GABA_A receptors contribute more to rate than temporal coding in the IC of awake mice

Abstract

Speech is our most important form of communication, yet we have a poor understanding of how communication sounds are processed by the brain. Mice make great model organisms to study neural processing of communication sounds because of their rich repertoire of social vocalizations and because they have brain structures analogous to humans, such as the auditory midbrain nucleus inferior colliculus (IC). Although the combined roles of GABAergic and glycinergic inhibition on vocalization selectivity in the IC have been studied to a limited degree, the discrete contributions of GABAergic inhibition have only rarely been examined. In this study, we examined how GABAergic inhibition contributes to shaping responses to pure tones as well as selectivity to complex sounds in the IC of awake mice. In our set of long-latency neurons, we found that GABAergic inhibition extends the evoked firing rate range of IC neurons by lowering the baseline firing rate but maintaining the highest probability of firing rate. GABAergic inhibition also prevented IC neurons from bursting in a spontaneous state. Finally, we found that although GABAergic inhibition shaped the spectrotemporal response to vocalizations in a nonlinear fashion, it did not affect the neural code needed to discriminate vocalizations, based either on spiking patterns or on firing rate. Overall, our results emphasize that even if GABAergic inhibition generally decreases the firing rate, it does so while maintaining or extending the abilities of neurons in the IC to code the wide variety of sounds that mammals are exposed to in their daily lives.

NEW & NOTEWORTHY GABAergic inhibition adds nonlinearity to neuronal response curves. This increases the neuronal range of evoked firing rate by reducing baseline firing. GABAergic inhibition prevents bursting responses from neurons in a spontaneous state, reducing noise in the temporal coding of the neuron. This could result in improved signal transmission to the cortex.

INTRODUCTION

Many animals, including humans, use acoustically complex vocalizations to convey information to members of their own species. Thus the auditory systems of animals have evolved to detect and process these salient signals. The neural mechanisms underlying the auditory system’s ability to discriminate among various vocalizations are not clearly understood. The auditory midbrain, however, is thought to be of particular importance for encoding vocalizations. In the midbrain, the inferior colliculus (IC) is an obligatory station in the ascending auditory pathway that integrates ascending, descending, and intrinsic excitatory and inhibitory connections (Winer and Schreiner 2005). Individual neurons in the IC selectively respond to features of vocalizations in a variety of species (Andoni et al. 2007; Holmstrom et al. 2010; Klug et al. 2002; Schneider and Woolley 2010; Šuta et al. 2003; Woolley and Portfors 2013). On one hand, selectivity to vocalizations in the IC relies on the absence of response to some vocalizations (even if their frequency content matches that of the neuron’s excitatory tuning curve) and on the variability of firing rate with the vocalization, both of which create specific spatial patterns of activation that differ from call to call (Klug et al. 2002). It has also been shown at the thalamocortical level that spike timings and patterns could provide a more efficient code for discriminating vocalizations at the single-cell scale (Huetz et al. 2009). There has been less focus on how such a temporal code is involved in response selectivity to complex sounds in IC neurons. Moreover, the mechanisms of selectivity to vocalizations rely on a microcircuitry that is not well understood in the IC.

This selectivity to vocalizations builds on the many modulatory serotonergic and dopaminergic inputs to the IC as well as the excitatory glutamatergic and inhibitory GABAergic and glycinergic projections (Winer and Schreiner 2005). In particular, inhibition plays a critical role in shaping selectivity in the IC for simple sounds (Burger and Pollak 1998; Pollak and Park 1993; Yang et al. 1992) and complex sounds, including vocalizations (Dimitrov et al. 2014; Mayko et al. 2012; Pollak 2013; Xie et al. 2005). However, the discrete contribution of GABAergic inhibition remains unclear. GABA is thought to play a pivotal role in the neural code of the IC, as it is present at levels among the highest found anywhere in the brain and because virtually every IC neuron receives significant GABAergic input, either from inside the IC or from massive ascending inhibitory inputs originating from the superior olivary complex (SOC) and the lateral lemniscus nucleus (LLN) (see reviews in Caspary et al. 1997; Ono and Ito 2018). Furthermore, most of the studies examining the role GABAergic inhibition plays in encoding sounds in the IC have focused primarily on the firing rate of the neurons and not on their spike timings and patterns.

We therefore investigated in detail how GABAergic inhibition in the IC affects responses of individual neurons to a variety of vocalizations, in particular examining these effects on both the rate and timing characteristics of the neurons. We used piggyback electrodes to iontophoretically apply GABA_A receptor (GABA_AR) antagonists onto single units in the IC of awake restrained mice while recording their responses to pure tones and mouse vocalizations. Previous studies have shown that some IC neurons respond to high-frequency vocalizations not matching the neuron’s frequency tuning, likely because of distortion products generated in the cochlea (Portfors et al. 2009; Portfors and Roberts 2014). Therefore, we used simulations of this distorted product in addition to the natural vocalizations emitted by mice to explicitly evaluate the impact of GABAergic inhibition on the components induced by nonlinear processing of the cochlea. Overall, we found that activating GABA_AR increased the range of evoked firing rate of IC neurons and reduced the spontaneous bursting behavior of neurons. Quite paradoxically, these two effects and the presence of GABAergic inhibition in general did not lead to improved vocalization-discriminating abilities of IC neurons.

METHODS

Subjects.

Subjects were CBA/CaJ mice between 2 and 12 mo old. This age range shows no age-related hearing loss that could affect the results of this study (Kobrina and Dent 2016; Ohlemiller et al. 2010; Zheng et al. 1999). Animals were group housed with same-sex littermates in standard acrylic cages until the day of surgery. Food and water were provided ad libitum, and lights were maintained on a reversed 12:12-h light-dark schedule so that all experiments were done during their active, dark period. All animal care and experimental procedures followed the guidelines of the National Institutes of Health and were reviewed and approved by the Washington State University Institutional Animal Use and Care Committee.

Animal preparation.

The surgical procedures used in this study have previously been described in detail (Muniak et al. 2012). Briefly, mice were anesthetized with isoflurane inhalation in order to mount a headpin onto the skull to stabilize the animal’s head during electrophysiological recordings. A midline incision was made on the scalp, and the skin was reflected laterally. A hollow metal head post was cemented to the skull over bregma, and a tungsten ground electrode was set into the right cerebral cortex. Using stereotaxic coordinates, we made a 1-mm² craniotomy over the left IC. At the end of surgery, the craniotomy was covered with bone wax. We applied lidocaine gel and triple antibiotic ointment to the exposed tissue and administered ketoprofen (0.05 mg/kg) for postsurgery analgesia. Experiments were conducted at least 1 day after surgery.

Electrophysiological recordings and drug application.

Electrophysiological experiments were performed in a dark, single-walled sound attenuation chamber. At the beginning of each experiment, animals were lightly sedated with acepromazine (5 mg/kg) to facilitate them being restrained in foam shaped to their body. The foam restraint was attached to a stereotaxic frame, and the mouse’s head was stabilized by screwing its head post into the stereotaxic apparatus. Acoustic stimulation was controlled from outside the sound-attenuating chamber. Experiments typically lasted between 1 and 4 h but never exceeded 5 h/day. The animal was immediately removed if it showed signs of discomfort or distress. Each animal was used for 1–3 experiment days. Between the recording sessions, exposed brain tissue was covered with a small amount of petroleum jelly and bone wax, lidocaine gel was applied to exposed muscles, and the mouse was housed individually.

