Binding site stoichiometry and the effects of phosphorylation on human α1 homomeric glycine receptors

Abstract

The kinetic properties of the human α1 homomeric glycine receptor were investigated. Receptors were expressed in HEK 293 cells, and glycine was applied to outside-out membrane patches with sub-millisecond solution exchange. The activation time course of the glycine response was used to investigate receptor stoichiometry. The unbinding of three strychnine molecules and the cooperative binding of two glycine molecules were required to activate the channel. The effects of phosphorylation on glycine receptor kinetics were investigated by pretreating cells with phosphorylators or with phosphatases. Phosphorylation accelerated desensitisation, but slowed deactivation and recovery from desensitisation. A chemical-kinetic model was developed that reproduced the experimental observations. The model suggests that only three binding sites on the glycine channel are functional, while the remaining two binding sites are ‘silent’, possibly due to strong negative cooperativity.

The glycine receptor (GlyR) is the main inhibitory neuroreceptor in the brain stem and spinal cord (Curtis et al. 1967). It mediates fast inhibitory synaptic neurotransmission by opening an anion-selective pore in response to released transmitter. The GlyR is a member of the ‘cys-loop’ superfamily of ligand-gated ion channels, which also includes the GABA_A receptor, the nicotinic ACh receptor and the 5-HT₃ receptor (Betz, 1990; Le Novere & Changeux, 2001).

The subunits of native GlyRs were chemically cross-linked, and the molecular mass of the resulting entity revealed that five subunits combine to form a heteromeric receptor. Further studies suggested a subunit stoichiometry of 3α:2β (Langosch et al. 1988; Kuhse et al. 1993). It is thought that the β subunit does not contribute to ligand binding, but has a role in receptor assembly (Kuhse et al. 1993). Homomeric GlyRs composed entirely of α1 subunits form functional channels with properties similar to those of native spinal cord GlyRs (Schmieden et al. 1989). Thus, the kinetics of homomeric receptors can be studied in a homogeneous population with known stoichiometry, and changes in the kinetic behaviour of the receptor produced by phosphorylation or site-directed mutagenesis can be more easily interpreted. This approach has been successfully applied to identify amino acids involved in agonist binding and channel gating in α7 homo-oligomeric neuronal nicotinic acetylcholine receptors (Palma et al. 1998; Arias, 2000) and in glycine receptors (Lynch et al. 1997).

Another interesting question that can be addressed with homomeric channels is whether they have five agonist binding sites, as suggested by their pentameric structure. Dose-response analysis of steady-state currents produced by recombinant α1 homomeric receptors and native receptors have resulted in Hill coefficient estimates ranging from 1.8 to 4.2 (Akaike & Kaneda, 1989; Bormann et al. 1993). Interpretation of published Hill coefficients is difficult because cooperativity between binding sites and the effects of receptor desensitisation during prolonged agonist application are largely ignored in these studies. A better approach for investigating glycine binding stoichiometry is to apply agonists using fast-solution exchange, then analyse the current activation time course (Clements & Westbrook, 1991). At low concentrations, the binding of glycine to the receptor is the rate-limiting step leading to the opening of the glycine channel. The degree of sigmoidicity of the current activation time course is related to the number of binding steps. This method has previously revealed that there are two glutamate binding sites on both NMDA and AMPA channels in cultured hippocampal neurons (Clements & Westbrook, 1991; Clements et al. 1998) and two glycine binding sites on native heteromeric GlyRs in the zebrafish Mauthner cell (Legendre, 1998). The fast-solution exchange method can also be used to reveal the number of competitive antagonist unbinding steps that are required before the channel can activate (Clements & Westbrook, 1994).

