Article

Activity-Dependent Transmission and Integration Control the Timescales of Auditory Processing at an Inhibitory Synapse Highlights

Authors

d

Cellular mechanism of long-lasting inhibition underlying echo suppression

Julian. J. Ammer, Ida Siveke, Felix Felmy

d

Physiologically relevant function of spillover and asynchronous release

Correspondence

Action potential suppression by multiple interacting pre- and postsynaptic mechanisms

In Brief

d

d

Long-lasting ionotropic inhibition in a fast spiking neuron with fast synaptic input

Ammer et al., 2015, Current Biology 25, 1562–1572 June 15, 2015 ª2015 Elsevier Ltd All rights reserved http://dx.doi.org/10.1016/j.cub.2015.04.026

[email protected]

An important transition in temporal processing occurs in a neural circuit that filters the distracting localization cues of echoes. Ammer et al. investigate the cellular basis of this transition and identify general mechanisms that dynamically adjust the duration of synaptic inhibition.

Current Biology

Article Activity-Dependent Transmission and Integration Control the Timescales of Auditory Processing at an Inhibitory Synapse Julian. J. Ammer,1,2 Ida Siveke,1 and Felix Felmy1,3,* 1Division of Neurobiology, Department Biology II, Ludwig-Maximilians University Munich, Großhaderner Straße 2, 82152 Planegg-Martinsried, Germany 2Graduate School of Systemic Neuroscience Munich, 82152 Planegg-Martinsried, Germany 3Bioimaging Center, Department Biology I, Ludwig-Maximilians University Munich, Großhaderner Straße 2, 82152 Planegg-Martinsried, Germany *Correspondence: [email protected] http://dx.doi.org/10.1016/j.cub.2015.04.026

SUMMARY

INTRODUCTION

To capture the context of sensory information, neural networks must process input signals across multiple timescales. In the auditory system, a prominent change in temporal processing takes place at an inhibitory GABAergic synapse in the dorsal nucleus of the lateral lemniscus (DNLL). At this synapse, inhibition outlasts the stimulus by tens of milliseconds, such that it suppresses responses to lagging sounds, and is therefore implicated in echo suppression. Here, we untangle the cellular basis of this inhibition. We demonstrate with in vivo wholecell patch-clamp recordings in Mongolian gerbils that the duration of inhibition increases with sound intensity. Activity-dependent spillover and asynchronous release translate the high presynaptic firing rates found in vivo into a prolonged synaptic output in acute slice recordings. A key mechanism controlling the inhibitory time course is the passive integration of the hyperpolarizing inhibitory conductance. This prolongation depends on the synaptic conductance amplitude. Computational modeling shows that this prolongation is a general mechanism and relies on a non-linear effect caused by synaptic conductance saturation when approaching the GABA reversal potential. The resulting hyperpolarization generates an efficient activity-dependent suppression of action potentials without affecting the threshold or gain of the input-output function. Taken together, the GABAergic inhibition in the DNLL is adjusted to the physiologically relevant duration by passive integration of inhibition with activity-dependent synaptic kinetics. This change in processing timescale combined with the reciprocal connectivity between the DNLLs implements a mechanism to suppress the distracting localization cues of echoes and helps to localize the initial sound source reliably.

Sensory information needs to be processed across multiple timescales to capture fast and fluctuating stimuli as well as slow changes in the sensory environment [1–3]. In the auditory system, temporal processing ranges from microseconds in the auditory brainstem to tens of seconds in the cortex [4, 5]. A particularly prominent change in processing timescales occurs between the first two stages of the binaural pathway. At the first stage, in the nuclei of the superior olivary complex (SOC), fast sub-millisecond integration is essential for sound localization [6]. At the next processing stage, the dorsal nucleus of the lateral lemniscus (DNLL), GABAA-receptor-mediated inhibition operates on a timescale of tens of milliseconds [7–12]. Importantly, this persistent inhibition (PI) outlasts the sound stimulus by tens of milliseconds, such that it is suited to suppress responses to echoes, and has therefore been suggested to be a neural correlate of the precedence effect [9–12]. Disruption of this inhibitory pathway impairs sound localization [13, 14], emphasizing its behavioral relevance. However, the cellular mechanisms that generate the PI are largely unknown. Most likely, the GABAA-receptor-mediated synaptic transmission has to be effective for a sustained period after the last presynaptic action potential (AP), as presynaptic activity does not persist after the stimulus [12, 15–17]. However, both synaptic and membrane time constants in mature DNLL neurons are too fast to directly explain the duration of PI in vivo [18]. Here, we identify the interaction of transmitter spillover and asynchronous release with passive integration of hyperpolarizing inhibition as a highly efficient mechanism to suppress APs for a duration that matches the PI in vivo. Thus, we propose that this activity-dependent interaction is the cellular mechanism of the PI and functions to suppress responses to echoes. RESULTS GABAergic Inhibition Is Ionotropic and Hyperpolarizing in the DNLL GABAergic inhibition in the DNLL originates from the opposite DNLL via the commissure of Probst and opposes incoming excitation from the SOC [7, 8, 17, 19–22] (Figure 1A). We determined the GABAergic reversal potential (EGABA) in the DNLL with

1562 Current Biology 25, 1562–1572, June 15, 2015 ª2015 Elsevier Ltd All rights reserved

A DNLL circuit

Figure 1. GABAA-Receptor-Mediated Inhibition in DNLL Neurons

C

perforated-patch

10 5 0 -5 -10

IPSP (mV)

B

ampl. (pA)

decay

(ms)

Co

ntr

ol SR

[Cli](mM)

mem. pot. (mV)

(A) Simplified DNLL circuit with major glutamatergic and GABAergic connections in red and blue. Minor -73mV projections were omitted for clarity. CP, commisEGABA sure of Probst; IC, inferior colliculus; LL, lateral lemniscus; LSO/MSO, lateral/medial superior olive. (B) Perforated-patch recording of a P25 neuron. -100 -50 The membrane potential was altered with conmem. pot. (mV) 10mV stant current injections, and IPSPs were stimu20ms lated via the commissure of Probst. Traces at E in vitro whole-cell F D resting membrane potential are in blue. -80 Control (C) IPSP amplitudes were fitted with a linear 6 SR function versus the holding potential to calculate 500 5 -90 the GABA reversal potential (EGABA). 4 250pA (D) Pooled data of measured reversal potential and 25ms 0 -100 3 corresponding estimated chloride concentration. Open symbols P18/19 n = 5; gray closed symbols 50ms P25–32 n = 3; average black closed symbols. EGABA [Cl]i (E) IPSCs stimulated via the commissure of Probst in control condition (single trials in gray; average in G in vivo whole-cell H I black) and during application of SR95531 (single cell1 40 Ipsi ear trials in light red; average in red) recorded with the cell2 cell3 current-clamp solution. Inset shows averages at expanded timescale. 20 10mV (F) Amplitude of IPSCs before and after applica3mV 100ms tion of SR95531 (n = 9). (G) Hyperpolarization evoked by auditory stimu0 25ms lation at the ipsilateral ear. Black trace is the 10 30 50 20dB re. thr. average of 20 repetitions in gray (cell_130717_01; level re. thr. (dB) 40dB re. thr. best frequency/stimulus frequency = 2,000 Hz; 35-dB sound level relative to threshold). (H) Decaying phases after stimulus offset of average responses for two different sound levels at 20 and 40 dB sound level relative to threshold. The decay phases of the hyperpolarization were fitted with a mono-exponential function (red lines). (I) Decay time constants of hyperpolarization versus sound level relative to threshold for n = 3 cells. Linear correlation coefficients cell1: r = 0.99, p = 0.01; cell2: r = 0.32, p = 0.6; cell3: r = 0.91, p = 0.03. Data are given as mean ± SEM.