Extracellular responses from well-isolated single units in the left IC were recorded with single-barrel glass pipette electrodes (A-M Systems, Sequim, WA) filled with 1 M NaCl (20–40 MΩ) and mounted onto a triple-barrel glass pipette electrode with its tip extending 10–25 µm from the end for microiontophoretic application of a GABA_AR antagonist. The tip of the triple-barrel electrode was broken to a diameter of ~30 µm. Two of the three barrels were filled with the GABA_AR antagonist gabazine (20 mM, pH 3.0, deionized water vehicle; Sigma), and the third barrel, acting as a sum channel, was filled with 1 M NaCl. Silver wires were inserted into all three barrels and connected to a microiontophoresis current generator (model 650; David Kopf Instruments, Tujunga, CA) to control the retention and ejection currents. We used iontophoresis currents for drug retention and application similar to those in previous studies [retain –15 nA; ejection range +10 nA to +40 nA (Mayko et al. 2012)]. We used the GABA_AR antagonist gabazine instead of the more frequently used antagonist bicuculline because of the strong nonspecific effects on calcium-dependent potassium channels that are present in the IC (Kelly and Caspary 2005; Kurt et al. 2006).

Electrodes were advanced with a hydraulic micropositioner (David Kopf Instruments, Tujunga, CA) located outside the acoustic chamber. Electrode penetrations were made dorso-ventrally through the central nucleus of the IC (ICC). Extracellular action potentials were amplified (model 2400; Dagan, Minneapolis, MN), filtered (band pass, 500–6,000 Hz; model 3364; Krohn-Hite, Brockton, MA), and sent through a spike enhancer (Frederick Haer, Bowdoin, ME) before being digitized (Microstar Laboratories, Bellevue, WA; 10,000 samples/s). Individual waveforms from neurons were displayed and saved for off-line analysis using custom-written software. Online analysis included visualization of raster plots and peristimulus time histograms.

Stimulus delivery.

Stimuli were output through a high-speed 16-bit digital-to-analog converter (Microstar Laboratories; 400,000 samples/s) and fed to a programmable attenuator (PA5; Tucker-Davis Technologies, Alachua, FL), a power amplifier (Parasound, San Francisco, CA), and a free-field speaker [EMIT high-energy speaker (Infinity Systems) or LCY-K100 Supreme Tweeter (Ying Tai Trading)]. Sounds were presented at 45° and 10 cm from the animal’s right ear. Speakers were regularly calibrated with a 1/4-in. calibrated condenser microphone (model 4135; Bruel and Kjaer, Denmark) placed at the position of the animal’s ear. For pure-tone stimuli produced by the EMIT speaker, sound pressure remained near constant from 6 to 50 kHz and there was a gradual decrease in sound pressure from 52 to 100 kHz of ~2.8 dB per 10 kHz. For pure-tone stimuli produced by the LCY-K100 Supreme Tweeter, increased voltage was applied on a frequency-specific basis to maintain a near-equal sound pressure level (SPL) across all frequencies. We observed no differences in neuronal response patterns between the two speakers. Harmonic component frequencies were at least 50 dB lower in intensity than the fundamental frequency.

Recording session.

Pure tones were used as search stimuli, focused on finding single units in the 10–50 kHz frequency representation of the ICC. Recordings were limited to neurons with response latencies typical of the ICC (Syka et al. 2000). When we isolated a single unit (signal-to-noise ratio > 4:1) we first collected data without applying any drugs (control condition).

Data collection began by audiovisual determination of a neuron’s characteristic frequency (CF) and minimum threshold (MT). We defined the CF as the frequency at which a neuron responded to at least 50% of the stimulus presentations at MT. MT was defined as the minimum intensity required to evoke a response to 50% of the stimuli at CF. We characterized the excitatory frequency response area (FRA) of each neuron by presenting pure tones (50-ms duration, 1-ms rise/fall time, 4/s, 10–20 repetitions) between 6 and 100 kHz in 2-kHz steps at three different intensities: 10 dB SPL below and 10 and 30 dB SPL above MT. We determined neuronal firing rates and temporal patterns by presenting a pure tone at CF and 10 dB SPL above threshold (50-ms duration, 1-ms rise/fall time, 4/s, 100 repetitions).

We then presented a suite of eight prerecorded mouse vocalizations of the “jump syllable” type (Mahrt et al. 2013). These vocalizations consist of multiple frequency components divided by jumps in frequency of 10–30 kHz. The vocalizations were selected from a catalog of over 21,000 male CBA/CaJ mouse “syllables” previously recorded when the males were interacting with a female mouse. All syllables were synthesized by using custom MATLAB code to remove background noise without changing natural frequency and amplitude modulations (Holmstrom et al. 2009, 2010). Vocalizations were presented with 1-ms rise/fall times at a rate of 4/s for 32 repetitions. Vocalizations were output so that the peak amplitude was the same across all vocalizations (~75–80 dB SPL), but relative amplitude relationships within each vocalization were maintained. These output intensities are the purported intensities at which mice naturally emit their vocalizations.

We finally presented a modified version of all jump syllables: previous studies have suggested that neurons with low CFs in the ICC of mice respond to high-frequency vocalizations because of distortion products generated in the cochlea (Portfors et al. 2009). To test this, we used the nonlinear mathematical model of the cochlea implemented by Portfors and Roberts (2014) and improved in Roberts and Portfors (2015) to preprocess the vocalization stimuli. The phenomenological cochlea model consists of two stages, a reverberation stage followed by a nonlinear amplification stage. This model creates a signal that more accurately represents stimulation of the auditory nerve. Typically, the calculated distortion products of these frequency jump syllables are narrowband, ranging between 10 and 25 kHz, and therefore match the mouse’s most sensitive hearing range at 10–30 kHz. In the following, we refer to these signals as “distorted.” Most neurons recorded in this study have CFs in the 5–20 kHz range, to stay outside natural vocalization frequency content while matching that of distorted vocalizations.

After the control stimulus protocol was complete, we iontophoretically ejected gabazine at low currents (e.g., 20 nA for gabazine). During drug application, firing rates were measured every 3 min by presenting a CF pure tone 10 dB SPL above threshold for 100 repetitions (as in control conditions) until there was a 25% increase in firing rate. If there was no increase in firing rate in the first 15 min, ejection currents were increased in 10-nA increments until maximum ejection current was reached (Mayko et al. 2012; Sanchez et al. 2008). Once the firing rate was stable for ∼2 min, we repeated the stimulus protocol exactly as in control conditions (drug condition). We then collected recovery data whenever possible to verify that changes in responses were due to drug application. We turned off the ejection current, and reapplied the retention current, for 1.5 times the ejection duration until the neuron had returned to within 25% of its control firing rate. We then repeated the stimulus protocol again. As the entire protocol for the control, drug, and recovery conditions took ~45 min, the signal-to-noise ratio of the neuronal recording often degraded such that obtaining full data in the recovery phase was not always possible.

Data analysis: quantification of responses to pure tones.

For each of the three intensity levels used, spectrotemporal receptive fields (STRFs) were obtained from single units by constructing poststimulus time histograms (PSTHs) for each frequency with 1-ms time bins. All spikes falling in the averaging time window (starting at stimulus onset and lasting 150 ms) were counted. Thus STRFs were matrices of 150 bins in x-axis (time) multiplied by 45–48 bins in y-axis (frequency). In the time dimension, they were smoothed with a uniform 3-bin window. The maximum firing rate for a given frequency is therefore computed on a 3-ms time window.

