GluA4 is indispensable for driving fast neurotransmission across a high-fidelity central synapse

Non-technical summary

Localization of sound sources in the azimuth, which makes use of interaural differences in timing and/or intensity of acoustic signals, is of vital importance for most mammals. Using the small differences in time of arrival and/or intensity at the two ears requires that propagation of electric pulses in the auditory system be temporally precise. In this study, we found that elimination of GluA4, a protein particularly abundant in auditory cells, significantly impairs their ability to faithfully transmit electric signals, leading to profound deficits in auditory responses to sound stimuli in mice. Therefore, we conclude that GluA4 is indispensable for enabling information flow with high fidelity in the auditory circuitry. Our work has identified GluA4 as a potential molecular candidate involved in human hearing deficits and disorders.

Abstract

Abstract

Fast excitatory synaptic transmission in central synapses is mediated primarily by AMPA receptors (AMPARs), which are heteromeric assemblies of four subunits, GluA1–4. Among these subunits, rapidly gating GluA3/4 appears to be the most abundantly expressed to enable neurotransmission with submillisecond precision at fast rates in subsets of central synapses. However, neither definitive identification of the molecular substrate for native AMPARs in these neurons, nor their hypothesized functional roles in vivo has been unequivocally demonstrated, largely due to lack of specific antagonists. Using GluA3 or GluA4 knockout (KO) mice, we investigated these issues at the calyx of Held synapse, which is known as a high-fidelity synapse involved in sound localization. Patch-clamp recordings from postsynaptic neurons showed that deletion of GluA4 significantly slowed the time course of both evoked and miniature AMPAR-mediated excitatory postsynaptic currents (AMPAR-EPSCs), reduced their amplitude, and exacerbated AMPAR desensitization and short-term depression (STD). Surprisingly, presynaptic release probability was also elevated, contributing to severe STD at GluA4-KO synapses. In contrast, only marginal changes in AMPAR-EPSCs were found in GluA3-KO mice. Furthermore, independent of changes in intrinsic excitability of postsynaptic neurons, deletion of GluA4 markedly reduced synaptic drive and increased action potential failures during high-frequency activity, leading to profound deficits in specific components of the auditory brainstem responses associated with synchronized spiking in the calyx of Held synapse and other related neurons in vivo. These observations identify GluA4 as the main determinant for fast synaptic response, indispensable for driving high-fidelity neurotransmission and conveying precise temporal information.

Introduction

At the vast majority of excitatory synapses in the mammalian brain, AMPA receptors (AMPARs) generate fast synaptic currents that drive neurotransmission. These receptors play essential roles not only in the formation, function and maintenance of synapses, but also in different forms of synaptic plasticity (Malenka & Bear, 2004; Turrigiano & Nelson, 2004; Sprengel, 2006; Shepherd & Huganir, 2007). Native AMPARs consist of various homo- and heteromeric combinations of four subunits (GluA1–4) with flip and flop splice variants, generating tremendous functional heterogeneity in gating kinetics, Ca²⁺ permeability and rectification of the current–voltage (I–V) relationship in different brain regions (Hollmann & Heinemann, 1994; Mosbacher et al. 1994; Partin et al. 1994; Geiger et al. 1995; Dingledine et al. 1999).

Slowly gating AMPARs, such as GluA1-dominant AMPARs in cortical principal neurons, have been shown to be critical for neuronal growth, long-term potentiation (LTP) and relevant cognitive functions (Malinow & Malenka, 2002; Shepherd & Huganir, 2007; Kessels & Malinow, 2009). In contrast, fast gating AMPARs with robust Ca²⁺ permeability and submillisecond gating kinetics are prominently present in subsets of neurons capable of firing action potentials at high frequencies (i.e. fast spiking neurons), such as auditory relay neurons, cerebellar granule cells, cortical interneurons and retinal bipolar cells (Silver et al. 1996; Geiger et al. 1997; Trussell, 1999; Jonas, 2000; Ravindranathan et al. 2000; von Gersdorff & Borst, 2002). Interestingly, these neurons also express high levels of Ca²⁺ binding proteins such as calbindin, calretinin and parvalbumin, which effectively buffer Ca²⁺ build-up during high-frequency synaptic activity (Hof et al. 1999; Felmy & Schneggenburger, 2004). Electrophysiological, immunohistochemical and molecular evidence suggests that GluA3/4 subunits play an important role in dictating the fast-gating phenotype of synaptic AMPARs and driving neurotransmission at remarkably high rates in these synapses. However, whether the GluA3 or GluA4 subunit or both are indeed the molecular substrate(s) of synaptic AMPARs, and how fast-gating AMPARs influence the functionality of the circuits involved in high-speed temporal information process in vivo remain elusive.

Extensive studies over the last two decades have accumulated strong evidence that auditory neurons use Ca²⁺-permeable AMPARs to generate fast EPSCs with submillisecond kinetics and drive high-fidelity neurotransmission in vitro in chicks, mice and rats (Raman & Trussell, 1992; Zhang & Trussell, 1994; Otis et al. 1995; Barnes-Davies & Forsythe, 1995; Golding et al. 1995; Gardner et al. 1999, 2001; Joshi et al. 2004; Koike-Tani et al. 2005, 2008), supporting the view that fast-gating AMPARs are critical for temporal processing in vivo (Oertel, 1999; Trussell, 1999). Single-cell RT-PCR analyses of GluA mRNAs from medial nucleus of the trapezoid body (MNTB) neurons in the auditory brainstem further suggested that GluA1–4 transcripts are all present and GluA4-flop transcripts are predominant (Koike-Tani et al. 2005). However, GluA transcript levels are not necessarily correlated with their protein products (Hall & Ghosh, 2008), and even if correlated, which subunit will preferentially target synaptic and extrasynaptic domains remains uncertain. In fact, a recent study examining subsynaptic distribution of AMPARs with immunogold particle labelling showed that GluA2/3 are the dominant subunit over GluA4 (Hermida et al. 2006), in sharp contrast to the conclusion from the analyses on GluA transcripts (Geiger et al. 1995; Koike-Tani et al. 2005). Furthermore, lack of specific blockers to readily distinguish different subunits of AMPARs has so far precluded us from obtaining unequivocal evidence to explicitly demonstrate the molecular basis of synaptic AMPARs and their functional roles in vitro and in vivo.

To specifically address these issues, we employ GluA3 and GluA4 knockout (KO) mice and demonstrate that at the calyx of Held synapse GluA4, but not GluA3, determines both the amplitude and decay kinetics of synaptic AMPARs, and enables the high-frequency firing mode of postsynaptic neurons in vitro. Deletion of GluA4 severely disrupts auditory brainstem responses (ABRs) evoked by rapidly presented sound stimuli in vivo. These results illustrate that GluA4 plays an indispensable role in conducting fast neurotransmission required for coding auditory timing signals.

