaptic clusters (Christie et al., 2002). This suggests that the receptors in the smaller clusters move into and are captured and clustered by the new synapse. This, moreover, is a two-way traffic. The presence of a presynaptic GABAergic terminal stabilises and reduces the lateral mobility of GABAAR clusters (Jacob et al., 2005), while the clustering of apposed GABAARs stabilises presynaptic terminals (Li et al., 2005; also summarised in Fig. 2). Addition of soluble β-neurexin to a neuronal co-culture to block endogenous neurexin interactions with other cleft proteins inhibited synaptic vesicle aggregation. β-Neurexins with splice inserts at site 4 (+S4), like α-neurexins, interact preferentially with neuroligins that lack a B site insert (e.g.NL2; Boucard et al., 2005; Chih et al., 2006) and promote greater clustering at GABAergic than at glutamatergic synapses, though they lack the near absolute exclusivity of α-neurexins.

The ability of NL2 to promote and strengthen GABAergic synapses is enhanced by network activity and both the release of GABA and the presence of postsynaptic GABAARs appear essential for normal synapse maturation (Chattopadhyaya et al., 2007; see Huang & Scheiffele, 2008 for review), for the targeting of GABAARs and for their stability at the synapse (Saliba et al., 2007). Overexpression of NL2 increases the amplitude of GABAergic IPSCs (but not of glutamatergic events), while pharmacological blockade of network activity prevents this synaptogenic effect (Chubykin et al., 2007). Synaptic activity also reduces the lateral movement of gephyrin-containing postsynaptic scaffold rafts, motion that is dependent upon actin and countered by microtubules (Hanus et al., 2006; and see below). NL2 also plays a role in long-term synaptic maturation during normal development. In cerebellar granule cells there is a developmental switch from α2/3- to α1-containing GABAARs. This switch is associated with the acquisition of faster receptor-channel kinetics. Overexpression of NL2 in cultured granule cells accelerated this change (assessed by Zolpidem efficacy and IPSC time course), promoting incorporation of α1-GABAARs at postsynaptic sites in immature cells (Fu & Vicini, 2009).

Crystal structure analysis of the extracellular adhesion domains of neuroligin identifies three distinct regions for partner association at the subunit surface: a C-terminal homodimer-forming region, a hydrophobic negatively charged cavity where two of the three fasciculin (a snake neurotoxin inhibitor of AChE) loops would nestle, and a negatively charged flat surface dominated by hydrogen-bond providers for apposition of the flat neurexin edge. As these three regions represent only 12% of the surface area it is proposed that additional extracellular binding domains exist (Leone et al., 2010).

Acting in concert with the neurexins and a wide range of other cleft and postsynaptic binding partners, NL2 is widely reported to be central to the formation and stabilisation of GABAergic synapses. Indeed it has even been proposed that GABAergic synapses can form in the absence of GABAA receptors provided NL2 is present (Patrizi et al., 2008). This is, perhaps, an extreme view, as many would define a synapse as a structure capable of transmission, but it would appear that GABAergic terminals are capable of ‘recognising’ NL2-containing membranes and making contact. However, deletion of α1-subunits, which results in a total loss of GABAARs in adult mouse (from postnatal day 18) cerebellar Purkinje cells, leads to aberrant asymmetric (i.e. excitatory in structure) synapses apposed to molecular layer dendritic spines instead of dendritic shafts (Fritschy & Panzanelli, 2006; Fritschy et al., 2006). Thus, even though apparently normal synapses form earlier in development, the maintenance of appropriate synaptic contacts, in the molecular layer, is receptor-dependent.

Intracellular binding partners for neurexins and neuroligins

The intracellular C-termini of presynaptic neurexins bind to synaptotagmin (Hata et al., 1993) and to PDZ domains of CASK (Hata et al., 1996), syntenin and Mint (Biederer & Südhof, 2000). Neurexins may also govern the concentration and perhaps the type(s) of Ca2+ channels at presynaptic release sites (O’Connor et al., 1993). Indeed, in the nonviable triple α-neurexin knockout mouse, N and P/Q Ca2+channels do not cluster at active zones, and action potential-triggered release fails (Missler et al., 2003; Zhang et al., 2005).

