Long-Lasting Sound-Evoked Afterdischarge in the Auditory Midbrain

Different forms of plasticity are known to play a critical role in the processing of information about sound. Here, we report a novel neural plastic response in the inferior colliculus, an auditory center in the midbrain of the auditory pathway. A vigorous, long-lasting sound-evoked afterdischarge (LSA) is seen in a subpopulation of both glutamatergic and GABAergic neurons in the central nucleus of the inferior colliculus of normal hearing mice. These neurons were identified with single unit recordings and optogenetics in vivo. The LSA can continue for up to several minutes after the offset of the sound. LSA is induced by long-lasting, or repetitive short-duration, innocuous sounds. Neurons with LSA showed less adaptation than the neurons without LSA. The mechanisms that cause this neural behavior are unknown but may be a function of intrinsic mechanisms or the microcircuitry of the inferior colliculus. Since LSA produces long-lasting firing in the absence of sound, it may be relevant to temporary or chronic tinnitus or to some other aftereffect of long-duration sound.

. The inhibitory neurons in the mammalian IC are exclusively GABAergic neurons, while glutamatergic excitatory neurons make up the majority of the neural population 13 . Thus, we identified the IC neurons activated by light as GABAergic, while we identified those whose response to sound was suppressed as glutamatergic neurons. The specific expression of ChR2 in inhibitory neurons ensures that the light activated neurons were GABAergic. However, it is possible that some GABAergic neurons are so heavily innervated by inhibitory inputs that light evoked inhibitory transmitter release suppresses their light-evoked firing as well as their response to sound. Thus, we used juxtacellular labeling to stain five light-suppressed neurons and determine if they expressed GAD67, a marker for GABA neurons (Fig. 1e). These light-suppressed neurons lacked GAD67 (Fig. 1e) consistent with the notion that that light-suppressed neurons are glutamatergic neurons (n = 5). We also juxtacellularly stained one light activated neuron, which was GAD67 positive (data not shown).
We found that 20% of glutamatergic (8/40) and 17% of GABAergic (6/36) neurons in the VGAT-ChR2 mice exhibited a long-lasting sound-evoked afterdischarge (LSA) and continued to fire after the sound terminated (Fig. 2a,c). The characteristic frequency (CF) and the threshold of the neurons with LSA were not significantly different from the neurons without LSA (Table 1; CF, p = 0.12; Threshold, p = 0.32; Kruskal-Wallis test). We tested 36 GABAergic neurons and 40 presumed glutamatergic neurons for LSA by using long duration, 30-60 s, continuous one-octave noise (see METHODS). All neurons were tested at 60 dB SPL. Time permitting, we also tested sound levels ranging from 30-90 dB SPL and durations ranging from 4-120 s. The LSA was stronger when the response during sound (RDS) was robust and the sound duration was long (Figs 2a-d and 3a). The minimum sound duration required to induce LSA was ~30 s (GABA, 28.3 ± 7.6 s, n = 6; nonGABA, 37.5 ± 7.5 s, n = 8, p = 0.42, Wilcoxon signed-rank test, Fig. 2f). In five GABAergic and five nonGABAergic neurons, we checked the minimal intensity required to evoke LSA using either a 30 or 60 s sound. The minimum intensity was 52.0 ± 6.6 dB (n = 5, range 30-70 dB) and 63.0 ± 3.0 dB (n = 5, range 60-75 dB) for GABAergic and nonGABAergic neurons, respectively, and not significantly different (p = 0.18, Wilcoxon signed-rank test). These levels were 20.0 ± 5.2 dB and 22.0 ± 6.8 dB above the threshold for GABAergic and nonGABAergic neurons, respectively (p = 0.75, Wilcoxon signed-rank test; Table 1). The number of spikes in the LSA and RDS was positively correlated (Fig. 2g, r = 0.51, p < 0.01, two-tailed Student's t test). Six neurons were tested for LSA with both continuous and AM noise. Most showed LSA to both stimuli (2/3 GABA; 2/3 nonGABA). One nonGABAergic neuron had LSA to AM sound but not to unmodulated sound. Neurons exhibiting LSA were restricted to the central nucleus of IC (ICC) in the VGAT-ChR2 mice (Fig. 3b). Most LSA neurons were in area 1 (9 of 12 neurons) in the ventrolateral part of ICC, a region shown recently to have high concentrations of glycinergic axonal inputs 14 (area 1, Fig. 3b). The remaining three neurons were in the adjacent area of ICC where GABAergic terminals are more prevalent than glycinergic terminals 14 (area 2, Fig. 3b).