Methods

Mouse slice preparation

Wild-type and transgenic mice were housed and bred in the facility certified by the Canadian Council of Animal Care and used for this study according to the protocol approved by the Animal Care Committee of the Hospital for Sick Children. The generation strategies of GluA3 (CD1 strain) and GluA4 (C57BL strain) knockout (KO) mice and confirmation of deletion of GluA3 or GluA4 genes and proteins had been previously described (Meng et al. 2003; Gardner et al. 2005). Homozygote breeding pairs of GluA3 and GluA4-KO mice appeared to be healthy and viable and hence were used to produce offspring for electrophysiological recordings. Brainstem slices were prepared from age- and strain-matched wild-type (WT) and mutant mice (postnatal day 16 to 19, P16–19), as previously described (Barnes-Davies & Forsythe, 1993). After decapitation with a small guillotine, brains were immediately immersed into semi-frozen artificial cerebral spinal fluid (ACSF) containing (in mm): NaCl (125), KCl (2.5), glucose (10), NaH₂PO₄ (1.25), sodium pyruvate (2), myo-inositol (3), ascorbic acid (0.5), NaHCO₃ (26), MgCl₂ (1) and CaCl₂ (2) at a pH of 7.3 when oxygenated (95% O₂ and 5% CO₂) followed by rapid dissection. Transverse slices of the auditory brainstem containing the MNTB were cut at a thickness of 200–250 μm using a Microtome (Leica VT1000S) and incubated at 37°C for 1 h prior to experimentation.

Electrophysiology

Whole-cell patch-clamp recordings were made from postsynaptic MNTB neurons with a MultiClamp 700A amplifier (Axon Instruments) at room temperature (20–22°C). ACSF was supplemented with bicuculline (10 μm) and strychnine (1 μm) to block inhibitory inputs and MK-801 (10 μm) to block NMDARs. EPSCs or action potential (AP) spikes were evoked by axonal stimulation with a bipolar platinum electrode placed near the midline of slices. Stimulation protocols were delivered by a Master-8 stimulator (A.M.P.I) and stimulation voltage was set at 30–50% above the threshold (usually 3–10 V). For voltage-clamp recordings of EPSCs, patch electrodes typically had resistances of 2.5–3 MΩ and were filled with intracellular solution containing (in mm): 97.5 potassium gluconate, 32.5 CsCl, 5 EGTA, 10 Hepes, 1 MgCl₂, 30 TEA plus 3 lidocaine N-ethyl bromide (QX314) and 0.2 spermine (pH 7.2). For current-clamp recordings, pipettes were filled with a solution including (in mm): 97.5 potassium gluconate, 32.5 KCl, 0.5 EGTA, 40 Hepes and 1 MgCl₂ (pH 7.3). Series resistance was 3–6 MΩ and compensated to 90%. In Fig. 6A, the current-command templates were generated by reversing the polarity of a typical train of EPSCs in response to 300 Hz stimuli from a WT or GluA4-KO neuron. After manually removing stimulation artifacts preceding the EPSCs, the digitized values were fed back into the amplifier as stimulation files (Axon Text File) through pClamp9 software (Axon Instruments) at the same frequency as their acquisition (50 kHz). Details for the current-clamp protocols are given in the text and figure legends. Data were filtered at 4 kHz, digitized at 50 kHz, acquired on-line and analysed with pClamp9 software or MiniAnalysis program (Synaptosoft). The decay time course of EPSCs (Fig. 1), short-term depression curves (Fig. 3) and the input–output relationship of steady-state potentials evoked by current steps (Fig. 5F, top panel) were fitted with a single exponential function. The steady state of synaptic depression was quantified as the average amplitude of the last five EPSCs during 300 Hz test trains (Fig. 3H). For analyses of depolarization-dependent activation of AP spikes in Fig. 5F (bottom panel), curve fittings were done with Clampfit and equations are given in the figure legend. Using 300 Hz trains of stimulation to deplete synaptic vesicles (SVs) from the readily releasable pool (RRP), we measured the size of the RRP based on a previously described method (Wesseling & Lo, 2002) by taking into account the variable rate at which SVs are recruited back into the RRP during the steady state of release in the late phase of stimulation trains. The advantage of this model is to prevent underestimating the contribution of the first few EPSCs to the RRP in a train of stimuli. First, the following equations were solved simultaneously:

(1)

(2)

where fe is the fusion efficiency, r(1) is the amplitude of the first response, r(∞) is the amplitude of the steady-state response, α is the rate of pool filling, S is the number of stimuli in the train and v is the frequency of stimulation. Then, the number of SVs from the RRP (N) can be derived from the following equation:

(3)

where w(S) is the vesicles released from the reserve pool during the train. Statistical tests of significance employed two-tailed, unpaired Student's t tests assuming unequal variances with a P value cut-off of <0.05. Data were expressed as the mean ± standard error of the mean (SEM) from a population of synapses (n). Reagents were purchased from Sigma, Tocris and Alomone Labs.

Figure 6. Functional rescue of high-frequency spike failures in GluA4-KO neurons with WT input template.

A, the current templates (top panels) were generated by reversing the polarity of two representative recordings of EPSCs from GluA4-KO (left) or WT (middle) synapse in response to a train stimulation (300 Hz, 100 ms) in 2 mm[Ca²⁺]_o. The top right panel displays the scaled KO template with the same initial amplitude as the WT template. Spikes were evoked in current-clamp configuration by injection of the simulated templates of EPSC trains into GluA4-KO neurons (bottom panels). B and C, summary plots of spike failure rates produced by the KO template, WT template or scaled KO template in GluA4-KO (B, n = 7) and matching WT (C, n = 7) mice. Note the KO template with small amplitude and severe depression induces much more spike failures in both groups, which can be fully rescued by the WT- but not the scaled KO template.

Figure 1. Predominant contribution of GluA4 subunits to fast and robust AMPAR-EPSCs at the calyx of Held synapse.

A, sample traces of single EPSCs evoked by axonal stimulation and recorded at −60 mV from WT (grey) and GluA3-KO synapses (black). Dotted line indicates normalized AMPAR-EPSC of GluA3-KO synapse to that of WT synapse, showing a slight difference in the decay kinetics of EPSCs from two synapses. B, summary plots of amplitude (a), 10–90 rise time (b) and decay time constants (c) of EPSCs obtained from WT and GluA3-KO synapses (n = 8 for each group). The decay time constants are estimated by fitting the decay phase of EPSCs with a single exponential function. C, AMPAR-EPSCs recorded from −60 mV to 60 mV with 20 mV increments in the presence of NMDA blocker MK-801 (10 μm) from WT (grey traces) and GluA3-KO (black traces) synapses. D, the peak amplitude of EPSCs at various holding potentials is plotted to show the voltage dependence of AMPARs in WT or KO neurons (n = 8 for each). E, the inward rectification index, i.e. the ratio of the amplitude of EPSCs at −40 mV to that at +40 mV, is summarized for WT (n = 7) and GluA3-KO (n = 8) groups. F–J, similar recordings and quantifications of AMPAR-EPSCs from GluA4-KO (black, n = 8–11) or age- and strain-matched WT (grey, n = 6–7) mice. Dotted line shows normalized EPSC of GluA4-KO synapse to that of WT synapse, illustrating a significant prolongation in the decay time course of EPSCs at GluA4-KO synapse (F). Note deletion of GluA4 gene dramatically reduces the amplitude (especially at negative potentials, H and I) and slows down the kinetics of evoked EPSCs (F and G) as compared to the GluA3-null group. The asterisks indicate statistical significance (P < 0.05) in this and all subsequent figures.

