Photo-antagonism of the GABAA receptor

Neurotransmitter receptor trafficking is fundamentally important for synaptic transmission and neural network activity. GABAA receptors and inhibitory synapses are vital components of brain function, yet much of our knowledge regarding receptor mobility and function at inhibitory synapses is derived indirectly from using recombinant receptors, antibody-tagged native receptors and pharmacological treatments. Here we describe the use of a set of research tools that can irreversibly bind to and affect the function of recombinant and neuronal GABAA receptors following ultraviolet photoactivation. These compounds are based on the competitive antagonist gabazine and incorporate a variety of photoactive groups. By using site-directed mutagenesis and ligand-docking studies, they reveal new areas of the GABA binding site at the interface between receptor β and α subunits. These compounds enable the selected inactivation of native GABAA receptor populations providing new insight into the function of inhibitory synapses and extrasynaptic receptors in controlling neuronal excitation.

T he precise coordination of our behaviour requires that we have adequate temporal control over neuronal excitation. The responsibility for this control falls largely to g-aminobutyric acid type A receptors (GABA A Rs). The timing, extent and cellular location of synaptic inhibition have a critical impact on neural network activity and therefore behaviour [1][2][3][4][5] . Under normal circumstances, inhibition will be regulated by endogenous factors, post-translational modifications and by plasticity mechanisms. It is therefore unsurprising that dysfunction to GABAergic inhibition is implicated in numerous neurological diseases [6][7][8] .
The strength (or macroscopic efficacy) of synaptic inhibition will depend on many factors, not least the number of GABA A Rs clustered at the postsynaptic membrane, and the mean probability of GABA channel opening. Receptor clustering will be affected by numerous signalling pathways, including GABA A R phosphorylation 9,10 ; while channel opening will be a function of the GABA concentration in the synaptic cleft and the activity of allosteric modulators, such as the neurosteroids 11 . Of equal importance for effective synaptic inhibition is the potential for different GABA A R isoforms with their attendant differences in physiological and pharmacological properties, to be targeted to specific domains (inhibitory synapses) in the same cell 12,13 .
To understand how this exquisite targeting of GABA A Rs to specific membrane domains in single cells relates to their impact on neural activity requires a method to modulate, irreversibly inactivate and/or to track the movement of such receptors. This can be partly achieved with fixed tissue by using receptor subtypespecific antibodies. Unfortunately this method will not allow any measure of real-time receptor dynamics 14 . By contrast, we can express GABA A R subunits that carry either mutations to critical structures (for example, ion channel) 15 , or are tagged with fluorophore labels 16 to reveal real-time dynamics in live cells. The latter approaches, although extremely useful, nevertheless require the expression and monitoring of recombinant receptor protein expressed in native cells, and thus, the behaviour of native GABA A Rs can only be ascertained by inference.
Here we take a different approach to enable the direct study of native GABA A Rs. This requires the design of novel ligands that can be attached, and irreversibly bound when appropriately activated, to native GABA A Rs. Using available knowledge of the interfacial GABA binding sites on the GABA A R 17 , we have developed a class of ligands that can photoinactivate GABA A Rs. These ligands have two major advantages over prior methods: first, we can track native GABA A Rs in situ without the need for recombinant receptor expression in neurons, and second, by choosing a ligand that occludes the GABA binding site, we can specifically inactivate populations of GABA A Rs in particular areas thereby gaining valuable insight into their function and trafficking, in addition to revealing the importance of membrane delimited inhibition.

Results
Designing a photoactivated GABA A R antagonist. We selected gabazine as the lead structure for synthesizing new photoactive reagents for several reasons: (i) It is a competitive GABA A R antagonist that binds to residues in the GABA recognition/ binding site preventing agonist-dependent receptor activation. This strategy of causing just inhibition was preferred to photoactive allosteric modulators (often anaesthetics 18,19 ), since these have multiple effects inducing inhibition and also concurrent activation and potentiation at GABA A receptors; (ii) gabazine exhibits partial negative allosteric modulation by inhibiting GABA A R activation by pentobarbital (barbiturate) and  alphaxalone (steroid) from their discrete binding sites on the receptor 20 ; (iii) gabazine contains an easily identified 'GABA structure' in the molecule that is unencumbered by other groups, unlike a similar GABA moiety in bicuculline 21 , which is another competitive GABA A R antagonist 22,23 ; and (iv), the phenoxy group on gabazine presents a chemically convenient site for attaching photoactivatable groups (Fig. 1a).