The intracellular domains of postsynaptic neuroligins bind to PSD-95 (Irie et al., 1997; Meyer et al., 2004) and related scaffolding proteins MAGUKs and S-SCAM, and probably also to proteins such as Shank, PICK1, GOPC and SPAR (Chih et al., 2005; Craig & Kang, 2007; Washbourne et al., 2004). In the neuroligin triple knockout, only the release of GABA and glycine are significantly compromised. However, with the absence of this postsynaptic protein, all synapses appear to display some disruption of presynaptic vesicular proteins, underlining the importance of trans-synaptic signalling and/or recognition. NL2 also binds gephyrin through a conserved cytoplasmic motif. Gephyrin is a postsynaptic scaffold protein found at many inhibitory synapses (Hanus et al., 2006; Fritschy et al., 2008), particularly those containing α2-GABAARs (Tretter et al., 2008).

The RhoGEF collybistin is vital for the translocation of gephyrin to synaptic sites (Harvey et al., 2004). Disruption of collybistin in mice leads to increased levels of anxiety and impaired spatial learning, which are associated with a selective loss of GABAARs in the hippocampus and basolateral amygdala (Papadopoulos et al., 2007). In humans, a missense mutation in the collybistin SH3 domain results in somatic and dendritic trapping of gephyrin and inhibitory receptors by collybistin aggregates, giving rise to a hyperekplexia, drug-resistant seizures and premature death (Harvey et al., 2004). Collybistin that has been activated by NL2 associates with the postsynaptic plasma membrane and promotes the subsynaptic clustering of gephyrin. This has led to the conclusion that complexes of NL2, collybistin and gephyrin are sufficient to generate the clustering of postsynaptic GABAARs, even in the absence of a presynaptic terminal (Poulopoulos et al., 2009). Whether, or for how long, such a structure would be stable remains to be determined. It has also been proposed that the interplay between neurexins and neuroligins is modified to maintain the balance between the excitation and inhibition that a neurone receives. Shifts in the location of NL1 and NL2 to excitatory synapses are associated with overexpression of PSD-95 and an increase in the ratio of excitatory to inhibitory synaptic currents; a decrease in this ratio follows knock-down of PSD-95 (Prange et al., 2004; Levinson et al., 2005; Levinson & El Husseini, 2005, for review).

Other cell adhesion molecules

There are many other cell adhesion molecules, some of which are found in the synaptic cleft. Some of these are selectively expressed at glutamatergic synapses, e.g. SynCAM (synaptic cell adhesion molecules), ephrins (ligands of class V-EPH-related - receptor protein-tyrosine kinases), ephrin receptors and netrin-G ligands (transmembrane protein ligands of secreted proteins that act as long-range cues for growth cones). Others, such as NCAM (neural cell adhesion molecule), localise to GABAergic synapses and promote their formation and/or stabilisation (Pillai-Nair et al., 2005). Postsynaptically derived dystroglycan accumulates at a subset of mature GABAergic synapses, but only after the formation and aggregation of presynaptic vesicles and the clustering of postsynaptic GABAARs, particularly of α2-subunit-containing GABAARs (Lévi et al., 2002). β-Dystroglycan is a binding partner for S-SCAM (synaptic scaffolding molecule) at inhibitory synapses, forming a tripartite complex with NL2 in vitro, with S-SCAM acting as a link between NL2 and β-dystroglycan (Sumita et al., 2007). α-Neurexins also bind dystroglycan, but only via LNS2 and LNS6 domains that lack splice inserts (Sugita et al., 2001).

Craig & Kang (2007) suggest that cell adhesion molecules may initiate interactions between putative pre- and postsynaptic elements and that neurexin–neuroligin interactions then recruit and stabilise other pre- and postsynaptic structures. We suggest that specific isoforms of these cleft-spanning proteins may signal synapse specificity during this process. There are two obvious ways in which different GABAAR subtypes could become clustered at different types of synapse: they could be selectively inserted at a particular postsynaptic site, in a highly specific way, or they could be inserted into the plasma membrane relatively randomly and find their way to an appropriate synapse by lateral diffusion, becoming stabilised there by a synapse class-specific assembly of proteins. The next section summarises some of what we know of GABAAR trafficking in this context.