To determine if LSA was due to the ChR2 transgenic strain, we tested CBA/J mice, a wild-type strain used in auditory research 15 . Recordings in CBA/J mice were made under urethane anesthesia to rule out ketamine, and its specific effects on NMDA receptors 16 , as a cause of LSA. In CBA/J mice, 15% (3/20) of IC neurons sampled showed an afterdischarge response that was essentially the same as that seen in the other conditions ( Fig. 2g-i, insets, Fig. 3a). The minimum sound duration to evoke LSA was 23.3 ± 6.7 s (n = 3), and the minimum sound intensity was measured in two neurons (30 dB and 60 dB). The peak and decay time of LSA in CBA/J mice were 7.3 ± 2.  Fig. 2i), respectively. Thus, the LSA was not due to the use of a transgenic mouse with ChR2 or the use of ketamine anesthesia.
Discontinuous sound could also evoke a LSA (Fig. 4a). We used 1 s narrowband noise bursts presented every 2 s in all neurons tested where LSA was first shown with continuous sound (50 repetitions, GABA, n = 5, non-GABA, n = 3). Interestingly, there was a gradual increase in interstimulus spikes (Fig. 4a,b) not seen in neurons noise (80 dB). Left lower, the white noise with light (30 ms, 50 mW). Blue box indicates light presentation. The right panel is the raster plot of the response to noise and light. (e) An image of a juxtacellularly stained neuron that was suppressed by light. Note that it is GAD67 (green signal, a marker for GABA) negative. All the stained neurons with light suppression were GAD67 negative (n = 5). lacking LSA (Fig. 4c). The interstimulus afterdischarge spikes began after 30 s of stimulation on average and continued until the termination of the 100 s train of noise bursts.

Discussion
The present results suggest that a subset of neurons in the auditory midbrain is under the influence of a powerful but unknown phenomenon that potentiates continued firing after the offset of a sound stimulus. IC neurons might be expected to adapt and stop firing during long duration sound stimulation due to the depression of excitatory synaptic inputs which was seen in the majority of IC neurons [17][18][19] . However, the neurons with LSA showed less accommodation than non-LSA neurons, and consequently, their rate of firing during the sound stimulus was more sustained. This result suggests a mechanism in these IC neurons that may compensates for any synaptic depression and help the IC neurons to fire persistently to long sound. However, the LSA phenomenon is unlike long term depression or potentiation 20 seen previously in IC neurons since it is not stimulus bound. It is also known that a subset of IC neurons has Ca 2+ -activated potassium conductances that are known to favor neural accommodation. Neurons with LSA are not likely to be the adapting or transient IC neurons with SK and BK currents, respectively, but instead the LSA neurons are more likely to be the sustained-regular or buildup-pauser neurons seen in previous in vitro studies 21 .