Chemistry of gabazine analogues. To maximize the prospects of obtaining high potency gabazine analogues, we took note of several key structure-function characteristics of ligands that bind effectively to the GABA binding site. As the carboxy-and aminoends of GABA are important for its engagement at the GABA binding site 24 , and the carboxyl side-chain of the GABA moiety in gabazine is crucial for antagonism 25 , we avoided making any modifications to these parts of the gabazine molecule. We also noted that the aromatic ring at position 6 on the pyridazine ring was important in affording gabazine its potency, and should therefore be retained [25][26][27] (Fig. 1a). Thus, we chose to concentrate on the phenoxy group as the point of attachment for the photoactivatable groups, having shown in initial synthetic studies that the incorporation of a benzyl group led to a further increase in potency (GZ-i1, Fig. 1a) 27 .
The following three types of photoactive groups were incorporated into gabazine: an aryl azide 28 (GZ-A1), a benzophenone 29 (GZ-B1) and an aryldiazirine 30 (GZ-D1; Supplementary Fig. 1a). A second truncated benzophenonegabazine analogue was also synthesized, where the phenoxy ring of gabazine was directly replaced by the benzophenone (GZ-B2; Fig. 1a; Supplementary Fig. 1b). When these photoactive groups are exposed to ultraviolet (UV) light (wavelength B300-370 nm) they respond by forming highly reactive intermediates. In the case of aryl azides and diazirines, this involves the loss of N 2 to afford a nitrene or carbene, respectively, while the benzophenones form a photoexcited state that behaves as a diradical. In each case, the reactive species can then react and covalently attach to nearby amino-acid residues in the GABA binding site.
Photoactive analogues are high potency inhibitors at GABA A Rs. We first assessed the gabazine analogues for their potency in antagonizing a GABA EC 50 response using the synaptic-type recombinant a1b2g2 GABA A receptor expressed in HEK cells. This would determine if the photoactive groups are accommodated by the GABA binding site. The synthetic compound, GZ-i1, is an intermediate between gabazine and its photoactive analogues. The simple addition of a phenyl ring increased the potency of gabazine by more than 30-fold 27 (Fig. 1b,c), in accord with the 20-fold increase in affinity (K i ) of GZ-i1 measured using Schild analysis (Fig. 2c,d).
Surprisingly, the relative potencies of the photoactive compounds, GZ-A1 (azide), GZ-B1 (benzophenone) and GZ-D1 (diazirine), were 1.5-to 30-fold higher than that of gabazine, with the exception of the truncated benzophenone, GZ-B2, which was equipotent (Fig. 1b,c). While these potency comparisons are dependent on the GABA concentration used, the affinities of the photoactive gabazine analogues are not as they are determined directly using a Schild analysis for competitive antagonism 31 (Fig. 2). The antagonist dissociation constants (k B , nM) decreased in the order: GZ-B2 (318)4Gabazine (300)4GZ-B1 (153)4GZ-D1 (132)4GZ-A1 (44)4GZ-i1 (13) ; Fig. 2d). Such a rank order was unexpected if the molecular volume of the photoactive sidechain was the major limiting factor for ligand binding. Thus, we concluded that these large photoactive groups in the phenoxy position of gabazine are fully accommodated at the GABA binding site. The increased affinity (lower k B ) of the analogues must therefore result from increased interactions between gabazine analogues and binding site residues either via H-bonds, cation-p interactions, or p-p stacking of aromatic rings.
Photoinactivation of recombinant GABA A receptors. The photoactive capabilities of the azide, benzophenone and diazirine groups on the gabazine molecule to covalently link to the GABA binding site were studied using whole-cell recording from HEK cells expressing a1b2g2S GABA A receptors. The gabazine analogues, GZ-A1, GZ-B1, GZ-B2 and GZ-D1, were selected, in conjunction with a photoactivation protocol involving UV exposure. The intensity and duration of exposure were titrated to ensure photoactivation of the compounds without perturbing cell health, ascertained by measuring the membrane leak current and access resistance. Control whole-cell GABA-activated currents, recorded before and after applying the photoactivation protocol (see Methods) in the presence of Krebs alone were unchanged (101.1 ± 1.8%; mean ± sem; n ¼ 7; Fig. 3a). This verified that under our conditions, UV light exposure did not damage cells or change GABA potency for a1b2g2 receptors 32 . Similarly, no reduction in the GABA-induced current was observed after applying the photoactivation protocol with gabazine (10 mM; 101.6±3.3%; n ¼ 7), indicating that the parent molecule has no innate photoreactivity, and that 3-5 min is sufficient, after UV exposure, for the antagonist to dissociate from the GABA binding site (Fig. 3b).