Figure 3. Exacerbated short-term synaptic depression in GluA4-KO synapses.

A and B, example recordings of AMPAR-EPSCs produced by a train of stimuli (A: 100 Hz, 100 ms vs. B: 300 Hz, 100 ms) from WT (top panels) and GluA4-KO synapses (bottom panels) in the absence or presence of CTZ (50 μm). Grey traces represent recordings in CTZ from the same WT and GluA4-KO neuron as shown in control solution, respectively. C–F, the absolute (C and D) and normalized (to the first response in a train stimulation; E and F) amplitude of EPSCs is summarized for all the groups (n = 6 for each). The decay rates of EPSC amplitude over stimulus time are fitted with a single exponential function and the values are given in the figures. G, the decay rates of the amplitude of EPSCs over 300 Hz test trains are shown (n = 6 for each group). Note, CTZ significantly slows down the depression rate at KO but not WT synapses. H, the steady state of synaptic depression, as defined as the average amplitude of the last five EPSCs during 300 Hz trains of stimuli, demonstrates the extent of short-term depression (STD) for all the groups.

Figure 5. Compromised spike fidelity during high-frequency neurotransmission in GluA4-KO mice.

A, postsynaptic action potentials (APs) recorded in response to repetitive stimulation at 300 Hz from WT (grey traces) or GluA3-KO (black traces) MNTB neurons in the extracellular solution containing 1 mm Ca²⁺ and 10 μm MK-801. B, a summary of the failure rates (normalized spike failures to the number of stimuli in a train) during 300 Hz stimulation obtained from WT (grey bar, n = 5) and GluA3-KO (black bar, n = 5) neurons. C, similar recordings from WT (grey traces) or GluA4-KO (black traces) synapses evoked by 100 Hz (left panels) or 300 Hz (right panels) stimulation protocols. Note that deletion of GluA4 gene fails to affect the lower frequency (100 Hz) neurotransmission but significantly increases the number of spike failures at higher frequency (300 Hz). D, plot summarizing the failure rates produced by 300 Hz stimulation in WT (grey bar, n = 5) and GluA4-KO (black bar, n = 9) mice. E, example traces of APs evoked by depolarizing currents (top panel) from a WT (middle panel) or GluA4-KO (bottom panel) synapse. F, summaries of steady-state potentials (top panel) measured within the last 5 ms range of the evoked potentials and the number of spikes (bottom panel) generated by various depolarization steps (as shown in E) in WT (grey, n = 7) and GluA4-KO (black, n = 7) neurons. The continuous lines in the top and bottom panels represent fits to a single exponential function and a Boltzmann function: f_V = I_max/(1+e^(V_mid-V)/Vc) + C, respectively. In this case, I_max is the theoretical value of the maximal number of spikes; V_mid indicates the depolarization current needed to produce half of the maximal number of APs; Vc describes the steepness of the Boltzmann curve.

Auditory brainstem responses (ABRs)

ABRs are short-latency far-field auditory evoked potentials that reflect the synchronous neural response of the auditory nerve and brainstem to the onset of sounds. ABRs are commonly used to assess hearing status (i.e. whether an animal shows hearing loss) and auditory brainstem synchrony in animals and humans (Henry, 1979; Hall, 1992; Melcher & Kiang, 1996). The recordings are averaged over hundreds of stimulus repetitions so that background noise is minimized. Mice were anaesthetized with ketamine (100 mg kg⁻¹) and xylazine (20 mg kg⁻¹) mixed in a 14% EtOH solution to eliminate head movement and placed on a heating pad located inside a sound-attenuating chamber (IAC). Temperature was monitored through a rectal thermometer and kept at ∼37°C. ABRs were recorded differentially from subdermal needle electrodes placed behind the left pinnea and at the vertex of the skull. A ground electrode was inserted into the left hind leg. Stimulus presentation and acquisition was controlled by Tucker Davis Technologies System 3 hardware and a PC equipped with custom Matlab-based software. Stimuli were played through two Radio Shack Supertweeters located 30 cm in front of the mouse's head.

Responses to 100 μs clicks and 5 ms tone pips of 8, 16 or 32 kHz (0.5 ms rise/fall time) presented at a rate of 10–50 s⁻¹ were analysed. Thresholds for eliciting an ABR were determined by presenting stimuli at successively decreasing levels. Stimuli were presented at levels beginning well above threshold and then descending in 10 or 20 dB steps until only a small response was visible. Stimulus levels were then presented in 5 dB decrements until the response was indistinguishable from background noise. Threshold was defined as the stimulus level that produced a response with magnitude of 2 times of standard deviation above background noise. Responses were amplified (ISO-80, World Precision Instruments) and filtered from 300 to 3000 Hz (Krohn-Hite Corp.). All responses were averaged over 300 stimulus repetitions. For multiple comparisons (wave 1–4), ANOVA was used with a Bonferoni correction for P values (SPSS software). The P criterion for significance was 0.0125.

Results

The calyx of Held MNTB synapse is a giant axosomatic synapse in the superior olivary complex of the auditory brainstem where interaural timing and level differences (ITD and ILD) are registered as primary cues for accurately localizing sound source in space. Such a structure provides a powerful excitatory drive for MNTB neurons to spike with minimal temporal jitter, and to project precisely timed inhibitory outputs to other superior olivary neurons where the ITD and ILD are detected and coded. Technically, this preparation also allows quantitative measurements of synaptic responses at the soma where adequate space-clamp can be achieved with minimal cable filtering effects. Previous work from our group and others has implicated a developmental subunit switch from a GluA1-dominant gating phenotype in immature synapses to a GluA3/4-dominant gating phenotype in mature synapses within the first two postnatal weeks, particularly following the onset of hearing at P11/12 (Joshi et al. 2004; Koike-Tani et al. 2005, 2008). Thus, the calyx of Held synapse is an ideal model to study the molecular composition of native AMPARs and their functionalities in a physiological context. We therefore performed all electrophysiological experiments in in vitro slices from P16–19 wild-type and GluA3- or GluA4-KO mice when the calyx of Held synapse is considered to be both morphologically and functionally mature. For ABR recordings, older mice (5–6 weeks) were used. This age group was chosen because some ABR component waves are not distinguishable in younger mice (Song et al. 2006), and C57BL mice (background strain) begin to show hearing loss at older ages (3–4 months).