gramicidin-mediated perforated-patch recordings (Figures 1B– 1D). EGABA was
91.3 ± 2.4 mV (n = 8; Figure 1D) and therefore 15 mV lower than the average membrane potential of
76.2 ± 0.8 mV (n = 55). Thus, similar to other mammalian auditory brainstem structures [23], GABAergic inhibition is hyperpolarizing in the DNLL with an estimated chloride concentration of 4.5 mM (Figure 1D). Next, we verified that inhibition evoked through the commissure of Probst is indeed mediated exclusively through GABAA receptors [21, 24]. Inhibitory postsynaptic currents (IPSCs) had a fast weighted decay time constant (tw = 4.8 ± 0.6 ms; n = 9) and were blocked by SR95531 (10–30 mM; Figures 1E and 1F). These IPSCs displayed no late component, similar to the evoked IPSPs (Figure 1B). Together, these findings exclude the activation of postsynaptic GABAB receptors. Thus, GABAergic inhibition in the DNLL is hyperpolarizing and mediated by ionotropic GABA receptors. Time Course of Hyperpolarization In Vivo In vivo, the GABAergic inhibition in the DNLL suppresses neuronal activity to lagging sounds in an intensity-dependent manner [10–12]. To test whether this activity-dependent regulation is reflected in the sound-evoked inhibition to the DNLL, we performed in vivo whole-cell current-clamp recordings. Ipsilateral sound stimulation led to a continuous hyperpolarization (n = 4; Figure 1G). One cell additionally received strong exci-

tation upon ipsilateral sound stimulation [17] and was therefore excluded from further analysis. In two of the three remaining cells, increasing the sound intensity resulted in a prolonged decaying phase of the hyperpolarization (maximal decay time constants: 36.9, 23.3, and 12.5 ms; Figures 1H and 1I). Thus, together with previous extracellular in vivo recordings [10–12], these results indicate that increasing sound intensity prolongs inhibition in the DNLL. IPSC Decay Depends on Presynaptic Activity One possible mechanism for the sound-intensity-dependent inhibition in vivo is a prolongation of the GABAergic IPSC decay following increased presynaptic activity [25]. To obtain an estimate of typical presynaptic firing rates, we recorded extracellular responses from DNLL neurons in vivo during monaural stimulation of the contralateral excitatory ear (Figures 2A and 2B). DNLL neurons responded with high firing rates of up to 170 Hz averaged across the full duration of the 200-ms stimulus. Higher firing rates of up to 430 Hz occurred in the initial part of the response (Figures 2A and 2B). Accordingly, we assessed whether presynaptic activity adjusts GABAergic decay kinetics in the mature DNLL in vitro by using in-vivo-like stimulation frequencies of 10–400 Hz (Figures 2C–2E). The weighted IPSC decay was similar for two pulses across all frequencies (4.4 ± 0.6 ms at 10 Hz to 5.5 ± 0.7 ms at 400 Hz;

Current Biology 25, 1562–1572, June 15, 2015 ª2015 Elsevier Ltd All rights reserved 1563

A in vivo extracellular

B

contra ear

200 ms first 25 ms first 12.5 ms

# of cells

# trials

15

50ms

10

# trials

5 0 0 0.5ms

5ms

250 response rate (Hz)

D

C frequency dependence 100Hz

400Hz

300pA 50ms

250Hz

300pA

20ms

100Hz

250Hz

30ms 500pA

10Hz

50ms

250pA

20ms

decay (ms)

F 2 p. 10 p. 20 p.

w

E stim # dependence

20

2 p. 10 p. 20 p.

15 10 5 0

50ms

0

200

400

freq. (Hz)

G IPSC summation

H IPSC depression 100Hz

250Hz

norm. IPSC ampl.

norm. env. ampl.

10Hz

2 1 0 0

5

10 15 20 Stim #

400Hz

1.5 1.0 0.5 0.0 0

5 10 15 20 Stim #

Figure 2. Inhibitory Decay Kinetics Are Modulated by Presynaptic Activity (A) Raster plot of extracellularly recorded single unit response to monaural auditory stimulation (best/stimulus frequency 960 Hz; 40 dB relative to threshold). (Top) Full length of stimulus is shown; orange box highlights first 12.5 ms. (Bottom) Shown is an enlarged view of first 12.5 ms and average extracellular spike waveform in black with SE in gray. (B) Histogram of maximal monaural response rates for n = 32 DNLL neurons in 50 Hz bins for the full length of the stimulus and the first 12.5 and 25 ms. (C) IPSCs were evoked at in-vivo-like stimulation frequencies of 10–400 Hz with 2–20 pulses. Shown are responses to 20 pulses at 100 (top) and 250 Hz (bottom). Artifacts are removed for clarity. (D) Aligned average decays after 20 pulses at different frequencies. Traces are averages of four repetitions. Inset shows last amplitude-normalized traces. (E) IPSCs in response to 2, 10, and 20 pulses at 250 Hz. The inset shows last amplitude-normalized traces.

ANOVA; p = 0.95; n = 9; Figure 2F), as well as for all pulse numbers tested at 10 Hz (4.4 ± 0.6 ms for 2 and 20 pulses at 10 Hz; ANOVA; p = 0.995; n = 9). However, an activity-dependent IPSC prolongation was observed with increasing pulse number at higher frequencies. At 400 Hz, the IPSC decay time constant increased 2-fold from 2 to 20 pulses (ANOVA; p = 0.004; n = 9). IPSCs did not sum at 10 Hz but displayed a 1.7-fold buildup during a 400-Hz train even though individual IPSC amplitudes showed depression during the stimulation trains (Figures 2G and 2H). Therefore, repetitive stimulation leads to IPSC summation and prolongation in an activity-dependent manner, thereby enhancing the inhibitory power at higher frequencies. At other inhibitory synapses, such a prolongation has been attributed to transmitter spillover [26–31], asynchronous release [30–35], and desensitization [36]. Spillover and Asynchronous Release If transmitter spillover between GABAergic synapses in the DNLL contributes to the IPSC prolongation, recruiting more presynaptic fibers should prolong the IPSC decay [29]. To test this prediction, we evoked IPSCs of different amplitudes for each cell (Figure 3A). The decay after single stimuli was unaffected by the increase in amplitude (Figures 3B–3D). In contrast, IPSC decays following 400-Hz trains were prolonged with increasing final amplitude (Figures 3B–3D), indicating an activity-dependent induction of spillover. Spillover can be pharmacologically verified by a change in IPSC decay upon application of a low-affinity antagonist [29, 30, 37]. 300 mM TPMPA, a low-affinity GABAA antagonist, reduced the IPSC amplitude of single pulses to 20% of control and following train stimulation to 40% of control (Figures 3E and 3F). In contrast to single pulses, the weighted decay following train stimuli was significantly accelerated (from 10.0 ± 1.1 to 8.7 ± 0.9 ms at 400 Hz; paired t test; p = 0.016; from 9.0 ± 0.9 to 7.9 ± 0.6 ms at 250 Hz; p = 0.04; Figure 3G). Conversely, low concentrations of a high-affinity antagonist should not influence the decay time constant in the case of spillover. Therefore, we applied sub-maximal concentrations of SR95531 (50–250 nM) as a control (Figures 3H–3J). As predicted, the weighted decay time constants remained unaffected although IPSC amplitudes were reduced to a similar extent (Figures 3I and 3J). Furthermore, the accelerating effect of TPMPA is not due to a possible weak agonistic effect on GABAB receptors as application of 5 mM baclofen resulted in similar final IPSC amplitudes following stimulation trains but did not affect the decay time constants (Figure S1). Thus, we conclude that GABA spillover contributes to the prolongation of the ionotropic IPSC decay following train stimulations.