One possible mechanism underlying LSA is slow afterdepolarization (sADP) [22][23][24] . The size of sADP is proportional to the duration of the depolarization in vitro 22 , consistent with our result that LSA was well correlated with RDS (Fig. 2g). Previous in vitro studies of IC in rat have shown plateau potentials evoked by brief synaptic stimulation; moreover, the sustained-rebound and pause-build neurons have a depolarization that extends beyond the duration of the synaptic inputs 25 . Neuromodulators are relevant to sADP 22,24,26,27 , and the receptors for modulators such as dopamine 28,29 , serotonin 30,31 , acetylcholine 32,33 and metabotropic glutamate (mGluR) 34,35 are found in the IC central nucleus where LSA is seen. Previous studies have shown that sADP could be due to the activation of TRP channels 26,36,37 and/or the inhibition of G-protein coupled inwardly rectifying K + (GIRK) channels 26 . In the IC, TRPC1, GIRK1, and GIRK3 were reported to be expressed in neurons 34,38 . One recent study 34 showed that both GABAergic and nonGABAergic neurons expressed TRPC1 channels and many of them co-expressed the group 1mGluR. It is possible that these channels shape sADP to generate LSA in IC.
A second possible mechanism underlying LSA is the recruitment of local microcircuits within the ICC. The primary disc-shaped neurons in ICC make extensive local axon collaterals within the fibrodendritic laminae that define the architecture of the nucleus [39][40][41] . Recently, glutamatergic ICC neurons were shown to make extensive synaptic inputs on neurons within the same layer 42 , and GABAergic neuron in IC may do so as well. Consequently, the ascending excitatory inputs to an ICC layer may stimulate multiple polysynaptic excitatory inputs within the same layer. The study of the ICC in the brain slice with voltage sensitive dye supports this view since it shows that localized responses to low-frequency stimulus trains spread when the stimulus frequency increases. This suggests the recruitment of silent microcircuits 43 . In the hippocampus, afterdischarges are seen as the rhythmic bouts of synchronized network activity in the clonic phase of seizure activity. These afterdischarges in CA3 pyramidal neurons require GABAergic synaptic transmission that becomes excitatory due to a transient collapse in the Cl − reversal potential 44 . Thus, prolonged activity in IC local microcircuits may cause a shift in the Cl − reversal potential of IC neurons and make normally inhibitory GABAergic synapses depolarize the neuron. This suggests that the long-duration acoustic stimulation might be particularly effective in recruiting these local circuits in the IC to produce LSA.
Mostly likely, LSA is a response property that emerges in the IC. Most neurons in the subcortical auditory pathway show a brief suppression of activity after a long-lasting acoustic stimulus [45][46][47] . However, the dorsal cochlear nucleus (DCN), does show a rebound afterdischarge after a brief period of post-stimulus depression in unanesthetized animals 48 . This afterdischarge may not continue beyond several hundred milliseconds, although it has not likely been examined with the long-lasting stimuli used here. The DCN buildup response does increase during the stimulus. This is shown as facilitation to the second of two successive 25 ms tones and with longer stimuli 41 . It is unclear whether a gradually increasing DCN response during the stimulus would result in a longer afterdischarge lasting into the range seen with LSA. Thus, it is unlikely that LSA in IC neurons is the result of a long-lasting excitatory input from the ascending auditory pathway. On the other hand, a suppressive aftereffect in the many inhibitory pathways that ascend to the IC might contribute to LSA by disinhibition. Although LSA has not been described previously in the auditory brainstem, the anatomical location of the neurons suggests that this response property could be related to some IC inputs more than others. Most LSA neurons (75%) were in area 1 of ICC. The inputs to ICC are organized into functional zones that receive different subsets of ascending inputs 49,50 , and the area corresponding to area 1 in the gerbil receives its most prevalent inputs from the superior olivary complex 51 . However, only 20% of IC neurons had LSA, they were distributed sparsely,  and they were likely surrounded by neurons without LSA. This suggests that both LSA and non-LSA neurons may share to same inputs, and makes it less probable that LSA is due to inputs from a single brainstem nucleus.
It is questionable that LSA was due to damage of the periphery. This is unlikely since the stimuli to do so are usually at higher sound levels and longer duration (e.g. 100 dB, 2 hr 52 ) than those used here. Even a one week exposure to continuous 80 dB sound produced little change in the auditory threshold in the rat 53 . In fact, a daily 6 hr exposure to 85 dB sound may be protective against traumatic noise exposure and may enhance the cochlea sensitivity 54 . These studies support the notion that LSA was not due to insult of the periphery. However, we cannot rule out a plastic change in the periphery or in the brainstem caused by our sound stimuli.