Peaks of significant response in STRFs were automatically identified with the following procedure: a positive peak in the STRF was defined as firing rates above the average level of the baseline activity plus 3 times the standard deviation of the baseline activity. The baseline activity was estimated from the first 9 ms of STRFs, which was a latency too short for containing evoked activities in our data. For each cell, we looked for the highest firing rate recorded across the STRFs associated with the three intensities tested, and it was called “max FR” in the study thereafter. An estimate of the range of evoked firing rate available to the cell was then computed as the log ratio (in dB) between the max FR and the baseline activity of the cell. Also, the frequency where the max FR was recorded defined the best frequency (BF) of the cell. Three measures were extracted from the peaks. First, the “first spike latency” was the latency of the shortest time where the peak at the BF was significant. Second, the “response duration” was the time difference between the first and last spikes of the significant peak at the BF. Finally, in the STRF corresponding to 30 dB SPL above threshold, the “bandwidth” parameter was defined as the maximum frequency range width (in octaves) where the peaks are significant.

Frequency response areas (FRAs) were formed by computing the maximum firing rate in a bin across time for each frequency and SPL. Thus FRAs were matrices of 48 bins in x-axis (frequency) multiplied by 3 bins in y-axis (SPLs). Two significance thresholds were defined for FRAs: 1) the average level of the baseline activity plus 10 times the standard deviation of the baseline activity and 2) half the maximum firing rate at the BF. FRAs were linearly interpolated by a factor of 2 in frequency and a factor of 41 in SPL so as to provide a more precise representation of FRAs. Although they were visually estimated before recordings, CF and MT values used in the study were more accurately computed from the linearly interpolated FRA. CF was defined as the lowest SPL (MT) where a point was significant.

Since none of the recorded neurons showed significant evoked response to pure tones above 49 kHz, we considered such responses as spontaneous activity in the following. In particular, the bursty behavior of cells was quantified by applying a nonparametric burst detector (Gourévitch and Eggermont 2007) on a time concatenation of all the trials associated to frequencies above 49 kHz.

Data analysis: quantification of responses to vocalizations.

Response duration to vocalizations was quantified as follows: PSTHs were first constructed as histograms of spikes across trials with a bin width of 1 ms and then smoothed by a rectangular 3-bin window. Thresholding was then applied to estimate response duration, using the same threshold as for STRFs. The temporal width of bin areas above threshold defined the response duration of a cell to a given vocalization. Maximum firing rate peak latency was also computed as a parameter of interest. Finally, to quantify the trial-to-trial reliability of neuronal responses to vocalizations, we used the spike-timing reliability coefficient (CorrCoef). This index corresponds to the normalized covariance between each pair of action potential trains recorded at presentation of a given vocalization and was calculated as follows:

$$\text{CorrCoef}=\frac{1}{N\left(N-1\right)}{\displaystyle \sum _{i=1}^{N-1}{\displaystyle \sum _{j=i+1}^N\frac{\text{σ}{x}_i{x}_j}{\text{σ}{x}_i\text{σ}{x}_j}}}$$

where N is the number of trials and σx_ix_j is the normalized covariance at zero lag between spike trains x_i and x_j, where i and j are the trial numbers. Spike trains x_i and x_j were previously convolved with a 1-ms-width Gaussian window. In a previous study (Huetz et al. 2009), it was shown that the CorrCoef was not influenced by fluctuations of firing rate.

To predict the PSTH that might be obtained in response to a given vocalization, the spectrogram of this vocalization was convolved by the STRFs for each frequency and then summed across frequencies to obtain an estimated firing rate expectancy as a function of time similar to a PSTH. This estimated PSTH was fitted to the actual PSTH in amplitude causing linear regression. The coefficient of determination (R²) was derived from this linear regression to obtain an estimate of the fitting accuracy between the linear prediction and the actual PSTH.

To quantify how well the vocalization discrimination can be inferred from the neuronal responses, we also estimated the mutual information (MI) between neural responses and stimuli. We used an indirect method (Nelken et al. 2005; Rolls et al. 1997; Schnupp et al. 2006) to build a confusion matrix and then to compute the amount of information (Shannon 1948) contained in the cortical responses to vocalizations. As this method is exhaustively described in Schnupp et al. (2006), we present only the main principles here.

The method relies on a pattern recognition algorithm that is designed to guess which stimulus evoked a particular response pattern by going through the following steps: for a given cell, a single response (test pattern) to a vocalization, say “syllable 94,” was extracted and represented as a PSTH with a 1-ms bin size. A mean response pattern was computed for each vocalization (excluding the test pattern for “syllable 94”). The test pattern was assigned to the vocalization of the closest mean response pattern (average squared difference between patterns, i.e., Euclidean distance). This operation was repeated for all the single responses available, generating a confusion matrix where each response was assigned to a given vocalization. From this confusion matrix, the MI was derived by the classical Shannon’s formula

$$\text{MI}={\displaystyle \sum _{x,y}p\left(x,y\right)\times {\text{log}}_2\left[\frac{p\left(x,y\right)}{p\left(x\right)\times p\left(y\right)}\right]}$$

where x and y are the rows and columns of the confusion matrix or, in other words, the values taken by the random variables “presented vocalization” and “assigned vocalization.” In a scenario where the responses would not carry any information, i.e., the assignments of each response to a mean response pattern would be equivalent to chance level (here 1/8 = 12.5% because we used 8 different vocalizations and each stimulus was presented the same number of times), the MI would be close to 0. At the other extreme, if all vocalizations were perfectly identified by a cell, the confusion matrix would be diagonal and the MI would tend to log₂(8) = 3 bits. In our case, whatever the vocalization duration, we selected the first 150 ms of the responses to evaluate MI, in order to use spike trains with the same duration. This procedure is for a spike timing-based code where binned response is taken into account. To account for a hypothetical firing rate-based code, we also applied the same methodology to the firing rate of a trial (instead of using the PSTH) to estimate the MI between firing rate (instead of spiking patterns) and stimuli.

Data reporting and statistical testing.

All statistical tests are Wilcoxon paired tests leading to a signed-rank statistics dubbed sr(n), where n is the data size, in the following. We used a significance level of P = 0.05 with Bonferroni correction for multiple tests when necessary.

RESULTS

Blocking GABA_A receptors decreases the range of evoked firing rates by increasing baseline rate.

We recorded 19 single neurons in the IC of 13 normal-hearing mice (7 females, 6 males) before and after pharmacologically blocking GABA_AR. Except when noted, all group results include those 19 cells. The most obvious effects of blocking GABA_AR are visible on both STRFs and FRAs of single neurons (see methods, Fig. 1A). First, the baseline activity (i.e., the spontaneous firing rate) clearly increased in the presence of GABA_AR blockers as shown by the more numerous spikes visible outside the significant peak area of STRFs or FRAs. This finding was confirmed by group results [Fig. 1Ci; Wilcoxon paired test (WPT), sr(19) = 12, P = 8.4e-4]. Second, the maximum firing rate of the neuron, which was obtained at the best frequency (BF), was unchanged when GABA_AR were blocked [Fig. 1E, center, WPT, sr(19) = 29.5, P = 0.29]. Thus the log ratio between the maximum firing rate available to the neuron and its baseline level, which can be seen as an estimate of the range of evoked firing rate of the neuron, significantly decreased in the presence GABA_AR blockers [Fig. 1Ciii; WPT, sr(19) = 182, P = 4.7e-4]. This result was also true at MT [WPT, sr(19) = 177, P = 1e-3], suggesting that at threshold the baseline activity increases faster than the evoked activity. If the significance threshold is built on baseline activity (see methods), the blocking of GABA_AR dramatically reduced the area of significant evoked activity in the FRA (Fig. 1Ai, right, red curves). However, the area of significant evoked activity remained unchanged if the significance threshold was equal to half the maximum firing rate (Fig. 1, Ai and Aii, right, black curves). In fact, the auditory threshold even improved in that case [Fig. 1Cv; WPT, sr(19) = 110.5, P = 2.8e-2].

