The desensitization gate of inhibitory Cys-loop receptors

Cys-loop neurotransmitter-gated ion channels are vital for communication throughout the nervous system. Following activation, these receptors enter into a desensitized state in which the ion channel shuts even though the neurotransmitter molecules remain bound. To date, the molecular determinants underlying this most fundamental property of Cys-loop receptors have remained elusive. Here we present a generic mechanism for the desensitization of Cys-loop GABAA (GABAARs) and glycine receptors (GlyRs), which both mediate fast inhibitory synaptic transmission. Desensitization is regulated by interactions between the second and third transmembrane segments, which affect the ion channel lumen near its intracellular end. The GABAAR and GlyR pore blocker picrotoxin prevented desensitization, consistent with its deep channel-binding site overlapping a physical desensitization gate.

T he presynaptic release of neurotransmitters is a prelude to their diffusion across the synaptic cleft and subsequent activation of postsynaptic ionotropic receptors, which are responsible for fast chemical neurotransmission. Agonist binding initiates the rapid opening of ligand-gated ion channels permitting a selective flow of ions through the pore. The sustained presence of the neurotransmitter will cause ligand-gated channels to transit from the active open-channel agonist-bound conformation to a desensitized shut-channel, agonist-bound state, thereby limiting current flow 1,2 . Receptor desensitization is a fundamental property of most ligand-gated ion channels and can have profound physiological consequences. These include the progressive reduction of postsynaptic current on repetitive synaptic neurotransmitter release [2][3][4][5] ; the predesensitization of receptors by low ambient concentrations of neurotransmitter due to spillover from active neighbouring synapses 6 ; and the slowing of synaptic current decays leading to their prolongation 5 .
In the central nervous system, neuronal activity results from a balance between excitation and inhibition. This is largely dictated by the activity of excitatory ionotropic glutamate receptors, and inhibition caused by GABA and glycine 7 activating GABA A (GABA A Rs) and glycine receptors (GlyRs). These latter receptors belong to the superfamily of pentameric Cys-loop receptors that also comprises excitatory nicotinic acetylcholine (nAChRs) and serotonin receptors (5HT 3 R) 7 . Differential desensitization kinetics of excitatory and inhibitory receptors has the potential to profoundly affect the spike firing profiles of neurons 5 and thus neural network activity. At the molecular level, desensitization of AMPA and kainate classes of ionotropic GluRs is relatively well understood and involves structural rearrangements at the dimer interface between adjacent extracellular agonist-binding domains [8][9][10][11] . However, for the Cys-loop GABA A Rs and GlyRs, although the structure-function studies have improved our understanding of how these receptors activate 7,12 , the molecular determinants underlying the process of desensitization have yet to be defined.
Here we show that for the inhibitory neurotransmitter GABA A and glycine receptors, the mechanism of agonist-induced desensitization is regulated by residue interactions between the transmembrane domains of the receptor, resulting in a timedependent constriction of the ion channel pore that reduces the agonist-activated membrane conductance. This constriction has the characteristics of a desensitization 'gate' that is discrete from the ion channel gate involved in ligand-gated channel opening.

Results
Desensitization profiles differ between GABA A receptors. To locate the molecular components responsible for desensitization, we selected two GABA A R isoforms that differ markedly in their response to supersaturating GABA concentrations (10 mM) by either desensitizing significantly (a1b2 heteromers; percentage of peak current desensitization (% Des) ¼ 86% and weighted decay time constant (t W ) ¼ 17 s) or minimally (r1 homomers, 19% Des and 23 s, Table 1, Fig. 1a). We then constructed chimeras to first explore the role of the extracellular domain (ECD) by replacing the ECD of a1 with the homologous section from r1 (r1 D260 -a1; Supplementary Fig. 1). The desensitization profile was similar to that observed for wild-type a1b2 receptors (Fig. 1a), consistent with published data 13 . Next we built chimeras containing M1 to M4 and the intracellular linkers (M1-M2 and M3-M4) from r1, coupled to the entire extracellular region of a1 or b2, corresponding to their ECDs and the external M2-M3 linkers (chimeras a1 EXT -r1 TM þ INT and b2 EXT -r1 TM þ INT , respectively; Supplementary Fig. 1). Co-expression of these chimeras supported large GABA currents with limited desensitization, comparable to the profile for wild-type r1 (Fig. 1a, see Table 1 for  all time constants and extents of desensitiszation data and  Supplementary Table 1 for GABA current amplitudes), thereby locating the main determinants of desensitization to the transmembrane and/or intracellular regions of the receptor.