Receptor trafficking

Several recent studies describe the trafficking, transport to the plasma membrane and subsequent fate of receptors. GABAARs containing a γ2-subunit appear destined for synapses; their surface expression is prolonged and internalization delayed by apposition to a GABAergic bouton. While surface clusters of GABAARs form in cultured neurones without GABAergic input (even apposed to glutamatergic terminals: Studler et al., 2005), clusters apposed to GAD (glutamic acid decarboxylase)-positive GABAergic terminals are larger, more stable and able (unlike inappropriately located clusters) to recruit postsynaptic density proteins such as gephyrin (Jacob et al., 2005). Could this property arise from specific “delivery” of GABAARs to postsynaptic plasma membranes by γ2-subunit binding partners?

Binding partners for γ-subunits

There are at least a few well characterized proteins that interact with this subunit, but the evidence for any of these proteins playing a role in targeting or insertion of GABAARs at the synapse is sparse. Not only are these proteins largely involved in the intracellular trafficking of GABAARs through secretory pathways, they localize away from the postsynaptic membrane and are often found to be associated with intracellular membranes. One such example is GABARAP, a member of the family of small microtubule-binding proteins, which was initially discovered as an interacting protein of the γ-subunits (Wang et al., 1999). GABARAP was shown to influence the levels of GABAARs expressed at the cell surface, as well as their channel properties (Leil et al., 2004; Chen et al., 2005; Chen & Olsen, 2007; Kawaguchi & Hirano, 2007), yet this protein co-localizes only with intracellular pools of GABAARs, within the Golgi apparatus and other associated intracellular membranes (Kittler et al., 2001). Within these same intracellular compartments GABARAP, and perhaps even GABAARs themselves, interact with NSF (N-ethylmaleimide-sensitive factor), a ubiquitous regulator of membrane fusion and trafficking (Kittler et al., 2001; Goto et al., 2005), as well as with the PRIP proteins (phospholipase-C related catalytically inactive proteins; Kanematsu et al., 2002), with gephyrin (Kneussel et al., 2000), and with proteins involved in vesicular transport and apoptosis. PRIP proteins are unlikely to play a role in synaptic targeting of GABAARs even though they can interact directly with γ-subunits (Kanematsu et al., 2006), because they compete with the γ-subunits for the same binding site on GABARAP (Kanematsu et al., 2002). Thus, the role of GABARAP and associated proteins in GABAAR targeting to the synapse is likely to be indirect, possibly through stabilizing the γ-subunit-containing intracellular pools of these receptors.

Another GABAAR binding protein that specifically associates with γ-subunits is GODZ (Golgi-specific DHHC zinc finger domain protein; Keller et al., 2004). This protein regulates palmitoylation of γ-subunits, and is required for the assembly of GABAARs and their transport to the cell surface (Fang et al., 2006). This protein is, however, also located away from the postsynaptic membrane, within the Golgi apparatus, and unlikely therefore to play a direct role in GABAAR-targeting to the synapse.

Synaptic proteins that interact with GABAARs

Paradoxically, direct association between GABAARs and proteins such as gephyrin that clearly co-localize with them at synapses has traditionally been difficult to demonstrate using biochemical approaches. Gephryn is highly enriched at GABAergic synapses, forming submembraneous aggregates due to its self-association into trimers and dimers mediated by its N-terminal G-domains and C-terminal E-domains (Sola et al., 2001; Schwarz et al., 2001). It is unclear whether gephyrin interacts with GABAARs directly, via low-affinity binding, such as its binding to the α2-subunit (Tretter et al., 2008), or indirectly, via as yet unidentified GABAAR-associated proteins, or both. While direct interaction with GABAARs remains to be confirmed in vivo, the role of gephryn in synaptic localization of GABAARs is strongly supported by prominent loss of α2- and γ-subunit-containing synaptic pools in gephyrin-knockout mice (Kneussel et al., 1999).