GluA4 but not GluA3 subunits are required to produce large and rapid EPSCs

To investigate the subunit composition of synaptic AMPARs, we first made whole-cell voltage-clamp recordings of EPSCs from postsynaptic principal neurons in the medial nucleus of the trapezoid body (MNTB) of brainstem slices acutely isolated from age- and strain-matched wild-type (WT) and GluA3- or GluA4-KO mice. All EPSCs were evoked in an all-or-none manner by stimulating afferent fibres, consistent with the fact that each MNTB neuron is innervated at the soma by a single axon. A specific NMDAR blocker, MK-801 (10 μm), was used to isolate AMPAR-mediated EPSCs (AMPAR-EPSCs) by eliminating the residual NMDA component. Figure 1 shows examples of AMPAR-EPSCs recorded at the holding potential of −60 mV from GluA3- (Fig. 1A) or GluA4-KO synapses (Fig. 1F) in comparison with those from control mice. We found that deletion of the GluA3 gene did not significantly affect the amplitude (Fig. 1Ba, 9.04 ± 0.68 nA for GluA3-KO vs. 9.49 ± 0.82 nA for WT, n = 8 for each group, P > 0.05) or 10–90 rise time (Fig. 1Bb, 0.21 ± 0.005 ms for GluA3-KO vs. 0.20 ± 0.004 ms for WT, P > 0.05) of evoked AMPAR-EPSCs. When the decay time course of EPSCs was fitted with a single exponential function, we noted an increase in the decay time constant at GluA3-KO synapses (Fig. 1Bc 0.53 ± 0.02 ms for KO vs. 0.40 ± 0.02 ms for WT, P < 0.05). We further studied the voltage dependence of AMPARs by changing the membrane potentials from –60 to +60 mV (in 20 mV increments) and found that both sets of I–V relationships were inwardly rectifying (with 0.2 mm spermine in the intracellular solution). Neither the reversal potential nor the amplitude of AMPAR-EPSCs at most potentials were affected by deletion of GluA3, except for those at +40/60 mV (Fig. 1C and D). In contrast, when the same experiments were performed using GluA4-KO mice (Fig. 1F–J), we found a marked decrease in the amplitude of evoked AMPAR-EPSCs (WT: 7.0 ± 0.6 nA vs. GluA4-KO: 4.4 ± 0.3 nA; n = 6 vs. 11, P < 0.05) and a significant slowdown in their rise time (WT: 0.22 ± 0.01 ms vs. GluA4-KO: 0.27 ± 0.01 ms; n = 6 vs. 9, P < 0.05) and decay time course (WT: 0.40 ± 0.03 ms vs. GluA4-KO: 0.92 ± 0.07 ms; n = 7 vs. 8, P < 0.05). The dotted lines in Fig. 1A and F represent EPSCs from GluA3- or GluA4-KO mice normalized to the same amplitude of EPSCs from WT synapses to illustrate the kinetic change in AMPARs at KO synapses. Further investigation of the amplitude of AMPAR-EPSCs at a wide range of holding potentials (−60 to +60 mV in 20 mV increment) showed that the I–V relationship at GluA4-KO synapses remained inwardly rectifying with much more notable reduction at negative potentials. To quantitatively describe the inward rectification of AMPARs in mutant mice, we calculated the ratio of the amplitude of EPSCs at −40 mV to that at +40 mV and revealed a decrease in the inward rectification index at both GluA3- (Fig. 1E, WT: 9.68 ± 1.24, n = 7 vs. GluA3-KO: 4.99 ± 0.72, n = 8; P < 0.05) and GluA4-KO synapses (Fig. 1J, WT: 5.60 ± 0.49, n = 6 vs. GluA4-KO: 3.12 ± 0.31, n = 11; P < 0.05). However, this apparent decrease appears to have resulted from a net increase in the current amplitude at positive potentials for GluA3-KO synapses, but a net decrease in the amplitude of AMPAR-EPSCs at negative potentials for GluA4-KO synapses. Different origins of the reduced inward rectification indicate that a minor compensation of synaptic AMPARs by GluA2 for the loss of GluA3 may have taken place, but not for the loss of GluA4. In either case, I–V relationships remain inwardly rectifying, suggesting synaptic AMPARs are largely composed of Ca²⁺-permeable GluA subunits (Kamboj et al. 1995).

Given the prominent difference in the amplitude and kinetics of evoked EPSCs between WT and GluA4-KO synapses, we subsequently focused our experiments on GluA4-KO mice. As evoked EPSCs may be a result of asynchronous summation of unitary quantal events, we next examined spontaneous miniature EPSCs (mEPSCs) at –60 mV from GluA4-KO and WT synapses as shown in Fig. 2A. Consistent with the evoked responses, the amplitude of mEPSCs for KO mice decreased by more than half compared to that for control animals (63.0 ± 6.2 pA for WT vs. 28.0 ± 2.2 pA for KO, n = 7 for each group, P < 0.05, Fig. 2B). Analyses of the cumulative probability of inter-event intervals demonstrated a significant decrease in the frequency of spontaneous release in GluA4-KO synapses (Kolmogorov–Smirnov test, P < 0.0001, Fig. 2C), implying that there are too few receptors at the postsynaptic site to make quantal events detectable and/or there is a reduction in the presynaptic release of SVs in GluA4-KO synapses. When the average mEPSCs recorded from WT and GluA4-KO synapses were overlaid, we revealed that the rise time course of mEPSCs was significantly prolonged by deletion of GluA4 (10–90 rise time: 0.17 ± 0.01 ms for WT vs. 0.23 ± 0.01 ms for KO, P < 0.05, Fig. 2D and E). The decay time constant of mEPSCs in GluA4-KO synapses (0.67 ± 0.03 ms) was also greater than that in WT synapses (0.41 ± 0.02 ms). These surprising results suggest that slow rise/decay time of evoked AMPAR-EPSCs at GluA4-KO synapses is probably of postsynaptic origin at the unitary quantal level, in sharp contrast to the conventional view that such changes typically reflect desynchronized release events evoked by an action potential. In other words, alterations in the unitary quantal release at GluA4-KO synapses largely account for changes in the size and kinetics of evoked EPSCs. Taken together, these lines of evidence demonstrate that GluA4 subunits play a dominant role over GluA3 in determining the amplitude and temporal profile of both evoked and spontaneous EPSCs at this synapse.

Figure 2. Postsynaptic deficits of unitary quantal events in GluA4-KO synapses.

A, examples of spontaneous mEPSCs recorded at −60 mV in the presence of MK-801 (10 μm) from WT (grey) and GluA4-KO (black) synapses. Superimposed traces are shown in the right panels. B and C, summary of the amplitude (B) and cumulative probability of inter-event intervals (C) of mEPSCs obtained from WT (grey, n = 7) and GluA4-KO (black, n = 7) synapses. D, examples of average mEPSCs from a WT (grey) or GluA4-KO (black) synapse. Dotted line illustrates the slow kinetics of mEPSCs at GluA4-KO synapses by scaling the mEPSC amplitude to that of WT. E, quantifications of 10–90 rise time and decay time constants of mEPSCs from both groups. The decay time constants are measured by fitting the falling phase of average mEPSCs of each neuron with a single exponential function.