(F) Summary plot of decay time constants versus stimulation frequency. The decays of the IPSCs after the stimulation trains were fitted with a bi-exponential function, and a weighted decay time (tw) constant was computed. Single cells are open symbols (n = 9); averages are closed symbols. (G) For the 20 pulse trains, the envelope amplitude was analyzed as total peak current after each individual stimulation pulse normalized to the first IPSC peak. (H) To estimate the synaptic depression during the train, the amplitude of the individual IPSCs was analyzed and normalized to the first IPSC amplitude. Data are given as mean ± SEM.

1564 Current Biology 25, 1562–1572, June 15, 2015 ª2015 Elsevier Ltd All rights reserved

decay (ms)

0

z 0H

2 1 0

3

z 0H 40

z 0H 25

2 1

20 10

z 0H 40

z 0H 40

25

0H

z

0

SP

0 25ms

SP

z 0H

decay (ms)

25ms

J Control low SR

w

Control low SR

ampl. (nA)

I

high affinity antagonist

500pA

40

25

0H

z

0

SP

25ms

H

10

z

500pA

4

20

0H

25ms

3

Control TPMPA

decay (ms)

ampl. (nA)

Control TPMPA

1 2 3 amplitude (nA)

G

F

low affinity antagonist

0

40

0.5 1.0 amplitude (nA)

w

E

10

0 0.0

SP

0 25ms

20

w

20

25

25ms

10

D

SP

500pA

20

-1

5ms

3ms

C 400Hz SP

slope (ms nA )

300pA

decay (ms)

B

amplitude dependence

w

A

Figure 3. GABA Spillover Contributes to Activity-Dependent IPSC Prolongation (A) GABAergic IPSCs evoked with different stimulation intensities in response to 1 (top) and 20 pulses at 400 Hz (bottom). Insets: amplitude-normalized IPSCs. (B) IPSC decays of the cell in (A) versus IPSC amplitude. (C) Slopes of linear fits as shown in (B) for single pulses (SP) (n = 17) and 400 Hz (n = 18). (D) IPSC decay versus amplitude for pooled data; symbols as in (B). SP: 50 IPSCs in 17 cells correlation coefficient r = 0.088, p = 0.54; 400 Hz: 57 IPSCs in 18 cells, correlation coefficient r = 0.59, p = 0.00002. (E) IPSC before (black) and during (red) application of 300 mM TPMPA in response to 400 Hz stimulation. Inset shows last amplitude-normalized traces. (F) Effect of TPMPA on IPSC amplitudes of 1 and 20 pulses at 250 Hz and 400 Hz. Bars represent average values; open symbols represent single cells (n = 11; paired t tests; SP: p = 0.0003; 250 Hz: p = 0.0009; 400 Hz: p = 0.002) (G) Effect of TPMPA on IPSC decay (paired t tests; SP: p = 0.43; 250 Hz: p = 0.040; 400 Hz: p = 0.016). (H) IPSC before (black) and during (blue) application of low concentrations of SR95531 for 400 Hz stimulation. Inset shows last amplitude-normalized traces. (I) Effect of low SR95531 on IPSC amplitudes of single (SP) and 20 pulses at 250 Hz and 400 Hz. Bars represent average values; open symbols represent single cells (n = 9–10; paired t tests; SP: p = 0.0004; 250 Hz: p = 0.001; 400 Hz: p = 0.004) (J) Effect of low SR95531 on IPSC decay (paired t tests; SP: p = 0.78; 250 Hz: p = 0.14; 400 Hz: p = 0.89). Data are given as mean ± SEM. Traces are averages of two to five repetitions. See also Figure S1.

Next, we tested the role of asynchronous release in the activity-dependent IPSC prolongation. To detect single asynchronous events more effectively, we used an internal solution with a high (50 mM) chloride concentration and repetitively evoked IPSCs with a 400-Hz stimulation (Figure 4A). We observed clear asynchronous events in all cells (n = 7). The probability of asynchronous events was significantly increased immediately after the stimulation (mean frequency in first 50 ms after stimulation: 95 ± 28 Hz; 200–250 ms after stimulation: 5 ± 1 Hz; paired t test; p = 0.016; Figures 4B and 4C). To interfere pharmacologically with asynchronous release, we applied the calcium chelator EGTA-AM to reduce the intracellular accumulation of calcium during trains [32–34, 38–40]. 200 mM EGTA-AM reduced IPSC amplitudes and accelerated weighted IPSC decays following train stimulation (from 7.4 ±

0.9 to 5.8 ± 0.7 ms at 250 Hz; paired t test; p = 0.054; and from 9.5 ± 1.4 to 6.7 ± 0.9 ms at 400 Hz; p = 0.011; Figures 4D–4F). As EGTA-AM also reduces overall transmitter release and thus reduces both spillover and asynchronous release, we repeated the application of EGTA-AM with prior block of the spillover component by TPMPA (Figures 4G–4I). Again, EGTA-AM decreased the IPSC amplitude (Figure 4H) and accelerated the weighted IPSC decay following train stimulations (Figure 4I; from 7.9 ± 1.0 to 7.1 ± 1.2 ms at 250 Hz; paired t test; p = 0.13; 8.2 ± 0.7 to 7.0 ± 0.9 at 400 Hz; paired t test; p = 0.04). From this additive effect of TPMPA and EGTA-AM, we conclude that both spillover and asynchronous release prolong the IPSC decay during increased presynaptic activity. In line with these activity-dependent mechanisms, the prolongation of IPSCs is even stronger when synaptic release probability is increased

Current Biology 25, 1562–1572, June 15, 2015 ª2015 Elsevier Ltd All rights reserved 1565

analysis of events

0.4

C freq. (Hz)

B

#events/trial

100pA

0.2

50

0.1 time (s)

25ms

0m

s

ms

0 0.0 0.0

20

25ms

100

0-5

1nA 1nA

0.2

200

0-2

A asynchronous release

1.5 1.0 0.5

10

z 40

10

w

0.5

0H

z 25

0H

SP

z 0H

1.0

I 20 decay (ms)

1.5

0.0

z 0H 40

z 0H 25

z 0H 40

z

0

25

25ms

20ms

TPMPA TPMPA+EGTA-AM

0H

100pA

2.0

SP

H TPMPA +EGTA-AM

ampl. (nA)

G

40

0H 25

SP

0

z

0.0 25ms

SP

500pA

20

decay (ms)

20ms

F

Control EGTA-AM

2.0

w

E Control EGTA-AM

ampl. (nA)