Desensitisation of ligand-gated receptors is a ubiquitous but poorly understood process, which interferes with dose- response and single channel studies. Typically, agonist-gated channels exhibit two or three kinetically distinct desensitised states (Clements & Westbrook, 1991; Jones & Westbrook, 1997; Clements et al. 1998; Legendre, 1998). Desensitisation can be modulated by phosphorylation agents (Lohse, 1993), but the effects of phosphorylation on the glycine receptor have generally been studied under steady-state conditions, or at the single channel level. Protein kinase C (PKC) potentiated the steady-state current evoked by glycine for α1 and α2 GlyR expressed in Xenopus oocytes (Nishizaki & Ikeuchi, 1995), and for native GlyR in the rat hippocampus (Schonrock & Bormann, 1995). However, PKC decreased steady-state glycine currents in rat spinal cord tissue in vitro studies (Vaello et al. 1994; Albarran et al. 2001). Activation of cAMP-dependent protein kinase (PKA) reduced the rapidly desensitising component of the response in acutely dissociated hypothalamic neurons (Agopyan et al. 1993) while it enhanced desensitisation in rat spinal cord tissue (Vaello et al. 1994). Interestingly, the potentiating actions of PKC and PKA in acutely dissociated trigeminal neurons were not additive, suggesting a cross-modulation of glycine responses by both kinases in this preparation (Gu & Huang, 1998). At the single channel level, PKA was found to increase the probability of glycine channel opening in spinal trigeminal neurons (Song & Huang, 1990). Finally, genistein, a tyrosine kinase inhibitor, was found to directly inhibit glycine-mediated currents in hypothalamic neurons (Huang & Dillon, 2000). Here, we investigate the kinetic properties of human α1 homomeric GlyRs using a rapid-solution exchange system, and then examine the effects of phosphorylation agents on these properties.

METHODS

HEK 293 cells were grown under standard culture conditions (Rajendra et al. 1994). An aliquot of cells was suspended and mixed with plasmids encoding the α1 subunit of the human GlyR and the CD4 membrane protein. They were transfected by electroporation at 250 V, 500 μF or 960 μF. The α1 GlyR plasmid was kindly donated by Dr Peter Schofield from the Garvan Institute, Sydney. The suspended cells were then plated on plastic coverslips (Nalgene Nunc, Rochester, NY, USA) coated with a mixture of rat tail collagen (Sigma, St Louis, MO, USA) at 0.5 mg ml⁻¹ and poly-d-lysine (Sigma) at 0.1 mg ml⁻¹. The coating increased the adherence of HEK 293 cells, which was important when pulling outside-out patches. Coverslips were placed in 24-well culture dishes and incubated in a humidified 5 % CO₂ incubator. Prior to electrophysiological recording, beads coated with a CD4 antibody (Dynal Biotech, Oslo, Norway) were added to the incubation medium for visual identification of transfected cells. Recordings were carried out 24–72 h after transfection.

Coverslips containing HEK 293 cells were transferred to the recording chamber and continuously perfused with a modified Ringer solution containing (mm): NaCl, 140; KCl, 5; CaCl₂, 2; MgCl₂, 1; Hepes, 10; glucose, 10; pH was adjusted to 7.4 with NaOH and osmolarity was adjusted to 300–310 mosmol l⁻¹. Recording pipettes were coated with Sylgard and fire-polished. Patch pipettes had resistances ranging from 3–5 MΩ and contained (mm): CsCl, 150; Hepes, 10; EGTA, 10; pH was adjusted to 7.4 with CsOH and osmolarity was adjusted to 300–310 mosmol l⁻¹.

Rapid drug application was performed using a four-chamber flow tube (VitroCom, Mountain Lakes, NJ, USA) attached to a piezoelectric translator (EXFO, Quebec, Canada), as previously described (Clements & Westbrook, 1991). The flow pipes had an individual tip diameter of approximately 40 μm and were arranged in a square pattern. Outside-out patches were positioned in the flow of the control solution and then switched to the drug solution by rapid movement of the flow tube. Current rise-times (20-80 %) of less than 1 ms were routinely obtained for the response to a saturating concentration of glycine (1 mm). Pulses of glycine were repeated every 10 s. This was long enough to permit complete recovery from desensitisation between each pulse. At the end of each experiment the membrane patch was removed by applying a brief pressure pulse to the recording electrode, and the solution-exchange time constant at the open tip was measured. The control solution was diluted by 2 % with H₂O, which generated a small offset potential at the open tip of the patch electrode, and this offset was used to measure the solution-exchange time (20-80 %). If the exchange time was > 1 ms, the data for that patch was discarded. Typical open tip solution-exchange time constants were of the order of a few hundred milliseconds (Maconochie & Knight, 1989).