GluA4 is important for alleviating frequency-dependent short-term synaptic depression

With the establishment of prominent roles for GluA4 in regulating the size and shape of AMPAR-EPSCs, we next asked how deletion of the GluA4 gene affects high-frequency synaptic transmission. We recorded EPSCs elicited by high-frequency train stimulation at 100 Hz and 300 Hz (100 ms in length), and noted that short-term depression (STD) typically developed with such stimulation protocols in a frequency-dependent manner. However, both the extent and rate of synaptic depression were dramatically exacerbated in GluA4-KO synapses as compared with WT synapses (Fig. 3A and B). When we plotted the amplitude of AMPAR-EPSCs in each train against the stimulus time, we found a use-dependent reduction in the EPSC amplitude which followed a single exponential time course (Fig. 3C and D). At either stimulation frequency, the time constants (τ) derived from the exponential fittings for GluA4-KO synapses were evidently faster than those for WT synapses (100 Hz: τ = 17.4 ms for KO and τ = 50.2 ms for WT; 300 Hz: τ = 6.9 ms for KO and τ = 16.0 ms for WT, n = 6 for each group). Figure 3E and F further illustrate such differences in STD by plotting normalized amplitude of EPSCs to that of the initial response in each train.

Given the observation that GluA4-KO synapses exhibit much slower decay time constants in individual evoked AMPAR-EPSCs, it is conceivable that deletion of GluA4 leads to cumulative desensitization of AMPARs during repetitive activity, contributing to faster and more severe STD at these synapses (von Gersdorff & Borst, 2002). To test this possibility, we applied cyclothiazide (CTZ, 50 μm) to block AMPAR desensitization and measured the extent and time course of STD (Fig. 3A and B). At 300 Hz, for instance, we found CTZ had very little effect on the amplitude of EPSCs at the end of stimulation trains in WT (1.18 ± 0.09 nA without CTZ vs. 1.17 ± 0.12 nA with CTZ) and GluA4-KO synapses (0.31 ± 0.07 nA without CTZ vs. 0.31 ± 0.05 nA with CTZ, Fig. 3H), but selectively retarded the development of STD at GluA4-KO ones, resulting in a slower time constant over train stimuli (τ = 7.0 ± 0.5 ms in control solution vs. τ = 10.5 ± 0.6 ms in CTZ, P < 0.05) without significantly affecting the WT synapses (control: τ = 16.4 ± 1.5 ms vs. CTZ: 21.1 ± 1.9 ms, P > 0.05; Fig. 3G). We interpret these observations as follows: the extent of synaptic depression is largely determined by presynaptic mechanisms (i.e. the size of the readily releasable pool of synaptic vesicles or RRP) (von Gersdorff & Matthews, 1997; Wang & Kaczmarek, 1998; Joshi & Wang, 2002; Taschenberger et al. 2002); however, cumulative desensitization is reflected by the time constants of STD during repetitive activity. Our results showing exacerbated STD in GluA4-KO synapses suggest that GluA4 subunits are particularly important for maintaining robust synaptic responses by minimizing desensitization during high-frequency neurotransmission.

Since significant differences in the extent and rate of STD remained between WT and KO synapses even after blocking desensitization with CTZ, we postulated that deletion of postsynaptic GluA4 may have affected presynaptic quantal parameters. We first measured the total quantal output from WT and GluA4-KO synapses by cumulative plots of AMPAR-EPSCs evoked by 100 ms trains of stimuli at 300 Hz, based on the model by Wesseling & Lo (2002) (see details in Method). The estimations showed that the total quantal output of GluA4-KO synapses decreased significantly (from 47.99 ± 2.04 nA for WT to 17.43 ± 2.43 nA for KO, n = 6 for both groups, P < 0.05, Fig. 4A and B). As postsynaptic receptor desensitization could lead to underestimation of the quantal output, we repeated the measurement in the presence of the desensitization blocker CTZ (50 μm). We found that the dramatic difference in the total quantal output between WT and GluA4-KO synapses remained although the absolute estimations for both WT and KO groups were larger (Fig. 4B, WT: 56.14 ± 2.19 nA, n = 6 vs. KO: 24.83 ± 3.60 nA, n = 6, P < 0.05). However, when the size of the total quantal output was normalized by the amplitude of mEPSCs, we revealed that the number of total SVs or RRP was comparable (WT synapses, 1048 SVs vs. GluA4-KO, 909 SVs). In contrast, we found the release probability (P_r), calculated by dividing the total quantal output by the first EPSC in a train, increased substantially from 0.15 ± 0.01 in WT synapses to 0.26 ± 0.01 in GluA4-KO synapses (P < 0.05), in line with the observation that only depression of EPSCs from GluA4-KO synapses was seen, while very little depression (or even slight facilitation in some cases) often preceded depression in evoked AMPAR-EPSCs from WT synapses in response to train stimuli (Fig. 3A and B). Such a difference in P_r was further illustrated in the paired-pulse ratio (PPR) at various intervals (i.e. the ratio of the amplitude of the second AMPAR-EPSC to that of the first). We found either in the presence or absence of CTZ, GluA4-KO synapses displayed a remarkable decrease in the PPR at all the time intervals (Fig. 4C and D). These results suggest that postsynaptic deficits due to lack of GluA4 subunits may not only affect the response of postsynaptic AMPARs, but also have an impact on presynaptic functions, probably via crosstalk between pre- and postsynaptic elements (see Discussion).

Figure 4. Presynaptic contribution to severe synaptic depression at GluA4-KO synapses.

A, cumulative AMPAR-EPSCs evoked by 300 Hz stimulation as a function of stimulus time for a WT (grey) and GluA4-KO (black) synapse are estimated with the model proposed by Wesseling & Lo (2002). The size of cumulative current from the readily releasable pool (RRP) is calculated through subtraction of the currents contributed by the replenished vesicle pool (RVP) from the total current at the end of the steady state of release during the train. B, the average estimations of the current generated from the RRP for WT and GluA4-KO synapses in the absence and presence of CTZ (50 μm) with axonal stimulation at 300 Hz, showing a significant reduction in the RRP at GluA4-KO synapses. C and D, paired-pulse ratios (PPRs), calculated by dividing the amplitude of 2nd EPSC by that of 1st EPSC at varied intervals between these two pulses, are summarized for WT and GluA4-KO neurons with (D) or without CTZ (C) (n = 6 for all groups). The example traces recorded at the time interval of 3.3 ms are illustrated. Note elimination of GluA4 substantially decreases the PPRs regardless of the stimulation interval or presence of CTZ.