Fig. 1.

Effects of blocking GABA_A receptors (GABA_AR) on neuronal responses to pure tones in the inferior colliculus. Ai: spectrotemporal receptive field (STRF; left) and excitatory tuning curve (right) for an individual neuron, obtained with pure tones. Top: normal condition [control (Ctrl)]. Bottom: GABA_AR blocked by microiontophoresis of gabazine. MaxFR, highest firing rate recorded across STRFs associated with the 3 intensities tested. STD, standard deviation. Aii: same as Ai for another neuron. B: minimum threshold (MT) as a function of the characteristic frequency (CF) for the 19 recorded neurons. C: group results of the effects of blocking GABA_AR on parameters characterizing the neural response to pure tones. Ci: baseline activity. FR, firing rate. Cii: bandwidth of the receptive field. Ciii: range of evoked FR computed as the ratio of the maximum FR to the baseline. Civ: 1st spike latency. Cv: auditory thresholds computed from tuning curves. The significance threshold was equal to half the maximum FR. Ci, Ciii, and Civ: the sound pressure level (SPL) used is that where the highest FR was recorded across the STRFs associated with the 3 intensities tested. Cii: 30 dB SPL above threshold. D: averaged temporal profiles of all neural responses to pure tones. To normalize the FR, this latter has been divided for each neuron by the maximum FR reached by the neuron, independent of condition. *P < 0.05 for a Wilcoxon paired test. E: population results on maximum FR. F: population data for significant peak duration. In D–F, data are split according to frequencies 1–2 kHz below the best frequency (BF) (left), at the BF (center), and 1–2 kHz above the BF (right). For some neurons whose BF was equal to the lowest stimulation frequency, there was no measurement for 1–2 kHz below BF. The SPL used is that where the highest FR was recorded across the STRFs associated with the 3 intensities tested.

Blocking GABA_A receptors increases neural excitability.

Although blocking GABA_AR did not increase the maximum discharge rate and duration of evoked responses at BF or below BF (Fig. 1, E and F, left and center; WPT, all P > 0.05), it did increase the evoked firing rate immediately above the BF [Fig. 1, D and E, right; WPT, sr(19) = 21.5, P = 3.1e-4]. The same numerical pattern was observed for the evoked peak duration, but the effect was not significant [Fig. 1F, right; WPT, sr(19) = 47.5, P = 0.056]. Reasons for such an asymmetry between results at the lower and upper parts of the FRA seem unclear, at least in our data. In any case, this increase in evoked firing rate above BF led to a slight increase in bandwidth [Fig. 1Cii; WPT, sr(19) = 28.5, P = 2.3e-2]. Blocking GABA_AR had no significant effect on the first spike latency of evoked responses at the BF [Fig. 1Civ; WPT, sr(19) = 42.5, P = 0.06]. Consistent with the increased excitability of neurons, we found that blocking GABA_AR triggered more burstiness in spontaneous firing of neurons (Fig. 2, A and B), consisting of spontaneous emission of a burst of several spikes within a few milliseconds. Using a nonparametric “rank surprise” burst detector (Gourévitch and Eggermont 2007), we were able to quantify that the duration of detected putative bursts and their rate of appearance both increased with GABA_AR blockade [Fig. 2B, left and center; sr(19) = 18, P = 2e-3 and sr(19) = 10, P = 6e-4, respectively]. Another measure of the burstiness of a neuron is the variation coefficient of interspike intervals (ISIs), relating the variance of ISI (higher if short and long ISIs coexist as in a succession of bursts) to its average value. The ISI variation coefficient was also found to increase after blockade of GABA_AR [Fig. 2B, right; WPT, sr(19) = 16, P = 1.5e-3].

Fig. 2.

GABAergic inhibition prevents the natural bursting behavior of neurons in the inferior colliculus. A: individual examples of spike trains associated with spontaneous activity in control condition (top) or after blockade of GABA_A receptors (bottom). Red lines indicate the extent of detected bursts of activity. B: group results on bursting time (left; % of total time occupied by bursts), bursting rate (center; rate of appearance of bursts), and variation coefficient of the interspike intervals (ISIs) (right). *P < 0.05.

Blocking GABA_A receptors alters duration but not temporal precision in response to natural sounds.

We also examined the role of GABA in the more natural context of auditory processing of conspecific vocalizations. We used high-frequency mouse vocalizations whose spectral energy is outside the frequency response area of the recorded neurons. Thus, theoretically, the evoked response of neurons as linearly predicted by convoluting the vocalization spectrogram with the neuron’s STRF (see methods) should be null (Fig. 3, Ai and Aii, bottom left). However, 8 of 19 recorded neurons significantly responded to the natural vocalizations (Fig. 3B). This might be explained, in part, by a frequency-distorted product created at the basilar membrane level, especially at the time of the frequency jumps in the syllables (Portfors et al. 2009; Portfors and Roberts 2014). This distorted product roughly corresponds to the spectral distance between vocalization components before and after the jump. By simulating this distorted product (see methods), we created an artificial distorted vocalization whose frequency content was in a lower frequency range, closer to the frequency area of recorded neurons and that therefore evoked a neural response in more cases (Fig. 3, Aii and Aiii, right). The fact that evoked responses to the distorted vocalization and to the natural vocalization were similar for some neurons suggests that part of the response to the natural vocalization could be due to a distorted product (Fig. 3Aii, compare left to right), as suggested in a previous study (Portfors and Roberts 2014). This model is not perfect though, as we observed that the opposite is not necessarily true: for some neurons where no evoked activity is observed in response to natural vocalizations, the simulated distorted product obtained from that vocalization would predict the contrary (Fig. 3Aiii, compare left to right). Still, those artificial distorted vocalizations also provide a set of complex stimuli within the neuron’s frequency range. We thus used them to study the effects of blocking GABA_AR. In the following, we refer to them as “distorted vocalizations” to differentiate them from “natural vocalizations.” We recorded responses of each neuron to eight natural and eight distorted vocalizations. Group results include responses to all natural and distorted vocalizations pooled together.

Fig. 3.

Some neurons in the inferior colliculus respond to natural and distorted vocalizations differently. A: response of a typical neuron to natural and distorted vocalizations. Ai: spectrotemporal receptive field (STRF) of the neuron. FR, firing rate. Aii, top: spectrograms for natural (left) and distorted (right) vocalizations. Bottom: poststimulus time histogram (PSTH) of the neural response to the vocalization is in blue. The linear prediction of neural response obtained from the convolution of the receptive field with the spectrogram is displayed in yellow. Correlation (R²) between the actual and predicted response is indicated. Dashed line is the significance threshold for significant peaks. Aiii: same as Aii for a different vocalization. B: % of neurons in the population that had a significant response to natural and/or distorted vocalizations.