Our primary measure of desensitization involved monitoring the decline of currents activated by maximal saturating concentrations of GABA. To ensure that the mutations did not adversely perturb the general structure, function or long-term trafficking of the receptor, we monitored both the dose-response curves and the recovery phase of the currents from desensitization for selected mutants. Overall, the concentration-response curves and EC 50 values were minimally affected by the mutations, though certain GABA receptor constructs did generate slightly lower EC 50 values, in accord with an increase in affinity that is often associated with desensitized receptors ( Supplementary  Fig. 2). Recoveries of the maximal peak GABA currents following desensitization were also similar for wild-type and singly mutated receptors, with complete recovery from maximal desensitization always achieved in mins ( Supplementary Fig. 3). In addition, we have reported both the rates and extent of desensitization for all mutants to gain an overall view of the desensitization process (Table 1).
Surprisingly, incorporating just M4 from the a1 subunit did not affect the desensitizing phenotype of r1 (r1 D433 -a1, Fig. 1b). We therefore targeted M3 in r1 and replaced six contiguous residues at the carboxyl-terminal (C-terminal; intracellular) end of M3 with the homologous residues from a1. This produced a receptor (r1 a1(D6-postM3) ) that desensitized profoundly within one second following activation by GABA (Fig. 1b). Within this six amino-acid cassette, substituting just one residue, r1 T349K , was sufficient to confer a rapid desensitizing profile on r1 receptors (Fig. 1b).
Interestingly, on the r1 a1(D6 À postM3) background, incorporating the M1-M2 intracellular linker from the b2 subunit considerably reduced desensitization towards that for wild-type r1 (chimera r1 b2(M1-M2 link) þ a1(D6-postM3) , Fig. 1b). Such a reversion was not observed after incorporating the a1 M1-M2 linker ( Supplementary Fig. 4). These data indicate that the intracellular end of M3 and the M1-M2 linker of b2 modulate desensitization, conceivably via an intersubunit (a-b) interaction, which is consistent with the close proximity of a1 M3 C-terminal end to the M1-M2 linker of an adjacent b2 subunit, as seen in our structural models (see below).
We subsequently examined this potential interaction in a1b2 GABA A Rs by first introducing the M1-M2 linker of r1 into b2 subunits (b2 r1(M1-M2 link) ). Co-expressing this chimera with wild-type a1 produced receptors that, unexpectedly, desensitized almost completely at a rate 50-fold faster than wild-type a1b2 receptors (Fig. 1c). Moreover, exchanging eight residues from the intracellular end of M3 in a1 and b2 with those from r1 resulted in a modest three-fold increase in the desensitization rate for a1 r1(D8-postM3) b2 r1(D8-postM3) (Fig. 1c). However, on this background, reintroducing the M1-M2 linker of r1 into b2 (a1 r1(D8-postM3) b2 r1(M1-M2 link þ D8-postM3) , produced relatively little effect on desensitization compared with the wild-type receptor (Fig. 1c). If the combined effect of these two modifications had been additive, we would have expected the double chimera to desensitize even faster than the a1b2 r1(M1-M2 link) receptors. Therefore, the non-additive nature of these mutations further supports an interaction between the intracellular end of M3 and the M1-M2 linker, which could represent a cassette of complementary residues conferring desensitizing properties on the GABA ion channel.
M2 and M3 interface strongly affects GABA A R desensitization. While the identified cassette critically modulates desensitization, it is not the sole determinant. Indeed, exchanging the entire transmembrane portion of M3 for a1 and b2 with r1, virtually eliminated desensitization when these two chimeric subunits were co-expressed (a1 r1 M3 b2 r1 M3 , Supplementary Fig. 5). We therefore compared the primary sequences for M3, and noted that the residue homologous to V338 in r1 is a conserved leucine in other GABA A R and GlyR subunits ( Supplementary Fig. 6). By substituting these leucines for valines (a1 L300V b2 L296V ), we created receptors that displayed very little desensitization (Fig. 2a). Notably, this was achieved with very little change in the GABA concentration-response curve for a1 L300V b2 L296V compared with that for wild-type receptors (Fig. 3), indicating the gating efficacy was minimally affected.