Currents were recorded using an Axopatch 200A amplifier (Axon Instruments, Union City, CA, USA). Series resistance compensation was 80–90 %. Data was low-pass filtered at half the sampling frequency, which was typically 10 kHz. Ensemble average responses were obtained from 10 to 200 responses. A holding potential of −20 mV was used, because this maximises the input resistance of HEK cells by reducing K⁺ channel activity (L. J. Gentet & J. D. Clements, unpublished observations). Data were analysed off-line using AxoGraph 4.0 (Axon Instruments). All data are presented as means ±s.e.m., unless stated otherwise.

A chemical kinetic modelling program (AxoGraph 4.0) was used to predict the current time course for a given reaction scheme and agonist/antagonist concentration time course. Each model was based on a chemical reaction scheme describing one or more agonist molecules binding to a receptor, and subsequent opening or desensitising of the bound receptor. AxoGraph was also used to optimise the parameters of each kinetic model to give the best fit between the predicted and observed current time course. The program uses a Simplex algorithm to adjust reaction rates and the number of channels so as to minimise the sum of square errors between the predicted current and the experimental data (Benveniste et al. 1990). In all models, the opening (α) and closing (β) rates were fixed to 1000 s⁻¹, in accordance with the reported mean open time of approximately 1 ms in single channel recordings (Langosch et al. 1994). This study investigated kinetic properties on time scales > 1 ms, so its main findings were not sensitive to the values chosen for these two parameters.

For each model, the goodness of fit was assessed using the χ² statistic, which was calculated as follows. At each time point, the difference between the data and the theoretical curve was divided by the s.e.m. at that point. These values were summed over all points to give the χ² statistic. The kinetic model was rejected if the χ² value exceeded the P= 0.05 level.

Non-stationary noise analysis was performed on 100–200 responses to brief pulses of agonist. The mean current (I) and variance (σ²) were calculated for each time point of the ensemble average. The plot of variance vs. mean current was fitted with the equation:

where i is an estimate of the average single channel current and N is an estimate of the number of channels in the patch (Sigworth, 1980). The mean open probability at the peak of the response (max P_o) was also estimated from the fitting procedure:

where I_peak is the current at the peak of the response.

RESULTS

Three strychnine unbinding steps are required before channel activation

We analysed the dissociation kinetics of strychnine from α1 GlyRs to study the binding site stoichiometry of the receptor (Clements & Westbrook, 1994). First, we applied pulses of glycine (1 mm) to outside-out patches and fitted the response using a kinetic model with three desensitisation steps to obtain desensitisation and resensitisation rates for each patch. Agonist binding is very fast at this concentration, so for simplicity only one glycine binding step was incorporated in the model. We then pre-equilibrated the GlyRs with a saturating concentration of strychnine (10 μM) and repeated the glycine pulses. Pre-equilibration with strychnine slowed the activation of the glycine response (Fig. 1A). Control responses had typical rise-times (20-80 %) of less than 1 ms, but pre-equilibration with strychnine increased the rise-time to several seconds. We fitted five different models to each ensemble average response to investigate the dissociation kinetics of strychnine from the GlyR. We added one to five identical and independent strychnine unbinding steps to the model generated from the control response recorded in the same patch. Desensitisation and resensitisation rates were fixed (Fig. 1B). The microscopic strychnine binding rate per site was fixed at 5 μM⁻¹ s⁻¹. Only the strychnine unbinding rate was free. The best fit was obtained using a model with three unbinding steps. This was the only model that provided an adequate fit to the data (P > 0.05) (Fig. 1B).

Figure 1. Strychnine dissociation kinetics.