GluA4-AMPARs are required for driving high-fidelity postsynaptic spiking

To further compare the physiological significance of GluA3 and GluA4 in response to high-frequency inputs, we recorded the action potentials (APs) from postsynaptic MNTB neurons in current-clamp configuration by applying 100 ms train stimuli at 100 Hz or 300 Hz to presynaptic axons. These experiments were done in the extracellular solution containing 1 mm calcium ([Ca²⁺]_o) to minimize saturation and desensitization of AMPARs (von Gersdorff & Borst, 2002) and MK-801 (10 μm) to block residual NMDARs. Figure 5A contrasts such two recordings from GluA3-KO and WT synapses in response to afferent stimulation at 300 Hz. In both groups of synapses, postsynaptic neurons usually locked their spike firing with inputs even at the very high frequency, in spite of some use-dependent reduction in their amplitude (Fig. 5B, failure rate: 1.3 ± 1.3% for WT vs. 0.7 ± 0.7% for GluA3-KO, n = 5 for each group, P > 0.05). In contrast, at GluA4-KO synapses we observed no failures at 100 Hz but a significant increase in the number of AP failures during the later phase of trains of EPSCs at 300 Hz. The average failure rate for the GluA4-KO group (24.0 ± 5.1%, n = 9) was 10 times higher than that for the control group (2.4 ± 1.5%, n = 5) (Fig. 5C and D). These failures at GluA4-KO synapses are probably caused by compound deficits including reduced amplitude and slowed decay of depolarizing synaptic conductance, with the former lessening the synaptic drive for spiking, and the latter impeding effective repolarization. Thus, GluA4-AMPARs are crucial for specifically driving high-frequency, but not low-frequency, firing.

An alternative possibility for the decline in the fidelity of neurotransmission in GluA4-KO synapses is that the intrinsic excitability of postsynaptic MNTB neurons could be affected by knocking out the GluA4 gene. To address this issue, we directly injected a series of current steps into MNTB neurons at the membrane potential of –70 mV (Fig. 5E). In both WT and GluA4-KO neurons, supra-threshold current injections usually generated repetitive spike discharges near the beginning part of current steps (Fig. 5E). We measured steady-state potentials within the last 5 ms of each step and plotted them against the amplitude of injected currents (Fig. 5F, top panel). We found that the I–V relationships for WT and GluA4-KO synapses virtually overlapped, suggesting that the steady-state input resistance of postsynaptic neurons was not affected by deletion of GluA4. In addition, plotting the number of evoked spikes at each current intensity revealed no significant difference in the maximal number of spikes between GluA4-KO (15.3 ± 3.2, n = 5) and WT (17.1 ± 3.7, n = 7) neurons. On the contrary, we noted a significant increase in the output gain of the input–output relationship in the GluA4-KO group, when fitted with a Boltzmann function, yielding Vmid (the depolarization current needed to produce half of the maximal number of spikes) of 0.70 and 0.38 nA, and slope factor of 0.16 and 0.06 for WT and GluA4-KO, respectively (Fig. 5F, bottom panel). This result indicates that GluA4-KO neurons are, in fact, more excitable than WT neurons at any given supra-threshold current. It is possible that deletion of GluA4 reduces synaptic drive for spiking, but MNTB neurons may have compensated for this deficit by increasing their intrinsic excitability and lowering the threshold for spiking. Hence, postsynaptic spike failures in response to high-frequency afferent inputs at GluA4-KO synapses cannot be attributed to any reduction in excitability but instead are accounted for by a deficiency in synaptic drive.

To demonstrate the role of GluA4 subunits in high-frequency firing independent of presynaptic factors, we performed functional rescue experiments in which postsynaptic neurons from GluA4-KO synapses were injected with realistic currents that are reminiscent of synaptic inputs at WT synapses. Such experiments completely uncoupled presynaptic mechanisms from postsynaptic ones as the currents were directly fed via patch electrodes into MNTB neurons at the soma. We first took two recordings of EPSCs in trains (300 Hz, 100 ms, 2 mm[Ca²⁺]_o) obtained from GluA4-KO and control synapses, and reversed the polarity of these currents to make these traces templates for current-clamp experiments. Subsequently, we evoked spikes from GluA4-KO neurons with both current templates and measured the number of elicited APs during the train. In GluA4-KO neurons, we found that injection of the KO template only produced a few APs in the early phase of the responses followed by multiple spike failures (Fig. 6A and B, failure rate = 52.86 ± 5.11% for KO, n = 5), similar to the firing pattern of postsynaptic neurons driven by afferent stimulation (Fig. 5C). In contrast, the WT template evoked high-frequency spikes with few failures in the same neurons (Fig. 6B, failure rate = 1.43 ± 0.99%, n = 5). When the same experiments were repeated in WT neurons, we found very similar results as shown in Fig. 6C, except that the KO template evoked an even fewer number of spikes (failure rate = 70.5 ± 6.10%, n = 7), in line with the result that WT neurons are less excitable than GluA4-KO neurons (Fig. 5F). When the KO template was scaled up to the same amplitude of the first peak in the WT template and used to generate APs at both WT and GluA4-KO neurons, it failed to rescue the firing deficits, suggesting not only the initial amplitude but also the steady-state amplitude of synaptic currents are important for faithfully triggering sustained postsynaptic spiking (Fig. 6B and C, failure rate produced by the scaled-KO template: 42.86 ± 4.96%, n = 7 for GluA4-KO neurons vs. 59.17 ± 6.29%, n = 4 for WT neurons). Collectively, these results further reinforced our interpretation that spike fidelity during high-frequency neurotransmission depends on the fast, robust and sustainable synaptic drive provided by GluA4-AMPARs, independent of differences in the intrinsic excitability of postsynaptic neurons between WT and GluA4-KO synapses.

Deletion of GluA4 impairs sound-evoked auditory brainstem responses in vivo

Knowing the critical role of the calyx of Held synapse in sound localization, we rationalized that tremendous deficits in high-frequency neurotransmission at GluA4-KO synapses, as shown from in vitro experiments, may result in compromised ability of the auditory brainstem to respond to sound stimuli in vivo. To this end, we recorded ABRs from anaesthetized WT (n = 19) and GluA4-KO mice (n = 12). As in other mature mice, ABR waveforms for WT and GluA4-KO mice consisted of multiple successive peaks occurring approximately 2–7 ms after stimulus onset, including stimulus air conduction time from the speaker to the ear (Fig. 7A). The neural origin of wave 1 and 2 is well established to be associated with synchronized neural firing of ipsilateral VIIIth nerve and the cochlear nucleus, respectively, while wave 3 and 4 are generated by both contralateral and ipsilateral spiking activity of downstream superior olivary complex including the calyx of Held MNTB synapse. Threshold was defined as the stimulus level that produced a response (any wave) with 2 standard deviations above the background noise. Variance analysis showed no significant difference in the ABR thresholds between WT and GluA4-KO mice at low and high frequencies (P > 0.05; Fig. 7B), suggesting GluA4-KO animals do not have hearing loss.

Figure 7. Deficits in sound-evoked ABRs in GluA4-KO mice.