We observed a large variation in effects of blocking GABA_AR on evoked responses in the IC, to both natural and distorted vocalizations. Most typical effects are shown as individual examples in Fig. 4 as follows: 1) a response appeared after blockade of GABA_AR, whereas there was no response at all in the baseline condition, even far outside the frequency response area of the cell (Fig. 4, 1st row); 2) the discharge increased in rate and duration after blocking GABA_AR (Fig. 4, 2nd row); 3) the response was not dramatically changed especially for peak amplitudes (Fig. 4, 3rd row), consistent with results from STRFs when stimulation was close to the BF (i.e., close to the maximum FR of the neuron), as it is for this distorted vocalization; 4) the response disappeared, while it could not be predicted from the effects on STRFs (Fig. 4, 4th row). Overall, when the distorted vocalization did not evoke a significant response during baseline conditions, it did evoke one after GABA_AR were blocked in 40% of cases. In contrast, when the distorted vocalization evoked a response during baseline conditions, its response disappeared in 22% of cases after GABA_AR were blocked (Fig. 5A). Those results are approximately the opposite for natural vocalizations. This discrepancy is explained by the lower levels of maximum evoked firing rate before and after gabazine for the natural vocalization compared with the distorted vocalization (Fig. 5B): the increase of baseline firing rate after gabazine (Fig. 1Ci), which is similar for natural and distorted vocalizations by definition, will favor the loss of significance of the smallest evoked firing rates and the emergence of a significant response for the largest firing rates. Furthermore, group results revealed that after GABA_AR were blocked maximum evoked firing rate was increased for both natural and distorted vocalizations [Fig. 5B, left; WPT, sr(38) = 5.5, P = 3e-7 and sr(69) = 385.5, P = 7e-6, respectively] as well as response duration [Fig. 5B, center left; WPT, sr(38) = 57, P = 1e-4 and sr(69) = 534, P = 1.6e-3] and peak latency [Fig. 5B, center right; WPT, sr(38) = 121.5, P = 4e-5 and sr(69) = 377, P = 1.5e-3]. Since the spectral content of the chosen mouse vocalizations did not excite the neuron specifically at the BF, these results are in agreement with those from Fig. 1D, right, which were obtained with pure tones. Figure 5C displays a scheme illustrating the summarized effects of blocking GABA_AR on the range of evoked firing rate of neurons: blocking GABA_AR increased baseline (Fig. 1B) and evoked firing rate (Fig. 1D, Fig. 5B) but not maximum firing rate at the BF (Fig. 1B), so that the range of evoked firing rate of the neuron was actually reduced on the side of excitation, above baseline.

Fig. 4.

Individual examples of the effects of blocking GABA_A receptors (GABA_AR) on the neural response to vocalizations in the inferior colliculus. Each row of panels corresponds to 1 neuron. From left to right: 1 given vocalization spectrogram; the receptive field in the control condition (Ctrl) followed by the blocked GABA_AR condition (gabazine); poststimulus time histograms of neurons in response to the chosen vocalization before (blue) and after (orange) blockade of GABA_AR; spike trains corresponding to each trial before (top) and after (bottom) blockade of GABA_AR. In this latter representation, the first trial is the first row, the last trial is the last row, 1 yellow point is 1 spike, and time is in the x-axis. FR, firing rate; STRF, spectrotemporal receptive field.

Fig. 5.

Group results for the effects of blocking GABA_A receptors (GABA_AR) on neural response to vocalizations in the inferior colliculus. A: % of neurons whose response appeared or disappeared after blockade of GABA_AR for natural and distorted vocalizations. B: parameters characterizing the neural response to natural (top) and distorted (bottom) vocalizations. From left to right: maximum firing rate (MaxFR) reached by the cell; significant peak duration; latency of the significant peak maximum; correlation coefficient measuring trial-to-trial reproducibility (CorrCoef). Ctrl, control. C: model summarizing the effects of gabazine on the firing rate of the neuron: by increasing the baseline and leaving the absolute maximum firing rate unchanged, blocking GABA_AR reduces the range of evoked firing rate of the cell. *P < 0.05.

Interestingly, the trial-to-trial reproducibility of evoked response measured by the CorrCoef coefficient (see methods), in other words the temporal jitter of the neuron, was not decreased by blocking GABA_AR for either natural or distorted vocalizations [Fig. 5B, right; WPT, sr(33) = 276, P = 0.94 and sr(67) = 888, P = 0.11, respectively]. Thus, although the presence of GABA_A decreases response duration and shortens peak latency (except at BF), making the evoked response sharper in time, it does not improve the temporal reliability of the response.

Blocking GABA_A receptors decreases response nonlinearity.

We then tested whether the inhibitory processing brought by GABAergic inhibition is adequately captured by the STRF, which is a linear characterization of the complex stimulus-response transformation. We compared how STRFs were able to predict evoked responses to vocalizations (Fig. 6). On the two individual examples shown in Fig. 6, Ai and Aii, the linear prediction showed little resemblance to the actual response to the distorted vocalization. On the contrary, blocking GABA_AR increased evoked firing rate and response duration in the STRFs of the control condition to such an extent that linear prediction of the evoked response to the vocalization better matched the actual response (Fig. 6, Ai and Aii, right). Group results confirmed these observations by showing that for both natural and distorted vocalizations the part of evoked responses to vocalizations that could be predicted from only the response to pure tones (i.e., STRFs) was much higher when GABA_AR were blocked [Fig. 6B; WPT, sr(38) = 86, P = 4e-5 and sr(54) = 279, P = 7e-5, respectively]. These results suggest that GABAergic inhibition imparts nonlinearities to an IC neuron’s response to complex sounds such as vocalizations.

Fig. 6.

Effects of blocking GABA_A receptors (GABA_AR) on linear processing of vocalizations in the inferior colliculus. Ai and Aii: individual examples. Ai: spectrogram of the vocalization (top left); spectrotemporal receptive field (STRF) in control condition (Ctrl; top center); poststimulus time histograms (PSTH) in response to the vocalization (bottom center); STRF after blockade of GABA_AR (top right); PSTH after blockade of GABA_AR (bottom right). The actual response is displayed in blue; the linearly predicted response is in yellow. Correlation between the actual and predicted response is indicated; top and bottom right same as top and bottom center, respectively, after blockade of GABA_AR. Aii: same as Ai for another neuron and another syllable. B: group results for the correlation between actual and predicted response for natural (top) and distorted (bottom) vocalizations. *P < 0.05.

Blocking GABA_A receptors does not affect vocalization discrimination.

The spectral and temporal processing properties of neurons allow them to “discriminate” vocalizations by providing a different response to different natural sounds. This capacity might be altered by the blocking of GABA_AR. To test this for each neuron, we used each neuron’s individual spiking patterns in response to the eight natural or distorted vocalizations to build a confusion matrix summarizing which putative stimulus evoked a particular response pattern and then to estimate from such matrices the mutual information (MI) between neural responses and vocalizations (see methods and Fig. 7, A and B). This method quantifies how well the vocalization identity can be inferred from the spiking patterns in neuronal responses (Schnupp et al. 2006). We applied the same methodology to the firing rate (instead of the spiking pattern) to also estimate the MI carried only by the firing rate.

Fig. 7.

