The dual‐gate model for pentameric ligand‐gated ion channels activation and desensitization

Abstract

Pentameric ligand‐gated ion channels (pLGICs) mediate fast neurotransmission in the nervous system. Their dysfunction is associated with psychiatric, neurological and neurodegenerative disorders such as schizophrenia, epilepsy and Alzheimer's disease. Understanding their biophysical and pharmacological properties, at both the functional and the structural level, thus holds many therapeutic promises. In addition to their agonist‐elicited activation, most pLGICs display another key allosteric property, namely desensitization, in which they enter a shut state refractory to activation upon sustained agonist binding. While the activation mechanisms of several pLGICs have been revealed at near‐atomic resolution, the structural foundation of desensitization has long remained elusive. Recent structural and functional data now suggest that the activation and desensitization gates are distinct, and are located at both sides of the ion channel. Such a ‘dual gate mechanism’ accounts for the marked allosteric effects of channel blockers, a feature illustrated herein by theoretical kinetics simulations. Comparison with other classes of ligand‐ and voltage‐gated ion channels shows that this dual gate mechanism emerges as a common theme for the desensitization and inactivation properties of structurally unrelated ion channels.

Introduction

Ionotropic receptors are responsible for fast chemical neurotransmission. Upon binding of their agonist, the transmembrane pore of these receptors quickly opens to enable the selective flow of permeant ions across the plasma membrane. Amongst ionotropic receptors, the superfamily of pentameric ligand‐gated ion channels (pLGICs) comprises excitatory serotonin receptors and nicotinic acetylcholine receptors (nAChRs), the latter contributing notably to higher brain functions such as cognition and reward (Changeux, 1990), as well as chloride‐permeable γ‐aminobutyric acid (GABA) type A receptors (GABA_ARs) and glycine receptors (GlyRs), which mediate fast inhibitory synaptic transmission in the central nervous system of vertebrates. This superfamily was formerly known as the Cys‐loop family, since all animal pLGICs contain a conserved disulfide bridge. However, following the discovery of bacterial homologues devoid of the corresponding cysteines (Tasneem et al. 2005; Bocquet et al. 2007), these receptors were regrouped under the generic name of pLGICs. More recently, the name of ‘Pro‐loop receptors’ has been proposed based on the presence of a strictly conserved proline in the ‘Cys‐loop’ region (Jaiteh et al. 2016). In line with their paramount physiological importance, pLGICs are primary targets for pharmacological treatments of a wide range of diseases: benzodiazepines, which positively modulate GABA_ARs, are used to treat anxiety and epilepsy (Galanopoulou, 2008; Nuss, 2015), while drugs targeting nAChRs are investigated as a potential cure for various diseases including Alzheimer's disease, schizophrenia and tobacco addiction (Taly et al. 2009).

A wealth of structural studies have recently improved our understanding as to how pLGICs activate at the near‐atomic level. Indeed, the structures of two prokaryotic pLGICs were solved by X‐ray crystallography almost a decade ago: the Erwinia ligand‐gated ion channel ELIC (Hilf & Dutzler, 2008), which is activated by a series of amino acids including GABA, and the proton‐activated Gloeobacter ligand‐gated ion channel GLIC (Bocquet et al. 2007, 2009). Several X‐ray structures of eukaryotic pLGICs followed in the past years: the C. elegans glutamate‐gated chloride channel (GluCl) (Hibbs & Gouaux, 2011), the mouse 5‐HT_3A serotonin receptor (Hassaine et al. 2014), the human β3 GABA_AR (Miller & Aricescu, 2014), the human α3 GlyR (Huang et al. 2015), and the α4β2 nAChR (Morales‐Perez et al. 2016), i.e. the first structure of a heteromeric pLGIC since the medium‐resolution cryo‐electron microscopy (cryo‐EM) structure of the Torpedo muscle‐type nAChR (Miyazawa et al. 2003). Of note, the zebrafish α1 GlyR has been the first member of the family to be examined by single‐particle cryo‐EM (Du et al. 2015).

All these data show an overall conservation of pLGIC architecture, where all five subunits are arranged in a ring‐like structure, with a pseudo five‐fold axis of symmetry coinciding with the ion channel (Fig. 1 A; Cecchini & Changeux, 2015). This stereotypical architecture includes the location of the orthosteric site where the transmitter binds: it is located at the N‐terminal extracellular domain (ECD), at the interface between adjacent subunits. The ECD of each subunit, formed by a β sandwich, is connected to the transmembrane domain (TMD), composed of four membrane‐spanning α‐helical segments, named M1–M4. The M2 segments line the pore, allowing the selective flow of permeant ions in the open conformation of the channel (Giraudat et al. 1986; Imoto et al. 1988; Leonard et al. 1988). Amidst this well‐ordered modular architecture, the M3–M4 loop is involved in the trafficking of the receptors to the plasma membrane, their anchoring at the synapse, and their modulation by intracellular interactions and phosphorylation (Smart & Paoletti, 2012; Langlhofer & Villmann, 2016). While the central portion of this cytoplasmic loop is a highly variable and flexible region, the post‐M3 and pre‐M4 regions fold into α‐helices called MX and MA, respectively (Hassaine et al. 2014). The latter is involved, in particular, in the ionic conductance of the channel (see below).

Figure 1. Structural overview and gating mechanism of pLGICs.

A, left, top view of GluCl bound to glutamate (pdb code 3RIF; Hibbs & Gouaux, 2011). One subunit is highlighted in green, coordinating a glutamate molecule shown in orange at its principal face. Note the M2 helices from all five subunits, which line the ion conducting pore. Right, side view of GluCl, which delimits the extracellular domain (ECD) in the extracellular space (Ext.) and the transmembrane domain (TMD). The plasma membrane is schematized in yellow. Note the absence of the intracellular domain (ICD) in the intracellular space (Int.) for this particular construct. B, schematic depiction of pLGIC activation. For the sake of clarity, only two simplified subunits are shown, omitting the M1, M3 and M4 segments of each subunit to retain only the M2 pore‐lining segments. In this scheme, the agonist (orange oval shape) binds to its ECD interfacial orthosteric site, which elicits a pre‐activation or ‘priming’ of the receptor by promoting an unbloomed conformation of the ECD. In other words, the primed conformation displays a higher affinity for the agonist than the resting conformation. The final activation step results from the opening of the channel gate, in the upper half of the pore, potentially concomitant with the twisting of the entire receptor. Note that this scheme is oversimplified, and does not address well‐known features of pLGIC activation, such as the multiplicity of agonist binding sites, nor does it distinguish between flipping, priming or catch and hold mechanisms.

Gating mechanism and permeation determinants of pLGICs

During fast synaptic transmission, vesicular fusion leads to the brief release of a high concentration of neurotransmitters (typically 1 mm), which remain in the synaptic cleft for a duration of ∼1 ms (Katz & Miledi, 1973; Kuffler & Yoshikami, 1975; Attwell & Gibb, 2005). Most neurotransmitter‐gated channels, and synaptic pLGICs in particular, have been selected by evolution for their fast activation and deactivation kinetics, which allows them to stay tuned for activation after a minimal time lapse and to follow high frequency vesicular fusions evoked by trains of action potentials (Papke et al. 2011). Fast deactivation kinetics notably involves a high dissociation rate for the agonist, which translates into a low apparent affinity for the agonist. On the contrary, extra‐synaptic types of pLGICs may display a higher apparent affinity for the agonist, as generally seen for GABA_ARs (Mortensen et al. 2011), making them able to react to low concentrations of agonists encountered during volume transmission (Vizi et al. 2010; Trueta & De‐Miguel, 2012).

Activation of pLGICs has long been analysed according to a minimal two‐state model, the receptor equilibrating between a resting (shut) state and an active (open) state. In particular, the Monod–Wyman–Changeux (MWC) model has been thoroughly used, whereby the receptor can readily visit both conformations in the absence of agonist, the latter simply shifting the conformational equilibrium towards the open state (Monod et al. 1965; Einav & Phillips, 2017). The strongest argument in favour of the MWC model resides in the spontaneous openings measured in the absence of agonist, initially described for mouse muscle‐type nAChRs (Jackson, 1984). Such spontaneous activity gives rise to robust constitutive currents in cells expressing mutant pLGICs endowed with strong gain‐of‐function phenotypes (Purohit & Auerbach, 2009; Colquhoun & Lape, 2012). Still, in the past decade, single‐channel studies performed on GlyRs and nAChRs identified intermediates between the resting and open states of the receptors, which we generically name here ‘pre‐active’ states. They carry a closed channel and are transiently stabilized by agonists. The pre‐active state called ‘flipped’ is partly stabilized by partial agonists, explaining why they elicit only a fraction of the response elicited by full agonists (Lape et al. 2008). The pre‐active states called ‘primed’ explain why low concentrations of agonist do elicit short‐lived open states (through stabilization of a partially primed state), which are kinetically distinct from the long‐lived ones occurring under higher concentrations (through stabilization of a fully primed state) (Mukhtasimova et al. 2009). More recently, a ‘catch and hold’ mechanism has been proposed, stipulating that the binding of agonists promotes a first conformational change leaving the binding site in a low affinity state (‘catch’), subsequently followed by another isomerization step resulting in a high affinity pre‐active state (‘hold’; Purohit et al. 2014; Nayak & Auerbach, 2017). This scheme explains the apparent paradox that structurally related agonists display markedly different association constants for the resting state. The ‘catch and hold’ model may be seen as a refinement over the idea behind the priming model, which might correspond to the ‘hold’ step; and the priming model may itself be seen as a refined version of the flipping model, in which conformational changes at distinct subunits are considered individually (Plested, 2014). In parallel, rate‐equilibrium free‐energy relationships suggest that pLGICs’ activation pathway contains multiple brief intermediates (Grosman et al. 2000), possibly indicating an even larger repertoire of pre‐active states accessible to the nAChRs. The concept of pre‐active states was further extended to GABA_ARs, since the positive allosteric modulators (PAM) benzodiazepines were shown to facilitate a pre‐activation step at GABA_ARs (Gielen et al. 2012; Dixon et al. 2015), similarly to the action of the allosteric modulator NS‐9283 at α4β2 nAChRs (Indurthi et al. 2016).