D calcium buffering

Figure 4. Asynchronous Release Contributes to Activity-Dependent IPSC Prolongation (A) IPSCs in response to 20 stimulation pulses at 400 Hz with 50 mM chloride in the internal solution. Shown are 22 trials in gray and the average in black. Inset shows an enlarged view of the decaying phase with a single trial highlighted in red. (B) (Top) The red trace from (A) after subtraction of the amplitude scaled average response. (Bottom) Shown is the mean histogram of detected events in 5-ms bins for the cell in (A). (C) Average frequency of events in 50-ms bins starting from 0 ms and 200 ms after stimulation offset (n = 7; paired t test; p = 0.016). (D) IPSCs before (black) and during application of 200 mM EGTA-AM (blue). Traces are averages of four repetitions. (E and F) Effect of EGTA-AM on (E) IPSC amplitude (n = 9
10; paired t tests; SP: p = 0.004; 250 Hz: p = 0.0009; 400 Hz: p = 0.0002) and (F) IPSC decay (paired t tests; SP: p = 0.23; 250 Hz: p = 0.054; 400 Hz: p = 0.011). (G) IPSCs recorded in 300 mM TPMPA (red) and during additional application of 200 mM EGTA-AM (blue). Traces are averages of four repetitions. (H and I) Effect of EGTA-AM after pre-application of 300 mM TPMPA on (H) IPSC amplitude (n = 7–9; paired t tests; SP: p = 0.006; 250 Hz: p = 0.001; 400 Hz: p = 0.0002) and (I) decay (n = 7–9; paired t tests; SP: p = 0.39; 250 Hz: p = 0.13; 400 Hz: p = 0.04). Data are given as mean ± SEM. See also Figure S2.

through elevation of the extracellular calcium concentration (Figure S2). Passive Integration Prolongs Inhibition How synaptic inputs influence the output of a neuron depends on the translation of synaptic currents into postsynaptic potentials. To investigate this translation, we recorded the same GABAergic input in voltage and current clamp in a given DNLL neuron (Figures 5A and 5B). IPSPs were generally slower than corresponding IPSCs (Figure 5C). This prolongation appeared stronger at higher stimulation frequencies (Figure 5C). Importantly, the difference between IPSP and IPSC decay correlated with the peak membrane potential of the IPSPs (Figure 5D). Thus, a larger peak magnitude of the IPSP causes a longer-lasting IPSP, which

cannot be explained by the response properties of DNLL neurons to hyperpolarizing current pulses (Figure S3). To investigate the dependence of IPSP duration on the membrane potential more systematically, we injected simulated inhibitory conductances (IPSGs) with different amplitudes and decay time constants spanning the physiological parameter space. For the same maximal conductance, slower conductance waveforms caused slower IPSPs (Figure 5E). Strikingly, increasing the maximal conductance of IPSGs with the same decay time constant also prolonged the resulting IPSP (Figure 5F), thereby leading to an equivalent relationship between IPSP prolongation and membrane potential as for the synaptic events (Figures 5D and 5G). Thus, the correlation between IPSP prolongation and membrane potential is caused by a

1566 Current Biology 25, 1562–1572, June 15, 2015 ª2015 Elsevier Ltd All rights reserved

A

B

C

D

E

F

G

H

I

J

K

L

M

Figure 5. Integration of GABAergic IPSCs Prolongs Inhibition (A and B) GABAergic inputs recorded in voltage and current clamp (VC and CC) in the same neuron. Traces show responses to 20 pulses at 100 Hz (A) and 400 Hz (B). Single trials are in gray; averages are in black. (C) IPSP versus IPSC decay. Grey symbols represent single values; black symbols represent averages across n = 8 cells (paired t tests; IPSPs versus IPSCs for 100 Hz: p = 0.009; 250 Hz: p = 0.001; 400 Hz: p = 0.002). (D) The tw difference between IPSP and IPSC versus peak membrane potential; linear correlation coefficient r =
0.79; p = 0.000005. (E) IPSPs in response to IPSGs (gGABA) with different decay kinetics. Voltage traces are averages of five sweeps. The bi-exponential decay consists of tfast = 5 ms, tslow = 20 ms, and fraction tfast = 0.7 (tw = 9.5 ms). (F) IPSPs in response to the bi-exponentially decaying conductance with amplitudes of 10 and 90 nS. Inset shows amplitude-normalized traces. (G) The difference in half-decay times of IPSPs/IPSGs (tIPSP/2,tIPSG/2) versus membrane potential. Dotted lines represent single cells; solid lines and closed circles represent average data of n = 14 cells. (H) tIPSP/2 versus conductance amplitude; same data as in (G). (I) Average tIPSP/2 versus conductance decay (tw IPSG) for the 10- and 90-nS conductance; n = 14. (J) Response of a DNLL neuron to a linear conductance ramp from 90 nS to 0 nS in 500 ms. The black trace represents average of ten single trials in gray. (K) The voltage response during the ramp in (J) versus the conductance of the ramp stimulus. Black trace represents the average of n = 8 neurons in gray, the neuron in (J) is shown in dark gray, and the blue line is the prediction computed from the steady-state potential (Vss) with Equation 1 with average gleak = 10.9 nS and Vm =
78.3 mV of the eight recorded neurons. (L) Mechanism of IPSP prolongation through the steady-state potential. Depicted is the non-linear steady-state potential Vss between gGABA and gleak as a function of gGABA. Inset: conductance stimuli used in (E). The filled symbols indicate time of peak and half maximal conductance. Note that the decay starts in the shallow region of the Vss for the 90 nS conductance and in the steep region for the 10-nS conductance. (M) Resulting voltage responses. Data are given as mean ± SEM. See also Figure S3.

conductance-amplitude-dependent mechanism (Figure 5H). This amplitude dependence was stronger for slower IPSGs (for t = 5 ms from 8.4 ± 0.3 ms at 10 nS to 11.7 ± 0.3 ms at 90 nS; Kruskal-Wallis; p = 2e-08; for t = 20 ms from 19.9 ± 0.8 ms at 10 nS to 36.7 ± 1.3 ms at 90 nS; Kruskal-Wallis; p = 2e-15; n = 14). Therefore, small differences in synaptic kinetics result in large differences in IPSP duration at large conductance amplitudes (Figure 5I). To understand the mechanism underlying the conductanceamplitude-dependent IPSP prolongation, we modeled a passive cell membrane with a capacitance, a leak conductance (gleak),

and a GABA conductance (gGABA) based on our experimental data. This simple model reproduced the experimental results well (Figure S3), implying that active membrane properties are not required for the IPSP prolongation, but the synaptic conductance amplitude determines the IPSP time course. An explanation could be synaptic saturation, which occurs for large conductances when the driving force for GABAergic current is lost near the peak of the IPSP as it approaches EGABA [41]. During the decay of the IPSP, the driving force increases again and together with the remaining gGABA generates a GABAergic current that delays the decay of the IPSP.

Current Biology 25, 1562–1572, June 15, 2015 ª2015 Elsevier Ltd All rights reserved 1567

A

B

C

D

E

F

G

H

Figure 6. Time Course of AP Suppression by GABAergic Inhibition (A) The excitatory conductance (gExc) was scaled to evoke APs with every EPSG pulse and shifted 5 ms relative to the GABAergic input stimulated with 20 pulses at 400 Hz (stim). AP probability (PAP) was computed from ten repetitions at each time shift. Effectiveness of AP suppression was quantified as the time of 50% AP probability (t50%PAP) after the last stimulation pulse. Inset: maximal t50%PAP for n = 10 neurons. See also Figure S4 for underlying IPSCs. (B) AP suppression by artificial gGABA scaled to final amplitudes of 10–90 nS with decay time constants of 5–20 ms and EGABA =
90 mV. Excitation was shifted relative to inhibition in 2.5-ms increments. Traces show AP suppression by 30 and 90 nS with t = 20 ms. PAP for t = 20 ms and gGABA from 10 to 90 nS.