At this point, we can only speculate on whether the LSA neural behavior results in a percept in the normal hearing animal. Auditory perception is assumed to be directly related to activity in the auditory cortex, but we do not know whether cortical activity is stimulated by IC neurons during the LSA response. The present results show that both GABAergic and glutamatergic IC neurons can exhibit LSA. If the IC neurons with LSA include both the glutamatergic and GABAergic tectothalamic neurons, they would both project to the medial geniculate body that, in turn, supplies the ascending input to auditory cortex. The firing rate of LSA is far higher than the normal spontaneous rate, and those higher rates are similar to those evoked by sound stimulation. Thus, if the tectothalamic pathway is driven by LSA, the auditory forebrain might misinterpret the LSA firing as sound-evoked firing. LSA may not have to be perceived to be useful. The LSA signal may be useful to stimulate the descending auditory pathways in response to prolonged sound stimulation.
Some sound stimuli can evoke auditory afterimages. One example is the Zwicker tone that is a pure tone-like auditory afterimage induced by a noise with a spectral gap or a low pass noise [55][56][57][58][59] . The frequency of the Zwicker tone is within the spectral gap or at a frequency above the low-frequency stimulus that produced it. This differs from stimulation at CF with narrow-band noise to induce LSA. The duration of the Zwicker tone is only seconds in most studies with the longest reported being 10 s induced by a 1 min noise 55 . In contrast, the time courses of LSA were diverse (Fig. 2i) and ranged from 17-32 s on average in glutamatergic and GABAergic neurons, respectively, but they could extend to several minutes. Despite these differences, the Zwicker tone and LSA have some properties in common. Both LSA and Zwicker tone became longer when the inducer sound was longer, and both appear to be central in origin and unrelated to changes in the peripheral auditory pathways 58,60 .
Another example of an afterimage is the temporary tinnitus that accompanies a temporary threshold shift induced by a narrow band noise at a high sound level (110-120 dB) 61,62 . Like LSA, temporary tinnitus is induced in normal hearing individuals and persists for up to 15 minutes after exposure to tone or noise 61 . Unlike LSA, but similar to the Zwicker tone, temporary tinnitus is seldom at the same frequency as the acoustic stimulation and hearing loss 61,62 . This suggests that within minutes some type of plastic change may take place in the central auditory pathway in addition to temporary hair cell damage in the cochlea.
LSA might play a role in chronic tinnitus. Chronic tinnitus induced by noise exposure is known to cause long-term plastic change in the auditory pathway 63 . In animal models of chronic tinnitus, hyperactivity in the DCN is well established [64][65][66] , and it may be caused by a reduction of potassium channel activity 67 and reduction in GABAergic inhibition 68 . Since the DCN is a major input to the central nucleus of the IC 69,70 , it supplies the ICC with a prolonged hyperactive, frequency-specific input after a noise-induced hearing loss, and there is a concurrent increase in spontaneous activity in the IC 71,72 . Such hyperactive input might induce LSA in some postsynaptic IC neurons, or it might induce LSA when frequency-specific hyperactivity is paired with sounds lower in frequency than the hyperactive, much like the low-pass noise that produces a higher-frequency Zwicker tone. Indeed, subjects with tinnitus are more likely to hear a Zwicker tone 73 . Thus, a prolonged LSA response in IC neurons induced by a mix of hyperactivity and sound stimulation might contribute to the perception of a phantom sound in tinnitus.

Methods
Ethical approval. All experiments were approved by the Animal Care and Use Committee at the University of Connecticut Health Center and done in accordance with institutional guidelines and with the NIH Guide for the Care and Use of Laboratory Animals. All efforts were made to minimize the number of animals used and their suffering.