A, the time course of strychnine dissociation is shown (open circles), together with model predictions based on the assumption of one to five identical and independent unbinding steps. The best fit was obtained with three strychnine unbinding steps (grey line). Predictions of the other models are also shown (black traces). B, summary of goodness of fit (χ²) for the various models (n= 7 patches). The 95 % confidence level for the χ² statistic is shown (horizontal line). C, Markov model incorporating three strychnine unbinding steps prior to receptor activation. The average unbinding rate for strychnine is shown (n= 7). Strychnine microscopic binding rate was fixed at 5 μM⁻¹ s⁻¹.

The value obtained for the strychnine unbinding rate was 1.1 ± 0.2 s⁻¹ (n= 7). This corresponds to a dissociation constant for strychnine of 73 ± 13 nm (n= 7), very similar to the value of 62 nm obtained by Schild analysis for native GlyR in culture (Kumamoto & Murata, 1996) and the value of 64 nm obtained by radioligand binding analysis in α1 GlyRs (Langosch et al. 1994). The results are summarised as a Markov model with three strychnine unbinding steps prior to the glycine binding step leading to the activation of the receptor (Fig. 1C). Our model requires three strychnine unbinding steps before channel activation, but this does not preclude the existence of more than three active binding sites. For example, there may be five binding sites and the channel may open while two strychnine molecules are still bound. Nonetheless, these results unambiguously indicate that there are at least three active binding sites on the wild-type (WT) human α1 homomeric GlyR.

Glycine binding is cooperative

To determine the number of glycine binding steps required for channel activation, we investigated GlyR activation kinetics at lower glycine concentrations. First, rapid pulses of a saturating concentration of glycine (1 mm) were applied to an outside-out patch, and the response was fitted with a model including three sequential desensitisation states. Next, a lower concentration of glycine (10 or 20 μM) was applied to the same patch (n= 7). In all cases, the activation time course was much slower than for responses to saturating concentrations of glycine, indicating that agonist binding had become the rate-limiting step in channel activation (Fig. 2A). The patches were pre-equilibrated in bicuculline, a very weak competitive antagonist at glycine receptors. This was done to compete off the glycine that inevitably contaminates salt solutions (typically 10–100 nm). The rise-time of the response to 1 mm glycine was not significantly slowed by pre-incubation in bicuculline, which dissociates in ≈1 ms (data not shown). The rise-time (20-80 %) of the response to 10 μM glycine was 34 ± 5 ms (n= 7). Also, the activation time course showed some degree of sigmoidicity in all patches (n= 7), which is evidence that more than one glycine molecule must bind to open the channel (Clements & Westbrook, 1991). The initial 50 or 100 ms of the activation time course was fitted with models including 1, 2 and 3 identical and independent steps, and with a model including two cooperative binding steps (Fig. 2C) (Clements et al. 1998). Desensitisation rates were fixed at the rates obtained from fitting the response to a saturating concentration of glycine. The precise time of onset of drug application was measured at the conclusion of each recording. The best fit to the data was obtained with the two cooperative binding steps model (Fig. 2B). For all but one patch (n= 6), this model provided an adequate fit to the data (P > 0.05). The degree of cooperativity, as measured by dividing the largest of the two unbinding rates by the smallest, was 12.5 ± 1.8 (n= 7). The 1, 2 and 3 independent and identical sites models could be rejected for all cells (P < 0.05, n= 7).

Figure 2. Glycine binding kinetics.

A, time course of glycine binding (open circles), together with model predictions based on the assumption of one to three identical and independent binding steps, or two cooperative binding steps. The best fit was obtained with two cooperative binding steps (grey line). Predictions of the other models are also shown (black traces). B, summary of goodness of fit (χ²) for the various models (n= 7). The 95 % confidence level for the χ² statistic is shown (horizontal line). C, Markov model including two glycine cooperative binding steps prior to receptor activation. The microscopic binding rates for both sites were fixed at 5 μM⁻¹ s⁻¹ and the unbinding rates were left as free variables.