For the azide-linked gabazine analogue, GZ-A1, the GABAinduced current was reduced irreversibly post-UV by B30% (to 71.3±6.8%; n ¼ 7; Fig. 3c,g). For the two benzophenone-linked gabazine analogues, the post-UV GABA current was irreversibly reduced by GZ-B1 (to 50.8±1.8%; n ¼ 12; Fig. 3d,g), but not by the truncated version, GZ-B2, lacking one phenyl ring (98.3±4.2%; n ¼ 7; Fig. 3e,g). In comparison, the diazirine-linked analogue, GZ-D1, irreversibly reduced GABA current by B20% (to 79.0±4.5%; n ¼ 7; Fig. 3f,g). The most efficacious molecule inducing irreversible block at the GABA binding site was therefore the 'extended' benzophenone-gabazine analogue, GZ-B1, which was selected for further characterization. The irreversible nature of the inhibition was evident from extended recording periods of at least 30 min post-UV exposure (Fig. 3h). The unchanging extent of inhibition and lack of recovery also re-affirmed that surface GABA A receptors in HEK cells are not replaced during this period 15 . Ablation of the agonist response was routinely achieved with successive cycles of UV exposure in the presence of 10 mM GZ-B1 (Fig. 3i). To ensure that some agonist response remained for the measurement of potencies, we used a single UV exposure cycle in the presence of GZ-B1.
GZ-B1 has lower potency at a3b3c2 and a4b3d GABA A receptors. To determine if GZ-B1 exhibited receptor subtype selectivity, we examined its inhibitory profile for 18 synaptic-and extrasynaptic-type GABA A receptors, selected because they are likely to be expressed in the central nervous system 33,34 . By varying the highly homologous b-subunits (b1-3) in synaptictype a1bxg2 receptors, GZ-B1 potency (IC 50 ) remained constant (analysis of variance (ANOVA); P ¼ 0.26; Fig. 4a,b). Conducting a similar examination with different a subunits in a1-6b3g2 receptors, GZ-B1 potency was significantly reduced at a3b3g2 compared with either a1b3g2 (Po0.001, ANOVA with Tukey-Kramer post hoc tests) or a6b3g2 (Po0.01; Fig. 4a,b). For the prospective extrasynaptic-type receptors, GZ-B1 potency significantly varied in the ab and abd subgroups (ANOVA, Po0.0001), being higher at a6b3 compared with a3b3 (Po0.001) and a4b3 receptors (Po0.01; Fig. 4c,d), and also higher at a6b3d compared with a4b3d receptors. Potency was unaffected by including the d-subunit with a1b2 or a6b3 receptors, but was reduced by its inclusion in a4b3 receptors (Po0.05). By comparison, potency was unaltered by incorporating either y or E subunits into a3b3 receptors (Fig. 4c,d). Comparing the selected synaptic and extrasynaptic GABA A receptors with a1b3g2 revealed significantly lower potencies for GZ-B1 at a3b3g2 and a4b3d receptors (ANOVA, Dunnett post hoc test, Fig. 4b,d).
Ligand docking using a GABA A receptor model based on AChBP. To understand how GZ-B1 binds within the GABA site, we first performed GOLD 35 docking simulations of GABA, gabazine and GZ-B1 with the a1b2g2 GABA A receptor modelled on the 2 Å resolution crystal structure of the unliganded acetylcholine binding protein (apo-AChBP, PDB ID: 2BYN). This template was initially selected because loop C, which caps the binding site when occupied by an agonist 36,37 , is uncapped, but not overtly displaced outwards, as observed when a large competitive antagonist is bound to the same site 36 . For antagonists of comparable size to gabazine and GZ-B1, such as methyllycaconitine, the positioning of loop C in AChBP is unchanged (PDB: 2BYR) 36 . The GABA binding site is located at b-a subunit interfaces surrounded by residues from six binding loops designated as: A, B, and C from the ' þ ' face of the b subunit and D, E and F from the ' À ' face of the a subunit 37,38 (Fig. 5a,b). From all the docking results, the most probable binding mode was selected based on its ranking, its similarity to GABA interactions with the GABA A R as reported in the literature and the frequency of its similarity to the other binding modes in the diverse docking solutions. Docking GABA, gabazine or GZ-B1 into the GABA site identified several charged residues potentially involved in binding (Fig. 5a,b). Some of these have been previously implicated in GABA binding 39 . By docking GABA, we identified two solutions (ranked 1 and 2) that predict two different binding modes whereby the carboxyl group of GABA formed H-bonds with R119 (a1, rank 1) or E155 (b2) and R207 (b2, rank 2) ( Supplementary   Fig. 2a). In addition, for the rank 1 solution, H-bonds are also formed with S156 (b2), G158 (b2), Y159 (b2) and Y205 (b2), and for the second ranked solution, H-bonds are formed with Y97 (b2) and a cation-p interaction with Y157 (b2). The interacting residues are spatially spread around the GABA binding site and hence we predict that GABA potentially binds to the receptor in two modes. Such interactions have been previously shown to be involved in GABA binding 40,41 .
