Inhibitory glycinergic neurotransmission in the mammalian auditory brainstem upon prolonged stimulation: short-term plasticity and synaptic reliability

Short-term plasticity plays a key role in synaptic transmission and has been extensively investigated for excitatory synapses. Much less is known about inhibitory synapses. Here we analyze the performance of glycinergic connections between the medial nucleus of the trapezoid body (MNTB) and the lateral superior olive (LSO) in the auditory brainstem, where high spike rates as well as fast and precise neurotransmission are hallmarks. Analysis was performed in acute mouse slices shortly after hearing onset (postnatal day (P)11) and 8 days later (P19). Stimulation was done at 37°C with 1–400 Hz for 40 s. Moreover, in a novel approach named marathon experiments, a very prolonged stimulation protocol was employed, comprising 10 trials of 1-min challenge and 1-min recovery periods at 50 and 1 Hz, respectively, thus lasting up to 20 min and amounting to >30,000 stimulus pulses. IPSC peak amplitudes displayed short-term depression (STD) and synaptic attenuation in a frequency-dependent manner. No facilitation was observed. STD in the MNTB-LSO connections was less pronounced than reported in the upstream calyx of Held-MNTB connections. At P11, the STD level and the failure rate were slightly lower within the ms-to-s range than at P19. During prolonged stimulation periods lasting 40 s, P19 connections sustained virtually failure-free transmission up to frequencies of 100 Hz, whereas P11 connections did so only up to 50 Hz. In marathon experiments, P11 synapses recuperated reproducibly from synaptic attenuation during all recovery periods, demonstrating a robust synaptic machinery at hearing onset. At 26°C, transmission was severely impaired and comprised abnormally high amplitudes after minutes of silence, indicative of imprecisely regulated vesicle pools. Our study takes a fresh look at synaptic plasticity and stability by extending conventional stimulus periods in the ms-to-s range to minutes. It also provides a framework for future analyses of synaptic plasticity.


INTRODUCTION
Fast and reliable synaptic transmission at high frequencies is one of the hallmarks of the auditory system. In particular, it is a key feature for processing interaural time and intensity differences, the two major features for computation of sound localization (reviews: Yin, 2002;Borst and Soria van Hoeve, 2012). Central auditory relay synapses can follow high-frequency inputs (>100 Hz) with great fidelity, conveying signals with a high degree of precision and reliability (review: Klug, 2011). A great deal of knowledge concerning synaptic transmission has been obtained from principal neurons in the medial nucleus of the trapezoid body (MNTB) and their giant presynaptic partner structure, the calyx of Held, between which phase-locked transmission occurs in a failure-free, one-to-one fashion up to a frequency of 800 Hz, at least over a short period of 20 ms (Taschenberger and von Gersdorff, 2000). MNTB neurons provide the major inhibitory input to neurons in the lateral superior olive (LSO), and the MNTB-LSO pathway is a model system for investigating inhibitory neurotransmission (Sanes and Friauf, 2000;Kandler, 2004;Noh et al., 2010). In contrast to the wellcharacterized calyx of Held-MNTB synapses, however, much less is known about the reliability of downstream MNTB-LSO synapses.
Despite the demands concerning stable signal transmission, synapses display substantial, mostly temporary, alterations in strength during repetitive use. Distinctive types of synaptic plasticity are exhibited, such as depression and facilitation, or combinations of both. In the auditory system, calyceal input to MNTB neurons has been extensively analyzed Taschenberger and von Gersdorff, 2000;Taschenberger et al., 2002;Trussell, 2002;von Gersdorff and Borst, 2002;Wong et al., 2003;Oleskevich et al., 2004). By contrast, in other auditory relay stations, short-term alterations are relatively unexplored (reviews: Bender and Trussell, 2011;MacLeod, 2011). Short-term depression (STD) in a brief time window of <1 s has been described at specialized terminals of auditory nerve fibers, the endbulbs of Held, which contact bushy cells in the cochlear nuclear complex. Here, STD occurs upon stimulation with 100-300 Hz over a period of 150 ms (Oleskevich and Walmsley, 2002;Manis, 2006, 2008;Wang et al., 2010Wang et al., , 2011. STD was also reported in GABAergic synapses between the inferior colliculus and the medial geniculate body (Venkataraman and Bartlett, 2013). In the auditory cortex, GABAergic IPSCs display STD in a layer-specific manner (Humberto et al., 2011). In the only LSO studies so far, STD in the inhibitory MNTB-LSO path was described in P11-15 mice at 23-25 • C with stimulus trains of 20 Hz lasting 1.05 s (Giugovaz-Tropper et al., 2011) and in P10-18 gerbils and P10-15 mice at 32 • C with stimulus trains of 100 Hz lasting 200 ms (Walcher et al., 2011).