Interestingly, three pLGICs, GLIC, GluCl and the zebrafish α1 GlyR, have been solved in several conformations, highlighting some key allosteric reorganization associated with channel opening (Prevost et al. 2012; Althoff et al. 2014; Sauguet et al. 2014; Du et al. 2015). Comparison of the structures solved in the absence and presence of agonist suggests a common mechanism of activation whereby agonists stabilize the ECD in a contracted (‘unbloomed’) conformation, and the entire receptor in a twisted conformation (Nemecz et al. 2016). Importantly, normal mode analysis and molecular dynamics simulations are consistent with this bloom and twist activation mechanism (Taly et al. 2006; Calimet et al. 2013). As a cautionary note, the interpretation of MD trajectories relies on the assignment of functional states to the structures used as starting points for the simulations, and it has been proposed that the initial GLIC and GluCl structures might represent desensitized states (Akabas, 2013). Still, recent molecular dynamics suggests that the main quaternary event concomitant with channel opening lies in the twisting (Martin et al. 2017). This quaternary motion is coupled to a tilt of the M2 pore‐lining helices, yielding a widening of the upper part of the channel that carries the activation gate. It consists in two or three rings of hydrophobic residues encompassing the 9′ and 13′ M2 residues that form a hydrophobic barrier to ion permeation, which is released upon channel opening (see Fig. 2 and Jaiteh et al. 2016 for numbering conventions). These structural features, including the binding sites for agonists and allosteric modulators, have been extensively described in recent review articles (Corringer et al. 2012; daCosta & Baenziger, 2013; Cecchini & Changeux, 2015; Nemecz et al. 2016), and will not be reviewed here in detail.

Figure 2. Location of molecular determinants of anionic pLGIC desensitization in the lower half of the TMD.

Top left, top view of the TMD of GluCl (pdb code 3RI5; Hibbs & Gouaux, 2011). One subunit is highlighted in green, showing the arrangement of transmembrane segments M1–M4 to the pore: M4 is the most distal segment, while M2 forms the pore‐lining α helix. The other subunits are coloured in grey. Top right, enlargement showing the proximity of the M3 and M1 helices of adjacent subunits. Bottom right, side view depicting the location of the M1–M2 linker, in the vicinity of the intracellular end of the M3 helix from the adjacent subunit. Note also that the cytoplasmic M3–M4 loop extends at the C‐terminal end of M3. Bottom left, rotated side view of the M2 and M3 segments of the subunit coloured in green. The M2 9′ and −2′ residues, part of the activation gate and the selectivity filter respectively, are shown in stick representation. Their mutation affects the gating and desensitization of pLGICs (see main text). Residues highlighted with the sphere representation are homologous to the residues whose mutation strongly modulates the desensitization of α1β2 GABA_ARs, α1β2γ2 GABA_ARs and α1 GlyRs in Gielen et al. (2015). Numbering of residues has been made according to Jaiteh et al. (2016).

It is noteworthy that the structural reorganizations underlying the above‐mentioned pre‐active states remain unknown. However, recent work on the prokaryotic GLIC recently identified and characterized structurally such an intermediate. Indeed, using the tryptophan‐induced fluorescence quenching method, Menny et al. (2017) managed to follow the structural dynamics of a fully functional GLIC protein reconstituted into liposomes. Data show that the agonist promotes a fast quaternary compaction of the ECD in concert with a key revolving motion of the M2–M3 loop at the ECD–TMD interface. This global pre‐activation step is followed by a slower opening of the channel to elicit the electrophysiological response. Interestingly, similar protein motions are found in a particular X‐ray conformation of GLIC, named the ‘locally closed‐LC2’ conformation, where the ECD has undergone the transition toward the active state‐like conformation, but where the TMD still remains in a resting state‐like conformation (Fig. 1 B; Prevost et al. 2012). The symmetrical nature of this ECD‐compacted LC2 conformation might be representative of a flipped (or fully primed) state.

Once the channel gate is open, ions flow according to their electrochemical gradient and the selectivity and conductance of the channel. The major determinant of ionic selectivity, namely the selectivity filter, lies at the cytoplasmic end of the pore (Imoto et al. 1988; Leonard et al. 1988), at the M2 −1′ level for most eukaryotic cationic pLGICs and at the M2 −2′ level for anionic ones, the latter featuring an insertion in this region. Indeed, cation‐permeable pLGICs harbour an acidic residue at the −1′ position, providing a favourable electrostatic environment for the coordination of positively charged ions, and mutating this residue, together with a proline insertion at position −2′, converts the cationic α7 nAChR into an anion‐permeable channel (Corringer et al. 1999). Moreover, other important determinants in the vicinity of the −1′/−2′ residue have been identified, such as the protonation of the buried 0′ basic residue (Cymes & Grosman, 2011) or the side‐chain orientation of neighbouring protonable residues (Cymes & Grosman, 2016). A 2.4 Å resolution structure of GLIC in its putative open conformation further revealed the organization of water pentagons, which coordinate permeant ions in the 2′/6′ region (Sauguet et al. 2013). Besides the canonical selectivity filter, two other regions have been shown to participate in the conductance of pLGICs. First, the extracellular vestibule, which prolongs the transmembrane pore in the inner part of the ECD, can provide an electrostatic environment through which permeant ions flow and thus participate in the ionic conductance (Hansen et al. 2008), although this region doesn't appear to affect ionic selectivity (Cymes & Grosman, 2016). Second, charged residues in the intracellular domain, located in the MA segment upstream of the M4 N‐terminus, line a putative lateral exit serving as an intracellular ionic portal (Hassaine et al. 2014). Their charge thus influences the passage of permeant ions, a mechanism that explains the differential conducting properties between 5‐HT_3A and 5‐HT_3B receptors (Kelley et al. 2003).

Desensitization of pLGICs: case study of the muscle‐type nAChR

Besides the agonist‐elicited activation, most pLGICs display another fundamental property: desensitization. Indeed, for most pLGICs, the sustained presence of the neurotransmitter causes the channels to transit from the active agonist‐bound conformation to an agonist‐bound shut‐channel one called the desensitized state, thereby decreasing current flow (Katz & Thesleff, 1957). Desensitization is thought to prevent the over‐activation of receptors in pathological conditions, and can also lead to the reduction of postsynaptic current upon repetitive synaptic neurotransmitter release (Changeux, 1990; Jones & Westbrook, 1996; Papke et al. 2011). Functional studies, mostly performed on muscle‐type nAChRs during the 1980s and the 1990s, highlighted a series of kinetic features of desensitization:

• (1)Desensitization kinetics are multiphasic: stopped‐flow binding experiments, performed on nAChRs from Torpedo marmorata membranes, directly revealed the existence of distinct ‘fast’ and ‘slow’ desensitized states, and that 20% of unliganded nAChRs are in a desensitized high‐affinity state in this preparation (Heidmann & Changeux, 1979, 1980; Boyd & Cohen, 1980). In parallel, electrophysiological recordings showed up to five temporal components of desensitization in the milliseconds to minutes time range (Feltz & Trautmann, 1982; Elenes & Auerbach, 2002; Papke et al. 2011). As a result, desensitization is classically portrayed as a two‐components phenomenon stemming from the existence of distinct ‘fast’ and ‘slow’ desensitized states (Edelstein et al. 1996). Of note, the desensitization kinetics of pLGICs is often highly variable, complicating experimental investigation (see Box 1).
• (2)The macroscopic desensitization rate (i.e. the rate of temporal decline of ensemble macroscopic currents) increases linearly with the open probability of muscle‐type murine nAChRs, suggesting that desensitization mainly proceeds from a fully liganded state (Dilger & Liu, 1992; Franke et al. 1993). Auerbach and Akk took this work further, and showed that desensitization of muscle‐type mouse nAChRs occurs mostly from the diliganded active state (Auerbach & Akk, 1998).
• (3)Concerning the recovery from desensitization following agonist removal, the main pathway involves first agonist dissociation from the desensitized receptor, and then the unbound desensitized receptor undergoing a rate‐limiting isomerization toward the resting state (Katz & Thesleff, 1957; Cachelin & Colquhoun, 1989; Dilger & Liu, 1992; Franke et al. 1993). This scheme is also largely applicable to GABA_ARs, for which functional recovery from desensitization is about fourfold slower than the dissociation of radiolabelled agonist (Chang et al. 2002). Still, this ratio enables some GABA_ARs to recover from desensitization through the liganded open state, which in turn allows the receptors to open after sojourns in desensitized states and prolongs deactivation kinetics after desensitization (Jones & Westbrook, 1995). While initial models suggested a direct transition from an agonist‐unbound desensitized state to an agonist‐unbound resting state, further modelling indicated that a transition through the active state cannot be discarded, even in the absence of measurable openings of the channel (Edelstein et al. 1996). Indeed, the unliganded openings are so brief that most would escape electrophysiological detection.

Box 1. The variability of desensitization

Variability of outside‐out patch‐clamp recordings

One hallmark of the pLGICs’ desensitization lies in its variability, which adds to the difficulty of investigating this process. Indeed, the kinetics of desensitization, as measured with fast theta‐tube applications to outside‐out patches, show high variability for the muscle‐type nAChRs, GABA_ARs and GlyRs (Papke et al. 2011; Papke & Grosman, 2014). This variability of desensitization kinetics is particularly prominent at α1 GlyR patches, some of which are well described by two‐ or four‐component fits, all patches including a fast 5 ms component, and most of them including an intermediate 100 ms component (Papke & Grosman, 2014). Since outside‐out recordings can be viewed as highly dialysed miniature whole‐cell patch‐clamp recordings, it is intriguing that such an observation is mirrored by intra‐cell variability when performing whole‐cell patch‐clamp recordings: the desensitization kinetics and the extent of desensitization increase over time after entering the whole‐cell configuration (unpublished personal observation for α1β2 GABA_ARs; see also Papke et al. 2011 for time‐dependent changes in the desensitization on‐rates in outside‐out patches). Unfortunately, the source for this huge variability is currently unknown (see below), which has considerably hampered research efforts to investigate the mechanisms of pLGICs’ desensitization.

Discrepancy in desensitization kinetics between outside‐out patch‐clamp recordings and two‐electrode voltage‐clamp recordings

When expressing GABA_ARs and GlyRs in Xenopus laevis oocytes, the observed desensitization kinetics measured by two‐electrode voltage clamp (TEVC) are much slower and less variable than the ones measured by outside‐out patch‐clamp recordings when expressing the same receptors in HEK or CHO cells: the fastest component for desensitization of α1β2γ2 GABA_ARs and α1 GlyRs is about 1.6 s and 1 s when measured by TEVC, respectively, and only account for 10–20% of the overall desensitization (Gielen et al. 2015). In contrast, outside‐out patches pulled from HEK cells expressing α1β2γ2 GABA_ARs and α1 GlyRs usually show two fast components in the 3–5 ms and 70–100 ms range, which account together for ∼70–75% of the desensitization amplitude (Papke et al. 2011). Such apparent discrepancy is usually explained by the intrinsic limitation of TEVC recordings of Xenopus laevis oocytes: the currents are rate‐limited by the solution exchange around the oocyte, which usually takes almost a second. Thus, if desensitization occurs on a much faster time scale, it could be entirely missed, and the peak currents recorded by TEVC should onlyreflect the equilibrium between active receptors and receptors in their fast‐desensitized state(s). However, two lines of evidence provide some arguments against this view:

• (1)Using super‐saturating concentrations of agonist, it is possible to elicit TEVC currents in Xenopus laevis oocytes with a 20–80% rise time of about 25 ms at GABA_ARs and GlyRs. In this case, a fast desensitization component could be recorded with a decay time constant of 75 ms at mutant α1^G256V GlyRs (Gielen et al. 2015), and a 70–100 ms component present at wild‐type α1 GlyRs should have been recorded, even if its amplitude would have been lowered compared to systems with a faster perfusion (Karlsson et al. 2011).
• (2)If we assume that most of the peak current is missed in TEVC recordings of Xenopus laevis oocytes because of desensitization, the apparent affinity for the agonist should be higher in TEVC recordings than in patch‐clamp recordings of small cells. The opposite is actually observed: α1β2γ2 GABA_ARs expressed in HEK, which do show prominent desensitization, display an EC₅₀ for GABA in the 5–10 μm range (Mortensen et al. 2004, 2011; Hernandez et al. 2017), while the same receptors expressed in Xenopus laevis oocytes lead to lower apparent affinities (EC₅₀ in the 40–150 μm range depending on the γ2 isoform; Downing et al. 2005; Campo‐Soria et al. 2006; Gielen et al. 2012, 2015). This questions whether whole‐cell patch‐clamp recordings from small cells or outside‐out patch‐clamp recordings could actually change the pharmacological response of pLGICs, potentially by increasing apparent desensitization.