Phosphorylation enhances desensitisation

We examined the effects of increased phosphorylation on the desensitisation and deactivation kinetics of recombinant human α1 homomeric GlyRs in outside-out patches from transfected HEK 293 cells. Some of the cells were incubated for 1 h prior to recording in 100 μM forskolin and 500 μM 3-isobutyl-1-methyl-xanthine (IBMX), which activates the cAMP-dependent phosphorylation pathway, or 100 nm phorbol-12,13-dibutyrate (PDBu), which activates PKC. There was a similar change in GlyR desensitisation kinetics following both treatments, and the results have been pooled. Application of a saturating concentration of glycine (1 mm) to an outside-out patch produced a desensitising response (Fig. 3A), which was well fitted with the sum of three exponential functions (n= 110). The time constant of the first component of desensitisation fell from 30.1 ± 2.6 ms (n= 57) for untreated cells to 18.8 ± 1.9 ms (n= 53) for cells that had been pre-incubated (Fig. 3B). Both the second and third desensitisation time constants were unaffected (235 ± 46 and 1082 ± 226 ms (n= 12) for control patches vs. 275 ± 28 and 1358 ± 142 ms (n= 21) for phosphorylated patches). The ratio of the steady-state to peak current fell from 16.9 ± 1.6 % for untreated patches (n= 57) to 13.5 ± 1.5 % for phosphorylated patches (n= 53). The deactivation phase following rapid washout of glycine was well fitted with two exponentials over a 500 ms range. The time constant of the first component was unaffected, but the second time constant increased from 93.0 ± 10.5 ms (n= 30) for untreated patches to 145.7 ± 21.7 ms (n= 16) for patches from pre-incubated cells (Fig. 3C). In summary, enhanced phosphorylation speeded up the early phase of desensitisation, but slowed the late phase of deactivation.

Figure 3. Effects of phosphorylation on the desensitisation and deactivation kinetics of the WT human homomeric α1 GlyR.

A, response to a 5 s pulse of 1 mm glycine for an untreated patch (black circles) and a patch pre-treated to enhance phosphorylation (grey circles). Both responses were optimally fitted with a chemical-kinetic model (black line and grey line). B, enhanced phosphorylation (grey bar) decreased the time constant of the first component of desensitisation. C, the first component of deactivation was unaffected by phosphorylation (grey bar), but the time constant of the second component was increased.

We next investigated the effects of dephosphorylation on the desensitisation and deactivation phases of the response to glycine. We compared the responses of two different outside-out patches pulled from the same untreated cell. The pipette solution for the second patch contained 20 units ml⁻¹ of alkaline phosphatases (Sigma). After pulling the patch we waited for 1 min before starting the recording, to permit the hydrolysis of phosphate groups on the GlyRs. Both the first time constant of desensitisation and the ratio of steady-state to peak current were significantly increased (287 ± 51 and 302 ± 76 %, respectively, n= 4) (Fig. 4A and B).

Figure 4. Effects of dephosphorylation on the desensitisation kinetics of the GlyR.

A, responses to a 5 s pulse of 1 mm glycine for an untreated patch (black circles) and a patch pulled from the same cell with alkaline phosphatases included in the patch pipette (grey circles). Model fits for both patches are shown (black line and grey line). B, dephosphorylation increased the first desensitisation decay time constant and the steady-state to peak ratio. The percentage change between the first and second paired patches is shown (n= 4 pairs). C, the Markov model used to fit the glycine responses included three sequential desensitisation steps, labelled fast, moderate and slow. The number of glycine binding steps was arbitrarily set to one.

Responses from outside-out patches pulled from untreated and pre-incubated cells were fitted with a kinetic model that included three desensitised states and one open state (Fig. 4C). The glycine microscopic binding rate was fixed at 5 μM⁻¹ s⁻¹, which is a reasonable estimate given the receptor's affinity and kinetic properties (Clements et al. 1998). The unbinding rate and the desensitisation and resensitisation rates were free variables. The data was well fitted over the whole desensitisation phase in all cases (n= 10). The average rates obtained for the free variables are shown in Table 1. The main effects of increased phosphorylation were an increase in the fast and moderate desensitisation/resensitisation rates. Interestingly, there was no significant change in the unbinding rate, despite the observed slowing of deactivation. This implies that recovery from desensitisation contributes to the slow component of GlyR deactivation, as has previously been reported for GABA_A receptors (Jones & Westbrook, 1995).