To describe more quantitatively how the trajectory of the IPSP is controlled by the synaptic conductance, we applied an artificial conductance ramp that decayed linearly from 90 nS to 0 nS in 500 ms (Figure 5J). With such a slow decay, the contribution of the membrane time constant to the voltage response becomes minimal. Therefore, the voltage response at every time point reflects the steady-state potential determined by the opposing currents flowing through the neuron’s leak channels and the artificial GABAA receptor channels. The rate of change in the voltage responses to the linear conductance ramp was slowest for the highest gGABA values and accelerated for lower values of gGABA (Figure 5J). Importantly, the resulting voltage responses were markedly non-linear due to synaptic saturation when approaching EGABA (Figure 5K) [41]. Thus, synaptic saturation not only controls the level of hyperpolarization but can control the speed of voltage responses. Therefore, we next assessed whether synaptic saturation prolongs IPSPs of the faster, physiological IPSG waveforms. To test this hypothesis, we calculated voltage responses to fast IPSGs with the steady-state potential (Equation 1; Figures 5L and 5M). The IPSP decay was again slower for larger conductances, because it started in the shallow part of the steady-state potential. Taken together, the mechanism underlying the conductance-amplitude-dependent IPSP prolongation in the DNLL is passive and relies on synaptic saturation. These results depict a general mechanism by which IPSPs can be prolonged in neurons with fast-membrane time constants. Inhibition In Vitro Matches PI Duration In Vivo Next, we determined the functional relevance of the activitydependent prolongation of IPSPs. In the DNLL, GABAergic inhibition after a leading tone suppresses the response to a lagging tone that is excitatory when played alone [10–12]. Accordingly, we assessed whether such a long-lasting inhibitory effect can be achieved by synaptically evoked inhibition in our in vitro preparation. Evoked GABAergic IPSPs were paired with an injection of an excitatory conductance train that reliably evoked an AP with every pulse when presented alone (Figure 6A). On average, APs were suppressed for 22 ± 3 ms by a conductance of 44.9 ± 6.5 nS (Figure S4) with a tw of 8.8 ± 1.0 ms (n = 9). The range of AP (C) t50%PAP for different decays versus gGABA. Dotted lines represent single cells; solid lines represent average data; n = 9. The blue circle represents average of stimulated GABA input, with tw = 8.8 ± 1.0 ms. (D) t50%PAP for 10 and 90 nS versus conductance decay. n = 10 for 0 ms decay; n = 9 for all other decays. Kruskal-Wallis for 10 nS: p = 0.000004; for 90 nS: p = 4e-15. Average stimulated value in blue with gGABA = 44.9 ± 6.5 nS. (E) Bi-exponential conductances simulating TPMPA and EGTA-AM effects on the IPSG decay (siTPMPA, tw = 8.4 ms; siEGTA, tw = 6.9 ms, blue) were compared to control conditions (siControl, tw = 10.1 ms, black). Traces show responses to siControl and siEGTA at gGABA = 90 nS. (F) t50%PAP for n = 10 neurons. Kruskal-Wallis for 10-nS data: p = 0.81; for 90nS data: p = 0.036; Dunn-Holland-Wolfe significantly different pair siEGTA versus siControl. (G) For siControl, EGABA was set to the resting potential of the neuron to produce shunting inhibition. Traces show responses to identical gGABA (siControl; 90 nS) with either hyperpolarizing (black) or shunting EGABA (green). (H) t50%PAP for n = 11 neurons. The 10-nS gGABA in shunting condition failed to reduce PAP below 0.5 in 10 out of 11 neurons and made computation of t50% PAP impossible (indicated by values below zero, red dotted line). Paired t test for 90 nS data: p = 9e-07. Data are given as mean ± SEM.

1568 Current Biology 25, 1562–1572, June 15, 2015 ª2015 Elsevier Ltd All rights reserved

A

B

C

D

E

Figure 7. Input-Output Transformation Depends on Strength and Dynamics of Inhibition and Excitation (A) The excitatory conductance (gExc) was altered to span the dynamic range of the input-output relation in DNLL neurons during constant inhibitory conductance (gray dashed box, const.) and during the decaying phase of gGABA (green dashed box, dec). Excitation was shifted relative to inhibition in 3.33-ms increments. (B) AP probability (PAP) for the neuron in (A) during the decaying phase of inhibition was computed from ten repetitions at each time shift for different values of gExc (4–18 nS) at gGABA = 50 nS. Time = 0 at start of the gGABA decay. (C) t50%PAP versus gExc for different values of gGABA. Dotted lines represent single cells (n = 8). Filled circles average data at gExc values with at least n = 6 data points. For very high or very low gExc, t50%PAP could not be determined as PAP was either constantly above or below 50%. (D) Input-output functions during the decay and constant inhibition. (Top) PAP during dec. condition averaged across all pulses for ten repetitions versus gExc for different gGABA is shown. (Bottom) PAP averaged from 60 stimulus presentations at each gExc value for different constant background gGABA is shown. All input-output functions were fitted with a sigmoid function of the form P(gExc) = PMax/(1 + exp((gExc50 – gExc) / crate)). (E) (Top) gExc50 for input-output functions versus gGABA (const. ANOVA: p = 7e10; dec. ANOVA: p = 0.4). (Bottom) A gain index was calculated as crate/Pmax to compare the steepness of the input-output functions for both conditions (const. ANOVA: p = 0.002; dec. ANOVA: p = 0.054). Single cells and average of n = 8 cells in condition dec. are in light and dark green; single cells and the average of n = 8 cells in condition const. are in gray and black. Data are given as mean ± SEM.