This discrepancy between outside‐out and TEVC recordings is also observed at α7 neuronal nAChRs, which display an unusually fast and complete desensitization (Bouzat et al. 2017). Outside‐out recordings performed with fast applications of ACh have been shown to result in currents decaying with a ∼400 μs desensitization time constant, which actually reflects the mean duration of single‐channel open time as measured in cell‐attached (non‐dialysed) patches, i.e. ∼350 μs, suggesting that desensitization is so fast that it shapes the single‐channel openings (Bouzat et al. 2008). Type 2 positive allosteric modulators (PAM), such as the prototypical PNU‐120596 molecule (Hurst, 2005), can then produce a massive potentiation of single‐channel activity by preventing desensitization (daCosta et al. 2011). In the absence of type 2 PAM, the desensitization kinetics of α7 nAChRs as measured in both outside‐out and cell‐attached patches should be far too fast to enablethe detection of currents in TEVC recordings of Xenopus laevis oocytes. However, such recordings enable the detection of robust α7 nAChR currents elicited by ACh alone, which desensitize with a time constant of ∼100 ms. These currents elicited by ACh alone can amount to up to 10–20% of the current elicited by ACh in the presence of PNU‐120596 (unpublished personal observation; see also Young et al. 2008). Such observations suggest that the desensitization of α7 nAChRs in Xenopus laevis oocytes measured in TEVC recordings is almost three orders of magnitude slower than the one measured in outside‐out or cell‐attached patches from HEK cells.

Basis of desensitization variability: a potential role for lipids?

When performing whole‐cell patch‐clamp recordings of HEK cells, one major effect lies in the dialysis of the intracellular medium – which is further enhanced in the outside‐out patch‐clamp conformation. It is thus possible that the dialysis of an intracellular component could be responsible for a gain of desensitization. In line with this hypothesis, the intracellular concentration of Ca²⁺ ions and the phosphorylation state of the receptors have been held responsible for modifying the desensitization rate of muscle‐type nAChRs (Huganir et al. 1986; Huganir & Greengard, 1990), and phosphorylation of the GABA_AR β1^S409 serine residue by PKA has been shown to decrease the macroscopic desensitization rate of β1‐containing GABA_ARs (Moss et al. 1992). The phosphorylation state cannot, however, account for the extremely variable kinetics and extent of desensitization of GLIC expressed in HEK cells (Laha et al. 2013), since this prokaryotic receptor lacks the M3–M4 intracellular loop and is devoid of phosphorylation sites. Moreover, deleting most of the M3–M4 intracellular loop does not reduce the variability in the apparent rate of desensitization at the α1 GlyR, arguing against a role of phosphorylation in this variability (Papke & Grosman, 2014).

An alternative hypothesis was put forward in 2011, in which the fast macroscopic decline of GABA_AR and GlyR currents in patch‐clamp experiments would not reflect actual desensitization of the receptors. Indeed, Karlsson et al. argued that such decline of current is actually due to the rather slow diffusion of chloride ions at the tip of the glass patch‐clamp electrode, which results in a gradual loss of electrochemical driving force at the plasma membrane (Karlsson et al. 2011). In this hypothesis, the decline of current upon sustained activation is due to a diminished unitary current through the ion channels, and not to a desensitization‐induced decrease in the receptors’ open probability. Still, α1β2γ2 GABA_ARs display a cluster behaviour at the single‐channel level (Mortensen et al.2004), which is usually interpreted as desensitization. The duration of these clusters is in the 100 ms range at saturating GABA concentrations, suggesting that a fast component of desensitization on this time scale is inherent to patch‐clamp recordings of α1β2γ2 GABA_ARs.

A third hypothesis would be that interactions with the glass pipettes might affect the distribution of lipids in the plasma membrane, these lipids playing a major role in the receptors’ desensitization. Negatively charged lipids, in particular, could interact with the negatively charged surface of glass pipettes through the bridging by divalent cations such as calcium ions. In this hypothesis, it would then be relevant to note that the lipid composition of neurons can vary depending on the subcellular component (Calderon et al. 1995), and that different recombinant systems vary in that respect too (Opekarová & Tanner, 2003). Lipid candidates for such a mechanism could include the phospholipid phosphatidylinositol 4,5‐bisphosphate (PIP₂), which is known, amongst other ion channel regulators, to gate inwardly rectifying potassium channels (Hansen et al. 2011), and to be required for the ivermectin‐elicited gating of GIRK channels (Chen et al. 2017); PIP₂ depletion also leads to the desensitization of TRPV1 receptors (Yao & Qin, 2009). Interestingly, PIP₂ is a component of the inner leaflet of the plasma membrane, and thus ideally located to interact with the desensitization gate. If this hypothesis were true, it is important to note that cell‐attached single‐channel recordings would also likely be affected, e.g. that the duration of single‐channel clusters may actually be decreased due to interactions between the glass pipette and the plasma membrane.

Mutations affecting the desensitization properties of pLGICs

Seminal site‐directed mutagenesis work revealed that mutations of the hydrophobic M2 9′ residue, most commonly a leucine, into small polar threonine or serine residues, almost fully ablate desensitization of α7 nAChRs (Revah et al. 1991); analysis of homologous mutations extended this result to the entire pLGIC family (Yakel et al. 1993; Bianchi & Macdonald, 2001). Since then, many mutations have been found to affect macroscopic desensitization. Strikingly, such mutations are scattered throughout the entire sequence of the receptors (see Zhang et al. 2013 for review). At the level of the ECD, for instance, the W55A mutation ablates the desensitization of α7 nAChRs (Gay et al. 2008), while the ECD–TMD interface is a hotspot for mutations with pronounced loss‐of‐desensitization phenotypes, as illustrated by a series of α7 nAChRs mutants (Bouzat et al. 2008; Wang & Lynch, 2011; Zhang et al. 2011).

In recent years, though, various studies have indicated an involvement of the cytoplasmic end of the pore in the desensitization of GlyRs. Analysis of patients with hyperekplexia has led to the investigation of mutations near the selectivity filter (M1–M2 loop) of the GlyR α1, such as the GlyR α1^I244A and α1^P250T, which produce a loss‐of‐function phenotype accompanied by strong increases in the desensitization kinetics (Fig. 2; Lynch et al. 1997; Saul et al. 1999; Breitinger et al. 2004). Further work showed the involvement of intracellular M3–M4 linker mutation or splice variant in desensitization (Papke & Grosman, 2014; Langlhofer & Villmann, 2016).

Recently, we constructed a series of chimeras between the α1β2 heteromeric and ρ1 homomeric GABA_ARs, which show contrasted desensitization properties (Gielen et al. 2015). It revealed that interactions between the M1–M2 linker and the intracellular end of M3 of an adjacent subunit shape macroscopic desensitization. In the course of our mutagenesis work, we then focused on the intracellular end of the M2–M3 interface, and found a series of residues whose mutation drastically increases desensitization, rendering it almost total and increasing its on‐rate kinetics by up to a hundred‐fold (Fig. 2; Gielen et al. 2015). We obtained similar results for GlyRs, indicating a conserved mechanism for anionic pLGICs.

It is noteworthy that, since desensitization mainly occurs downstream of the receptor's activation (Auerbach & Akk, 1998), a mutation specifically affecting the receptor's microscopic activation kinetics could profoundly impact the macroscopic desensitization kinetics. Hence, for most of the above‐mentioned mutations, which also strongly alter the activation of pLGICs, the effect on desensitization remains ambiguous. However, the single‐site mutations at the intracellular end of the M2–M3 interface of GABA_ARs minimally affect their GABA and benzodiazepine dose–response curves (Gielen et al. 2015), supporting that they selectively affect desensitization.

A desensitization gate at the intracellular end of the pore, distinct from the activation gate

Given the location of those M2–M3 interface mutations, we hypothesized that desensitization might involve the closure of the pore at its intracellular end. To test that hypothesis, we used the pore blocker picrotoxin, which binds in this M2 −2′/2′ region (Fig. 3 A; Hibbs & Gouaux, 2011).

Figure 3. Picrotoxin prevents the desensitization of α1 GlyRs.