A, representative ABRs from both WT (grey) and GluA4-KO (black) mice in response to 94 dB stimuli presented at the frequency of 50 s⁻¹, showing the abnormalities of ABR waveform in GluA4-KO mice. B, comparison of ABR-derived thresholds for tones and clicks recorded from WT (grey) and GluA4-KO (black) mice, indicating that the GluA4-KO animals did not have a hearing loss. C, summary plot of the amplitude of ABRs obtained from WT (grey) and GluA4-KO (black) mice. Note that ABR waves 3 and 4 are significantly smaller in size in GluA4-KO mice, particularly at fast stimulus presentation rates (50 s⁻¹).

A visual inspection of ABRs measured for click stimuli indicated that waves 3 and 4 of ABRs in GluA4-KO mice were much smaller and slightly delayed in time compared to WTs, while waves 1 and 2 appeared to be normal (Fig. 7A). Thus, we measured individual wave amplitudes for ABRs evoked by 94 dB click stimuli presented at rates of 10 s⁻¹, 30 s⁻¹ and 50 s⁻¹ to quantify the differences in ABR waveform between WT and KO animals. Click stimuli briefly presented at high sound levels (for instance, 94 dB) evoked the clearest ABR waveforms with good signal to noise ratios and therefore were used in this study. It should be noted that prolonged exposure to a constant stimulus at high levels could produce hearing loss but our stimuli were very brief and repeated for ∼1 min at a time. As expected from synchronized spiking activity of different subsets of neurons along the brainstem pathway, more synchronous waves should have larger amplitudes and shorter peak latencies (Hall, 1992). Variance analysis (at the rate of 50 s⁻¹) revealed a significant decrease in the amplitude of waves 3 and 4 in GluA4-KO animals as compared to WT ones (wave 3: 3.94 ± 0.27 μV for KO vs. 5.53 ± 0.34 μV for WT; wave 4: 2.32 ± 0.30 μV for KO vs. 4.24 ± 0.40 μV for WT; n = 12 vs. 19; P < 0.001, Fig. 7C). Measurements of ABR wave latency also demonstrated that the latencies of waves 3 and 4 were significantly longer in GluA4-KO than in WT animals (wave 3: 4.64 ± 0.04 ms for KO vs. 4.40 ± 0.05 ms for WT; wave 4: 5.59 ± 0.04 ms for KO vs. 5.40 ± 0.04 ms for WT; n = 12 vs. 19; P < 0.005). Stimulus presentation rates affected both wave amplitudes (P < 0.0001) and latencies (P < 0.0001) for both groups of mice. ABR wave amplitudes decreased while latencies increased with rising stimulus rates in GluA4-KO compared to WT mice (data not shown). Although GluA4 subunits are known to be abundantly expressed in different stations along the auditory pathway, selective attenuation of waves 3 and 4, but not waves 1 and 2, in Glu4-KO mice indicates that GluA4 subunits are dominant and indispensable for synaptic AMPARs to drive synchronized firing in specific subsets of neurons, particularly the calyx of Held MNTB synapse and others downstream to the superior olivary complex during high-frequency sound presentation.

Discussion

By studying synaptic transmission at the calyx of Held synapse and auditory function in genetically knockout mice, we have provided compelling evidence that GluA4, but not GluA3, subunits are the main determinant for fast and robust synaptic transmission mediated by AMPARs. Deletion of GluA4 leads to prominent deficits in the magnitude and kinetics of synaptic responses as well as the coupling of excitatory inputs to postsynaptic spiking during high-frequency neurotransmission. Such cellular deficiencies probably underlie abnormalities at system levels, as evidenced by phenotypic deficits in the specific ABR components that reflect synchronized spiking of neurons in MNTB and other downstream nuclei in the superior olivary complex when acoustic signals are presented. Although GluA3 subunits may make a minor contribution to synaptic AMPARs in WT synapses, our results clearly demonstrated the indispensable role of GluA4 in gating fast neurotransmission that is essential for precisely preserving temporal information in auditory brainstem circuits.

We and others have previously shown that there is a significant increase in the amplitude of evoked EPSCs and acceleration in their decay kinetics in the developing calyx of Held synapse (Taschenberger & von Gersdorff, 2000; Iwasaki & Takahashi, 2001; Joshi & Wang, 2002; Joshi et al. 2004; Koike-Tani et al. 2005). On the basis of immunohistochemistry and fast perfusion experiments on outside-out patches from mouse MNTB neurons as well as single-cell RT-PCR data in the literature (Geiger et al. 1995; Joshi et al. 2004; Koike-Tani et al. 2005), we proposed that a subunit switch from slowly gating GluA1 to fast-gating GluA3/4 subunits underlies the apparent changes in evoked EPSCs, while GluA2 remains virtually absent throughout the maturation process (Joshi et al. 2004; Koike-Tani et al. 2005, 2008). Furthermore, single-cell RT-PCR analyses of GluA mRNAs from rat MNTB neurons showed that GluA1–4 transcripts all exist but GluA4-flop transcripts are dominant and developmentally upregulated (Koike-Tani et al. 2005, 2008). However, these observations were confounded by findings using a subtype-specific immunogold labelling approach (Hermida et al. 2006), in that GluA2/3 were found to be predominant over GluA4 at synaptic domains of mature rat MNTB neurons, implying a large number of GluA4 transcripts may not be necessarily translated or transported to the synaptic sites. Our current study clearly disputes this view and has provided direct and explicit evidence that functional receptors composed mainly of GluA4 subunits are targeted to synaptic sites and govern synaptic strength.

Given clear cellular and functional phenotypes of GluA4-KO mice, we suggest that GluA4 is the main molecular determinant of native synaptic AMPARs, and its predominant roles cannot be compensated by other GluA subunits as it shares little redundancy with them. This is supported by the fact that GluA3-flop subunits, which can also form remarkably fast-gating AMPARs, fail to boost the amplitude and accelerate the decay time course of AMPAR-EPSCs in GluA4-KO synapses. Synaptic currents at GluA4-KO synapses remain much smaller in size than those at WT synapses and are inwardly rectifying, indicating an inability of GluA2 to replace GluA4 as otherwise one would expect a linearized I–V relationship in the presence of spermine (Bowie & Mayer, 1995; Donevan & Rogawski, 1995; Kamboj et al. 1995). In contrast, AMPAR-EPSCs in GluA3-KO synapses show only marginal changes in synaptic properties, which appear to be partially compensated by GluA2 as shown by the attenuated inward rectification. Although it may be argued that native AMPARs are composed of heteromeric GluA3 and GluA4 subunits and deletion of GluA4 prevents GluA3 from insertion to the synaptic sites, our observations from GluA3-KO synapses (Figs 1 and 5) indicate that GluA3 is unlikely to be a major contributor to synaptic AMPARs and readily compensated by GluA4 if deleted. Other accessory proteins such as TARPs, cornichon and CKAMP44 are unlikely to be involved in gating native synaptic AMPARs at the calyx of Held synapse because these proteins, when co-assembled with GluAs, would be expected to reduce inward rectification and/or slow down the gating kinetics of AMPARs (Tomita et al. 2005; Schwenk et al. 2009; von Engelhardt et al. 2010). This was clearly not the case as AMPAR-EPSCs display robust inward rectification while their decay kinetics accelerate throughout development of WT mice (Joshi et al. 2004). In fact, the characteristics of AMPAR-EPSCs from GluA4-KO synapses are reminiscent of those from immature WT synapses where AMPARs are dominated by slowly gating GluA1 subunits (Joshi & Wang, 2002; Joshi et al. 2004; Koike-Tani et al. 2005), again reinforcing the notion that GluA4 is a crucial substrate for the synapse to make a developmental gating switch, and once removed, it cannot be readily substituted by other GluAs or accessory proteins.