Animals.
We used thirty transgenic mice expressing channelrhodopsin (VGAT-mhChR2-YFP, Tg(Slc32a1-COP4*H134R/EYFP)8Gfng/J; #14548, Jackson Labs) of either sex (Postnatal day 1.5-4 months). A colony of transgenic mice backcrossed against a C57B/6 background was established. Breeding pairs consisted of one hemizygous and one wild-type or opposite sexed hemizygous. Transgenic offspring had a high expression of EYFP in their brains, and neonates (P0 -2) were phenotyped by visible fluorescence in the brain under blue light. In additional experiments, we used six female CBA/J (#656, Jackson Labs) mice.
Sound system. Acoustic stimuli were generated by a TDT System 3 (TDT, Tucker Davis Technologies, Alachua, FL) under the control of custom software (Brian Bishop, UCHC) written in MATLAB (Mathworks, Portola Valley, CA). All sounds were delivered by a closed system that included electrostatic speakers (TDT EC1) coupled to small metal tubes inserted into the external auditory meatus. Binaural acoustic crosstalk was minimal 74 . The sound system was calibrated from 100-100000 Hz. The calibration was performed at the end of the metal tubes with a 1/4″ microphone (Type 4135, Brüel & Kjaer, Naerum, Denmark).

Surgical Preparation.
VGAT-ChR2(H134R)-EYFP mice were anesthetized with a mixture of ketamine (100 mg/kg), xylazine (20 mg/kg) and acepromazine (10 mg/kg), and maintained in an areflexive state with isoflurane (0.5-1%) mixed with oxygen during the surgery and recording. CBA/J mice were anesthetized with urethane (0.9-1.3 g/kg, Sigma). Body temperature was monitored and maintained at > 35 °C by a DC temperature controller (FHC, Bowdoin, ME). Vital signs also were monitored (MouseOx Plus, Starr Life Science Corp, PA), and the surgery and recordings were done in a double-walled sound attenuating chamber (IAC, Bronx, NY).
The surgical procedure was described previously 17,74 . After the craniotomy, the auditory brainstem response (ABR) to a click (0.5 ms) was measured to verify normal hearing. The threshold of the ABR was around 30 dB SPL (VGAT-ChR2, left, 32.7 ± 1.3 dB, right, 32.3 ± 1.1 dB; CBA/J, left, 31.7 ± 3.1 dB, right, 31.7 ± 1.7 dB). Mice were used for experiments only when the ABR threshold was less than 40 dB.
Electrophysiology. Single cell extracellular recordings were obtained using glass pipettes filled with 0.01 M PBS (pH 7.4) with 2% Neurobiotin (4-7 MΩ ). Glass pipettes were made from borosilicate glass capillaries (34502-99, Kimble Chase, Vineland, NJ). The signals were amplified, bandpass filtered from 300 to 4000 Hz and sampled at 10 kHz with a Multiclamp 700B Amplifier, Digidata 1440A digitizer and Clampex 10.2 system (Molecular Devices). The voltage signals were recorded in current clamp mode. In parallel with recording the signals, the spike times were extracted using a window discriminator and recorded with the TDT System 3 and MATLAB software.
Optogenetic identification of GABAergic and nonGABAergic neurons. After the single unit was isolated in the VGAT-ChR2 mice, we identified the neuronal type optogenetically (Fig. 1). Light was generated by a blue laser (MBL-ΙΙΙ-473 nm-200 mW, CNI, China) and delivered through an optical fiber (400 μ m). The fiber tip was placed several millimeters above the brain surface using a micromanipulator. The light stimulus was a 30 ms light pulse (10-50 mW/mm 2 at the fiber tip). Light pulses were given every four seconds. Light evoked firing in GABAergic neurons and suppressed spikes in other neurons. We judged that the neuron was suppressed when the light reduced sound evoked spikes by more than 50%. Five neurons with spike suppression by light were juxtacellularly stained and immunohistochemistry confirmed they were not GABAergic and did not contain GAD67 (Fig. 1e).