From the gabazine docking, we examined the top 2 ranked solutions (rank 1 and 2). Rank 1 only had one H-bond interaction between the carboxyl group of gabazine and R119 (a1). However for rank 2 the key carboxyl group formed H-bonds with the receptor residues, R207 (b2) and E155 (b2), and the aromatic ring was also engaged in a cation-p interaction with R119 (a1) ( Supplementary Fig. 2b). These interactions were also evident with the top 2 solutions for GABA docking elevating rank 2 as a potential binding mode compared with the other docking solutions. In addition, based on the root mean squared deviation (r.m.s.d.) measure, rank 2 was found to be part of a cluster of similar binding modes. The cluster contained 24% (12/50) of the diverse docking solutions, including ranks 3 and 4 ( Supplementary Fig. 2c).
Ligand docking using a GABA A receptor model based on GluCl. The predicted binding mode for GZ-B1 obtained from the first stage of docking involved H-bonding with R207 (b2), E155 (b2) and R119 (a1) (Fig. 5f). This binding mode was similar in 32% (16/50) of the diverse docking solutions, including ranks 2, 3 and 4, representing the most populated binding mode ( Supplementary Fig. 2e). Intriguingly, the two-stage docking protocol predicted a similar binding mode to that observed using the AChBP template and the scaffold-match constraint. This identified an H-bond between the activated oxygen of the benzophenone group and R84 (a1) (rank 1, Fig. 5g). However, interactions with D162 (b2) and D163 (b2) were not predicted to occur either from two-stage docking or from region-constraint docking.
The docking results predicted that GABA and gabazine are bound completely within the GABA site behind loop C, whereas the benzophenone group of GZ-B1 projects up along the b-a subunit interface and out from under loop C, before re-entering the interface and terminating near a new cavity between b and a subunits (Fig. 5d,f). This cavity is predicted to penetrate through to the external vestibule located above the ion channel. The intersubunit space around the cavity is considered unimportant for GABA activation of the receptor, but its volume is such that competitive antagonists with additional moieties can be accommodated without impeding binding. Another interesting observation is that among the unconstrained docking results, the aromatic ring of GZ-B1 was always orientated towards the extracellular domain in 68 and 84% of the solutions based on AChBP and GluCl, respectively. This preferred orientation of GZ-B1 within the GABA binding site is also supported by the proposed binding mode (Fig. 5d,f).
Residues outside the GABA binding site interact with GZ-B1.
The impact of the b2 E155Q mutation on ligand binding is difficult to interpret as it clearly affects the ability of the ion channel to remain shut in the absence of agonist. To verify that the other mutations are only locally affecting the GABA binding site and not introducing major conformational perturbations into the receptor, we examined allosteric modulation of the GABA A receptor. Specifically, benzodiazepine-induced modulation was unaffected (Fig. 6e).
Photoactivated GZ-B1 irreversibly binds to a1-R84. The importance of a1-R84, b2-D162 and b2-D163 for irreversible binding following photoactivation of GZ-B1 was investigated using near-saturating concentrations of GZ-B1 before and after UV. We also examined a1-R119 as a likely candidate to engage in irreversible bond formation given its close proximity to the photoactivated oxygen of the benzophenone group in GZ-B1.
To assess the ability of photoactivated GZ-B1 to reduce synaptic inhibition, we recorded from cultured cerebellar granule cells and monitored whole-cell GABA currents and spontaneous inhibitory postsynaptic currents (sIPSCs; Fig. 8a). Responses to rapidly applied GABA (1 mM) were depressed to a similar degree, after a single UV exposure, to those observed for recombinant a1b2g2 GABA A receptors. No recoveries were observed  over 40-45 min following GZ-B1 photoactivation (Fig. 8b).