Here, we have functionally analyzed glycinergic MNTB-LSO synapses with the aims to characterize their performance and to determine their limits. To do so, IPSCs were recorded from mouse LSO neurons in brainstem slices at two ages (P11 and P19), two temperatures (37 and 26 • C), and various frequencies (1-400 Hz). Moreover, "marathon experiments" involved very prolonged stimulation (20 min) comprising up to 30,600 stimuli, many-fold more than routinely used (cf. Galarreta and Hestrin, 1998). The rationale for this strategy is based on several arguments. First, continuous auditory stimulation is virtually ubiquitous in natural environments (Lalor et al., 2009). Second, brief tone bursts, commonly used stimuli for in vivo recordings, recruit bursts of action potentials (APs) at frequencies between 50 and 500 Hz (Klug, 2011). Finally, highly active conditions of prolonged stimulation are a better approximation of the in vivo situation than standard slice experiments. Our novel stimulus protocols address synaptic depression in both the millisecondsto-seconds (STD) and the tens of seconds-to-minute range, for which we use the term "synaptic attenuation" as the counterpart to the term synaptic augmentation (Thompson, 2000;Regehr, 2012). Concerning the mechanisms underlying STD and synaptic attenuation, we provide some evidence in favor of presynaptic processes.

ANIMALS
Experiments were performed on C57BL/6N mice of either sex at postnatal day (P)11 ± 1 or P19 ± 1. They were in accordance with the German law for conducting animal experiments and followed the NIH guide for the care and use of laboratory animals.
Patch pipettes were pulled from borosilicate glass capillaries (GB150(F)-8P, Science Products) with a horizontal puller (P-87, Sutter Instruments). They were connected to an Axopatch-1D patch-clamp amplifier (Molecular Devices) and had resistances of 3-7 M when filled with intracellular solution (in [mM]: HEPES 10; EGTA 5; MgCl 2 1; K-gluconate 140; Na 2 ATP 2; Na 2 GTP 0.3). The Cl − concentration of 2 mM was slightly lower than the native 5 mM previously determined for P9-11 via perforated-patch clamp recordings (Ehrlich et al., 1999) and was chosen to increase the driving force for inward-directed Cl − flow. With 2 mM [Cl − ] i , E Cl was −112 mV at 37 • C and −108 mV at 26 • C. Liquid junction potential was calculated using the program JPCalc (Barry, 1994); it amounted to 15.4 mV and was corrected offline. Data were sampled at 10 kHz, low-pass filtered at 5 kHz, and analyzed using ClampEX and ClampFit 8.2.0.235 (Molecular Devices) and Origin software (Origin Lab). Capacitive current transients were electronically compensated online, and the whole cell capacitance was estimated from this compensation. Throughout the experiments, the series resistance was monitored and ranged from 6 to 20 M . It was routinely compensated by 60-80%. If it exceeded 20 M , recordings were excluded from further analysis.
Recordings were performed in voltage clamp mode at a holding potential of −70 mV. To evoke inhibitory postsynaptic currents (IPSCs), a theta glass electrode (TST150-6, WPI) with a tip diameter of 10-20 μm was filled with ACSF and placed lateral to the MNTB ( Figure 1A). Monopolar pulses (100 μs duration) were applied through a programmable pulse generator (Master 8, A.M.P.I.) connected to a stimulus isolator unit (A360, WPI). The amplitude of the stimulus pulses ranged from 0.1 to 4 mA and was set to achieve stable synaptic responses with a jitter in amplitude of <50 pA.