A, side view of the M2 segments from GluCl (pdb code 3RI5), showing only two distal subunits for clarity. Picrotoxin (PTX) binding site is delimited by the −2′ and 2′ residues, shown in sticks. The 9′ residue position serves as a reference to pinpoint the bottom end of the activation gate. The 4′ and −3′ residues delineate the bottom of the M2–M3 interface, which bears critical determinants of the desensitization of anionic pLGICs (see main text and Fig. 2). B, depiction of a scheme in which desensitization of anionic pLGICs involves a desensitization gate overlapping the PTX binding site. Binding of PTX thus prevents the entry into the desensitized state. Moreover, PTX cannot associate to or dissociate from the resting state of the pore. This is fully consistent with previous studies, which showed that picrotoxin is trapped in the resting state of the receptor (Bali & Akabas, 2007), under the activation gate (Hibbs & Gouaux, 2011; Rossokhin & Zhorov, 2016). C, two‐electrode voltage clamp recording from an oocyte expressing α1 GlyRs. Supersaturating glycine (10 mm) elicits a current that desensitizes over a 2 min‐long application. This current can be blocked by PTX (500 μm). Co‐application of glycine and PTX for 3 min, allowing equilibration between the various allosteric states, yields a pronounced rebound current upon wash‐out of PTX. Data taken from Gielen et al. (2015). D, kinetic model corresponding to the scheme depicted in panel B. The receptor can be found in its agonist‐free resting state (R), and in agonist‐bound resting (AR), open (AO) or desensitized (AD) states. P denotes PTX‐bound states. Except for ρ, all values are taken from Gielen et al. (2015); ρ values greater than 1 reflect the stabilization of the resting state by PTX. It should be noted here that our model does not address many fine aspects of pLGIC gating: (1) Only one agonist binding site was included, whereas pLGICs usually require the binding of two or three agonist molecules for full activation (Lape et al. 2008; Corradi et al. 2009; Rayes et al. 2009; Gielen et al. 2012). No pre‐activation step is included either. Fine details of receptor activation by low concentrations of agonists (or by partial agonists) would not be recapitulated in our model. (2) Channel opening requires the binding of the agonist in our model, while it is well known that unliganded pLGICs can spontaneously open. Such spontaneous gating is, however, extremely rare at wild‐type pLGICs (Purohit & Auerbach, 2009), which is the reason why we decided not to include it. (3) Our scheme for agonist binding, activation and subsequent desensitization is linear, the agonist being thus unable to dissociate from the desensitized state in our model. This is in contradiction with what is known from desensitization recovery, and our model thus could not reflect what happens during desensitization recovery in the absence of the agonist, for example during the wash‐out of a desensitizing application of agonists. (4) Only one desensitized state is included, corresponding to a slow component of desensitization. Accounting for the detailed multiphasic components of desensitization is thus out of reach in our model. E, kinetic model, in which the PTX‐bound receptor can desensitize (ADP state). PTX is trapped in the desensitized state, akin to what happens in the resting state. Such a scheme would reflect the hypothesis that desensitization and activation might involve the same physical gate. All rates are kept identical to the ones from panel D. F, simulations performed with the kinetic model from panel D, for ρ values of 1 (grey trace), 10 (blue trace) and 100 (red trace). Note the simulations with ρ values of 1 and 10 display pronounced rebound currents akin to the one seen in experiments. Note also that further increase in ρ leads to slower apparent recovery of the current upon PTX wash‐out, and to a much reduced rebound current (see main text for discussion). Simulations were performed with QuB (Nicolai & Sachs, 2013). G, simulations performed with the kinetic model from panel E, for ρ values of 1 (grey trace), 10 (blue trace) and 100 (red trace). None of these simulations can recapitulate the pronounced experimental rebound current.

We suspected that picrotoxin could prevent desensitization by a foot‐in‐the‐door mechanism, i.e. by physically hindering the constriction of the channel, and thus investigated a potential inhibition of desensitization by picrotoxin binding (Fig. 3 B). Such a mechanism is clearly apparent for the α1 GlyR: following prolonged co‐application of saturating concentrations of the agonist glycine and picrotoxin, wash‐out of picrotoxin reveals a prominent rebound current (Fig. 3 C). This rebound current is perfectly reproduced in a kinetic scheme, in which picrotoxin binding fully prevents desensitization (Gielen et al. 2015).

These results demonstrate that the activation and desensitization gates are distinct, at least for the slow‐desensitization component that is investigated here. A similar idea was speculated about early on by Auerback and Akk from single‐channel recordings of muscle‐type nAChRs, assessing primarily fast desensitization components (Auerbach & Akk, 1998; Purohit & Grosman, 2006). Altogether, our results are consistent with picrotoxin preventing desensitization by a foot‐in‐the‐door mechanism, the lower part of the channel fulfilling two functions: the selectivity filter and the desensitization gate.

The detailed analysis of picrotoxin block of anionic pLGICs requires the inclusion of an effect on activation

In addition to the rebound current described above, picrotoxin binding results in a loss of apparent affinity for the agonist by approximately one to two orders of magnitude, i.e. a so‐called ‘competitive’ rightward shift in the dose–response curve for the agonist at both GABA_ARs (Smart & Constanti, 1986; Goutman & Calvo, 2004; Qian, 2004) and GlyRs (Lynch et al. 1995; Wang et al. 2006, 2007). At sub‐activating agonist concentrations, picrotoxin also promotes agonist dissociation by stabilizing the resting conformation, as shown by voltage‐clamp fluorometry of GABA_ARs (Chang & Weiss, 2002). It could thus be argued that the rebound current observed upon picrotoxin wash‐out could actually reflect the stabilization of the resting state over the active state, similarly to competitive antagonists (Xu et al. 2016), rather than preventing desensitization.

To investigate these possibilities, we built simplified kinetic models of picrotoxin block of GlyRs containing a minimal number of steps: (1) agonist binding, (2) channel opening, (3) desensitization, and (4) picrotoxin binding. Figure 3 shows illustrative traces assuming that picrotoxin stabilizes the resting channel, i.e. that picrotoxin decreases the microscopic opening rate by a factor ρ. In a first model, picrotoxin binding prevents the desensitization of α1 GlyRs (Fig. 3 D), while it can bind and be trapped in desensitized receptors in the second model (Fig. 3 E). The first model results in non‐distinguishable rebound currents for the values ρ = 1 and ρ = 10. However, the rebound current is strongly decreased in the case in which ρ = 100 (Fig. 3 F). Indeed, in this situation and with our set of kinetic values, picrotoxin largely stabilizes the resting shut state of the channel (i.e. the opening rate β becomes smaller than the shutting rate α), but cannot dissociate from it. Picrotoxin thus presents an ‘auto‐trapping’ phenomenon at high ρ values, which slows down its apparent dissociation rate and limits the amount of rebound current. Of note, a 10‐fold decrease in gating efficacy of human α1 GlyRs upon picrotoxin binding is consistent with previous studies (Wang et al. 2006). On the other hand, no set of parameter with the second model can reproduce a prominent rebound current with ρ values in the 1–100 range (Fig. 3 G), further arguing that picrotoxin indeed prevents the desensitization of inhibitory pLGICs. Moreover, if the channel adopted an identical conformation in the resting and desensitized states, picrotoxin should actually stabilize the desensitized state over the open one in that second model (i.e. picrotoxin binding should decrease d ⁻ in Fig. 3 E). Such mechanism should lead to an increased desensitization in our protocol, resulting in a large slow component of current recovery after picrotoxin wash‐out, unlike what is observed experimentally.

The pore‐block mechanism by picrotoxin, which not only prevents desensitization but also stabilizes the channel in a resting state, probably accounts for the complex interplay between picrotoxinin (i.e. the most active part of picrotoxin, which is an equimolar mixture of picrotoxinin and picrotin) and the allosteric potentiator ivermectin at zebrafish α1 GlyRs. The structure of this receptor has been solved in three different conformations by cryo‐EM: apo, glycine‐bound and glycine‐ plus ivermectin‐bound, the latter proposed to represent either a desensitized or a partially open state (Du et al. 2015). In the presence of ivermectin, the zebrafish α1 GlyR was shown to desensitize on a time scale of several minutes. Interestingly, in the presence of an approximate EC₅₀ concentration of glycine, 1 mm picrotoxinin inhibits the zebrafish α1 GlyR by ∼80%, but inhibition drops to ∼20% in the presence of ivermectin (Du et al. 2015). It is likely that ivermectin simply reduces the apparent affinity for picrotoxinin by decreasing the likelihood for the channel to visit its picrotoxinin high‐affinity state, namely its resting state, without any effect of ivermectin on desensitization. This hypothesis is illustrated in the kinetic model from Fig. 4 A and B, leading to simulations that account well for experimental observations (Fig. 4 C).

Figure 4. Interplay between ivermectin and picrotoxinin at zebrafish α1 GlyR: a plausible kinetic scheme.

A, kinetic scheme proposed for the dual modulation of the zebrafish α1 GlyR, whose cryo‐EM structure was solved in Du et al. (2015). In this model, the binding of the agonist glycine (A) to the receptor (R) promotes its transition to the glycine‐bound active state (AO), which can subsequently desensitize (AD). Ivermectin (Iv) is hypothesized to stabilize the open state over the resting state, without affecting the receptor's microscopic desensitization: Iv increases the opening rate β by a factor γ, and decreases the shutting rate by a factor δ. Picrotoxinin (P) acts accordingly to Fig. 3 D: it prevents desensitization and promotes the resting state of the receptor by decreasing the opening rate of the channel by a factor ε. Note that the active states of receptors bound to glycine alone or to glycine and ivermectin are the sole ion conducting states (other states being either inactive or blocked by picrotoxinin). These two states are boxed in green. Note also that the effects of ivermectin and picrotoxinin are fully additive: when bound to both ivermectin and picrotoxinin, the channel has a microscopic opening rate of β.γ/ε, and a microscopic shutting rate given by α/δ. Such a hypothesis precludes any direct interaction between these two modulators. Finally, this kinetic model does not aim at portraying all features of GlyR functioning (see Fig. 3 D legend). B, parameters used for kinetic simulations based on the scheme from panel A. Note that these parameters were chosen arbitrarily – their exact values have not been determined for the zebrafish α1 GlyR crystallized construct. The only constraints were to satisfy a peak apparent glycine affinity of 0.26 mm, and to provide a macroscopic desensitization profile comparable to the one from Du et al. (2015). C, simulation based on the kinetic model and parameters from panels A and B. Currents elicited by 0.3 mm glycine alone are inhibited up to ∼90% by 1 mm picrotoxinin (PTX), and are potentiated approximately twofold by 5 μm ivermectin. Note that currents elicited by co‐application of glycine and ivermectin are inhibited by picrotoxinin by only ∼20%. D, simulation indicating that, with the kinetic model and parameters from panels A and B, the application of 10 mm glycine (trace in grey) elicits the same peak current and the same desensitization as the co‐application of 0.3 mm glycine and 5 μm ivermectin.

Such a kinetic model, in which ivermectin does not affect the microscopic desensitization step, predicts that ivermectin will not increase desensitization under high glycine concentrations (Fig. 4 D). In that hypothesis, it is unclear why the zebrafish α1 GlyR adopts two different conformations under glycine alone or glycine plus ivermectin conditions (Du et al. 2015). Importantly, recent structural and modelling work suggests that the bona fide open structure of pLGICs most likely resembles the open structures of GLIC and GluCl, rather than the much larger pore conformation of the cryo‐EM structure of the zebrafish α1 GlyR bound to glycine alone (Gonzalez‐Gutierrez et al. 2017). A contrario, recent MD simulations suggest that the wide open zebrafish α1 GlyR structure is conductive, while the ivermectin‐ and glycine‐bound α1 GlyR structure should not conduct ions (Trick et al. 2016). In these simulations, however, it is unclear whether the simulated conduction properties of the wide‐open zebrafish α1 GlyR match the experimentally derived values, nor is it possible to infer how much structural change is required to convert the ivermectin‐ and glycine‐bound α1 GlyR structure into a conductive conformation. Further work will thus be required to assign a functional annotation to the zebrafish α1 GlyR conformations.