Gephyrin interacts with a number of other proteins including collybistin, a guanylate exchange factor (GEF) for Cdc42 (Kins et al., 2000), cytoskeletal protein tubulin (Prior et al., 1992), tubulin-associated protein dynein light chain (DLC; Fuhrmann et al., 2002), the actin-binding proteins profilin I and II (Mammoto et al., 1998), actin-associated proteins Mena and VASP (Giesemann et al., 2003) and a glutamate receptor-associated protein GRIP-1 (Yu et al., 2008). Of these, the gephyrin–collybistin interaction is the best characterized (Harvey et al., 2004; and see above). This correlates well with the phenotype of collybistin-knockout mice. These mice have increased levels of anxiety and impaired spatial learning associated with a selective loss of GABAARs in the hippocampus and basolateral amygdala (Papadopoulos et al., 2007).

Reversible low-affinity interactions between GABAARs and gephyrin at the synapse may be necessary for the observed high mobility of GABAARs within the plane of the plasma membrane (Jacob et al., 2005; Lévi et al., 2008). Using a variety of imaging techniques, GABAARs have been shown to diffuse rapidly, in and out of synaptic contact regions (Jacob et al., 2005; Thomas et al., 2005; Bogdanov et al., 2006). However, while measured diffusion rates for single-particle-labelled γ2-subunits, within the plane of the plasma membrane, are fast for extrasynaptic GABAARs, they are significantly slower once receptors enter synapses (Lévi et al., 2008). In this way the number of GABAARs at the synapse can be modulated without altering the number of receptors at the cell surface (Jacob et al., 2005; Thomas et al., 2005), providing a mechanism for rapid changes in the efficacy of synaptic transmission. Clearly, GABAARs are ‘trapped’ and stabilised by synapses, probably by interactions with proteins in the postsynaptic density and/or synaptic cleft. The refinement of this process at individual synapses to ensure the clustering of specific receptor subtypes appears not to involve the intracellular binding partners so far identified, as most exhibit little or no α-subunit specificity.

Surface expression of GABAARs

Another important aspect of GABAAR regulation at the neuronal cell surface relates to the overall levels of expression of these receptors and thus their availability for recruitment to specific synapses. The number of GABAARs at the cell surface is determined by their rate of insertion into the plasma membrane following their synthesis and assembly within the ER, their maturation within the Golgi apparatus, and their rate of removal from the plasma membrane by endocytosis (Arancibia-Carcamo & Kittler, 2009). What remains unclear is whether newly synthesized receptors are inserted directly into the postsynaptic membrane, or only following lateral diffusion from extrasynaptic sites.

A number of proteins associated with GABAARs have been implicated in the maturation of GABAARs following their synthesis through the secretory pathways. Within the ER, newly synthesised GABAARs associate with the chaperone proteins BiP (immunoglobulin binding protein) and calnexin (Connolly et al., 1996b), which may provide important quality control, or with PLIC-1 (a ubiquitin-like protein; Bedford et al., 2001). PLIC-1 was demonstrated to interact directly with all GABAARs and subunits, stabilizing receptor assemblies and protecting them from proteosome-dependent degradation. In addition, the interaction with PLIC-1 promotes the insertion of GABAARs into the plasma membrane (Saliba et al., 2008). Another GABAAR-associated protein that is implicated in the maturation of newly synthesised receptors within the Golgi apparatus is BIG2 (brefeldin A-inhibited GDP/GTP exchange factor 2), which directly associates with β-subunits and co-localizes with GABAARs within the trans-Golgi network (Charych et al., 2004).

Again, despite the identification of a growing number of proteins that influence the insertion of GABAARs into the plasma membrane, no well-characterised mechanisms that differentiate between synaptic and extrasynaptic insertion and none that can be predicted to distinguish between GABAAR subtypes have yet been identified.

Receptor internalisation

Are ‘old’ receptors internalised directly at synaptic sites, or only following lateral diffusion to extrasynaptic sites? This poses additional questions. For example, if all extrasynaptic, γ2-containing GABAARs are removed from the surface away from the synapse, are different receptor subtypes removed independently, or are several different subtypes removed by the same endocytosis process? If the former, different types of GABAARs must either exist in different extrasynaptic domains, where they associate with molecules involved in internalisation, or the structural differences provided by their different subunit compositions must account for the differential binding of proteins involved in internalisation.