To understand how M3 residues could affect receptor function, we constructed a three-dimensional (3D) model of the a1b2 GABA A R (Fig. 2a) based on the structure of the C. elegans glutamate-gated Cl À channel (GluCl) in the open conformation 14 , which shares high sequence homology with GABA A Rs and GlyRs ( Supplementary Fig. 6). In this model, the side chain of the M3 leucine (a1 L300 ) is orientated towards M2 near the 4 0 position (Fig. 2b,c; Supplementary Fig. 7), and therefore any conformational changes to M3 may be detected in M2 by a proximal glycine residue (a1 G258 ) that is highly conserved in all GABA A R and GlyR subunits ( Supplementary Fig. 6).
We next assessed the importance of the highly conserved M3 valine (a1 V296 ), which is located one a-helical turn above a1 L300 ( Supplementary Fig. 6). Together these residues effectively straddle the 4 0 glycine (G258) in M2 (Fig. 2b). Significantly, their mutation to leucines in a1 V296L b2 V292L strongly disrupted desensitization, further suggesting that rearrangements to the lower part of the M2/M3 intrasubunit interface underlies the process of receptor desensitization (Fig. 2b).
Exploring this interface further, we examined the consequences of mutating a1 V251 and a1 N307 . Valine 251 is located at the À 3 0 position intracellular to M2, bordering the M1-M2 linker, and facing M3 at the level of a1 N307 (Fig. 2b,c). Mutating a1 V251 and the homologous b2 S247 to isoleucines (Fig. 2b), alanines or aspartates (Table 1), produced a 7-to 14-fold increase in the rate of desensitization, while phenylalanine mutants (a1 V251F b2 S247F ) did not yield a resolvable current; however, the co-expression with their wild-type counterparts (a1 V251F b2 and a1b2 S247F ) created receptors desensitizing more than 20-fold faster than wild-type a1b2 counterparts (Fig. 2b, Table 1). It is possible that the double mutant receptors are functional but may desensitize too rapidly to be resolved. Indeed, we used excised patch recordings (see Methods) but still failed to resolve any current, which may reflect a desensitization rate that is faster than the rate of ion channel opening.
Similar results were obtained with the adjacent residue a1 N307 and the homologous b2 N303 . Conservatively exchanging these asparagines for glutamines did not affect receptor function, while a1 N307S b2 N303S and a1 N307V b2 N303V receptors desensitized 15and 100-fold faster, respectively, compared with wild-type a1b2 (Fig. 2c, Table 1). Co-expression of a1 N307D and b2 N303D failed to elicit current, but co-expression with their respective wild-type subunits revealed 4-to 40-fold increases in desensitization kinetics ( Table 1). The lack of current, apparent with these double mutants, could thus be due to their profoundly enhanced rate of desensitization. Overall these results are in complete accord with residues located at the M2/M3 interface in the lower part of the membrane, playing a key role in receptor desensitization.
To investigate the importance of some of the key residues at the M2-M3 interface, and to ensure that the use of Xenopus oocytes was not altering the desensitizing profiles of the GABA-activated currents, we explored the impact of several receptor mutants when expressed in HEK cells. As observed with the oocyte-based experiments, the expression of a1 V251F b2 generated currents that desensitized faster, by B10-fold compared with those for the a1b2 wild type ( Supplementary Fig. 8, Supplementary Table 3). Again, similar current profiles were evident by co-expressing the nearby residue, a1 N307V with the homologous b2 N303V , in HEK cells giving a sevenfold increase in the desensitization rate (Supplementary Fig. 8, Supplementary Table 3).