Table 1.

Optimum reaction rates obtained from kinetic model fits

	dfast	rfast	dmod	rmod	dslow	rslow	Unbinding(s−1)
Untreated	10.2 ± 3.8	3.2 ± 1.3	1.3 ± 0.8	0.4 ± 0.2	0.4 ± 0.1	2.5 ± 0.7	41.4 ± 7.3
Enhanced phosphorylation	46.7 ± 6.7	27.2 ± 5.2	2.9 ± 0.9	2.0 ± 0.7	0.9 ± 0.2	0.4 ± 0.1	52.3 ± 7.2

Values are means ±s.e.m.; n= 5

Recovery from desensitisation is affected by phosphorylation

The experimental protocol we used to investigate desensitisation and deactivation does not provide any information about the rate of recovery from the slow component of desensitisation. To investigate the affect of enhanced phosphorylation on this parameter, we applied paired pulses of 1 mm glycine at increasing interpulse intervals. The initial pulse was 500 ms long, which causes some receptors to enter a slow desensitised state. A second pulse was applied after a delay ranging from 200 to 4200 ms in increments of 500 ms. Rundown of the response, which sometimes occurred between paired pulses, was compensated by normalising each paired pulse response at the peak of the initial pulse (Fig. 5). However, if the rundown exceeded 10 % between the initial and final paired pulses the data were discarded. The amount of desensitisation at the end of the initial pulse was similar for both phosphorylated and untreated patches (phosphorylated patches: 34 % ± 8 %, n= 4; untreated patches: 36 % ± 9 %, n= 4). The time course of recovery from desensitisation was measured by plotting the second peak amplitude as a function of the interpulse interval. For both phosphorylated patches and untreated patches, the recovery could be well fitted with one exponential (Fig. 5C). The time constant was 0.66 s for untreated patches and 1.99 s for phosphorylated patches. Thus, phosphorylation slows recovery from desensitisation by stabilising the channel in the desensitised state.

Figure 5. Recovery from desensitisation is affected by phosphorylation.

A, paired-pulse recovery for an untreated patch. Data are shown as black circles and model fit is shown as a black line. B, paired-pulse recovery for a patch with enhanced phosphorylation. Data are shown as grey circles and model fit is shown as a black line. C, time course of recovery of the peak response for untreated patches (black squares, n= 4) and phosphorylated patches (grey squares, n= 4). The data points were fitted with a single exponential (black line and grey line) to obtain time constant values for recovery of 0.66 s for untreated patches and 1.99 s for phosphorylated patches. D, cyclic reaction scheme used in the kinetic model of recovery from desensitisation.

In order to refine our kinetic model of the α1 GlyR, the model parameters were adjusted to simultaneously fit all eight paired-pulse responses. The deactivation phases of the pulses were not well fitted by a kinetic model that had three sequential desensitisation steps. As noted earlier, the deactivation phase is influenced by desensitisation (Jones & Westbrook, 1995). Changing the sequential desensitisation to a cyclic desensitisation model, the data were well fitted over the entire time course (n= 3 patches) (Fig. 5A and B). We used a cyclic reaction scheme, which included two sequential desensitisation steps and one monoliganded desensitisation state (Fig. 5D).