Although a variety of proteins have been demonstrated to regulate internalization of GABAARs, these proteins do not show sufficient specificity in their binding to GABAAR subunits to promote subtype-specific internalization. They bind to all β- and/or all γ-subunits, suggesting a more ubiquitous role in the internalization of GABAARs. It is well established that GABAARs undergo a ligand-independent constitutive internalisation through clathrin/dynamin-dependent endocytosis, which requires the AP2 adaptor complex (Tehrani & Barnes, 1997; Tehrani et al., 1997; Kittler et al., 2000). GABAAR α-2/4/5-, β1-3-, γ1-3- and δ-subunits all associate directly with the μ2-subunit of AP2 (Kittler et al., 2000, 2005, 2008; Smith et al., 2008). Blocking these interactions leads to an increase in GABAAR cell surface levels and enhances spontaneous GABAergic currents.

Receptor degradation

Internalised GABAARs are believed to have one of two possible fates: they can be recycled and re-inserted back into the plasma membrane or they can undergo degradation and thus removal from the cell. In cultured neurones, 50% of GABAARs internalised in response to GABA treatment undergo degradation with an approximate half-life of 4 h. The other 50% display a half-life of ∼24 h (Borden et al., 1984; Borden & Farb, 1988).

GABAARs that have been constitutively endocytosed in heterologous expression systems appear to undergo considerable recycling and re-insertion into the plasma membrane (Connolly et al., 1999). It has also been suggested that recycling of GABAARs occurs in cultured neurones (van Rijnsoever et al., 2005; Kittler et al., 2000, 2004). GABAARs that undergo constitutive endocytosis were shown to associate with an intracellular subsynaptic pool upon internalisation (van Rijnsoever et al., 2005), which suggests that GABAARs may shuttle rapidly between this intracellular pool and the surface. Interestingly, this intracellular pool was unaffected by the addition of GABAAR agonists or antagonists, or of benzodiazepines (van Rijnsoever et al., 2005), i.e. there may be differential regulation of GABAARs that are internalised by ligand-dependent and by ligand-independent mechanisms.

As internalised receptors can have these two fates: being recycled back to the cell surface or targeted for degradation, there must be a signal that allows the sorting of GABAARs into these two pools. It has been suggested that the decision for this sorting occurs through an interaction between GABAARs and HAP1 (Huntington-associated protein 1). HAP1 was shown to interact directly with the β-subunits. This interaction stabilizes endcytosed receptors by inhibiting degradation and facilitates receptor recycling to the cell surface, leading to an overall increase in the number of GABAARs (Kittler et al., 2004). Internalized receptors can also be stabilized by an interaction between the γ2-subunit and CAML (calcium-modulating cyclophilin ligand), which also appears to promote recycling of endocytosed receptors.

Known postsynaptic GABAAR binding partners do not account for synapse-specific receptor clustering

A large number of proteins that can be found at GABAergic synapses and/or that associate with or bind to GABAARs have been identified. To date, attempts to find specific binding partners for the intracellular domains of GABAARs have been more successful than attempts to find extracellular domain partners. Some of these postsynaptic proteins associate with GABAA receptors and subunits in the ER or Golgi apparatus, some act as chaperones for the receptors, and others interact with each other to form the postsynaptic density, anchoring and stabilising GABAARs and inhibiting their internalisation and degradation. Finally, there are the proteins that promote GABAAR internalisation and degradation. However, with the possible exception of radixin, which is reported to bind directly and selectively to the α5-subunit, anchoring these GABAARs to the cytoskeleton (Loebrich et al., 2006), none that would mediate selective insertion, sequestration, capture or stabilisation, of a specific α-subunit-containing GABAAR subtype, has yet been identified (Chen & Olsen, 2007, for review). Much of this review has necessarily focussed on proteins that are manufactured in the postsynaptic neurone. To explain the highly selective clustering of GABAAR subtypes at the synapses made by the axons of individual presynaptic GABAergic neurones, it may be necessary to invoke the huge diversity of presynaptic cleft-spanning proteins and their postsynaptic interactors. The extracellular domains of all ionotropic amino acid receptors are very large and complex. This size and complexity has been preserved through the development of many species and must therefore be assumed to confer some benefit and imply some important function(s) beyond the support of transmitter or modulator binding sites.