Effects of blocking GABA_A receptors on the ability of neurons to discriminate vocalizations. A, left: poststimulus time histogram (PSTH) of a given neuron in response to the 8 natural vocalizations used in the study. The neuron is the one used in the second row of Fig. 4. FR, firing rate. Center: spike trains for each trial (from top to bottom) in response to the third vocalization. Right: each trial is assigned to the vocalization whose PSTH resembles most (i.e., has the smallest Euclidean distance from) the spike train recorded during the trial. The assigned vocalization can be considered as the most likely that was presented to elicit such a spike train. Bi: matrix of confusion counting the occurrences of trials assigned to natural vocalizations as a function of the actual presented vocalization. If the matrix is diagonal, the neural responses are perfectly separable between vocalizations, and the latter can be discriminated by the neuron. The neuron is the same as in A. The mutual information (MI) estimated from this confusion matrix is indicated and measures the ability of the neuron to discriminate between the 8 natural vocalizations on the basis of its fine temporal spiking patterns. Ctrl, control. Bii: same as Bi for distorted vocalizations. Ci: same as Bi, but the Euclidean distance is not between spike trains and PSTHs but between the FR of the trial and the FR in response to the vocalization computed on a 150-ms time interval. The confusion matrix and the associated MI reveal the ability of the neuron to discriminate between the vocalizations on the basis of FR and not the spiking patterns. Cii: same as Ci for distorted vocalizations. D: group results for MI for natural (left) and distorted (right) vocalizations, using spiking patterns to construct the confusion matrices. E: same as D, using FR instead of spiking patterns to build the confusion matrices.

In the two individual examples shown in Fig. 7, Bi and Bii, we observed very similar confusion matrices, and therefore close values of MI carried by spiking patterns, before and after blocking GABA_AR. Group results confirmed that, based on temporal patterns, there was no significant effect of blocking GABA_AR on the discrimination abilities of neurons between vocalizations, either natural or distorted [Fig. 7D; WPT, sr(16) = 36, P = 0.1 and sr(17) = 41, P = 0.3, respectively]. Based on the firing rate, the MI was much lower than that carried by temporal patterns (Fig. 7, Ci and Cii). Although the confusion matrices could look slightly altered by blocking GABA_AR, group results showed that there was also no significant effect of blocking GABA_AR on the discrimination abilities of neurons between vocalizations, either natural or distorted [Fig. 7E; WPT, sr(16) = 40, P = 0.15 and sr(17) = 61, P = 0.98, respectively].

DISCUSSION

We iontophoretically blocked GABA_AR on IC neurons in awake restrained mice to evaluate how inhibition affects neuronal responses to pure tones and natural and distorted mouse vocalizations. The main results can be summarized as follows: 1) Frequency response areas slightly broadened above BF and reached a lower SPL threshold. 2) The spontaneous firing rate of neurons (baseline activity) increased, as did their burstiness when GABA_AR were blocked. 3) Although the baseline activity and the weak evoked response of neurons increased with GABA_AR blocked, the maximum firing rate at BF did not. Thus the range of evoked firing rate of the neurons (the ratio between their maximum and baseline response) decreased. 4) About half of the neurons responded to vocalizations whose frequency content did not match the neuron’s frequency tuning, whereas 75% responded to artificial distorted vocalizations matching the neuron’s receptive field. Blocking GABA_AR increased response strength, peak latency, and duration to both types of vocalizations. 5) Responses to such complex sounds were better predicted by the receptive field of the neuron and therefore were more linear when GABA_AR were blocked. Despite these effects, blocking GABA_AR neither altered the trial-to-trial reproducibility of the neurons in response to vocalizations nor modified the ability of neurons to discriminate vocalizations based on either the firing rate or the temporal patterns of the evoked response.

GABAergic inputs to the ICC.

In this study, we recorded neurons of the ICC while blocking their GABAergic inputs. In mammals, sources of GABAergic inputs to the ICC mostly originate from the ipsilateral and contralateral IC; several nuclei of the superior olivary complex (SOC), namely, the superior paraolivary nucleus, medioventral periolivary nucleus, and lateral superior olive; and the dorsal and ventral parts of the lateral lemniscus nucleus (NLL) (Ono and Ito 2018). Inhibitory inputs from the SOC are thought to convey binaural and temporal information as well as inhibition at the termination of the sound stimulus. The dorsal part of the NLL would convey binaural information as opposed to the ventral part, which is rather sensitive to the temporal structure of sounds. Although there are massive descending projections from the auditory cortex to the IC, such connections mostly target the dorsal and lateral parts of the IC and not the ICC (Stebbings et al. 2014; Torii et al. 2013). They are also mostly glutamatergic. These results suggest that the direct top-down influence on our results should be limited even if there is also evidence of functional corticofugal modulation of the ICC (reviews in Bajo and King 2013; Suga 2012), perhaps through indirect routes. Note that the awake condition seemed particularly relevant here, as anesthetics have been shown to affect the response rate (and therefore the maximum firing rate), the temporal response pattern, as well as the spontaneous activity of neurons from the IC (Astl et al. 1996; Bock and Webster 1974; Duque and Malmierca 2015; Kuwada et al. 1989). This is probably because of an increased GABA-mediated inhibition (Garcia et al. 2010; MacIver 2014).

Contribution of GABA to the range of evoked firing rate.

Our results of broader tuning curves, increased firing rates outside BF, and longer peak latencies to pure tones and vocalizations were all expected in light of previous studies in the IC (Fuzessery and Hall 1996; LeBeau et al. 2001; Pollak and Park 1993; Sivaramakrishnan et al. 2004; Yang et al. 1992). However, the broadening was moderate, probably because GABAergic and glutamatergic neurons in the IC share similar response properties, particularly FRA contours, such that GABAergic inhibitory inputs are aligned primarily within the IC target neurons’ excitatory inputs (Ono et al. 2017; Palombi and Caspary 1996a).

Because we did not observe changes in the maximal firing rate (at BF) while blocking GABA_AR, it is likely that GABAergic inputs are not normally acting to suppress the presynaptic excitatory input at BF in these IC neurons. However, we found that baseline firing rate significantly increased during GABA_AR blockade, the increase being faster than the evoked firing rate, even at threshold. This suggests that GABAergic inputs normally act to provide contrast in the firing rate between the baseline and the evoked response. If baseline is considered as “off” for the information transmitted, these data support the idea that GABAergic inputs modulate the “off” part of the signal, essentially improving the signal-to-noise ratio ofa neuron’s output. Additional evidence is in Fig. 1, Ai and Aii, where we show two different measures of significance threshold (red vs. black tuning curves). If we hypothesize that activity important to the brain is a function of the maximum firing rate of a neuron, the area of significant evoked activity does not change if the threshold is reduced to half the maximum firing rate. However, if spontaneous activity is seen by the brain as an “off” reference for the firing rate, then the area of significant activity becomes smaller and so does the range of firing rate values available to the neuron to code the signal features. Consistent with this view is the increased excitability of the neuron for synaptic inputs weaker than at the BF. Indeed, blocking GABA_AR also increases bandwidth and improves auditory threshold along with the increases in firing rate above BF and in response to vocalizations.