suppression of 13–38 ms matches the reported values of PI in vivo [10–12]. To systematically dissect the contributions of the activitydependent conductance amplitude and decay on AP suppression, we injected artificial conductances resembling a typical IPSC train with variable decays (Figure 6B). Increasing the conductance amplitude for a given decay caused longer and more-effective AP suppression (Figures 6B and 6C; e.g., from 16.6 ± 2.7 at 10 nS to 56.7 ± 3.6 ms at 90 nS for t = 20 ms; n = 9; Kruskal-Wallis; p = 0.0000002). Slower decays generally produced longer AP suppression (e.g., 14.8 ± 1.6 for t = 5 ms versus 46.2 ± 3.4 ms for t = 20 ms at 50 nS; Kruskal-Wallis; p = 0.00005). The difference in the duration of AP suppression between the different underlying conductance decays was enhanced for large conductance amplitudes (Figure 6D). To evaluate the contribution of the membrane time constant to AP suppression, we applied a conductance waveform with instantaneous decay. This conductance waveform caused only very short AP suppression (4.8 ± 1.1 ms at 90 nS; Figure 6D), suggesting that the membrane time constant alone plays a minor role in PI. To decompose the contributions of spillover and asynchronous release to the PI, we compared the effect of conductance waveforms reflecting the IPSC kinetics under control, TPMPA, and EGTA-AM conditions (Figure 6E). The different waveforms produced similar AP suppression for a 10-nS conductance (siControl: 5.6 ± 1.0 ms; siTPMPA: 5.1 ± 1.0 ms; siEGTA: 5.1 ± 1.0 ms; Figure 6F). For a 90-nS conductance, AP suppression differed significantly between conditions (siEGTA 29.0 ± 2.4 ms; siTPMPA 32.0 ± 2.7 ms; siControl 39.3 ± 2.5 ms; n = 10; Kruskal-Wallis; p = 0.036). The 3-ms difference in weighted decay time constants between siControl and siEGTA translated into an 10-ms difference in AP suppression. Thus, small differences in IPSC decay can have large effects on AP suppression. The activity-dependent prolongation of the IPSC mediated by spillover and asynchronous release is therefore ideally suited to gradually adjust the AP suppression time in the DNLL. Next, we examined whether the prolonged hyperpolarization that results from synaptic saturation is crucial for the long inhibition in the DNLL or whether a shunting effect by the GABAergic conductance is sufficient to generate the long-lasting AP suppression (Figures 6G and 6H). For a 10-nS conductance, shunting inhibition failed to suppress APs effectively in 10 out of 11 cells. For 90 nS, shunting inhibition reduced AP suppression from 38.4 ± 1.3 ms to 20.4 ± 1.4 ms (n = 11; paired t test; p = 9e-07) compared with hyperpolarizing inhibition. Thus, the long-lasting hyperpolarizing IPSPs are essential to effectively suppress APs for a duration explaining PI in vivo. PI Depends on the Level of Excitation and Inhibition The level of incoming excitation from the SOC evoked by the lagging tone in vivo is expected to change depending on the contralateral sound level. As this might affect the duration of PI, we characterized the interaction of the GABAergic inhibition with variable levels of excitation (Figure 7). Varying the conductance amplitude for the excitatory conductance train (gExc) at a fixed value of inhibitory gGABA (siControl decay; Figure 7A) strongly affected the duration of AP suppression (Figures 7B and 7C). Nevertheless, even for high values of gExc, the initial APs were suppressed with realistic values of gGABA (11.8 ± 1.8 ms AP

Current Biology 25, 1562–1572, June 15, 2015 ª2015 Elsevier Ltd All rights reserved 1569

suppression for gExc = 18 nS and gGABA = 90 nS). Thus, the effective duration of PI in vivo depends on how gGABA and gExc co-vary upon auditory stimulation. The impact of inhibition is often described as a change in the neuronal input-output functions during steady-state inhibition [42–44]. Because the decaying phase of inhibition is the relevant part for the PI, we compared the effect of steady-state and dynamic inhibition on the DNLLs’ input-output functions (Figures 7A, 7D, and 7E). During the decaying phase of inhibition, APs early in the train were strongly, whereas those late in the train were slightly inhibited (Figure 7B). As a consequence, the threshold of the input-output function was not altered and the gain only slightly affected (Figures 7D and 7E). In contrast, constant inhibition affected all APs in the train equally and therefore led to a large additive threshold shift and a change in gain (Figures 7D and 7E). Taken together, constant inhibition strongly affects the threshold and gain of the input-output function, whereas the decay of inhibition predominantly sets the time window during which APs are transmitted but barely impacts the input-output characteristics of DNLL neurons. In this way, the information about sound source location following the PI is preserved in the rate code generated in the SOC. DISCUSSION Here, we identify the cellular mechanisms that implement a stimulus-dependent prolongation of inhibition in the DNLL [9–12]. We explain how synaptic and integrative properties of DNLL neurons in vitro lead to AP suppression in a graded, activity-dependent manner, matching the in vivo description of PI. These data strongly suggest that the interplay between the activity-dependent IPSC prolongation produced by spillover and asynchronous release and the non-linear integration of hyperpolarizing inhibition is crucial to achieve the physiologically relevant time course. This interplay is required as the fast kinetics of a single IPSC and the membrane time constants in the mature DNLL cannot explain the duration of PI in vivo [18]. Both spillover and asynchronous release contribute to the long-lasting inhibition mediated via the commissure of Probst. At the synaptic level, these mechanisms could be linked by assuming a loose micro-domain coupling between calcium influx and vesicle release [45] in agreement with the IPSC amplitude reduction by EGTA-AM. Loose micro-domain coupling is ideal to support calcium buildup and an increase in release probability during train stimulation. In turn, an increase in release probability favors the generation of spillover and asynchronous release. Thus, micro-domain coupling might constitute a structural basis for the generation of the activity-dependent prolongation of GABAergic IPSCs in DNLL neurons. The activity-dependent IPSC prolongation is a prerequisite for the effective prolongation of hyperpolarizing IPSPs by passive integration that is crucial to achieve long-lasting inhibition. This dynamic prolongation is caused by the non-linear effects introduced by synaptic saturation when the IPSP approaches EGABA. Through this process, the level of hyperpolarization is translated into the inhibitory time course and AP suppression. To our knowledge, this mechanism describes a novel non-linear synaptic computation in addition to the well-known non-linear effects of inhibitory conductance on the membrane time constant and

amplitude of postsynaptic potentials [41] or multiplicative effects on the input-output functions [44]. At the circuit level, this passive integration mechanism can be exploited in two ways. First, the number of simultaneously activated presynaptic inhibitory neurons defines the size of the compound inhibitory conductance and therefore prolongs the IPSP. Second, a frequency-dependent buildup in synaptic conductance of a single presynaptic neuron also prolongs the IPSP. Therefore, an increase in sound intensity, which recruits more presynaptic neurons and leads to higher firing rates [46, 47], most efficiently exploits synaptic saturation to prolong the IPSP. Through their impact on the IPSP shape, the decay and amplitude of gGABA control the suprathreshold responses in the DNLL. Spillover and asynchronous release add substantially to the activity-dependent AP suppression, despite their apparently small effects on the level of IPSC decays. Furthermore, hyperpolarizing inhibition generates far longer and more-effective AP suppression than mere shunting inhibition, indicating a vital function of the non-linear relaxation of the membrane potential caused by synaptic saturation. Taken together, inhibition in the DNLL is adjusted to the physiologically relevant duration by passive integration of hyperpolarizing inhibition with activity-dependent synaptic kinetics. Together with previous work [18, 25], our findings elucidate the cellular physiology of the neuronal circuitry generating PI. Excitatory signals of ascending binaural pathways are amplified in an NMDA-receptor-dependent manner generating high firing rates in DNLL neurons [18, 25, 48]. The high firing rates in turn prolong the synaptic output of these neurons by pre- and postsynaptic mechanisms, based on asynchronous release and spillover. This activity-dependent prolongation is enhanced by the passive integration of hyperpolarizing, GABAA receptormediated IPSCs in DNLL neurons. These interacting mechanisms help to slow sensory processing down from the microsecond to the millisecond timescale from the SOC to the DNLL and the inferior colliculus. Overall, the properties of this GABAergic input combined with the reciprocal inhibition between the two DNLLs implement a flexible gating mechanism for direct sounds and echoes [49, 50].