c2 subunit M2 and M3 affects a1b2c2 GABA A R desensitization. To initially simplify our structural approach to exploring desensitization, we used a1b2 GABA A Rs. However, native synaptic GABA A receptors will contain a single copy of the g2 subunit 15 . To probe desensitization in the abg receptor, we examined the effect of mutations at the intracellular end of the g2 subunit. We selected g2 V262 and g2 H318 for mutation since these are homologous to a1 V251 and a1 N307 , respectively, which have a profound effect on desensitization of ab receptors ( Supplementary Fig. 6). While coexpressing wild-type a1, b2 and g2L subunits produced receptors that desensitized similarly (% Des ¼ 80%, t W ¼ 15.5 s) to wild-type a1b2, mutating g2L V262F and g2L H318V increased the rate of desensitization by 12-and 7-fold respectively, and strongly increased the extent of desensitization (% Des ¼ 98.7 and 96.3, respectively; Fig. 4a). These mutations did not generally perturb receptor function as GABA potency was minimally affected at a1b2g2L V262F and a1b2g2L H318V compared with wild-type a1b2g2L GABA A Rs ( Supplementary Fig. 9), suggesting that receptor gating was unaffected. Moreover, we also examined   Table 1 for values of t w and extent of desensitization (% Des) in all figures. The subunit chimeras are depicted by a colour code: green (a1), red (b2) and grey (r1) (right column). Numbering refers to the position of the interface between the two subunits in the chimera. TM/M, transmembrane domain, INT, intracellular loops.

M4
M4 another key pharmacological facet of g2-containing GABA A Rs involving modulation by benzodiazepines 16 . Diazepam-induced potentiation of low concentration (EC 1-2 , 1-3mM) GABA currents at wild-type a1b2g2L, and mutant a1b2g2L V262F and a1b2g2L H318V receptors, remained unaltered (2.53 ± 0.20-fold (a1b2g2L); 2.14 ± 0.16-fold (a1b2g2L V262F ); and 2.51 ± 0.05-fold (a1b2g2L H318V ), n ¼ 3), again suggesting that the mutations have neither affected gating, nor assembly, nor the folding of the receptors. It is important to note that, since there is only one copy of the g2 subunit per a1b2g2 receptor pentamer, there is also only one copy of the g2 subunit mutation, further emphasizing the importance of the cytoplasmic end of the M2/M3 interface on the desensitization of Cys-loop receptors.
A similar mechanism accounts for the desensitization of GlyRs. If the identified residues in GABA A Rs form a part of the generic mechanism for receptor desensitization at inhibitory Cys-loop receptors, then we should be able to replicate the effects in GlyRs. Applying saturating concentrations (10 mM) of glycine to GlyR a1 homomeric receptors (GlyRa1) evoked peak currents that desensitized by B72% with a t w of 11 s (Table 1, Fig. 4b,c). As mutations to the M1-M2 linker of GlyRa1 have been reported to strongly affect desensitization 17 , we focused on the intracellular end of M3, as we had done with the GABA A R, and exchanged six consecutive residues with those from r1. This chimera (GlyRa1 r1(D6-postM3) ) desensitized by B98% (t w ¼ 0.9 s, Fig. 4b), establishing the intracellular end of M3 in GlyRs as a key component for desensitization.
We then examined the effect of single mutations towards the cytoplasmic end of M2 and M3. GlyRa1 L298 is the homologous M3 leucine to GABA A R a1 L300 . Although the mutant GlyR-a1 L298V did not affect desensitization (Table 1), mutating the highly conserved M2 glycine in GlyRa1 G256V , significantly increased desensitization (% Des ¼ 99.4%, t W ¼ 0.8 s, Fig. 4c) while minimally affecting the apparent affinity for glycine ( Supplementary Fig. 2b). Furthermore, mutating the M3 V294 (GlyRa1 V294L ; homologous to GABA A R a1 V296 ) virtually abolished desensitization (Fig. 4c). At the intracellular end of M2 and M3, we then targeted GlyRa1 A249 (M2) and GlyRa1 N305 (M3), which are homologous to GABA A R a1 V251 and a1 N307 . In accord with our GABA A R data, the mutant GlyRa1 A249F and GlyRa1 N305V displayed near-complete desensitization with rates 13-and 60-fold higher, respectively, than wild-type GlyRa1 (Fig. 4b). This is consistent with our results for a1b2 GABA A Rs, and indicates that similar determinants and therefore mechanisms of desensitization apply to both GABA A Rs and GlyRs, involving structural rearrangements at the cytoplasmic end of the M2/M3 interface.
Locating the desensitization gate using a channel blocker.