In the IC, changes in the maximum evoked firing rate of neurons have been investigated mostly in the context of aging (Khouri et al. 2011; Palombi and Caspary 1996b) or noise exposure (Bureš et al. 2010). In control studies, it was hypothesized that GABAergic circuits could adjust the gain needed for coding complex signals over a wide dynamic range (Palombi and Caspary 1996a). However, the contribution of GABA to the range of evoked firing rate is, to our knowledge, rarely suggested in the literature despite the great number of studies focusing on blocking GABA receptors. In fact, while the effects of GABA on spontaneous firing rate are in line with previous results from the literature (Faingold et al. 1989; Pollak and Park 1993), the stability of maximum firing rate was in disagreement with a similar study in the mustached bat (Pollak and Park 1993), which found that over half of the units more than doubled their maximum firing rate. However, the population was different since Pollak and Park sampled the units around a homogeneous neural population with BFs at 60 kHz, the specific dominant frequency of the echolocation calls. Without showing group results, a fraction of individual examples of neurons shown in Sivaramakrishnan et al. (2004) also display increases in the maximum firing rate. It is possible that the increase observed in those two studies stems from strong nonspecific effects of bicuculline on calcium-dependent potassium channels that are present in the IC (Kelly and Caspary 2005; Kurt et al. 2006). It is also possible that our gabazine application did not reach saturating effects. However, we think that the greatest difference with our study lies in the way the maximum firing rate is computed. In the above studies, the maximum firing rate actually is the firing rate at the BF averaged over the duration of the stimulus, i.e., between 20 and 75 ms. In our study, we wanted to assess the effects of blocking GABA_AR on the highest probability of discharge reached by the neuron. Even if we spanned the firing rate within the range −10 to +30 dB around threshold only, mainly to avoid lengthening the protocol too much and thus risking the loss of the neuron, we could indeed reach the maximum firing rate, or close to it, because the dynamic range of neurons is typically between 20 and 40 dB in the IC (Aitkin 1991; Bureš et al. 2010; Dean et al. 2005). Given that IC neurons can fire with a precision as short as 1 ms, we estimated our maximum firing rate on a 3-ms time interval, at the BF of the neuron. The range of maximum firing rate observed (Fig. 1E) shows that there was still potential for this highest probability of spiking to increase, but blocking GABA_AR did not alter it. As shown in Fig. 1D, the mean firing rate was increased over the time course of the stimulation at BF, and above BF, when blocking GABA_AR, but we found that the highest probability of discharge by the neuron at the BF, which often occurs at the onset, was not. This result is combined with those from Le Beau et al. (1996), who found that the largest change in firing rate after bicuculline occurred in the sustained part of the response and not at the onset. In the context of a firing rate neural code, the time interval in which a firing rate matters to the brain can extend to dozens of milliseconds (Ince et al. 2013). However, in the context of a temporal neural code, involving, e.g., fast spectral integration, phase locking, and temporal pattern spiking, which largely correlate to fine amplitude modulations of acoustic signals and which are crucial to processing in the IC (Chase and Young 2006; Rees and Langner 2005; Rodríguez et al. 2010; Voytenko and Galazyuk 2007), the probability of discharge within a few milliseconds is the most salient cue. What we show here is that GABA seems to contribute to the range of discharge probabilities of the neuron, not by decreasing the highest discharge probabilities (at the BF) but by decreasing the weaker evoked activities above BF or after onset, as well as the spontaneous activity.

However, the range of evoked firing rate as we computed it can only be estimated for nonzero levels of spontaneous activity, although it is not the case for some neurons in the IC (see, e.g., such individual examples in Figs. 12–14 in Pollak and Park 1993). Also, we did not test enough SPL levels to identify nonmonotonic neurons reliably, and we cannot exclude that results related to the range of evoked firing rate may differ between monotonic and nonmonotonic neurons. Many studies have shown that nonmonotonic neurons could be functionally distinct subgroups of glutamatergic neurons (Aitkin 1991; LeBeau et al. 2001; Ono et al. 2017; Yang et al. 1992). However, nonmonotonicity would stem, at least in part, from a stronger inhibition at higher intensities (Liu et al. 2019). It is therefore possible that blocking inhibition at any suprathreshold SPL would in fact result in the same highest discharge probability as long as the excitation remains stable for a range of high SPL, as in the example of Fig. 3 in Palombi and Caspary (1996a).

Temporal effects of GABA.

Regular and temporally precise firing is a characteristic feature of most neurons in the IC (Rees et al. 1997). Given that inhibition participates in shaping the discharge pattern of most IC neurons (Fu et al. 2013; Kuwada et al. 1997), GABA is thought to have major contributions to the temporal dynamics of neuron discharges in the IC. First, we found that the time course of neural responses was modified, with an increase of the later portion of the response to pure tones or with modified PSTHs in response to vocalizations. Consistently, half of neurons show a modified time course of their response (PSTH) after GABA_AR are blocked by bicuculline (Le Beau et al. 1996). It is possible that mechanisms contributing to such modification of PSTHs rely on the reduction of adaptation processes that occur after microiontophoretic application of gabazine (Pérez-González et al. 2012). PSTH alterations could also stem from changes in temporal patterns induced by GABA_AR blockade. Such changes were observed in 35% of neurons in the IC of the horseshoe bat (Vater et al. 1992). Interestingly, in the study by Le Beau et al. (1996), half of the units that changed their temporal patterns in the presence of bicuculline became chopper units. Among the possible mechanisms are the interplay between excitation and inhibition as well as intrinsic membrane properties (Kuwada et al. 1997), and also the fact that evoked excitatory postsynaptic currents are sometimes followed by long-lasting oscillatory currents revealing patterns of altered excitability (Covey et al. 1996). Many currents could be kept subthreshold by GABA and therefore revealed by blocking it. This point of view could explain that even in absence of stimulation (or, as in our study, in the presence of a subthreshold stimulation) the increase in baseline activity resulting from blockade of GABA_AR was accompanied by an increase of the bursting behavior of many of the neurons. In fact, as it is known that bursting activity can be a vector of information transmission (Lisman 1997; Sherman 2001), one role of GABA in the IC could be to silence neurons in order to avoid transmitting information absent from a stimulus or in the presence of a subthreshold stimulus. In other words, GABA could act to reduce false alarms carried by isolated bursts. The salience of this hypothesis depends on the neural code used postsynaptically to the IC. If the significance of the neural response is based on a fraction of the maximum firing rate, GABA would hardly affect it and would actually even degrade the auditory threshold (Fig. 1Cv). However, if the spikes from isolated bursts or even any particularly strong spontaneous activity actually act as false alarms, then GABA would dramatically improve the auditory threshold and therefore the neural sensitivity, as we saw in the examples of Fig. 1, Ai and Aii (red curves). From this point of view, GABA would have more effects on the spiking probability than on its timing. Consistently, we did not find any effects of blocking GABA_AR on the CorrCoef value, i.e., on the trial-to-trial reproducibility of neural responses to vocalizations. This latter result is in line with our previous investigations where both glycine and GABA receptors were blocked (Dimitrov et al. 2014). It suggests that there is no additional time jitter when GABA_AR are blocked and that GABA does not contribute to the extraordinary submillisecond precision of the sound encoding in the IC (Garcia-Lazaro et al. 2013; Luo et al. 2018; Voytenko and Galazyuk 2008). One could speculate that maintaining precise timing while reducing false alarms is especially important for the IC, which shows much better phase locking compared with the auditory cortex (Joris et al. 2004) and therefore provides a more precise and isomorphic representation of the auditory signal. Finally, increased neural activity might allow the formation of unwanted auditory objects such as tinnitus (Bauer et al. 2008; Coomber et al. 2014; Eggermont 2015; Mulders and Robertson 2009).

We also observed an increase in response duration when blocking GABA_AR in the context of stimulation using vocalizations, which is in line with a previous study using pure tones in the bat (Koch and Grothe 1998). With regard to response duration, one must note that an important characteristic of many neurons in the IC is to be tuned in duration (Aubie et al. 2012). Duration tuning is generated by the temporal dynamics of excitatory and inhibitory inputs to neurons in the IC (Casseday et al. 1994). When GABA_AR are blocked, duration tuning is typically abolished (Casseday et al. 1994), which confirms the contribution of GABA to the information coded by the response duration.