For normalization, 40 control stimuli were presented at 1 Hz during a baseline period and the peak amplitudes of the 40 IPSCs were averaged (= control value, set to 100%; cf. Galarreta and Hestrin, 1998). Two protocols were employed for determining the synaptic performance of the MNTB-LSO connection. For protocol 1, pulse trains were applied for 40 s each, with stimulation frequencies from 1 to 400 Hz in random order, and a 3-min pause was introduced between each pulse train ( Figure 1B 1 ). For protocol 2, which assessed the performance over 20 min, 10 trials of stimulation were applied, alternating between 50 Hz stimulation for 60 s and 1 Hz stimulation for another 60 s (Figure 1B 2 ). 1 Hz stimulation for 60 s was chosen to describe the time course of recovery, rather than eliciting single test IPSCs at varying FIGURE 1 | Whole-cell patch clamp recordings from LSO neurons located in the central region of the nucleus upon electrical stimulation of their input fibers from the MNTB. (A) Schematic view of the circuit as seen in a coronal brainstem slice. A theta glass electrode was used for focal electrical stimulation and placed at the lateral border of the MNTB to reach the maximum number of connecting fibers (depicted in magenta). Glycinergic transmission was pharmacologically isolated with the depicted drugs. (B 1 ,B 2 ) Scheme of the two protocols used for prolonged and very prolonged stimulation, comprising up to 30,600 pulses.
time intervals (Giugovaz-Tropper et al., 2011). No pause was introduced between trials. In some control experiments, recordings were obtained from MNTB neurons in the current-clamp mode, and antidromic APs were evoked in the LSO by electrical stimulation of MNTB axons.

DATA ANALYSIS
Data analysis was performed with Mini Analysis 6.0.3 (Synaptosoft). At the beginning of each experiment, the control value of glycinergic IPSC peak amplitudes was obtained for each neuron (see above). The following peak amplitudes underwent statistical analysis: amplitudes of the 1st, 2nd, and 10th IPSC (IPSC 1 , IPSC 2 , IPSC 10 ), amplitudes after the 1st and the 40th s (IPSC 1 s , IPSC 40 s ), and averaged amplitudes over several 10-s periods, namely 30-40 s, 50-60 s, and 110-120 s (IPSC 30−40 s , IPSC 50−60 s , IPSC 110−120 s ). When individual IPSC peak amplitudes were ≤5% of the control value, the event was considered a failure. A failure was also attributed if the peak amplitude was <15 pA, which equaled the noise level. For such failures, the peak amplitude was set to 0 pA. For kinetic measurements, the decay time constant τ (decay to 37% of peak amplitude) was determined by fitting a single exponential to the decay phase of the IPSC.

STATISTICS
Statistical analysis was performed on data sets if n ≥ 7 (Winstat, R. Fitch Software). Outliers (more than four times standard deviation above/below mean) were excluded. Such outliers occurred in the range of 5.3-11.8% (median: 6.2%, mean: 8.0%). In absolute numbers, n gross = 19 became n net = 18 in the best case, and n gross = 17 became n net = 15 in the worst case. Samples with a Gaussian distribution (Kolmogorov-Smirnov) were compared using a paired or an unpaired two-tailed Student's t-test. A Wilcoxon-signed rank test was used for unequal variances. Equality of variances in unpaired samples Student's t-tests was assessed with an F-test. Values in bar charts are presented as mean ± s.e.m. Significance values were as follows: * p < 0.05, * * p < 0.01, * * * p < 0.001.