Monitoring sIPSCs before and after an identical UV cycle in the presence of 10 mM GZ-B1 (Fig. 8c) revealed up to a 90% reduction in synaptic current amplitude, which did not recover during the recording (B45 min; Fig. 8d). This level of inhibition indicates that the synaptic receptors are highly sensitive to inhibition by photoactivated GZ-B1. The lack of recovery (both whole-cell GABA currents and sIPSCs) suggests that membrane insertion of GABA A receptors from intracellular stores must be relatively slow.
Tracking photolabelled GABA A receptors. The specific and irreversible binding of GZ-B1 to neuronal GABA A receptors provided a means to label such receptors with fluorophores. We exploited this using a variation of GZ-B1 incorporating a  (d,f insets) Subunit interface surfaces (b2 is blue; a1 is green) are shown with the benzophenone of GZ-B1 protruding from underneath loop C and settling in a cavity above the GABA binding site. The unconstrained binding modes for GZ-B1 (d,f) predicted an interaction with R207 (b2), E155 (b2) and R119 (a1). For AChBP, the scaffold-match-constrained binding mode for GZ-B1 (e) predicted H-bond formation with R84 (a1) and E155 (b2) and a cation-p interaction with R119 (a1). For GluCl, the scaffold-match-constrained binding mode for GZ-B1 (g) predicted H-bond formation with R84 (a1) and a cation-p interaction with R119 (a1). The H-bonds are shown as spring representation. Cation-p interactions are depicted as dashed black lines. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms5454 ARTICLE polyethylene glycol linker attached to biotin ( Supplementary  Fig. 3a) designed to not interfere with photoactivation of GZ-B1 and its binding to GABA A receptors. This moiety readily reacts with streptavidin-coated highly fluorescent quantum dots (QD 655 ; Fig. 9a). By subsequently exposing these molecules to UV light, we labelled and then tracked the surface mobility of irreversibly inactivated GABA A receptors on hippocampal neurons ( Supplementary Fig. 3b,c,d; Fig. 9).
GABA A receptors labelled with GZ-B1 exhibited both confined and mobile trafficking profiles in hippocampal neurons as expected for receptors that are confined at inhibitory synapses and for those that reside in the extrasynaptic domain (Fig. 9c). For comparison with GZ-B1, we also labelled separate GABA A receptors with QDs on a1 subunits via a primary antibody to an external epitope (Fig. 9b). By tracking receptor mobility labelled with GZ-B1 or anti-a1 antibody, we determined the diffusion coefficients (D; Fig. 9d). The median D value after tracking individual QDs for anti-a1-labelled receptors (0.08; n ¼ 788) (Fig. 9e) was significantly reduced for GZ-B1-biotin-labelled receptors (0.07; Kolmogorov-Smirnov test, Po0.001; n ¼ 446 QDs). This probably reflects a1 subunit-containing GABA A receptors predominantly located at synapses, which have lower D values, compared with GZ-B1-biotin-tagged receptors, which will include synaptic as well as the faster moving extrasynaptic GABA A receptor populations. The mean square displacement plots for GABA A receptors labelled with GZ-B1 (black) and anti-a1 antibody, revealed no significant difference in the confinement of the receptors. This is likely, as the ensemble of diffusion coefficients will include a mixed population of various synaptic and extrasynaptic receptors.
The utility of the GZ-B1-QD label is also emphasized in studying receptor internalization. Transfected hippocampal neurons expressing enhanced green fluorescent protein were labelled with GZ-B1-biotin-streptavidin-QD 655 and incubated at 37°C from 0 up to 60 min before fixation ( Supplementary Fig. 4). Under these conditions, we followed the trafficking itineraries of receptors as they internalized into subcellular compartments (Supplementary Fig. 4; Supplementary Movie 1). Overall, the GZ-B1-QD complex forms a very useful label for tracking GABA A receptor movement.

Discussion
Dynamically regulating the number of GABA A receptors at inhibitory synapses is a vital component of synaptic plasticity with implications for the long-term control of neuronal excitability, and for dysfunctional inhibitory transmission during neuropathological states. Monitoring the trafficking of synaptic receptors often requires antibodies recognizing an innate epitope, or a modified receptor structure to incorporate an epitope that is either recognized by selective antisera 48 , or is an inherent fluorophore 49 . Further modifications can enable the receptor to be coupled to a quantum dot 50,51 or carry a mutation that is recognized by another ligand 15 . Although useful, such methods cannot be easily adapted to study native receptors. To address this problem, we devised a method that irreversibly inactivates native GABA A receptors, using a new class of photoactivated GABA A receptor antagonists. These can be used to investigate inhibition in various membrane domains and by linking the photoactivated antagonists to fluorophores, we can simultaneously investigate both receptor function and receptor trafficking.