EXPERIMENTAL PROCEDURES All experiments were performed with protocols approved by the Regierung von Oberbayern in accordance with the German law (Tierschutzgesetz) governing animal welfare. In vitro recordings were performed in acute brain slices of Mongolian gerbils (Meriones unguiculatus) of postnatal day (P) 18–36 at 35 C in extracellular recording solution containing 125 NaCl, 25 NaHCO3, 2.5 KCl, 1.25 Na2HPO4, 1 MgCl2, 1.2–2 CaCl2, 25 glucose, 0.4 ascorbic acid, 3 myo-inositol, and 2 Na-pyruvate (pH 7.4) when bubbled with carbogen. All potentials reported in the manuscript were corrected for liquid junction potentials (LJP). For current-clamp and voltage-clamp recordings in Figures 1 and 5, the pipette solution consisted of (in mM): 145 K-gluconate; 4.5 KCl; 15 HEPES; 2 Mg-ATP; 2 K-ATP; 0.3 Na2-GTP; 7.5 Na2-phosphocreatine; and 5 K-EGTA (pH 7.3; LJP 17 mV), for voltage-clamp recordings with high internal chloride concentrations (Figure 4) in mM: 105 Cs-gluconate; 26.7 CsCl; 10 HEPES; 20 TEA-Cl; 5 Cs-EGTA; 3.3 MgCl2; 2 Na2-ATP; 0.3 Na2-GTP; 3 Na2-phosphocreatine; and 5 QX-314 (pH 7.2; LJP 11 mV), and for all other voltage-clamp recordings of (in mM): 140 Cs-gluconate; 15 HEPES; 3 MgATP; 0.3 Na2-GTP; 5 Na2-phosphocreatine; 5 Cs-EGTA; 2.5 QX-Cl; and 2 TEA-Cl (pH 7.3; LJP 17 mV). 100 mM Alexa 488 or 568 were added to the internal solution to control for cell type and location. For perforated-patch

1570 Current Biology 25, 1562–1572, June 15, 2015 ª2015 Elsevier Ltd All rights reserved

recordings, the pipette was tip filled with a solution containing (in mM) 130 KCl, 10 HEPES, 5 NaCl, 3 MgCl2, and 0.5 K-EGTA (pH 7.3; LJP 3.4) and then back filled with the same solution containing 50 mg/ml Gramicidin A (Merck) and 100 mM Alexa Fluor 568. Electrophysiological in vivo recordings (AZ 55.2-1-54-2531-57-05 and AZ 55.2-1-54.2531.8-211-10) were performed on adult (3-month-old) Mongolian gerbils (extracellular recordings from 14 animals; whole-cell recordings using the current-clamp solution from four animals). Electrophysiological data were analyzed with custom written procedures in IGOR Pro or Matlab. The steady-state potential between reversal potentials of gleak and gGABA was computed with equation Vss = ðgleak Eleak + gGABA EGABA Þ=ðgleak + gGABA Þ:

(Equation 1)

For more information, please refer to the Supplemental Experimental Procedures. SUPPLEMENTAL INFORMATION Supplemental Information includes Supplemental Experimental Procedures and four figures and can be found with this article online at http://dx.doi.org/ 10.1016/j.cub.2015.04.026. AUTHOR CONTRIBUTIONS

9. Yang, L., and Pollak, G.D. (1994). The roles of GABAergic and glycinergic inhibition on binaural processing in the dorsal nucleus of the lateral lemniscus of the mustache bat. J. Neurophysiol. 71, 1999–2013. 10. Yang, L., and Pollak, G.D. (1998). Features of ipsilaterally evoked inhibition in the dorsal nucleus of the lateral lemniscus. Hear. Res. 122, 125–141. 11. Burger, R.M., and Pollak, G.D. (2001). Reversible inactivation of the dorsal nucleus of the lateral lemniscus reveals its role in the processing of multiple sound sources in the inferior colliculus of bats. J. Neurosci. 21, 4830– 4843. 12. Pecka, M., Zahn, T.P., Saunier-Rebori, B., Siveke, I., Felmy, F., Wiegrebe, L., Klug, A., Pollak, G.D., and Grothe, B. (2007). Inhibiting the inhibition: a neuronal network for sound localization in reverberant environments. J. Neurosci. 27, 1782–1790. 13. Ito, M., van Adel, B., and Kelly, J.B. (1996). Sound localization after transection of the commissure of Probst in the albino rat. J. Neurophysiol. 76, 3493–3502. 14. Kelly, J.B., Li, L., and van Adel, B. (1996). Sound localization after kainic acid lesions of the dorsal nucleus of the lateral lemniscus in the albino rat. Behav. Neurosci. 110, 1445–1455. 15. Aitkin, L.M., Anderson, D.J., and Brugge, J.F. (1970). Tonotopic organization and discharge characteristics of single neurons in nuclei of the lateral lemniscus of the cat. J. Neurophysiol. 33, 421–440.

J.J.A. performed in vitro recordings, computational modelling, and analysis; I.S. performed in vivo recordings and analysis; F.F. supervised research; and J.J.A., I.S., and F.F. designed experiments and wrote the paper.

16. Kuwada, S., Fitzpatrick, D.C., Batra, R., and Ostapoff, E.M. (2006). Sensitivity to interaural time differences in the dorsal nucleus of the lateral lemniscus of the unanesthetized rabbit: comparison with other structures. J. Neurophysiol. 95, 1309–1322.

ACKNOWLEDGMENTS

17. Siveke, I., Pecka, M., Seidl, A.H., Baudoux, S., and Grothe, B. (2006). Binaural response properties of low-frequency neurons in the gerbil dorsal nucleus of the lateral lemniscus. J. Neurophysiol. 96, 1425–1440.

We thank Benedikt Grothe for discussions and the generous support; Martin Stemmler for discussions; George Pollak, Mike Burger, Christian Leibold, and Delwen Franzen for discussions and comments on the manuscript; and Na Li for technical help with in vivo recordings. This work was funded by DFG FE789/2-1, DFG CRC870 (http://www.dfg.de), BMBF Bernstein Fokus 01GQ0981 (http://www.bmbf.de), and the Elisabeth and Helmut Uhl Foundation (http://www.eh-uhl-stiftung.org/). Received: February 10, 2015 Revised: March 25, 2015 Accepted: April 14, 2015 Published: May 21, 2015

18. Ammer, J.J., Grothe, B., and Felmy, F. (2012). Late postnatal development of intrinsic and synaptic properties promotes fast and precise signaling in the dorsal nucleus of the lateral lemniscus. J. Neurophysiol. 107, 1172– 1185. 19. Glendenning, K.K., Brunso-Bechtold, J.K., Thompson, G.C., and Masterton, R.B. (1981). Ascending auditory afferents to the nuclei of the lateral lemniscus. J. Comp. Neurol. 197, 673–703. 20. Oliver, D.L., and Shneiderman, A. (1989). An EM study of the dorsal nucleus of the lateral lemniscus: inhibitory, commissural, synaptic connections between ascending auditory pathways. J. Neurosci. 9, 967–982. 21. Chen, L., Kelly, J.B., and Wu, S.H. (1999). The commissure of probst as a source of GABAergic inhibition. Hear. Res. 138, 106–114.