Structural data corroborate the desensitization gate model

In agreement with the above‐mentioned functional work, recent X‐ray studies provide a structural counterpart to the analysis of the desensitization gate. Miller and Aricescu published in 2014 the structure of the homomeric β3 GABA_AR in complex with its agonist benzamidine (Miller & Aricescu, 2014). This structure shows a wide open activation gate in the M2 9′/13′ region, but a hydrophobic constriction at the −2′ proline, where the pore radius narrows down to 1.6 Å, thereby precluding the flow of chloride ions whose Pauling radius is 1.8 Å. More recently, several pLGICs have been solved in similar conformations: the human α3 GlyR in complex with both glycine and a positive allosteric modulator, AM‐3607 (Huang et al. 2017), as well as a GLIC–(GABA_A α1) and a (GABA_A β3)–(GABA_A α5) chimera, the former carrying a α1^G258V mutation promoting desensitization (Laverty et al. 2017; Miller et al. 2017). All these structures show a conserved pore conformation and were assigned to a desensitized state, since they correspond to agonist‐bound shut states. In addition, they account for the above functional work that identified the desensitization gate near the binding site for picrotoxin, which comprises the −2′ residue.

Comparing these structures to the putative open GLIC and GluCl suggests that, during the transition from the active to the desensitized state, a symmetrical tilt and rotation of the M2 helices narrows down the cytoplasmic constriction while widening the upper part of the channel (Fig. 5 A). At the level of the 9′ residue, this latter ‘pull and twist’ motion generates a rotation of the side‐chain away from the lumen of the pore to point towards a neighbouring M2 segment (Fig. 5 A). This local motion probably contributes to the marked ‘gain of function’ phenotype observed when the 9′ residue is mutated to a more hydrophilic residue. Indeed, such mutations are expected to stabilize preferentially the active conformation with a 9′ side chain facing the polar aqueous environment, consistent with functional studies (Bianchi & Macdonald, 2001).

Figure 5. Structural rearrangements at the level of the selectivity filter and the activation gate during desensitization.

A, left, side view of the M2 segments from the presumably open states (O) of GLIC (coloured red; pdb code 4HFI) and the C. elegans GluCl (coloured light pink; pdb code 3RI5), superimposed with the presumably desensitized states (D) of the human α3 GlyR (coloured blue; pdb code 5TIN) and β3 GABA_AR (coloured light cyan; pdb code 4COF), showing only two distal subunits for clarity. Note the difference in the backbone conformations between the O and D structures. Middle, top view of the pore at the level of M2 −2′ residues (shown in sticks), which are part of the selectivity filter, for the structures mentioned above. Note the reduction in the pore diameter of D structures compared to O structures. Right, top view of the pore at the level of 9′ residues (shown in sticks), which are part of the activation gate, for the structures mentioned in panel A. Note the increase in the pore diameter of D structures compared to O structures, and the difference in the M2 9′ side‐chain orientations, which point towards the neighbouring M2 segments for D structures. B, left, side view of the M2 segments from the presumably open state (O) of GLIC and the presumably desensitized state (D) of the human α3 GlyR, superimposed with the structure of the α4β2 nAChR (coloured green; pdb code 5KXI). Middle, top view of the pore at the level of the selectivity filter. Note that it might be difficult to assign a functional status to the α4β2 nAChR structure on this basis, especially when taking into account potential rotamers of the nAChR −1′ glutamate (see main text). Right, top view of the pore at the level of 9′ residues (shown in sticks). Note that the conformation of the α4β2 nAChR structure superimposes well to the α3 GlyR in a putative desensitized state.

Does this mechanism of desensitization also pertain to cationic pLGICs such as nAChRs? The α4β2 nAChR structure may illustrate this proposal (Morales‐Perez et al. 2016): the M2 α‐helices are well superimposable when comparing the α4β2 nAChR structure to the structures of anionic pLGICs in putative desensitized conformations (Fig. 5 B), including the orientation of the M2 9′ residue. This is consistent with the α4β2 nAChR structure being representative of a desensitized state. The M2 −1′ nAChR glutamate residues, the major contributor to the selectivity filter, form the most constricted part of the channel, leading to a pore diameter akin to the ones observed for putative desensitized structures of anionic pLGICs (Miller et al. 2017). However, the relationship between pore diameter and conduction property is complicated by two issues in this case: first, the constriction at −1′ is hydrophilic and even negatively charged, and it is unclear by which mechanism such a pore, which would provide a microenvironment favourable for cations, would impair their permeation. Second, the M2 −1′ glutamates are fixed in a symmetrical conformation in the crystal structure, whereas the residue side‐chains are dynamic and certainly adopt a variety of rotameric conformations in solution, potentially widening the effective pore diameter (Rossokhin & Zhorov, 2016). This point is all the more important since the rotameric conformations of the M2 −1′ glutamates have been proposed to control the conductance of nAChRs (Harpole & Grosman, 2014). Further work is thus required to fully understand the exact determinants for nAChR desensitization.

Thus, similar desensitization mechanisms might indeed be conserved amongst both anionic and cationic pLGICs, with the existence of distinct activation and desensitization gates. Consistent with this idea, diverse pore‐blockers have been shown to differentially stabilize the resting, active and desensitized states of both anionic pLGICs and muscle‐type nAChRs (see Box 2). The desensitization gate model may thus be extended to the entire pLGIC family, where the full gating cycle would include (1) a pre‐activation step involving the ECD ‘unblooming’; (2) channel activation gate opening, in the upper half of the pore, concomitant with the ECD–TMD interface rearrangement and the whole receptor ‘twisting’; and (3) channel desensitization resulting from the constriction of the desensitization gate at the intracellular end of the pore (Fig. 6).

Figure 6. Proposed model for the gating and desensitization of pLGICs: a dual gate mechanism.

Schematic depiction of pLGIC activation and desensitization. It expands on the iconography from Fig. 1 B to include a desensitization step, in which the intracellular end of the pore constricts during desensitization, thus forming a physical desensitization gate distinct from the activation gate.

Box 2. Pore‐blockers as allosteric modulators differentially stabilizing distinct states of the channel

Pore‐blockers have proven useful pharmacological tools to discriminate between the various allosteric states of the channel. In cases where the binding site for a given pore‐blocker displays a state‐dependent conformation, the allosteric states of the channel will impact the affinity for the pore‐blocker. Examples are plentiful in the pLGIC literature, such as picrotoxin, which prevents the desensitization of anionic pLGICs (see main text). This conclusion is actually reminiscent of experiments performed with another pore blocker, t‐butylbicyclophosphorothionate (TBPS): the binding of radiolabelled TBPS to GABA_ARs is decreased under desensitizing conditions (Othman et al. 2012).

Pore‐blockers have been used historically to study allosteric transitions at muscle‐type nAChRs; tetracaine preferentially binds to agonist‐unbound resting compared to agonist‐bound desensitized states (Middleton et al. 1999), which shows that these two states are distinct. On the other hand, the pore‐blocker chlorpromazine binds with high association on‐rates to the open state of nAChRs, but rapid chlorpromazine binding is prevented under desensitizing conditions (Heidmann & Changeux, 1984, 1986). This result is all the more interesting since chlorpromazine binding occurs quite deep in the pore, with an involvement of M2 2′ and 6′ residues (Giraudat et al. 1986; Revah et al. 1990; Chiara et al. 2009). It is thus tempting to speculate that chlorpromazine acts on nAChRs in a similar manner as picrotoxin acts at anionic pLGICs, with the binding of this pore‐blocker competing with the closure of a desensitization gate.

More recently, the pore‐blocking properties of choline at muscle‐type nAChRs gave credence to the two‐gate model proposed by Auerbach & Akk: choline induces longer single‐channel openings, while leaving the desensitization properties unaffected. This led to the conclusion that choline‐binding in the pore interferes with an activation gate, while leaving the desensitization gate unaffected, and thus that these two molecular entities are distinct (Purohit & Grosman, 2006). Nevertheless, another recent study suggests on the contrary that the binding of choline has only minimal effect on the closing rate – i.e. on the duration of single‐channel openings (Lape et al. 2009). Further work is required to fully understand the interactions between choline and the pore of nAChRs in their different functional states.

This use of pore‐blockers is not restricted to the study of pLGICs: tetraethylammonium binds at the level of the selectivity filter of potassium channelsand prevents slow inactivation by a foot‐in‐the‐door mechanism (Choi et al. 1991; Kurata & Fedida, 2006), while ketamine and memantine differentially affect the desensitization of NMDA receptors (Glasgow et al. 2017).

Pore‐blockers thus don't only act as mere plugs, and their state‐dependent affinity might even be used to design clinically relevant drugs, e.g. to target preferentially extrasynaptic receptors over synaptic ones, as proposed for the action of memantine at NMDA receptors (Xia et al. 2010).

Reconciling the docosahexaenoic acid‐bound structure of GLIC with a dual gate model

Recently, an X‐ray structure of the bacterial pLGIC GLIC in complex with the fatty acid docosahexaenoic acid (DHA) revealed a new pore conformation, with a profile intermediate between that of the putative resting and open states of the channel (Basak et al. 2017): the activation gate has transitioned towards the open channel conformation, with the exception of the M2 9′ residue that still shapes a hydrophobic constriction in the middle of the pore.

To propose a functional annotation of this apparently shut state, Basak et al. performed a series of electrophysiological experiments showing that (1) the co‐application of DHA with an activating acidic solution produces no alteration of the fast time response to protons, but a slow and progressive inhibition as if DHA favoured a desensitization process, (2) DHA produces a slight decrease in the EC₅₀ of activation by protons, and (3) DHA fails at inhibiting the proton‐elicited currents of the I9′A mutant of GLIC.

Based on these data, it was proposed that DHA stabilizes the desensitized state of GLIC, which would account for the decrease in the proton EC₅₀ and the I9′A mutant phenotype, since this mutation is believed to prevent desensitization. However, the X‐ray structure of the GLIC–DHA complex looks more like an intermediate state in the activation pathway than a desensitized state as discussed above. We thus investigate an alternative possibility herein that DHA may stabilize an intermediate pre‐active state.

Figure 7 shows a purely theoretical kinetic scheme of a receptor, in which the resting state binds an agonist, then enters a pre‐active state and can subsequently activate. In this scheme, binding of an allosteric inhibitor that selectively stabilizes the pre‐active state (Fig. 7 A and B) produces both an increase in the pre‐activation constant and a decrease in the activation constant (Fig. 7 C). It will drive the receptor away from the active state to lower the open probability (Fig. 7 D), but also promotes the overall population of agonist‐bound state to increase the apparent affinity for the agonist (i.e. it decreases the agonist EC₅₀) (Fig. 7 D). Therefore, an allosteric inhibitor increasing the apparent affinity for the agonist can do so by promoting either a pre‐active or a desensitized state.

Figure 7. A theoretical model: an allosteric inhibitor selectively stabilizing an agonist‐bound pre‐active state would increase the apparent affinity for the agonist.