Gabazine is an ideal lead compound due to its high affinity for the GABA binding site, its suitability for chemical synthesis, and the ease by which structural modifications can be made 25,26 . By attaching photoreactive groups to the phenoxy-end of gabazine, away from the GABA backbone, we found that these analogues The extended benzophenone analogue, GZ-B1, proved the most effective at irreversibly blocking a1b2g2 GABA A receptors following UV photoactivation, with near-saturating concentrations blocking B50% of GABA A receptors in an irreversible manner after only one cycle of UV. Although submaximal, this is more than sufficient for functional and trafficking studies of GABA A receptors 15 . A similar level of inhibition was also reported for the photoactive glutamate receptor inhibitor, ANQX, on AMPA receptors 54 . However, for experiments that demand complete inhibition of GABA currents, several cycles of UV exposure can achieve this; although synaptic GABA currents can be virtually abolished by very few cycles of UV activation of GZ-B1. The reason why the block becomes more effective with successive UV exposure, most likely relates to the photochemical properties of the benzophenone group, which, unlike the azide and diazirine groups, does not lose N 2 upon photoexcitation and thus can readily revert back to its ground state. This feature is advantageous since it allows the benzophenone group to have repeated attempts at covalent binding during successive periods of photoactivation.
The GABA concentration-response curves with GZ-B1 after photoactivation revealed a non-competitive depression compared with the competitive inhibition noted with reversible binding of GZ-B1 in the absence of UV. This is the expected behaviour of an irreversible antagonist at the agonist binding site, whereupon the GABA EC 50 remains largely unaffected.
Once Cys-loop receptor agonists, such as GABA, are accommodated at their binding site, loop C is proposed to close, capping the binding site 36,37,55 , whereas no movement of loop C is observed with larger ligands of comparable size to gabazine and GZ-B1 (ref. 36). For the GZ-B1 molecule, computational docking analysis revealed that the benzophenone group extends along the b-a subunit interface to a region outside the recognized GABA binding site. Interestingly, aligning the primary sequences of a and b subunits along this part of the interface identified a lack of homogeneity for the a-subunits ( Supplementary Fig. 5), which could underlie the slightly different potencies of GZ-B1 at some GABA A receptors. However, the activity of GZ-B1 at both synaptic-and extrasynaptic-type GABA A receptors suggests it can be considered as a broad spectrum photoactive antagonist.
While mutating these residues did not affect GABA binding, they were important for the reversible binding of GZ-B1, since a combined mutation, a1-R84Q and b2-R207Q caused a 41,000fold loss of potency. We identified a1-R84 as the most important binding partner for the UV-activated GZ-B1 molecule, over b2-D162, b2-D163 and a1-R119. This suggests that GZ-B1 is optimally irreversibly bound in just one conformation at the binding site, with suboptimal binding conformations occasionally adopted. However, we should emphasize that docking solutions represent energy-minimized snapshots of the most prevalent three-dimensional (3D) orientations of the bound ligand. Nevertheless, the bound ligand, as well as the amino-acid side-chains at the binding site, will be constantly undergoing Brownian motionlike movement during covalent binding of GZ-B1. Thus, while the photoactivated benzophenone may, most commonly, associate with a1-R84, it could, at different times, associate with a1-R119, b2-D162 or b2-D163. These residues may play key roles in the energy-minimized positioning of GZ-B1 at the binding site, that is, by controlling the efficiency of the covalent attachment.
Applying GZ-B1 to cerebellar granule cells indicated that synaptic GABA A receptors are very susceptible to inhibition and that this inhibition was irreversible over the time course of our recordings (usually 440 min). The level of inhibition was higher than that for whole-cell GABA currents. However, this does not involve changes to the affinity of the antagonist for the GABA A receptors. By simulating synaptic and whole-cell GABA currents, the brief GABA concentration transient (B1 ms) and synaptic receptor occupancy expected at inhibitory synapses resulted in a higher level of block compared with that for longer whole-cell applications (B4 s) and correspondingly longer duration receptor occupancies.