REFERENCES 1. Mauk, M.D., and Buonomano, D.V. (2004). The neural basis of temporal processing. Annu. Rev. Neurosci. 27, 307–340. 2. Kiebel, S.J., Daunizeau, J., and Friston, K.J. (2008). A hierarchy of timescales and the brain. PLoS Comput. Biol. 4, e1000209. 3. Baden, T., Nikolaev, A., Esposti, F., Dreosti, E., Odermatt, B., and Lagnado, L. (2014). A synaptic mechanism for temporal filtering of visual signals. PLoS Biol. 12, e1001972. 4. Brand, A., Behrend, O., Marquardt, T., McAlpine, D., and Grothe, B. (2002). Precise inhibition is essential for microsecond interaural time difference coding. Nature 417, 543–547. 5. Ulanovsky, N., Las, L., Farkas, D., and Nelken, I. (2004). Multiple time scales of adaptation in auditory cortex neurons. J. Neurosci. 24, 10440– 10453. 6. Grothe, B., Pecka, M., and McAlpine, D. (2010). Mechanisms of sound localization in mammals. Physiol. Rev. 90, 983–1012. 7. Adams, J.C., and Mugnaini, E. (1984). Dorsal nucleus of the lateral lemniscus: a nucleus of GABAergic projection neurons. Brain Res. Bull. 13, 585–590. 8. Shneiderman, A., Oliver, D.L., and Henkel, C.K. (1988). Connections of the dorsal nucleus of the lateral lemniscus: an inhibitory parallel pathway in the ascending auditory system? J. Comp. Neurol. 276, 188–208.

22. Kelly, J.B., van Adel, B.A., and Ito, M. (2009). Anatomical projections of the nuclei of the lateral lemniscus in the albino rat (Rattus norvegicus). J. Comp. Neurol. 512, 573–593. 23. Friauf, E., Rust, M.B., Schulenborg, T., and Hirtz, J.J. (2011). Chloride cotransporters, chloride homeostasis, and synaptic inhibition in the developing auditory system. Hear. Res. 279, 96–110. 24. Wu, S.H., and Kelly, J.B. (1996). In vitro brain slice studies of the rat’s dorsal nucleus of the lateral lemniscus. III. synaptic pharmacology. J. Neurophysiol. 75, 1271–1282. 25. Porres, C.P., Meyer, E.M.M., Grothe, B., and Felmy, F. (2011). NMDA currents modulate the synaptic input-output functions of neurons in the dorsal nucleus of the lateral lemniscus in Mongolian gerbils. J. Neurosci. 31, 4511–4523. 26. Alger, B.E., and Nicoll, R.A. (1982). Pharmacological evidence for two kinds of GABA receptor on rat hippocampal pyramidal cells studied in vitro. J. Physiol. 328, 125–141. 27. Overstreet, L.S., and Westbrook, G.L. (2003). Synapse density regulates independence at unitary inhibitory synapses. J. Neurosci. 23, 2618–2626. 28. Farrant, M., and Nusser, Z. (2005). Variations on an inhibitory theme: phasic and tonic activation of GABA(A) receptors. Nat. Rev. Neurosci. 6, 215–229.

Current Biology 25, 1562–1572, June 15, 2015 ª2015 Elsevier Ltd All rights reserved 1571

29. Balakrishnan, V., Kuo, S.P., Roberts, P.D., and Trussell, L.O. (2009). Slow glycinergic transmission mediated by transmitter pooling. Nat. Neurosci. 12, 286–294.

40. Chang, C.Y., and Mennerick, S. (2010). Dynamic modulation of phasic and asynchronous glutamate release in hippocampal synapses. J. Neurophysiol. 103, 392–401.

30. Tang, Z.Q., and Lu, Y. (2012). Two GABAA responses with distinct kinetics in a sound localization circuit. J. Physiol. 590, 3787–3805.

41. Koch, C. (1999). Biophysics of Computation: Information Processing in Single Neurons. (New York: Oxford University Press).

31. Nerlich, J., Keine, C., Ru¨bsamen, R., Burger, R.M., and Milenkovic, I. (2014). Activity-dependent modulation of inhibitory synaptic kinetics in the cochlear nucleus. Front. Neural Circuits 8, 145.

42. Chance, F.S., Abbott, L.F., and Reyes, A.D. (2002). Gain modulation from background synaptic input. Neuron 35, 773–782.

32. Lu, T., and Trussell, L.O. (2000). Inhibitory transmission mediated by asynchronous transmitter release. Neuron 26, 683–694. 33. Hefft, S., and Jonas, P. (2005). Asynchronous GABA release generates long-lasting inhibition at a hippocampal interneuron-principal neuron synapse. Nat. Neurosci. 8, 1319–1328. 34. Best, A.R., and Regehr, W.G. (2009). Inhibitory regulation of electrically coupled neurons in the inferior olive is mediated by asynchronous release of GABA. Neuron 62, 555–565. 35. Manseau, F., Marinelli, S., Me´ndez, P., Schwaller, B., Prince, D.A., Huguenard, J.R., and Bacci, A. (2010). Desynchronization of neocortical networks by asynchronous release of GABA at autaptic and synaptic contacts from fast-spiking interneurons. PLoS Biol. 8, pii: e1000492. 36. Jones, M.V., and Westbrook, G.L. (1995). Desensitized states prolong GABAA channel responses to brief agonist pulses. Neuron 15, 181–191. 37. Szabadics, J., Tama´s, G., and Soltesz, I. (2007). Different transmitter transients underlie presynaptic cell type specificity of GABAA,slow and GABAA,fast. Proc. Natl. Acad. Sci. USA 104, 14831–14836.

43. Mitchell, S.J., and Silver, R.A. (2003). Shunting inhibition modulates neuronal gain during synaptic excitation. Neuron 38, 433–445. 44. Silver, R.A. (2010). Neuronal arithmetic. Nat. Rev. Neurosci. 11, 474–489. 45. Eggermann, E., Bucurenciu, I., Goswami, S.P., and Jonas, P. (2012). Nanodomain coupling between Ca2+ channels and sensors of exocytosis at fast mammalian synapses. Nat. Rev. Neurosci. 13, 7–21. 46. Yang, L., and Pollak, G.D. (1994). GABA and glycine have different effects on monaural response properties in the dorsal nucleus of the lateral lemniscus of the mustache bat. J. Neurophysiol. 71, 2014–2024. 47. Kelly, J.B., Buckthought, A.D., and Kidd, S.A. (1998). Monaural and binaural response properties of single neurons in the rat’s dorsal nucleus of the lateral lemniscus. Hear. Res. 122, 25–40. 48. Kelly, J.B., and Kidd, S.A. (2000). NMDA and AMPA receptors in the dorsal nucleus of the lateral lemniscus shape binaural responses in rat inferior colliculus. J. Neurophysiol. 83, 1403–1414.

38. Atluri, P.P., and Regehr, W.G. (1998). Delayed release of neurotransmitter from cerebellar granule cells. J. Neurosci. 18, 8214–8227.

49. Meffin, H., and Grothe, B. (2009). Selective filtering to spurious localization cues in the mammalian auditory brainstem. J. Acoust. Soc. Am. 126, 2437–2454.

39. Maximov, A., and Su¨dhof, T.C. (2005). Autonomous function of synaptotagmin 1 in triggering synchronous release independent of asynchronous release. Neuron 48, 547–554.

50. Mysore, S.P., and Knudsen, E.I. (2012). Reciprocal inhibition of inhibition: a circuit motif for flexible categorization in stimulus selection. Neuron 73, 193–205.

1572 Current Biology 25, 1562–1572, June 15, 2015 ª2015 Elsevier Ltd All rights reserved