FREQUENCY-DEPENDENT SYNAPTIC DEPRESSION OF GLYCINERGIC IPSCs IN THE MILLISECONDS-TO-SECONDS RANGE
In a first series of experiments, synaptic current responses from LSO principal neurons were obtained by stimulating axons in the MNTB at P11, and glycinergic IPSCs were pharmacologically isolated at a holding potential of −70 mV ( Figure 1A). Under these conditions, IPSCs comprised outward currents with a fast rising phase (<1 ms) and short decay times (<5 ms), corroborating that they were mediated through glycine receptors (GlyRs). The performance of the glycinergic MNTB-LSO synapses was assessed by prolonged stimulation for 40 s with frequencies of 1-400 Hz (in random order), inserting a 3-min pause in between (Figure 1B 1 ). As spontaneous discharges in this frequency and time range have been described for mouse MNTB neurons in vivo at P11-12 and thereafter (Sonntag et al., 2009), our stimuli resembled physiological conditions. We were further motivated to stimulate the MNTB-LSO synapses for prolonged periods with 1-400 Hz because mean spontaneous firing rates in this domain (25-100 Hz) have been recorded in vivo from auditory nerve fibers in several species, such as cats (Liberman, 1978), gerbils (Hermann et al., 2007), and chinchillas (Temchin et al., 2008).
Representative current traces from a P11 LSO neuron at 37 ± 1 • C are illustrated in Figure 2A. At 1 Hz, every stimulus pulse elicited an IPSC (Figures 2A 1 ,B 1 ). As the stimulus frequency was increased, IPSC peak amplitudes declined, resulting in STD (Figures 2A 2 -B 1 ). There was no indication of a short-term facilitation. Despite the frequency-dependent STD, the neuron displayed a detectable IPSC to each of the first 40 stimulus pulses, even at 333 Hz. STD was most prominent during the first 10 pulses (Figure 2B 1 ), whereas the peak amplitudes appeared to be quite stable after 5 s (Figure 2B 2 ). With time, however, some pulses remained unresponded, i.e., failures occurred (Figure 2B 2, five values at 100 Hz touching x axis after >10 s).
The STD behavior shown exemplarily in Figure 2 was observed in the whole ensemble of P11/37 • C LSO neurons The last IPSC during the 40-s-period is shown on the right and depicted by its pulse number (#40). The 200 pA calibration bar is also valid for panels (A 2 ,A 3 ). (A 2 ,A 3 ) IPSCs during 10 and 100 Hz stimulation, respectively, from the same neuron as in panel (A 1 ). Similar to (A 1 ), the very last IPSC during the 40-s-period is shown on the right (#400 and #4000, respectively). (B 1 ) Time course of IPSC peak amplitudes from three LSO neurons at eight stimulation frequencies, ranging from 1 to 333 Hz (for color code, see inset). Peak values to the first 40 pulses (#1-40) are plotted. (B 2 ) Like (B 1 ), but depicting the time course of IPSC peak amplitudes during the complete 40-s-periods. For each stimulation frequency, peak amplitudes were sampled at 1-s-intervals. Notice the frequency-dependent short-term depression (STD).
(n = 4-22, depending on stimulus frequency; Figure 3). At each stimulus frequency and for virtually every neuron, the first IPSC (IPSC 1 ) had the highest peak amplitude ( Figure 3A, see Methods for normalization to 100%). This was quantified by calculating the ratio between the mean peak amplitude of IPSC 2 and IPSC 1 (IPSC 2 /IPSC 1 ). For example, at 1, 10, and 100 Hz, IPSC 2 /IPSC 1 ratios amounted to 0.84, 0.85, and 0.80, respectively (Table 1 and Figure 3B 1 ). After the initial decline between IPSC 1 and IPSC 2 , STD continued during the first 10 pulses, such that IPSC 10 was of significantly lower peak amplitude than IPSC 2 at stimulus frequencies ≥2 Hz (e.g., IPSC 10 /IPSC 2 ratio at 50 Hz: 0.59; Table 1, Figure 3B 1 ). Because of the prolonged stimulation period of 40 s, we extended the quantification from a pulse-based to a time-based analysis. During the first second of stimulation, peak amplitudes declined significantly and in a frequency-dependent manner >2 Hz (e.g., IPSC 1 s /IPSC 2 at 50 Hz: 0.54; Table 1, Figure 3B 2 ). No further decline occurred between IPSC 1 s and IPSC 40 s , except for stimulus frequencies of 50 and 100 Hz, implying that stable inhibitory response amplitudes are maintained in the seconds-to-minutes range up to 10 Hz at P11/37 • C ( Table 1).