In the IC, inhibition arrives earlier than excitation and may therefore play a role as a mechanism for delaying the first spike latency in IC neurons (Covey et al. 1996; Voytenko and Galazyuk 2008). However, bicuculline (and strychnine) were found to have no significant effect on latency in general (Fuzessery et al. 2003) except for a few long-latency neurons that showed marked reductions in latency (Le Beau et al. 1996). Furthermore, we did not observe any change in first spike latency after blocking GABA_AR, but we did observe a prolonged peak latency in response to vocalizations. This latter effect is likely to stem from the parallel increase in evoked firing rate, which delays the time to reach the peak firing rate. Conversely, it might induce a modification of the neural representation of the temporal envelope of complex sounds.

Effects of GABA on complex sound encoding.

The mouse presents an interesting conundrum to understanding vocalization selectivity because many mouse vocalizations that elicit a neuronal response contain spectral energy that does not match the neuron’s frequency receptive field (Portfors et al. 2009; Woolley and Portfors 2013). We hypothesize that the spectral characteristics of a particular type of mouse vocalization, the frequency-jump syllables, create distortion products in the cochlea that then stimulate neurons in the IC that have response curves below the frequencies of these vocalizations. Distortion products are a well-known phenomenon of the mammalian cochlea (Plomp 1965; Robles et al. 1991; Robles and Ruggero 2001). The extent to which they are used in auditory processing is not completely understood, and their occurrence has even been referred to as an epiphenomenon (Warren et al. 2009). There is some evidence that distortion products may be used in the perception of pitch (McAlpine 2004). In addition, single-unit recordings and BOLD fMRI experiments have provided evidence that distortion products may be utilized by neurons in the rodent auditory system to encode high-frequency social vocalizations (Gao et al. 2015; Portfors et al. 2009; Portfors and Roberts 2014; Roberts and Portfors 2015). In the present study, using vocalizations that we had previously recorded from male mice in our laboratory (Mahrt et al. 2013), we wanted to assess three questions: 1) Can neurons in the IC respond to low-frequency distortions created by vocalizations and presented alone? 2) Does GABA modify the neural response to vocalizations? 3) Does GABA elicit or silence a neural response to spectral contents that do not match the neurons’ receptive fields?

To account for this, we applied a dynamic model of the cochlea (Portfors and Roberts 2014; Roberts and Portfors 2015) to filter our natural vocalizations and we isolated vocalization-induced distortion products. We found that the natural overlap of components in frequency-jump syllables creates high intensity distortions at low frequencies (Fig. 3, Aii and Aiii). Interestingly, quadratic distortions (F2 − F1) were the highest-intensity distortion products in mouse vocalizations and fell between 10 and 25 kHz. These quadratic distortions are within the 10–30 kHz frequency range that is overrepresented in the mouse auditory system (Egorova et al. 2001; Portfors et al. 2011). Indeed, these distortions elicited a significant response in our neurons more often than natural ones.

We previously showed that inhibition could improve the selectivity to vocalizations based on firing rate (Mayko et al. 2012) but not based on spiking patterns (Dimitrov et al. 2014). However, these results were obtained by blocking both GABA and glycine receptors. In this study, we isolated specific effects of GABA and used a sophisticated method to assess the selectivity to vocalizations taking into account either spike timing-based or firing rate-based neural codes. When comparing neural responses to distorted vocalizations with those to natural vocalizations, we found very similar effects of blocking GABA_AR: the evoked firing rate, response duration, and the peak latency all increased, but the trial-to-trial reproducibility was not modified. These effects are consistent with those found on amplitude-modulated sounds (Zhang and Kelly 2003) and suggest that, as stated above, the neural representation of complex sounds is shaped quite heavily by GABAergic inhibition. However, blocking GABA_AR did not degrade the abilities of neurons to discriminate between vocalizations when using either a spike-timing based code (Fig. 7D) or a firing rate-based code (Fig. 7E). This result is true in average, but one must note that paired values shown in Fig. 7, D and E, display many positive or negative variations of MI when GABA_AR are blocked. We were unable to find significant correlations of such variations with the many other parameters studied (data not shown), but we cannot rule out that subpopulations of IC neurons show their MI significantly modified by GABA. Note that the information carried by firing rate was much lower (by a factor of 10) than that carried by spiking patterns. This result is in agreement with several studies in the IC and in the auditory cortex of mammals or birds, suggesting that temporal patterns are far more efficient than firing rate to code natural sounds (Dimitrov et al. 2014; Huetz et al. 2009; Narayan et al. 2006; Schnupp et al. 2006).

We also used a computational model that linearly predicts a neuron’s response to a stimulus based on the neuron’s excitatory tuning curve to test whether GABA could affect the predictive ability of the model. We found that blocking GABA_AR significantly and dramatically improved the match between the actual and predicted neural responses to both natural and distorted vocalizations. Note that the increase is of the same magnitude for natural vocalizations whose spectral content lies outside the neurons’ excitatory receptive fields as for distorted vocalizations within the neuron’s frequency range. This improved prediction, which also holds for vocalizations when either GABA receptors alone (Klug et al. 2002) or both GABA and glycine receptors (Mayko et al. 2012) are blocked, suggests that GABAergic inhibition adds nonlinearity to the neural response to complex sounds. In other words, the combination of excitatory synaptic inputs and GABAergic inputs makes the neural response to complex sounds less predictable when only knowing the spectrotemporal receptive field of the neuron. This result likely builds upon the complex temporal interaction between inhibitory and excitatory inputs in the IC, the inhibition arriving both earlier and later than the excitation (Voytenko and Galazyuk 2008). Note that the neurons we recorded all had long latencies, a feature common to the IC and possibly related to such early inhibition (Covey et al. 1996; Park and Pollak 1993) as well as NMDA receptors (Blitz and Regehr 2003; Kelly and Zhang 2002; Sanchez et al. 2007). Further investigation will be required to extend our results to other populations of neurons in the IC.

Understanding how inhibition shapes selectivity for vocalizations in the auditory midbrain expands our understanding of how the brain filters salient features from conspecific communication sounds. Neuromodulatory agents, such as GABA and glycine, have the critical role of balancing excitatory input streaming from ascending and descending pathways in the brain. The results from this study are important for understanding how single neurons in the IC process natural sounds and discriminate between them in the awake condition. In light of our results, GABAergic inhibition clearly contributes nonlinearly to the temporal shaping of the neural response to complex sounds even if it does not modify the spike timing code carried by neurons about those complex sounds. Importantly, the role of GABA in the extension of the range of evoked firing rate of cells, beyond the well-known decrease of evoked and baseline firing rates, may open the gate for new investigations into the factors contributing to the neural code in the IC.

GRANTS

B. Gourévitch and W. Bakay were supported by a Junior Grant from the French National Research Agency to B. Gourévitch (ANR-15-CE37-0007-01). C. V. Portfors and E. J. Mahrt were supported by National Science Foundation (IOS-0920060) and National Institutes of Health (NIDCD R01 DC-013102) grants to C. V. Portfors.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

C.V.P. conceived and designed research; E.J.M. and C.E. performed experiments; B.G. and E.J.M. analyzed data; B.G., E.J.M., and C.V.P. interpreted results of experiments; B.G. prepared figures; B.G., E.J.M., and W.B. drafted manuscript; B.G., E.J.M., W.B., and C.V.P. edited and revised manuscript; B.G., E.J.M., W.B., and C.V.P. approved final version of manuscript.