The significance of selective receptor clustering

The interneurones that innervate α1-GABAARs, including the PV-containing basket cells in cortical regions, contribute to rhythm generation and synchrony, while enhancing their inhibitory outputs is anticonvulsant and sedative. PV-positive basket cell boutons on pyramidal cell somata and axon initial segments are also frequently positive for the M2 muscarinic receptor, although the somata of these interneurones rarely express these receptors (Hájos et al., 1998). Cortical CCK-containing interneurones may also contribute to rhythm generation (Klausberger & Somogyi, 2008), but in many other ways are very different; they uniquely receive direct 5-hydroxytriptamine (5-HT) input, via excitatory ionotropic 5-HT3 receptors. They also, near uniquely, express presynaptic cannabinoid type 1 receptors (CB1R). CB1R activation and CCK both decrease the inhibition produced by these interneurones (Katona et al., 1999; Hájos et al., 2000; Neu et al., 2007; Freund, 2003 for review) and both CCK analogues and cannabis are reported to induce panic attacks, whereas increasing the release of 5-HT, which activates these interneurones, reduces the attacks. These interneurones and the α2-GABAARs they innervate would therefore appear ideally placed to control anxiety. Indeed, enhancement only of the inhibition mediated by α2/3-GABAARs reduces behavioural indices of anxiety (Möhler et al., 2002; McKernan et al., 2000; Whiting, 2006). The potential, therefore, for nonsedative anxiolytic therapies with reduced tolerance and withdrawal and for selective partial agonists as anticonvulsants with reduced dependence is driving development of new benzodiazepine site ligands.

The α5-GABAARs that are activated by dendrite-preferring interneurones in cortical regions do not appear to contribute to the sedative or anxiolytic effects of benzodiazepines. This is, perhaps, not surprising when it is remembered that these receptors are activated by very different types of interneurones. Disrupting or blocking these α5-GABAARs enhanced cognitive performance in rats in hippocampal-dependent learning tasks (Collinson et al., 2002; Chambers et al., 2003, 2004), with α5-GABAARs being implicated as control elements of the temporal association of threat cues in trace fear conditioning (Crestani et al., 2002). Moreover, selective blockade of these receptors in people has been reported to block alcohol’s amnestic effect (Nutt et al., 2007). Interest in partial α5-GABAAR inverse agonists as cognitive enhancers is therefore growing.

Clearly, if these different GABAAR subtypes were randomly distributed over the synaptic and extrasynaptic regions of their postsynaptic targets, the very specific effects on behaviour and cognition that enhancing or disrupting their activity has, would simply not be possible. Why there is such specificity remains to be determined, as it would be unreasonable to propose that it is designed to allow the development of anxiolytic and cognitive-enhancing drugs, convenient though this may prove. It may be that elusive endogenous benzodiazepine site ligands do indeed exist and are able to modulate these GABAARs differentially. That this is at least a possibility is indicated by the partial inverse agonist activity, at synaptic receptors in situ (File et al., 1986; King et al., 1985; Thomson et al., 2000), of benzodiazepine site ligands that act as pure antagonists in expression systems. Whatever the (currently unknown) advantage(s) that may be conferred by the selective clustering of GABAARs, it is dependent upon some hitherto unidentified mechanism that we tentatively propose resides with the cleft-spanning proteins that emanate from the presynaptic bouton and their cleft and postsynaptic binding partners.


The work reported here from our own laboratories was funded by the Medical Research Council, the Biotechnology and Biological Sciences Research Council, the Wellcome Trust, the Engineering and Physical Sciences Research Council (COLAMN), EU Framework 6 (FACETS), Novartis Pharma Basel and Glaxo Smith Kline.

BZ1, BZ2 and BZ3

benzodiazepine (binding site) type 1, 2 and 3


Calcium/calmodulin-dependent serine protein kinase




endoplasmic reticulum


GABAA receptor


α5-subunit-selective partial inverse agonist


inhibitory postsynaptic current


inhibitory postsynaptic potential


laminin neurexin sex hormone binding protein


metabotropic glutamate receptor type


neural cell adhesion molecules


neuroligin 2




Oriens lacunosum moleculare


postsynaptic density




Regulating synaptic membrane exocytosis protein 1α