A, putative free energy diagram highlighting the effect of an allosteric inhibitor I, which would selectively stabilize the pre‐active state of a receptor (F) over its resting (R) or active (O) states in the presence of the agonist (A). Note that the free energy of the AF state is decreased by δG in the presence of I. B, translation of the free energy diagram from panel A into a linear kinetic model for the activation of the receptor by the agonist and its inhibition by I. In the presence of I, note the dependence of the pre‐activation and activation microscopic rates on the value of δG and those of δG ^f and δG ^o, the two latter reflecting the effect of the inhibitor's binding on the transition states during the AR → AF and the AO → AF transitions, respectively. C, equilibrium constants for the AR ↔ AF and AF ↔ AO equilibriums are noted as F and E, respectively. Note that the value of those constants, in the presence of the inhibitor I, depends solely on two parameters: the value of the constants in the absence of I, and the value of δG. In other words, the values of the pre‐activation and activation equilibrium constants do not depend on δG ^f or δG ^o. D, expression of the maximal peak open probability (P _o,max) and of the apparent EC₅₀ for the agonist (EC_50,A) in control condition and in the presence of a saturating concentration of the inhibitor I (see Gielen et al. 2012 for further details). The binding of I causes both a decrease in the open probability, meaning that I is indeed an inhibitor, and a decrease in the apparent agonist EC₅₀, or in other words, I increases the apparent affinity for the agonist.

As an illustration, Fig. 8 provides a kinetic model of GLIC (Fig. 8 A and B), which expands the one presented in Fig. 7 by adding a desensitization step from the active state. This kinetic model reproduces all aspects of GLIC activation by protons and its inhibition by DHA. First, DHA co‐application produces an increase in the rate and the extent of current loss upon prolonged proton applications (Fig. 8 C). Second, the concentration–response curve for DHA yields an IC₅₀ value broadly consistent with the inhibition measured experimentally (Fig. 8 D). Finally, and most importantly, DHA increases the apparent affinity of GLIC for protons when measured either at peak or steady state currents (Fig. 8 E and F).

Figure 8. The selective stabilization of a pre‐active state can recapitulate the effects of DHA on wild‐type GLIC.

A, proposed kinetic model of wild‐type GLIC, expanding on Fig. 7 B and including a desensitization step to reach the agonist‐bound desensitized state AD. For the sake of simplicity, the inhibitor DHA (I) is presumed to stabilize selectively the agonist‐bound pre‐active state AF by decreasing the AF → AR (f ⁻) and AF → AO (β) microscopic rates, although similar results would be obtained with modified f ⁺ and α rates, as long as the equilibrium constant f ⁺/f ⁻ and β/α are conserved. Note that the model probably oversimplifies many aspects of the gating of GLIC: for example, the model only contains one single binding site for protons and DHA, whereas the homomeric GLIC probably contains at least five proton binding sites, and harbours five DHA sites. Unliganded openings and proton dissociation from the desensitized state are not portrayed here, although they probably occur as for other pLGICs. B, parameters for kinetic simulations of GLIC modulation by DHA. Note that the exact values for all the microscopic steps are unknown, and were chosen arbitrarily. C, simulation of GLIC currents elicited by pH 4.5 applications, highlighting the inhibitory effect of DHA co‐application, which increases the rate and the extent of current loss upon prolonged proton applications. This could be misinterpreted in terms of an increased rate of desensitization, but only reflects the slow on‐rate of DHA association to GLIC in the kinetic model. D, theoretical concentration–response curve for DHA inhibition in simulated currents. The effect of DHA co‐application is assessed by its effect on the ratio between the steady‐state and the peak current elicited by a pH 4.5 application. With such measurement, simulations yield an apparent DHA IC₅₀ of 9.7 μm. E, simulation of GLIC currents elicited by increasing proton concentrations (pH 7.0 to pH 3.5), either in control conditions (black trace), or in the continued presence of 50 μm DHA (red trace). F, left, normalized concentration–response curve for the proton‐elicited peak currents in control condition (continuous black line, filled black circles; pH₅₀ = 5.0) or in the continued presence of 50 μm DHA (continuous red line, filled red triangles; pH₅₀ = 5.2). Right, normalized concentration–response curve for the proton‐elicited steady‐state currents in control condition (dashed black line, open black circles; pH₅₀ = 5.2) or in the continued presence of 50 μm DHA (dashed red line, open red triangles; pH₅₀ = 5.6). Note that DHA increases the apparent affinity for protons in these simulations, 1.6‐fold and 4‐fold in the cases of peak and steady‐state responses, respectively.

Another argument for DHA stabilizing a desensitized state is the inability of DHA to inhibit GLIC 9′ mutants. However, M2 9′ mutations not only ablate desensitization, they actually stabilize the open state of the pore over both resting and desensitized shut states (Bianchi & Macdonald, 2001). Figure 9 depicts a kinetic model of GLIC functioning, whereby the M2 9′ mutation selectively stabilizes the open state over all the other states (Fig. 9 A and B). Two schemes are then considered: in scheme I, the inhibitor DHA selectively stabilizes the desensitized state; whereas in scheme II, the inhibitor DHA selectively stabilizes the pre‐active state (Fig. 9 C). In both schemes, the effects of DHA and the M2 9′ mutation are considered additive. Using the set of parameters from Fig. 8, the 9′ mutation is predicted to fully prevent DHA inhibition in both schemes (Fig. 9 D).

Figure 9. The selective stabilization of a pre‐active state can recapitulate the effects of DHA on GLIC 9′ mutants.

A, free energy diagram highlighting the putative effect of the M2 9′ I to A mutation on GLIC. In this model, the M2 9′ mutation selectively stabilizes the open state (AO) over the resting (AR), pre‐active (AF) and desensitized (AD) shut states (continuous black line, compared to the wild‐type in dashed grey line). Two schemes are then considered to account for the DHA inhibition: in scheme I (blue line), the inhibitor DHA selectively stabilizes the desensitized state over all other states and decreases its free energy by δG _d, whereas in scheme II (red line), the inhibitor DHA selectively stabilizes the pre‐active state and decreases its free energy by δG _f. In both schemes, the effects of DHA and the M2 9′ mutation are considered additive. B, left, translation of the free energy diagram from panel A into a linear kinetic model for the activation of GLIC by the agonist proton (A) and its inhibition by DHA. Right, translation of the free energy effects of the M2 9′ I to A mutation into kinetics effects, highlighting the increased efficacy of gating and the decreased equilibrium constant for desensitization. C, left, in scheme I, DHA binding displaces the AO ↔ AD equilibrium towards the AD state. In other words, the affinity of GLIC for DHA is increased in the desensitized state. Right, in scheme II, DHA affects equally the AR ↔ AF and the AF ↔ AO equilibriums, displacing them towards the AF state. In other words, the affinity of GLIC for DHA is increased in the pre‐active state. D, simulated GLIC currents elicited by pH 4.5 applications, highlighting the inhibitory effect of DHA co‐application on wild‐type GLIC (black trace) in both scheme I (left) and scheme II (right). Assuming that the M2 9′ mutation selectively stabilizes the open state, the inhibitory effect of DHA is predicted to be lost for the 9′ GLIC mutant, both in scheme I (blue trace, left) and in scheme II (red trace, right). Parameters for the simulation are identical to Fig. 8 B.

As a conclusion, the whole set of experiments are equally accounted for assuming a stabilization of either the desensitized (scheme I) or the pre‐active state (scheme II) by DHA. However, the two schemes make quite different predictions regarding the short‐term effect of DHA on the proton‐elicited response, since in scheme II DHA should strongly affect the peak response, while in scheme I DHA should only affect the downstream process of desensitization. As illustrated with the kinetic models shown in Figs 8 and 9, if we equilibrate GLIC with DHA at neutral pH, and then apply acidic pH in the continued presence of DHA, we observe no inhibition (versus robust inhibition) of the peak proton‐elicited current in scheme I (versus scheme II) (Fig. 10 A and B). Incidentally, such an experiment has been performed, and DHA pre‐application is indeed shown to elicit robust inhibition of the GLIC peak response elicited by protons (Basak et al. 2017), consistent with DHA affecting a pre‐activation step. Moreover, with simple linear schemes as the ones presented here, and if we assume that desensitization kinetics remain slower than activation kinetics, the agonist concentration–response curve for peak responses should not be affected by a drug modulating desensitization. The DHA‐induced increase in apparent affinity for protons, as measured with peak responses (Basak et al. 2017), is thus another argument in favour of DHA modulating the pre‐activation step (Fig. 8 F).

Figure 10. The inhibition of GLIC peak currents by pre‐applications of DHA favours an effect of DHA on the pre‐activation, rather than the desensitization, of GLIC.

A, wild‐type GLIC currents simulated according to scheme I of Fig. 9, i.e. with DHA promoting desensitization. Note that DHA pre‐application fails to inhibit the peak current elicited by pH 4.5. B, wild‐type GLIC currents simulated according to scheme II of Fig. 9, i.e. with DHA selectively stabilizing the pre‐active state AF. Note that pre‐application of 10 μm and 50 μm DHA inhibit the peak current elicited by pH 4.5 by 25% and 55%, respectively. All these simulations are performed with the same parameters as in Fig. 8 B.

Of course, it is still possible that more complex kinetic models, in which direct resting‐to‐desensitized transitions are possible in the absence of agonist, for example, could account for these experimental results. However, the presently developed simple kinetic models argue that the DHA‐bound structure of GLIC actually represents that of a pre‐active state. Interestingly, electron paramagnetic resonance measurements suggest that the M4 segments undergo an outward movement during desensitization, and double electron–electron resonance experiments indicate that the distance between M4 segments increases during desensitization (Basak et al. 2017). These results do not agree with the DHA‐bound structure representing a desensitized state, since the M4 segments are superimposable in the DHA‐bound and in the putative open and resting states of GLIC (Basak et al. 2017). Thus, the DHA‐bound structure may well represent an intermediate pre‐active shut state, and this interesting one‐of‐a‐kind structure could help in delineating the activation transition pathway of pLGICs. However, a definitive answer to the pre‐activation versus desensitization hypotheses of DHA modulation might only be provided by single‐channel recordings: an effect on desensitization should decrease the mean cluster duration, while an effect on pre‐activation should decrease the intra‐cluster open probability.

Global allosteric reorganization associated with desensitization

As expected for an allosteric process, the constriction of the desensitization gate occurs in concert with a global reorganization of the protein structure. So far, several local motions have been inferred from a variety of experimental approaches: (1) at the bottom of the TMD, the marked phenotypes of mutations at the interfaces between the M2 and M3 helices suggest that important reorganizations concern the ring of helices adjacent to M2; (2) at the opposite end of the pore, X‐ray crystal structures of pLGICs in a putative desensitized state also suggest that the upper part of the pore widens during desensitization. Fully consistent with that idea, nuclear magnetic resonance measurements made on the prokaryotic ELIC suggest both that the intracellular end of the pore constricts, and that its extracellular end expands during desensitization (Kinde et al. 2015).