Phosphorylation prolongs brief pulse responses in WT GlyR

A brief pulse (3 ms) of high concentration glycine (1 mm) evoked a current with a rapid rise (< 1 ms) and a decay that could be well fitted with two or three exponential functions (Fig. 6A). The time course of these responses resembled IPSCs evoked in rat dorsal horn neurons at room temperature (Takahashi et al. 1992) but were substantially slower than IPSCs recorded from cat spinal motoneurones at 37 °C (Stuart & Redman, 1990). Pretreatment with IBMX or forskolin slowed the decay phase of the responses, while the rising phase remained below 1 ms (Fig. 6A-C). The first decay time constant for deactivation was unaffected (5.4 ± 0.4 ms, n= 37 for control patches, and 4.9 ± 0.6 ms, n= 17 for patches with enhanced phosphorylation). However, both the second and third decay time constants increased from 19.4 ± 1.0 and 70.3 ± 4.0 ms for control patches to 24.0 ± 1.9 and 123.2 ± 10.1 ms for phosphorylated patches (Fig. 6B). Moreover, the current after 200 ms increased from 3.1 ± 0.4 % of peak for control patches to 10.8 ± 1.5 % of peak for phosphorylated patches (Fig. 6C). Thus, phosphorylation substantially prolonged current decay following a brief pulse of glycine. In part, this is due to a larger number of receptors entering then exiting a desensitised state, which can prolong the response to a brief pulse of agonist (Jones & Westbrook, 1995). There was no difference between the average peak amplitude of the responses observed in control patches (600 ± 70 pA, n= 37) and in phosphorylated patches (630 ± 130 pA, n= 17), suggesting that the channel open probability at the peak is unaffected by phosphorylation.

Figure 6. GlyR deactivation kinetics following a brief pulse of agonists.

A, average response to a 3 ms pulse of 1 mm glycine for an untreated patch (black circles) and a patch with enhanced phosphorylation (grey circles). Model fits are shown as black and grey lines. B, comparison of the decay time constants obtained by fitting three exponential functions to the deactivation phase for untreated patches (black, n= 37) and phosphorylated patches (grey, n= 17). C, comparison of residual current after 200 ms of decay for untreated patches (black, n= 37) and phosphorylated patches (grey, n= 17). D, non-stationary noise analysis for a control patch obtained from 110 traces. Ensemble variance is plotted against ensemble mean at each time point, and the scatter plot is fitted with a parabolic function. The open probability for this patch was 0.84, and the single channel conductance was 25 pS.

Non-stationary noise analysis was performed on five control patches, which were chosen for their high signal-to-noise ratio and the absence of rundown. The peak open probability (max P_o) was 0.80 ± 0.05 (n= 5) (Fig. 6D). The mean single-channel conductance showed some patch-to-patch variability with an average value of 30 ± 15 pS. As expected, this value is smaller than the main 86 pS conductance state reported for homomeric α1 GlyRs (Bormann et al. 1993; Langosch et al. 1994), because it represents a weighted average across all the subconductance states of the channel.

Proposed Markov model for the WT human homomeric α1 GlyR

To summarise our results, we propose a Markov model that explains all the observed kinetic properties of α1 GlyRs (Fig. 7). This model has only a single open state, despite previously published evidence that the receptor can open to several conductance levels. Under our experimental conditions these subconductance states were kinetically indistinguishable, so it was a valid approximation to combine them into a single open state in the Markov model. With this simplification, the closing rate can be set 2.5 times slower than the opening rate, to account for the max P_o of ≈0.8 obtained from non-stationary noise analysis. It is possible that more than two glycine molecules can bind to the receptor, but extrapolation of the negative cooperativity effects would reduce the affinity of the third binding step by a factor of ≈150, such that no significant binding of glycine would occur. Three sequential strychnine unbinding steps are also included in the model, to account for its unbinding time course. Three desensitisation steps are included in accordance with a triphasic decay of the desensitisation phase of the response. Two desensitisation states are arranged in a cyclic reaction scheme with the mono and doubly liganded glycine activation steps to account for the recovery from desensitisation observed in paired-pulse experiments.

Figure 7. Proposed Markov model for α1 human GlyR.

The model includes three sequential strychnine unbinding steps, two cooperative glycine binding steps, two sequential desensitisation steps (R_d), one cyclic desensitisation step and an open state (*). Rates fixed during curve fitting are in italic.