The frequency-dependent depression behavior was also evident in the mean peak amplitudes obtained during the last 10 s of recording, when a steady-state scenario appeared to be present (30-40 s; Figures 3B 2 ,C). Peak amplitudes declined significantly with stimulus frequency up to 100 Hz, with no further decline at higher frequencies. The low mean amplitudes (<10% of the control) at ≥100 Hz are indicative of a massive depression in the P11/37 • C MNTB-LSO connections in the seconds-to-minute range when the stimulus frequency is >50 Hz (Figures 3B 2 ,C).
Concomitant with the frequency-dependent decline of peak amplitudes, LSO neurons were not able to respond reliably (i.e., failure-free) to high frequency stimulation, even during the first 40 pulses (Table 2, Figures 3A 1 -A 8 ). Virtually 100% fidelity (no failures) was observed up to 50 Hz, because only a single failure occurred in a total of 2840 events, namely at 10 Hz ( Figure 3A 4 ). At 100 Hz, however, 27% of the neurons (4 of 15) displayed failures, and the average failure rate amounted to 0.8% (5 failures in 600 events), still a low value (Figure 3A 6 ). The effect became more pronounced at even higher frequencies (333 Hz: 16.3% failure rate, 26 failures in 160 events, 3 of 4 neurons, Figure 3A 8, Table 2). When analyzing the failure rate during the last 40 pulses, 100% fidelity was present only up to a stimulus frequency of 10 Hz ( Table 2). At 50 Hz, 46% of the neurons (10 of 22) displayed failures with an average failure rate of 21% (188 failures in 880 events). Both the number of neurons with failures and the failure rate increased with stimulus frequency. Every neuron displayed failures at ≥100 Hz, and at 333 Hz, the failure rate amounted to 89.4% ( Table 2). In summary, 50 Hz appears to be the maximal stimulus frequency to which P11/37 • C MNTB-LSO synapses can continually respond within a time frame of milliseconds to tens of seconds.

TEMPERATURE-DEPENDENCY OF STD IN THE MILLISECONDS-TO-SECONDS RANGE
Next, we analyzed temperature effects on synaptic transmission. To do so, P11 LSO principal neurons were analyzed as described above, yet recordings were obtained at 26 ± 2 • C. Representative recordings are illustrated in Figure 4A. Similar to the scenario at 37 • C, IPSCs were reliably evoked at 1 Hz ( Figure 4A 1 ). IPSC amplitudes declined when the interpulse interval was shortened, Notice the decline to <20% at frequencies ≥100 Hz. Black frame depicts the last 10 data points in each trace that underwent statistical analysis shown in panel (C). (C) IPSC peak amplitudes decreased with increasing stimulation frequency. N numbers in the inset of (B 1 ) correspond to all panels. * * p < 0.01; * * * p < 0.001.