To understand how the identified residues regulate the desensitization in GABA A Rs and GlyRs, we compared the crystal structure of GluCl, with those of ELIC and GLIC, the prokaryotic homologues of Cys-loop receptors in putative shut and open conformations, respectively [18][19][20] . While receptor activation involves an increase of the pore diameter between the 9 0 and 20 0 positions, there is a corresponding constriction of the pore at its cytoplasmic end, between À 3 0 and 4 0 (Fig. 5a) 14,18 . We hypothesized that desensitization could be an extension of the conformational rearrangements that occur during the activation process, which is consistent with desensitization proceeding from the open state of the channel 21,22 . This implies that desensitization would involve a gate that is located proximal to the cytoplasmic end of the channel.
To corroborate our desensitization gate model for inhibitory Cys-loop receptors, we used the pore blocker, picrotoxin (PTX), since its binding site is generally considered to be located deep in the pore (between À 2 0 and 2 0 ) precisely where the pore constricts during the activation process ( Fig. 5a) 14,23 . After applying a low GABA concentration (0.3 mM) to a1b2 receptors, enabling the resting (shut) state(s) of the receptor to remain significantly populated, we co-applied a saturating concentration (50 mM) of PTX, which blocked the GABA current. Subsequent washout of PTX while still in GABA revealed an extremely slow dissociation of the pore blocker (Fig. 5b), consistent with it being trapped in a resting (shut) state of the channel 23 . Repeating this protocol now with a saturating GABA concentration (10 mM) to deliberately ARTICLE drive receptors into the desensitized state, the apparent dissociation of PTX was much faster (Fig. 5c). This implies that PTX is not trapped in a desensitized state and that the desensitization gate must be distinct from the activation gate (Model 1, Fig. 6a).
Two simple schemes may explain the fast dissociation of PTX in the presence of saturating GABA. First, PTX binding/ unbinding could be the same to open and desensitized receptor states. However, when GABA is saturating, the peak current after washing out PTX is clearly larger than the steady-state current before its application, which is not predicted by equal binding/ unbinding to both states (Model 2, Fig. 6b). Second, PTX binding could be prevented in the desensitized state (Model 3, Fig. 6c). The recovery of current following PTX removal could then reflect two distinct processes: PTX dissociation from its binding site and subsequent receptor desensitization. For this scheme, we would expect the recovery current to exhibit a peak followed by a plateau. However, such behaviour is not obvious from our recordings with the GABA A R (Fig. 5c), probably because PTX dissociation is much slower than the rate of desensitization of GABA A Rs (Fig. 6c, Supplementary Table 2).
To gain a clearer view as to where PTX is binding in relation to our proposed desensitization locus, we examined GlyRs, where PTX dissociation is notably much faster 24,25 . During saturating glycine concentrations, PTX blocked the plateau current (Fig. 7a). Washing out PTX now revealed a clear peak-plateau current profile as expected theoretically, indicating that PTX had blocked the receptor in an open conformation, a profile that is accurately predicted only by Model 3 (Fig. 7b-d). These observations are consistent with a mechanism whereby the PTX-binding site spatially overlaps the desensitization gate. Furthermore, as desensitization leads to an increase in the agonist occupancy, such a model could account for the negative allosteric interaction between the receptor agonist and PTX 24,25 .

Discussion
The present view of desensitization at Cys-loop receptors is that the molecular determinants are most likely contained within the ECD 26 or the ECD-TMD coupling interface 27,28 . Our study indicates that a different domain is involved whereby the cytoplasmic end of the ion channel forms a physical desensitization gate in inhibitory Cys-loop receptors. While this may initially appear contradictory to earlier studies, a role for residues in the ECD or the ECD-TMD interface in setting macroscopic desensitization can be entirely reconciled with our new view. First, in the event that channel opening precedes desensitization, a mutation affecting the efficacy of gating could ultimately affect the macroscopic rate of desensitization without actually modifying the microscopic rate of desensitization. Interestingly, the ECD-TMD coupling interface is critical in setting the efficacy of channel gating. Accordingly, exchanging residues in this region between nAChRs and 5HT 3 Rs not only switches the desensitization properties of the receptors, but also their single-channel open lifetimes 27 .