DISCUSSION

A physiological role for phosphorylation of theGlyR

Phosphorylation of WT α1 GlyRs affects their desensitisation and deactivation kinetics. Both cAMP-dependent, and PKC-mediated phosphorylation modulated the kinetics of the glycine response, suggesting that both pathways converge on a common residue. Consistent with this, the potentiating actions of PKC and PKA on GlyRs are not additive in acutely dissociated trigeminal neurons (Gu & Huang, 1998). A putative site for PKC-mediated phosphorylation has been identified at serine 391 in the large intracellular loop of the α subunit between M3 and M4 (Ruiz-Gomez et al. 1991). Supporting evidence has been reported for the GABA_A receptor, where both PKA and PKC phosphorylate the homologous serine 409 residue of the β subunit (Moss et al. 1992).

We observed cell-to-cell variability in the rate of desensitisation and the ratio of steady-state to peak current, in outside-out patches pulled from untreated HEK 293 cells. This could be explained by variations in the degree of phosphorylation at different stages of a cell's cycle. The mean desensitisation of α1 GlyRs in outside-out patches pulled from Xenopus oocytes is less than 50 %, compared to over 80 % in our study (De Saint Jan et al. 2001), and this discrepancy may be due to different levels of constitutive PKA activity in the two cell types. Supporting this suggestion, GABA_A receptor properties are affected by the different levels of constitutive PKA activity found in three cell lines (Angelotti et al. 1993).

Changes in constitutive phosphorylation levels could modulate glycinergic synapses. We found that enhanced phosphorylation slowed the deactivation kinetics following a brief pulse of agonist, which mimics a synaptic pulse. Slower deactivation would lead to a stronger inhibitory synaptic signal, and would enhance integration during a train of inhibitory activity.

Comparison of WT GlyR and GABA_A receptor kinetics

The WT GlyR exhibits multiphasic desensitisation and deactivation. Multiphasic desensitisation has also been observed in GABA_A receptors in the hippocampus of guinea-pigs (Celentano & Wong, 1994). Three desensitisation steps were identified and their estimates of desensitisation and resensitisation rates were generally similar to the ones we obtained for the WT GlyR. The only exception was recovery from the fast component of desensitisation, which was significantly slower for the GlyR. This indicates that hippocampal GABA_A receptors would recover from desensitisation much more rapidly than GlyRs and this could have significant implications for the processing of repetitive presynaptic activity at synapses where glycine and GABA_A receptors are co-localised (Bohlhalter et al. 1994).

Deactivation was also observed to be multiphasic in GABA_A receptors in rat hippocampal neurons (Jones & Westbrook, 1997). Decay time constants for GABA_A receptors were significantly larger than for WT GlyRs. Thus, although GABA_A and GlyR share some similarity in their desensitisation properties, their deactivation properties are distinct. The predominance of GlyRs in the spinal cord, where fast synaptic inhibition plays a crucial role in the co-ordination of movement may be related to its faster deactivation kinetics.

Stoichiometry and structure of WT GlyRs

In the native heteromeric GlyR, only three binding sites are available since the stoichiometry of such receptors is believed to be 3α:2β (Langosch et al. 1988). In contrast, the homomeric α1 GlyR has five potential binding sites. Results obtained from our study of strychnine dissociation kinetics suggest that only three sites are functional at any time. This result can be interpreted by assuming that the last two binding sites become unavailable after three ligand molecules have already bound. This could be due to steric hindrance from the bound ligand molecules, or because of conformational changes in the receptor. An alternative structural explanation is that the homomeric GlyR has only three exposed binding sites, because the five α1 GlyR subunits are not arranged symmetrically around the central pore.

A more likely explanation of our findings is that all five binding pockets are initially exposed, but they exhibit strong negative cooperativity for the binding of both glycine and strychnine. Under this assumption, the binding of two glycine molecules would induce structural changes that inhibit further binding even at millimolar concentrations of glycine. In this view, when the first two sites are occupied, the third, fourth and fifth sites are still available but are of such low affinity that they are effectively ‘silent’. Similarly, the binding of three strychnine molecules would render the subsequent binding of agonist or antagonist molecules highly improbable. Such negative cooperativity could be a consequence of allosteric interactions between binding sites. The difference in the number of functional glycine binding sites and the number of functional strychnine binding sites can be accounted for under this scheme, since the molecules are of different size and would produce different conformational changes in the receptor.