Although GluA4 subunits are thought to be exclusively localized to postsynaptic MNTB neurons (Hermida et al. 2006), it is surprising that deletion of GluA4 has also affected presynaptic properties including P_r and depletion of SVs. Takago et al. (2005) reported that there are low levels of presynaptic AMPARs in immature calyces. These receptors can be activated by exogenous agonists, leading to an inhibition of voltage-gated Ca²⁺ currents via G-protein-coupled signalling and consequently attenuation of neurotransmitter release. Deletion of GluA4 would be expected to eliminate the effect of presynaptic inhibition (if any) and therefore enhance the strength of synaptic transmission. This is opposite to what we observed in GluA4-KO synapses, which show dramatic reduction in the amplitude of AMPAR-EPSCs, suggesting other mechanisms may account for the presynaptic changes. Previous work has shown that the relative abundance of GluA4 at different postsynaptic densities on the soma of the same cell is influenced by presynaptic inputs (Rubio & Wenthold, 1997), suggesting pre- and postsynaptic elements operate as a unit in which extensive crosstalk takes place in an activity-dependent manner. In fact, emerging evidence suggests activity-dependent clustering of glutamate receptors is through interaction between AMPARs and trans-synaptic molecules, such as n-cadherin, neurexin–neuroligin, Narps and NP1 (Song & Huganir, 2002; Gerrow & El-Husseini, 2007; Tai et al. 2008; Zhou et al. 2011). In line with this model, elimination of GluA4 resulted in slower rise/decay time course at the unitary quantal level, suggesting a change in the geometrical distribution of synaptic AMPARs which may not be aligned to the release sites and cause slower response to unitary quantal release (Fig. 2). Recent investigation with cultured neurons have led to the proposal that the trans-synaptic signalling can reversely modulate presynaptic properties (Ripley et al. 2011). Our current study has provided the first evidence from the native central synapse that AMPARs are critically involved in regulating P_r and short-term depression (Figs 3 and 4). We postulate that Ca²⁺-permeable AMPARs such as GluA4 may serve as ideal channels to generate activity-dependent Ca²⁺ transients for driving Ca²⁺-dependent transcription and post-translational regulation of receptors and their trafficking in postsynaptic sites as well as feedback control of presynaptic release machinery. Future experiments are required to address if and how a lack of GluA4 in postsynaptic MNTB neurons impairs trans-synaptic signalling cascade important for presynaptic functions.

Deletion of GluA4 does not affect ABR thresholds in mice, indicating that the animals do not exhibit a hearing loss in the classic sense. The fact that waves 1 and 2 remain intact in the GluA4-KO animals can be interpreted as follows: neurotransmission along the auditory pathway upstream to the calyx of Held MNTB synapses is not significantly affected by deletion of GluA4 or readily compensated by other GluAs. Specifically, wave 1 of ABRs, which reflects the activity of the VIIIth nerve, is normal in GluA4-KO mice, suggesting that there is unlikelyto be any abnormal inputs from the VIIIth nerve to the cochlear nucleus. However, wave 2 of ABRs is a compound response from different types of neurons in the cochlear nucleus, some of which may use GluA3/4-dominant AMPARs while others employ AMPARs containing GluA2 subunits (Gardner et al. 1999, 2001). Hence, little change in wave 2 of ABRs in GluA4-KO mice does not rule out the possibility that deficits or abnormal compensation in subsets of cochlear neurons, particularly globular bushy cells which give rise to the calyx of Held and innervate MNTB neurons, could contribute to the phenotypic deficits in later waves that we observed from GluA4-KO mice. In contrast, ABR components, wave 3 and 4, in GluA4-KO animals are clearly degraded, as reflected by smaller amplitudes and longer latencies, and these deficits were further exacerbated when the stimulus presentation rate was increased. Consistent with the observation that action potential failures in MNTB neurons are particularly evident at high stimulation frequency in vitro, these in vivo deficits from GluA4-KO mice are at least partially, if not fully, accounted for by reduced spiking synchrony in the calyx of Held MNTB synapses and other downstream subgroups of neurons in the superior olivary nuclei along the critical auditory brainstem timing pathway (Henry, 1979; Achor & Starr, 1980; Wada & Starr, 1983a,b,c; Melcher & Kiang, 1996; Melcher et al. 1996). Perceptual abilities that depend on the accurate encoding of temporal information in an acoustic signal include speech perception, sound localization in the horizontal plane, and auditory scene analysis. Indeed, human listeners with auditory dys-synchrony show deficits in temporal processing and speech perception (Zeng et al. 1999; Kraus et al. 2000; Zeng & Liu, 2006). Our results in this study raise the possibility that the GluA4 gene is potentially a candidate for genetic screening of human mutations underlying human auditory deficits.

In conclusion, we have demonstrated that GluA4 subunits are indispensable for preserving the fidelity of high-frequency neurotransmission in vitro and temporal conduction of high-frequency sound in vivo. The most notable contributions of GluA4 subunits include boosting the size of both evoked EPSCs and mEPSCs, speeding up their decay kinetics and attenuating short-term synaptic depression during high-frequency neural activity. The fidelity of postsynaptic spiking critically depends on brief but robust synaptic drive mediated by GluA4. Rapid deactivation of GluA4-AMPARs helps lower the build-up of plateau potential and shorten the refractory period of APs, while fast recovery from desensitization of GluA4-AMPARs alleviates cumulative desensitization to minimize synaptic depression and strengthen synaptic potentials during repetitive activity (Joshi et al. 2004). Therefore, GluA4 is of vital importance for providing rapid and robust synaptic response, ultimately ensuring accurate conveyance of temporal codes required for the functionality at this auditory synapse and perhaps other fast central synapses (Silver et al. 1996; Eliasof & Jahr, 1997; Geiger et al. 1997).

Glossary

Abbreviations

ABR

auditory brainstem response

ITD

interaural timing difference

ILD

interaural level difference

KO

knockout

MNTB

medial nucleus of the trapezoid body

RRP

readily releasable pool

STD

short-term depression

SV

synaptic vesicle

Author contributions

Y.-M.Y. designed and performed the in vitro electrophysiological recordings, analysed the data and wrote the manuscript. J.A. contributed to some of the recordings and maintained the colonies of knockout animals. A.M.L. performed in vivo electrophysiological recordings of ABRs. M.N., K.T. and R.L.H. created GluA4-KO mice and Z.J. generated GluA3-KO animals. L.-Y.W., R.L.H. and B.J.M. supervised this study and edited the manuscript. All authors approved the final version.