Second, our proposal for a constriction (gate) of the channel lumen near its intracellular end during desensitization is likely to be accompanied by conformational rearrangements at the extracellular end that should be constrained by residues in the ECD-TMD coupling interface. Considering that M2 helices can exhibit rigid-body motion during gating 12,29 , we would predict that the pore may widen at its extracellular end during desensitization. This could be consistent with conformational rearrangements at the ECD-TMD interface of GABA A receptors during desensitization, as reported by voltage-clamp fluorometry 28 . Moreover, such an effect has recently been proposed for GLIC, using electron paramagnetic resonance (EPR) spectroscopy to measure the distance between spinlabelled cysteines introduced along the ion channel pore 30 .
Interestingly, EPR predicts that desensitization might involve a mid-membrane gate with the intracellular end of the channel remaining mostly fixed. However, there are caveats to the interpretations of EPR data. First, in the narrowest part of the GLIC channel, spin labels might occlude the pore, thus acting as pore blockers, and as for PTX-bound GABA receptors, desensitization could conceivably be prevented by the labelling. Unfortunately, the irreversible inhibition by pore-blocking spinlabels would also render these receptors unsuitable for electrophysiological studies. Second, in the confined space of an ion channel pore, it is likely that the motion of the spin labels is also constrained. Thus, conformational changes around the spin labels could cause dynamic reorientation, making the extraction of distance information quite challenging 31 . For these reasons, EPR spectroscopy could be confounding when studying conformational rearrangements in the constricted space of an ion channel.
It is also important to note that a previous study showed that M1 and M2 were important in setting the desensitization properties of g2 and d subunit-containing GABA A receptors 3 . While the molecular mechanisms were unclear, these results are interpretable in the light of our present study. The importance of M2 is clearly demonstrated by our mutagenesis data, and from our proposal regarding the location of a desensitization gate. Moreover, as there are extensive interactions between the TMDs, M1 would be expected to affect the conformation of both M2 and M3, and could also influence desensitization by affecting the M1-M2 linker. We should add a cautionary note to emphasize that perturbations to receptor structure can potentially have far reaching ramifications for predicting the consequences of mutagenesis to a functional outcome. For this reason, it is hard to predict the effects of residue substitution in the M2-M3 linker when manipulating the side-chain volume, charge and hydrophobicity. Nevertheless, structure-function studies of Cys-loop receptor activation, using similar methodologies to those employed here have proved to be very precise tools in identifying and attributing key parts of the receptor involved in function 12,29 , as evidenced by recent corroborating data between receptor function and the crystal structure of the apo form of GluCl 32 .
The molecular mechanisms underlying the desensitization of inhibitory Cys-loop receptors, which may conceivably extend to the entire Cys-loop family, appear strikingly different from the mechanisms responsible for desensitization of ionotropic GluRs. This is unsurprising, since from a structural point of view, Cys-loop receptors are quite distinct from ionotropic GluRs by forming pentamers with the ECD of each subunit interacting to form a ring, and the agonist binding sites located at interfaces between adjacent ECDs 7,33 . By contrast, ionotropic GluRs are tetramers harbouring discrete intrasubunit-binding sites 7,8 . Moreover, Cys-loop receptor subunits contain four transmembrane segments, which differ from the three transmembrane segments and one re-entrant P-loop topology of ionotropic GluR subunits.
The cytoplasmic end of the pore in Cys-loop receptors also contains the ion selectivity filter 34 . The conformational changes proposed to underlie desensitization here are reminiscent of the collapse of the selectivity filter that is considered responsible for slow inactivation of voltage-gated K þ and Na þ channels [35][36][37] .  ARTICLE Specific mutations near the P-loop of these voltage-gated channels profoundly impact on slow inactivation, in a similar manner to the effects of the mutations described in this study for Cys-loop receptor desensitization 38,39 . Corroboration for the slow inactivation model of voltage-gated ion channels was obtained using pore-blockers, such as tetraethylammonium, which reduced slow inactivation of Shaker K þ channels by a 'foot-in-the-door' mechanism, essentially by binding near the selectivity filter to prevent its physical collapse 38,40 . By analogy, we used the Cys-loop Cl À channel blocker PTX to prevent GABA A R and GlyR desensitization, which accorded with there being a spatial overlap between the desensitization gate and the PTX-binding site in the ion channel. Of note, a recent study showed that, in the absence of GABA, PTX is trapped in the resting state and not in the desensitized state 23 , though prolonged GABA applications were not investigated, and it therefore remains possible that PTXbound receptors might desensitize in the presence of saturating concentrations of GABA. Indeed, all the models we depict (Fig. 6a-c) are consistent with the data in ref. 23. To extend this further, our experimental results show that PTX actually prevents desensitization even in the presence of saturating concentrations of GABA. Overall, our study identifies a new potential molecular mechanism in GABA A Rs and GlyRs that underpins one of the most fundamental parameters in pharmacology, that of receptor desensitization. We also establish a strong parallel between Cys-loop receptor desensitization and the slow inactivation of voltage-gated K þ and Na þ channels, even though the structure and membrane topology of these tetrameric channels is completely different from pentameric Cys-loop receptors.