Such movement of the extracellular end of the TMD is further expected to involve ECD–TMD interface rearrangements during desensitization. Several lines of evidence support this idea. First, marked changes in desensitization kinetics are observed upon mutation at this level (Bouzat et al. 2008; Zhang et al. 2011; reviewed in Zhang et al. 2013). Second, photoaffinity labelling experiments performed on Torpedo nAChRs revealed a differential labelling at the ECD–TMD interface of the δ subunit in the resting, active and desensitized states (Yamodo et al. 2010). Third, voltage‐clamp fluorometry experiments on the α1 GlyRs show that variations in the fluorescence signal parallel the time course of desensitization onset when the fluorescent reporter is introduced at specific positions of the ECD–TMD interface. In contrast, when introduced higher up in the ECD, fluorescence does not report the active to desensitized transition (Wang & Lynch, 2011), suggesting that the ECD remains in a similar conformation in the active and desensitized states. Such a mechanism implies that the conformation of the orthosteric site, and thus the intrinsic affinity for the agonist, would be similar in the active and desensitized states, as previously proposed in various kinetic schemes (Edelstein et al. 1996; Auerbach & Akk, 1998). It should be noted here that the ECD conformation of the α4β2 nAChR in a putative desensitized state differs significantly from the ECD conformation of the β3 GABA_AR, showing a distinct degree of ECD twist. This led Hibbs and colleagues to propose that the β3 GABA_AR structure reflects a partially desensitized state, while the α4β2 nAChR structure would correspond to a fully desensitized state (Morales‐Perez et al. 2016). However, it remains possible that such differential ECD conformations reflect a fundamental difference between nAChRs and anionic pLGICs in their ECD unbloomed states in general. Interpreting distinct structural conformations as distinct functional states would require the comparison of such conformations obtained on the same receptor, and future work is required to understand whether distinct desensitized states of a given pLGIC might differ in their ECD conformation.

Towards a full transition pathway of desensitization?

All of the above‐mentioned symmetrical structures are probably relevant to the most stable slow desensitized states. However, the multiple temporal components of desensitization underlie the occurrence of a cascade of conformational changes. One extreme case could stem from concerted rearrangements of all subunits into distinct symmetrical desensitized conformations. At the other end of the spectrum, one could speculate that each subunit rearranges independently during desensitization. In this hypothesis, the various components of desensitization time constants could reflect the entry into various asymmetrical states, each subunit displaying a potentially distinct set of microscopic desensitization rate constants (Prince & Sine, 1999; Yamodo et al. 2010; Kinde et al. 2015). Such a scheme could account for multiphasic desensitization decay of heteromeric receptors, and even for homomeric pLGICs (see Fig. 11). One key parameter in this model is actually the number of subunits required to be in their desensitized state in order to prevent the ionic flow. Investigation of such a speculative model would necessitate further work, although some evidence suggests that functional desensitization requires a conformational change at either one or two subunits in α7 nAChRs (daCosta & Sine, 2013).

Figure 11. Independent transition of identical subunits into a unique desensitized state can result in multiphasic desensitization profiles in homomeric ligand‐gated ion channels.

A, in this illustrative model of a trimeric homomeric receptor (R) gated by an agonist (A), the agonist‐bound receptor can activate (AO state), and each subunit can subsequently undergo desensitization. AD_ij denotes a state where subunits i and j are in their desensitized state. One major assumption of this model is that all subunits behave independently, i.e. the microscopic rates for the AO ↔ AD_i and AD_i ↔ AD_ij equilibriums are the same regardless of the identity of subunits i and j, the desensitization on‐ and off‐rates being noted d ⁺ and d ⁻, respectively. In other words, the desensitization state of one subunit does not influence the desensitization kinetics of another subunit. Finally, the pore is considered as non‐conducting, i.e. desensitized from a functional point of view, as soon as one single subunit has entered its desensitized state. As a result, only the AO state is conducting. B, arbitrary parameters chosen for the kinetic model from panel A. C, left, simulation of currents based on the model and parameters from panel A and B, elicited by a saturating 10 mm application of agonist. Middle and right, a one‐component fit (dashed blue line) does not faithfully reproduce the simulated current (continuous grey trace), unlike a two‐component fit (dashed red line). The slow component reflects the entry of receptors in states where several subunits have desensitized.

Pharmacological modulation of desensitization

Desensitization provides an intrinsic second‐order regulatory mechanism of the activity of pLGICs, potentially endowing them with additional possibilities of neuromodulation in physiological conditions (Heidmann & Changeux, 1982). In particular, desensitization appears well suited to affect pLGIC signalling during volume transmission, which involves low tonic concentrations of neurotransmitters (Vizi et al. 2010; Trueta & De‐Miguel, 2012). For instance, a significant proportion of α7 nAChRs are found extrasynaptically, at a remote distance from the locus of ACh release (Brumwell et al. 2002; Jones & Wonnacott, 2004), in both neuronal and non‐neuronal microglial cells and astrocytes (Shytle et al. 2004; Duffy et al. 2011). However, α7 nAChRs display the fastest desensitization among pLGICs, since they desensitize fully within ∼1–100 ms (see Box 1 for the discussion of variability, and Bouzat et al. 2018 for review). Most agonist‐bound extrasynaptic α7 nAChR should thus be massively desensitized during volume transmission. Interestingly, a metabotropic role of α7 nAChRs has been proposed (King & Kabbani, 2016); an attractive hypothesis would be that such signalling could occur from the desensitized state, in the absence of any ionic flow.

From a pharmacological point of view, the α7 nAChRs are particularly interesting, since they are the target of a series of allosteric effectors that bind at the TMD and display a large spectrum of pharmacological activities (Bouzat et al. 2018): (1) type I PAMs potentiate the peak response to ACh while minimally affecting desensitization; (2) type II PAMs cause a massive potentiation of α7 nAChR currents while preventing their desensitization; (3) allosteric activators produce a non‐desensitizing response in the absence of orthosteric agonists; (4) negative allosteric modulators (NAM) inhibit the response, although we don't currently know if they favour a desensitized state or another shut state; and (5) silent allosteric modulators (SAMs) don't affect the ACh‐elicited current, but can competitively displace the previously mentioned PAMs or NAMs. Given that subtle atomic differences can convert a PAM into a NAM or a SAM (Gill‐Thind et al. 2015), it is likely that these molecules bind into the same site. Site‐directed mutagenesis data support that this site is located in the TMD (Young et al. 2008; daCosta et al. 2011), and recent 3D models support the idea that α7 allosteric modulators bind in the vicinity of the desensitization gate (Newcombe et al. 2018). However, the precise location of the binding site of α7 allosteric modulators within the 3D structure remains to be established. Such detailed understanding could be of great therapeutic interest: animal studies have highlighted the potential of type II PAMs of α7 nAChRs in schizophrenia (Hurst, 2005), ischaemia (Kalappa et al. 2013) and cognitive enhancement (Callahan et al. 2013). Alzheimer's disease being characterized by both cognitive decline and neuroinflammation, the combination of neuroprotective and procognitive properties makes these molecules potential candidates in the treatment of this debilitating neurodegenerative disease (Bouzat et al. 2018).

Besides nAChRs, modulation of desensitization by endogenous compounds might also affect the signalling properties of other pLGICs, such as the GABA_ARs. Indeed, these key players of the excitation–inhibition balance in the brain of vertebrates are the target of neurosteroids, which act as endogenous allosteric modulators. The two previously mentioned GLIC–(GABA_A α1) and (GABA_A β3)–(GABA_A α5) chimeras were actually constructed in order to provide a structural platform for the analysis of the neurosteroid modulation. The binding site of potentiating neurosteroids is located at the bottom end of the TMD, at the intracellular end of the groove between the M3 and the M1 segments of adjacent subunits (Laverty et al. 2017; Miller et al. 2017), while inhibitory neurosteroids likely bind in an intra‐subunit site, at the intracellular end of the M3–M4 interface (Laverty et al. 2017). Given that these sites are in the vicinity of the desensitization gate, one could expect that neurosteroids impact desensitization. Interestingly, previous work suggests that the potentiating neurosteroid tetrahydro‐deoxycorticosterone (THDOC) slows down the recovery from desensitization of native GABA_ARs from cerebellar granule cells, while leaving the desensitization on‐rate kinetics unaffected (Zhu & Vicini, 1997). However, the main effect of potentiating neurosteroids is to enhance the gating efficacy of GABA_ARs, leading to an increase of single channel activity under low GABA concentrations (Twyman & Macdonald, 1992) and to the increased macroscopic efficacy of partial agonists (Bianchi & Macdonald, 2003). These results might indicate that potentiating neurosteroids stabilize both the open and desensitized states over the resting state, leaving the microscopic desensitization rates unaffected. On the other hand, the inhibitory neurosteroid pregnenolone sulfate (PS) has been suggested to increase the desensitization rate of GABA_ARs (Shen et al. 2000). However, PS binding to GABA_ARs is state‐dependent (Eisenman et al. 2003), complicating the macroscopic analysis of PS inhibition during co‐application of GABA_AR agonists and PS. Still, single channel recordings show that PS shortens the duration of single‐channel clusters of α1β2γ2 GABA_ARs while minimally affecting the intra‐cluster open probability (Akk et al. 2001). This would be fully consistent with PS exerting its inhibitory effects by enhancing GABA_AR desensitization.

On a more general note, drugs affecting the desensitization, or desensitization‐modifying allosteric modulators (DAMs) of ligand‐gated ion channels might hold significant therapeutic promise. Indeed, like all allosteric modulators, they would respect the ‘biological rhythms’ of the receptors by only fine‐tuning their response to specific physiological patterns of endogenous agonist release. Moreover, allosteric binding sites are much less conserved than the critical orthosteric sites: consequently, designing subtype‐selective allosteric drugs is much easier than designing subtype‐selective orthosteric ligands. For these two reasons, allosteric modulators should yield significantly broader therapeutic windows than orthosteric ligands or even allosteric activators, which can activate the receptors in the absence of orthosteric agonists. One particular well‐known illustration is the benzodiazepine class of drugs: these compounds potentiate GABA_ARs by increasing their apparent affinity for GABA (Gielen et al. 2012), and they have replaced barbiturates in most prescriptions due to the toxicity of this latter class of molecules, which can directly activate GABA_ARs (Mathers & Barker, 1980; Parker et al. 1986; Rho et al. 1996; López‐Muñoz et al. 2005). Compared to other allosteric modulators, those that specifically affect the desensitization of pLGICs might even display milder functional effects and thus safer therapeutic windows: desensitization occurring downstream of activation, DAMs should minimally affect the peak agonist‐concentration curve, and should only modulate the activity level in the sustained presence of the agonist. Such an effect could be particularly beneficial for the treatment of pathologies affecting high frequency release or abnormal extracellular tonic levels of neurotransmitters.