In conclusion, by generating a new photoactivated gabazine analogue, GZ-B1, we can use UV photoactivation to irreversibly inactivate native GABA A receptors both within and outside inhibitory synapses in addition to studying their trafficking without the need to having to use expression-tagged recombinant receptors or antibody-based labelling procedures. By determining where the photoactivated molecule is likely to bind, we have also mapped residues in a new region of the interface between b and a subunits just above the GABA binding site.
Methods cDNA constructs. Murine a1 and b2 subunits and all point mutants were cloned into the plasmid pRK5, and verified by full-insert sequencing.
Cell culture and expression of recombinant GABA A receptors. HEK cells (ATCC, UK) were maintained in Dulbecco's Modified Eagle's Medium supplemented with 10% v/v fetal calf serum, 200 mM L-glutamine and 100 U ml À 1 of penicillin/Streptomycin at 37°C (95% air/5% CO 2 ). Cells were plated onto poly-L-lysine coverslips and transfected with cDNAs encoding enhanced green fluorescent protein and murine a1-6, b1-3, g2S, d, e and/or y GABA A receptor subunits using a calcium-phosphate method. Cells were used for electrophysiology experiments after 16-48 h (ref. 34).
Chemistry of gabazine analogues. To synthesize the photoreactive analogues, we developed a highly concise general strategy ( Supplementary Fig. 1a). Suzuki-Miyaura coupling of 4-hydroxybenzeneboronic acid (referred to as '1' in Supplementary Fig. 1a) with 3-amino-6-chloropyridazine afforded a biaryl building block (2) 27 , which could then be reacted with the appropriate benzyl bromide to attach the photoactivatable groups. Finally, N-alkylation and mild deprotection of the allyl group afforded the products (3; either GZ-A1,-B1 or -D1) in just 4 steps and with good overall yields. The only exception to this strategy involved the synthesis of the truncated analogue GZ-B2, in which the boronic acid of the benzophenone was used directly, resulting in just a 3-step synthesis (Supplementary Fig. 1b; Supplementary Table 3, Supplementary Information-Chemistry).
Photoactivation was performed using a Rapp OptoElectronic JML-C2, with a band-pass filter of 240-400 nm and an optic fibre located in the bath 1-2 mm from the recorded cell. A single cycle of an optimized photoactivation protocol consisted of 10 flashes (2-s interval), capacitance 2 setting (C2) at 150 V. After UV exposure in the presence of the antagonist, the cell was left to rest for 3-5 min while washing with recording solution, to ensure that only covalently bound antagonist would remain in the binding site.
Analysis of whole-cell current data. GABA concentration-response relationships were analysed by normalizing GABA currents to the response induced by a maximal, saturating GABA concentration (I max ) and subsequently fitting with the Hill equation: where EC 50 represents the concentration of the agonist ([A]) inducing 50% of the maximal current evoked by a saturating concentration of the agonist and n represents the Hill coefficient.
Antagonists were evaluated for their potency by constructing inhibitionconcentration relationship curves and fitting the data using: where the IC 50 is the antagonist concentration ([B]) causing half-maximal inhibition of the GABA (EC 50 )-induced response. When complete inhibition was not attained, the above equation was modified to: where I min represents the residual GABA current remaining with a saturating concentration of antagonist, and I max represents the control peak GABA-activated current.
The IC 50 values obtained from individual experiments were converted to pIC 50 values ( ¼ À Log (IC 50 ). Mean pIC 50 values±s.e.m. of at least four experiments were subject to statistical analyses (ANOVA and Student's t-test). Potency histograms have two y axes for mean pIC 50 values±s.e.m., and the IC 50 transform (note: error bars refer only to the pIC 50 ).
The competitive antagonism caused by gabazine and its analogues was analysed according to the Schild method 31 . Full GABA concentration-response curves were obtained in control Krebs in each HEK cell and then one or more curves were established in up to four concentrations of gabazine or one of its analogues. The curves were tested for parallelity and the dose ratios for GABA were calculated from the respective GABA EC 50 s. The mean dose ratios for each antagonist concentration (B) allowed the dissociation constant (k B ) to be determined using the transformed Schild equation: The slope of the Schild plot (log (DR À 1) versus log [B]) was tested to ensure its slope did not deviate significantly from unity. The slopes were then constrained to 1 and the intercept on the abscissa ('dose ratio-1') was used to ascertain the pA 2 ( ¼ Àlog k B ). The level of spontaneous activity observed with mutant GABA A receptors containing the b2 E155Q mutation was determined as, the maximal inhibition of channel activity observed in the presence of 1 mM picrotoxin (I PTX À max ), divided by the total range of channel activity (I PTX À max þ I GABA À max ) (ref. 57).