In this context, it is notable that a recent crystal structure for the GABA receptor b3 homomer has been solved. Although it is considered not to be a physiological form of the GABA A receptor, being unable to be activated by GABA 41 , the crystallographic data also alludes to a constriction at the intracellular end of the ion channel pore as being representative of a desensitized state 42 .
Our model suggests that the desensitization of Cys-loop receptors involves a rearrangement of the M2/M3 interface. This concept would accommodate the effects of modulators binding within the M1/M2/M3 intrasubunit cavity of nAChRs, which prevent desensitization [43][44][45] . Our results therefore provide a rationale for the mechanism of action of these drugs, and could help in designing pharmaceutical compounds to modulate Cys-loop receptor function. This is likely to be beneficial since Cys-loop receptors are involved in many pathological conditions, such as addiction, anxiety, depression, neurodegenerative diseases and seizures [46][47][48] .

Methods
Molecular biology. Murine GABA A R a1, b2 and g2 subunits and human r1 subunits were subcloned into pRK5 at the EcoRI site. The human GlyRa1 subunit was subcloned into pRK5 between the EcoRI and the NotI sites. Most GABA A chimeric receptors were obtained by amplification following a four-step procedure: first, a PCR was performed to build the amino-terminal end of the construct with the forward primer SP6 (5 0 to 3 0 sequence: CACATACGATTTAGGTGACACTA TAG) and a reverse primer overlapping the junction between the parental DNAs, followed by DNA purification (QIAquick PCR purification kit, Qiagen). Second, a PCR was performed to build the C-terminal end of the construct with the reverse primer P5 (CAGACATGATAAGATACATTGATGAG) and a forward primer overlapping the junction between the parental DNAs, followed by DNA purification. Third, the PCR products were assembled through a final PCR step using SP6 and P5, followed by DNA gel extraction (QIAquick gel extraction kit, Qiagen). Fourth, the end product was subcloned into pRK5 with the restriction enzymes ClaI and SalI (NEB), using T4 DNA ligase (Roche). See Supplementary Table 4 for the primer sequences and the DNA templates used for the chimeras molecular biology. Single-point mutants and some chimeric receptors with small sequence exchanges (less than eight residues) were constructed with a single PCR reaction using Phusion DNA polymerase (Thermoscientific), followed by DNA gel extraction, 5 0 phosphorylation using polynucleotide kinase (NEB) and subsequent ligation. See Supplementary Table 5 for the primer sequences and the DNA templates used for these mutant receptors.
Two-electrode voltage clamp and analysis. Xenopus laevis oocytes expressing recombinant receptors were prepared, injected with complementary DNAs (cDNAs; at 1-30 ng mL À 1 ), voltage clamped and recorded as described previously 16 . All compounds were purchased from Sigma. EDTA (10 mM) was added to the OR2 solution when recording from oocytes expressing GlyRs to chelate contaminating Zn 2 þ , which potentiates GlyR function 49 . Recordings were digitized at 500 Hz and filtered at 50 Hz. Desensitizing currents were induced by 1 min agonist applications. The design of our recording chamber and the use of supersaturating concentrations of agonist enabled us to elicit currents with 20-80% rise times of 22 and 15 ms for wild-type a1b2 GABA A Rs and GlyRa1, respectively. The extent of desensitization was determined as 1 À I Residual I peak ; where I peak is the agonist-induced peak current and I Residual the residual of this current remaining at the end of the agonist application. Weighted decay time constants for desensitization were determined by fitting the desensitizing phase with two or three exponential components (Clampfit ver 8). Experiments that assessed the time period for recovery of the receptor from desensitization were performed by first driving the receptor into a profound desensitized state (30 s exposure to millimolar GABA or glycine concentrations), then monitoring the recovery of peak currents (2 s exposures) repeated at 10, 20, 30, 45, 60, 90, 120, 300 and 600-s intervals after the initial desensitizing exposure to the agonist.