Acoustic stimuli. After the cell type identification, the neuron's best frequency (BF) was determined by 100 ms tone bursts at 70-80 dB. The characteristic frequency (CF) and threshold were also determined by reducing the sound intensity (5 dB step). BF was defined as the frequency where the neuron showed the strongest response. CF was defined as the frequency where the lowest sound level could evoke the response. After determining the BF and CF, we asked whether the neuron had an afterdischarge. In most neurons, we used a 60 dB, 1 octave noise centered at the CF that was either continuous (30-120 s) or discontinuous sound (1 s every 2 s repeated 50 times, or 5 s every 6 s repeated 20 times). We first examined 60 dB sounds and then repeated the stimulation at a higher and/or lower intensity level. In some neurons, we also examined LSA using amplitude modulated (AM) sound with either a sinusoid or a raised sine modulation envelope 75 . We first determined the best modulation frequency using AM one-octave noise (1 s every 2 s) with modulations 2-512 Hz (1 octave steps). Next, LSA was examined by using 30-120 s AM sound at the best modulation frequency. In all LSA experiments, sounds were delivered at least 20 s after the afterdischarge activity returned to the baseline level. Data analysis. Data were analyzed with Clampex 10.2 and MATLAB. Spike times were extracted off-line using a fixed threshold. The threshold was set at > 5 standard deviations above the baseline. After extracting the spike times, we constructed peristimulus time histograms (PSTH) with a bin size of 500 ms. The PSTH was low-pass filtered (Boxcar, 3 points) to measure the temporal character of the response during sound (RDS) and LSA. To evaluate the duration of the RDS, we measured the width between 25% rise and decay points of the filtered PSTH. When the 25% decay point was not in the RDS, we measured the width between the 25% rise point and sound termination point. For LSA, we measured the peak and decay time. To measure the peak point, we first measured the preceding drop (dip) of the response after sound termination (Fig. 2a,c), and detected the earliest peak point after the dip (Fig. 2a,c). The peak time was the interval between the sound termination and peak point. The LSA decay time was measured as the interval between 10 and 90% points in the decay phase. To evaluate the size of RDS and LSA, we counted the number of spikes in each. The number of RDS spikes was measured as the number of spikes between the sound start and termination, while that of LSA spikes was measured as the number of spikes between the sound termination and LSA termination. The LSA termination was measured as the point where the spike rate returned to the baseline level. The RDS firing rate (Fig. 2b,d) was measured by dividing the number of RDS spikes by sound duration. We measured an accommodation index (AI) to compare the number of spikes in response to the last two seconds of the sound vs. the number of spikes during the first two seconds. For this measurement, we only used the RDS to unmodulated sound. To analyze the responses to discontinuous sound (Fig. 4), we constructed the unfiltered PSTH with a bin size of 1 s. The RDS and the interstimulus responses were measured and compared (Fig. 4b,c).

Histology.
After the single cell recording, the recording site was marked with Neurobiotin by current injection (200 nA, 50% duty cycle of 500 ms, 5 min). In five light suppressed neurons, we performed juxtacellular labeling after the cell type identification, and the recorded neuron was stained by current injection (1-10 nA, 50% duty cycle of 500 ms, 5-20 min). After completion of the electrophysiological recording, animals were given additional anesthesia (ketamine/Xylazine/Acepromazine, 200 mg/kg + 40 mg/kg + 20 mg/kg) and were perfused transcardially and fixed.

Statistical analysis.
All the data are given as mean ± standard error of the mean. For some data, we calculated the correlation coefficient (r), which was statistically tested by the two-tailed Student's t test. When two parameters had an r with significance and r was > 0.5, we performed a regression test and plotted a regression line. CFs and thresholds were tested by Kruskal-Wallis test. All other statistical analyses used the Wilcoxon signed-rank test. Criteria for significance were defined as P < 0.05.