Homology modelling and computational docking. Murine a1, b2, g2 subunits were aligned to the subunit sequence of AChBP and GluCl using the T-COFFEE server 58 with manual adjustment. Based on the alignment, two 3D homology models of the a1b2g2 GABA A receptor were built with MODELLER 59 using the crystal structures of AChBP (PDB ID: 2BYN) at 2.02 Å resolution and of GluCl (PDB ID: 3RHW) at 3.26 Å resolution. The GABA A receptor a1 subunit exhibits 22% and 31% sequence identity with those of AChBP and GluCl, respectively. In comparison with AChBP and GluCl, the GABA A receptor b2 subunit shares 22% and 36% sequence identity.
Initially, our docking studies were performed on the GABA A receptor homology model derived from AChBP. First, GABA, gabazine and GZ-B1 were docked into the GABA binding site of the homology model. The binding site cavity was defined such that all the receptor residues defined within a sphere of 10 Å radius from the a-carbon of Y157 (b2) were included. Hermes version 1.4.1 interface and GOLD version 5.0.1 (ref. 35) were used to initiate docking. The genetic algorithm settings in GOLD were automatically optimized with maximum search efficiency. During the first stage, all the ligands were docked into the binding site and were kept fully flexible during docking. Ten residues within the binding cavity were selected and their side-chains were allowed full flexibility during docking: F64 (a1), R66 (a1), R119 (a1), Y97 (b2), F98 (b2), E155 (b2), Y157 (b2), Y159 (b2), F200 (b2), Y205 (b2) and R207 (b2). For each of the ligands, 50 diverse docking solutions were generated using the GoldScore scoring function with default parameters. From our homology models, we identified a new cavity at the b-a subunit interface (located higher up than the GABA binding site), which could feasibly accommodate large ligands. To further explore the potential binding residues found in the new cavity, we performed a second stage of docking only for the GZ-B1 case, using GOLD with the 'scaffold-match constraint' (starting from the selected binding mode obtained from the first stage of docking without any constraints). The scaffold-match constraint was used to maintain a fragment at an exact specified position in the binding site with the geometry of this fragment remaining unaltered during docking. All the atoms in GZ-B1 molecule, except the benzophenone group, were retained as a scaffold.
All the docking studies on GZ-B1 described above (two-stage docking and region-constraint docking) were also applied to the GABA A receptor homology model derived from GluCl. For the two-stage docking, we included an 'H-bond constraint' in addition to a scaffold-match constraint. The new constraint was added to promote H-bond interaction between the acceptor oxygen atom of the benzophenone in GZ-B1 and the donor nitrogen atoms of side-chain of R84 found in the newly identified cavity.
For analysing the results, all the H-bond interactions were identified using GOLD. We also analysed cation-p interactions, which are considered to be important for drug-receptor binding and are energetically comparable to H-bond interactions 61 . If the distance between the cation and the centroid of the p system is within 6 Å, and the angle between the line joining the cation, and that the centroid and the normal to the aromatic plane at the centroid is between 0 and 90°, we accepted this as a cation-p interaction 62 . The r.m.s.d. was used as a measure to compare different binding modes. For r.m.s.d. calculation, we only used the scaffold atoms of gabazine and GZ-B1 (those forming the rings and connecting them). Two binding modes with r.m.s.d. less than or equal to 2.5 Å were considered to be similar.
Tracking GABA receptor mobility. The mobilities of GABA A receptors in cultured hippocampal neurons were tracked using QDs photo-linked to GABA A receptors via GZ-B1-biotin (see legend to Supplementary Fig. 4). Cells were treated with 0.5 mM GZ-B1-biotin (previously incubated for 3 min with 25 pM QD 655 -streptavidin; Life Technologies) and either not exposed (control) or UV exposed (40 s) followed by washing of cells in Krebs solution.
Mobilities were also studied using GABA A a1 subunits tagged with QDs via a primary antibody against a1 (gift from Jean-Marc Fritschy, Zurich; incubation in 1 mg ml À 1 for 2 min) and a secondary antibody containing biotin (Millipore; incubation in 5 mg ml À 1 for 2 min) and QD 655 -streptavidin (25 pM; 1 min incubation). Trajectories were analysed using the ImageJ plug-in, SpotTracker 2D/3D and MatLab.