To resolve even faster channel kinetics, we have performed rapid-application perfusion experiments with outside-out patches pulled from Xenopus laevis oocytes expressing GABA A Rs. Using a piezo-driven theta tube, we achieved 20-80% solution exchange times o100 ms. At wild-type a1b2 GABA A Rs, a 3-s exposure of 10 mM GABA led to a current that desensitized almost completely and is described by three exponential components of B10, 100 and 500 ms, consistent with published results obtained with recombinant receptors expressed in HEK cells 2 . Although the 10-ms component cannot be detected using two-electrode voltage clamp (TEVC), we should have resolved the 100-ms component using oocytes, given the 22-ms rise time for GABA-activated currents, and, because exponentials of this order were detected with some of our mutant receptors. However, we never resolved such a fast component at wild-type a1b2 GABA A Rs (fastest component was B1-2 s). Furthermore, the almost complete extent of fast desensitization in most outside-out patches did not concur with the large currents obtained with TEVC after a few seconds of exposure to GABA. Interestingly, recent results demonstrate that, in whole-cell patch-clamp recordings, GABA-and glycinemediated currents decay much faster than their respective membrane conductances in the continuous presence of the agonist 50 . Since the decay rates of the currents were matched by the rates of change of internal Cl À concentration, such an observation led the authors to speculate that the fast decay of currents, initially thought to reflect desensitization, actually reflects a decrease in the driving force for Cl À ions. This effect is likely to be present in most patch-clamp experiments performed on small cells or outside-out patches, especially under high series resistance conditions. The oocyte is therefore a reliable model system to study the Cys-loop receptor desensitization. Of note, the slow conductance decay rates measured in ref. 50 are consistent with the desensitization kinetics we measure in TEVC at wild-type receptors.
Finally, as mentioned above, we used outside-out patches pulled from oocytes that expressed mutants showing increased rates of desensitization under TEVC. Unfortunately, we could not resolve any current from patches containing a1b2 r1(M1 À M2 link) or a1 V251F b2 receptors, even though the 'parent' oocytes exhibited strong functional expression (45 mA peak GABA current). Moreover, using outside-out patches from oocytes expressing the mild mutant a1 G258A b2 G254A , we rarely recorded currents, and only then of a few pA decaying fully within 100 ms. For all these reasons, we discarded the fast-perfusion experiments.
HEK whole-cell patch-clamp recording. HEK cells were transiently transfected using the calcium phosphate precipitation method with cDNAs pre-mixed in a 1:1 ratio, with 4 mg total cDNA applied to each per 22 mm coverslip. HEK cells were voltage clamped and recorded from as described previously 51 . Recordings were digitized and filtered at 6.5 kHz. Desensitizing currents were induced by 8 s applications of GABA (3 mM). The use of a rapid U-tube perfusion system and the use of supersaturating concentrations of agonist enabled us to elicit currents with average 20-80% rise times of o8 ms for wild-type a1b2 GABA A Rs. The extent of desensitization and weighted decay time constants were determined as for TEVC, up to 6 s after the start of GABA application.
3D molecular modelling and illustration. Modeller (ver 9.7 (ref. 52)) was used to build 3D homology models of the a1b2 heteromeric GABA A R based on the crystal structure of GluCl (pdb 3RHW (ref. 14)). Structures were visualized with Pymol 53 ; pore radii were calculated with MOLE 54 .
Kinetic modelling. Channelab (ver 2, Synaptosoft, GA) was used to build the virtual recordings in Figs 6 and 7. The binding and gating rate constants are broadly consistent with previously published values for GABA A Rs and GlyRs 51,55 . The desensitization rates were chosen to account for the profile of our current recordings. We modelled the slow component of receptor desensitization, since this is the component that is observed in our long duration applications of agonist, which were used to gather receptors in the desensitized state. The dissociation rate for PTX used for GABA A Rs predicted a slow recovery of the current consistent with our recordings in low concentrations of GABA. The dissociation constant for PTX (0.4 mM for GABA A Rs and 10 mM for GlyRs) are consistent with published IC 50 values 24,56 .