Comparison with other ion channels: high prevalence of a two‐gate mechanism

The desensitization gate model for pLGICs (Fig. 12 A) is reminiscent of the C‐type slow inactivation of voltage‐gated K⁺ and Na⁺ channels. Indeed, this slow inactivation is thought to involve the collapse of the pore P‐loop, which shapes the selectivity filter (Fig. 12 B; Cuello et al. 2010; Payandeh et al. 2012; Zhang et al. 2012; Li et al. 2017; Pau et al. 2017), at a remote distance from the activation gate facing the intracellular end of the channel. Such a model has been proposed initially for the prototypical voltage‐gated Shaker potassium channel and the prokaryotic KcsA channel, although it has also been shown to be responsible for the run‐down of distantly related TRPM2 channels (Toth & Csanady, 2012). Still, some caution needs to be exerted before generalizing: the analysis of pore‐blocker kinetics suggests that the prokaryotic MthK potassium channel only has one gate located at the selectivity filter level, the canonical activation gate being constitutively in its open conformation (Posson et al. 2013). Early observations that led to the C‐type inactivation model included the location of inactivation‐enhancing mutations surrounding the P‐loop (López‐Barneo et al. 1993; Kurata & Fedida, 2006), as well as the effect of pore‐blockers, which bind at the level of the selectivity filter and prevent desensitization by a foot‐in‐the‐door mechanism (Choi et al. 1991; Kurata & Fedida, 2006). This provides an interesting parallel to the desensitization‐enhancing effects of GABA_AR and GlyR mutations at the intracellular end of the M2–M3 interface, as well as the effects of picrotoxin, which is thought to prevent the collapse of the desensitization gate of anionic pLGICs (Gielen et al. 2015). Such similarities might seem surprising, given that pLGICs and voltage‐gated channels adopt totally unrelated structural organizations: the latter are tetramers, whose core pore‐forming domain comprises two transmembrane helices surrounding the re‐entrant P‐loop domain. It thus appears that structurally unrelated ion channels have converged towards a mechanism in which, after the activation through the opening of an activation gate, a topologically distinct desensitization/inactivation gate has been selected to limit ion flow under the sustained presence of the activating stimulus (Fig. 12 A and B). Such functional convergence might be extended to other classes of ion channels, including the trimeric ATP‐gated P2X channels. Indeed, the X‐ray crystal structure of the human P2X₃ receptors has been solved recently in three different conformations, presumably reflecting the resting, active and desensitized states of the receptor (Fig. 12 C; Mansoor et al. 2016). P2X₃ activation involves the stretching of the top half of the pore‐lining α helices: the transition into a 3₁₀ helical pitch produces a kink, which results in the opening of the channel and which is stabilized by an intracellular cap domain. Upon cap unfolding, desensitization would involve the recoiling of the pore‐lining helices, albeit in a different conformation compared to the resting state (Fig. 12 C; Mansoor et al. 2016). Of note, the extracellular ATP‐binding domain of P2X₃ receptors remains in the same conformation in the active and desensitized states.

Figure 12. Desensitization and inactivation mechanisms across structurally unrelated families of ligand‐ and voltage‐gated ion channels.

See main text for full discussion. A, schematic depiction of pLGIC activation and desensitization. For the sake of clarity, only two simplified subunits are shown, omitting the M1, M3 and M4 segments of each subunit to retain only the M2 pore‐lining segments in the transmembrane domain (TMD). Agonist binding occurs at the interface between two adjacent extracellular domains (ECD) and activation involves the opening of the pore in its upper half, while desensitization corresponds to the constriction of the desensitization gate at the level of the selectivity filter, at the intracellular end of the pore. Note the widening of the upper part of the pore during desensitization, which is probably accompanied by a rearrangement of the ECD–TMD interface. B, schematic depiction of the slow C‐type inactivation of tetrameric voltage‐gated potassium and sodium channels. For clarity, only two subunits (or repeat domains in the case of eukaryotic sodium channels) are shown, and voltage‐sensing domains are omitted. Activation is thought to open a gate at the intracellular end of the pore, while C‐type inactivation presumably involves the collapse of the P‐loop, which also forms the selectivity filter. Similar mechanisms have been suggested for the run‐down of the structurally related ATP‐ and calcium‐gated TRPM2 channel. C, schematic depiction of human P2X₃ receptor activation and desensitization. Similarly to pLGICs, these trimeric ATP‐gated receptors bind their agonist at the interface between the ECDs of adjacent subunits. Activation of human P2X₃ receptors involves the stretching of the pore‐lining TM2 helix, owing to a change in its helical pitch. This active state is stabilized by a cap domain. Unfolding of this cap domain presumably enables the recoiling of the TM2 segment into an α helix, thereby resulting in a desensitized state structurally distinct from the resting state at the pore level. D, schematic depiction of ionotropic glutamate receptor activation and desensitization. Here again, only two subunits are shown for clarity, out of four. The agonist binds to the interlobe cleft of the agonist binding domain (ABD) of individual subunits, which results in the closure of this clamshell‐like domain. Since the ABDs form dimers through their upper lobes, this closure results in an increased distance between the lower lobes, which are directly connected to the TMD. This movement results in the opening of the channel activation gate. Upon desensitization, the ABD dimers dissociate, which completely rearranges the ABD layer and releases the tension exerted on the pore in the active state. In this model, the TMDs of iGluRs adopt the same conformation in the resting and desensitized states. Note that this model applies to the AMPA and kainate subfamilies of fast‐desensitizing iGluRs. The desensitization of NMDA receptors might well depart from this view (see main text).

This dual gate model for the activation and desensitization/inactivation of ion channels has a potential major outlier: the family of ionotropic glutamate receptors (iGluRs), which mediate fast glutamatergic neurotransmission in the central nervous system of vertebrates. They comprise the fast‐desensitizing kainate and AMPA receptors, whose activation kinetics are fast enough to follow trains of glutamate release occurring during high frequency stimulations (Attwell & Gibb, 2005), and the NMDA receptors, which require the binding of both glutamate and glycine for measurable activation (Johnson & Ascher, 1987; Clements & Westbrook, 1991). These tetrameric glutamate‐gated receptors adopt very different structural topologies and activation mechanisms compared to pLGICs (Smart & Paoletti, 2012; Plested, 2016): each subunit is composed of two extracellular domains, namely the N‐terminal domain (NTD) and the agonist binding domain (ABD), one TMD resembling an inverted potassium channel, and one C‐terminal cytoplasmic domain involved in the trafficking of the receptors at the plasma membrane. The NTD and the ABD are clamshell‐like bilobed domains, the latter binding the agonist in its interlobe cleft. The agonist‐elicited closure of individual ABDs is then directly coupled to the opening of the TMD. Almost two decades of functional and structural work has revealed that most, if not all, structural determinants of the desensitization of kainate and AMPA receptors are located in their extracellular part. Indeed, desensitization involves the dissociation of ABD dimers and a complete rearrangement of the extracellular architecture of kainate and AMPA receptors. Such dissociation would relieve the constraint exerted by the agonist‐bound ABD on the TMD, allowing the pore to shut in a seemingly resting‐like conformation (Fig. 12 D; Sun et al. 2002; Dawe et al. 2013; Meyerson et al. 2014, 2016; Plested, 2016). Similar structural rearrangements also occur at NMDARs during their inhibition by allosteric modulators binding to the NTD of glutamate‐binding subunits (Gielen et al. 2008; Zhu et al. 2016).

In this iGluR desensitization scheme, the pore might adopt only two possible conformations, either an open conformation in the active state, or a shut conformation identical in the resting and desensitized states. However, several points need to be considered before drawing any firm conclusion. First, the resolution of recent iGluR structures might be too low in some parts of the TMD to fully discard potential slight differences in the resting and desensitized states. Interestingly, recent cryo‐EM data with higher resolution highlighted the structure of the channel open‐state of the AMPA receptor, revealing the existence of a conformational change at the selectivity filter during activation (Twomey et al. 2017). This selectivity filter is constricted in the resting and desensitized states, thereby providing a secondary gate, which suggests that a two‐gate mechanism might occur at iGluRs. Second, these cryo‐EM structures are obtained from iGluRs whose cytoplasmic domain has been deleted, which, in association with the detergent solubilization, might affect the differential stability of various pore conformations. Third, a structural identity between the resting and desensitized states of kainate receptors might appear contradictory with some literature highlighting a metabotropic role of kainate receptors. Owing to such a metabotropic role, kainate receptors can modulate GABA release by CA1 interneurons (Rodríguez‐Moreno & Lerma, 1998) and can produce long‐lasting inhibition of postspike potassium currents (I _sAHP) in CA1 pyramidal cells (Melyan et al. 2002) in a protein kinase C (PKC)‐dependent manner, independently of any ionic flow. The dependency of metabotropic signalling on agonist concentration seems to involve the desensitized state(s) of the receptors: the kainate‐induced inhibition of I _sAHP occurs with a kainate IC₅₀ of ∼15 nm in the pyramidal cells, which express GluK2‐containing kainate receptors (Melyan et al. 2002). Such a concentration is consistent with the apparent kainate affinity for the desensitized state of recombinant GluK2 kainate receptors expressed in HEK cells (IC₅₀ ∼31 nm; Jones et al. 1997). In the hypothesis that the metabotropic signalling of kainate receptors is transduced through their desensitized state, it would be expected that the desensitized and resting states differ in their intracellular conformations, thus requiring a differential TMD conformation. Last but not least, NMDA receptors can undergo some desensitization (Sather et al. 1992), albeit usually much slower and more limited than at AMPA and kainate receptors, through a calcium‐dependent phosphorylation by calcineurin (Tong & Jahr, 1994; Tong et al. 1995). Recent work suggests that the pore‐blockers ketamine and memantine differentially impact the desensitization of NMDA receptors, hinting towards a two‐gate desensitization mechanism in this subfamily of iGluRs (Glasgow et al. 2017). The dual gate model for the activation and desensitization/inactivation of ion channels might thus be the rule rather than the exception.

Additional information

Competing interests

None declared.

Author contributions

Both authors have read and approved the final version of this manuscript and agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. All persons designated as authors qualify for authorship, and all those who qualify for authorship are listed.

Biographies

Marc Gielen did his PhD with Pierre Paoletti (Paris) on the pharmacological and structure–function properties of NMDA receptors, followed by a postdoctoral stay with Trevor Smart (London), during which he notably identified the desensitization gate of inhibitory Cys‐loop receptors. Since 2015, he is a CNRS researcher in the Pasteur Institute in Paris, in the laboratory of Pierre‐Jean Corringer, where he combines electrophysiology, molecular biology, biochemistry and structural biology tools to study synaptic channel receptors.

Pierre‐Jean Corringer trained as a chemist and did his PhD (Paris) and post‐doctoral fellowship (Brighton) in organic synthesis. He then joined the Pasteur Institute as a CNRS researcher to work on nicotinic acetylcholine receptors, and contributed to the discovery of bacterial homologues of these neurotransmitter receptors. In 2008, he created his own research group, to decipher the allosteric mechanisms of bacterial and eukaryotic homologues by combining structural, electrophysiological and fluorescence approaches.