yet IPSC 1 to IPSC 40 were evoked reliably, even at a stimulus frequency of 200 Hz (Figures 4A 2 -B 1 ). However, prolonged stimulation (40 s) resulted in drastically reduced peak amplitudes after 5 s at stimulus frequencies ≥50 Hz, and many failures occurred at 100 and 200 Hz (Figure 4B 2 ). Since the IPSCs decay time was >3 ms at 26 • C (cf. Figure 8), we refrained from performing experiments with stimulus frequencies ≥333 Hz. The analysis across the ensemble of P11/26 • C LSO neurons (n = 4-18) revealed similarities as well as differences in STD behavior compared to 37 • C (Figure 5). Similarly to 37 • C, the mean IPSC 2 amplitude was significantly lower than the IPSC 1 amplitude at each frequency, with IPSC 2 /IPSC 1 ratios amounting to 0.88, 0.69, and 0.61 at 1, 10, and 100 Hz, respectively ( Table 1). Also similarly, further depression occurred from IPSC 2 to IPSC 10 (e.g., IPSC 10 /IPSC 2 ratio at 50 Hz: 0.61; Table 1, Figure 5B 1 ), and the amount of depression was generally higher than at 37 • C. The most prominent difference between 26 and 37 • C became obvious only upon prolonged stimulation. For example at 100 Hz, the decline of peak amplitudes between IPSC 2 and IPSC 1 s (obtained after 1 s) was more pronounced at 26 • C than at 37 • C (IPSC 1 s /IPSC 2 ratio: 0.24 vs. 0.39). The greater reduction with time at lower temperature was corroborated by the IPSC 40 s /IPSC 1 s ratios. For example at 50 Hz, the ratio was 0.07 at 26 • C, yet 0.61 at 37 • C ( Table 1). In line with this, the mean peak amplitudes obtained within 30-40 s at 50 Hz/26 • C were drastically lower than those at 10 Hz/26 • C, and there was no further decline beyond 50 Hz (Figures 5B 2 ,C). Together, a major temperature effect is that MNTB-LSO synapses can continually respond to 50 Hz stimulation at 37 • C, yet only up to 10 Hz at 26 • C.
Concerning the failure rate, P11/26 • C MNTB-LSO synapses were able to respond with 100% fidelity to the first 40 pulses only up to a frequency of 5 Hz, because at 10 Hz, 1.1% failures occurred in 2 of 11 neurons (5 of 440 events; Figures 5A 1 -A 4 , Table 2). This was again in contrast to 37 • C (0.3% failures) and further emphasized by a more than 10-fold higher failure rate at 100 Hz (26 • C: 8.3%; 37 • C: 0.8%, Figure 5A   difference to the corresponding failure rates at 37 • C ( Table 2). Together, these data show that 5-10 Hz is the highest stimulus frequency to which P11/26 • C MNTB-LSO synapses can continually respond within a time frame of milliseconds to tens of seconds.

AGE-DEPENDENCY OF STD IN THE MILLISECONDS-TO-SECONDS RANGE
Although synaptic transmission in the rodent MNTB-LSO pathway appears to be quite mature by hearing onset (Kim and Kandler, 2003;Sonntag et al., 2011), some maturation steps still take place thereafter (Sanes and Friauf, 2000;Awatramani et al., 2005). To take this into consideration, we addressed synaptic transmission more than 1 week after hearing onset which, in mice, occurs at about P10 (Mikaelian and Ruben, 1965;Ehret, 1976;Ehret and Romand, 1992). We recorded from P19 LSO principal neurons at 37 • C, essentially following the stimulus paradigms described above (highest stimulus frequency: 400 Hz). Figure 6 shows representative data from two neurons. Similar to P11/37 • C, amplitudes declined in a time-and frequency-dependent manner. When analyzing the sample of all LSO neurons (n = 7-15), STD was consistently observed, progressing up to 200 Hz (Figure 7, Table 1). In slight contrast to P11/37 • C, the mean IPSC 2 peak amplitude was significantly lower than that of IPSC 1 at only three of nine frequencies. For example, at 10 Hz, the IPSC 2 /IPSC 1 ratio amounted to 0.78 (p = 0.006). At P11/37 • C, the corresponding ratio was 0.85. A significant decline between IPSC 2 and IPSC 10 did not occur systematically. If there was one, it was moderate (e.g., IPSC 10 /IPSC 2 at 100 Hz: 0.59; Figure 7B 1 , Table 1) and similar to P11/37 • C. STD between IPSC 2 and IPSC 1 s was significant only at ≥10 Hz (Figure 7B 2 , Table 1). After the first second, peak amplitudes were quite stable and in a steady state, as evidenced by the finding that a statistically significant decline between IPSC 1 s and IPSC 40 s was evident only at 100 and 200 Hz (e.g., IPSC 40 s /IPSC 1 s at 100 Hz: 0.53; Table 1). Thus, 1-40 s depression at P19 was slightly less pronounced than at (A 2 ,A 3 ) IPSCs during 10 and 100 Hz stimulation, respectively, from the same neuron as in panel (A 1 ). Similar to (A 1 ), the last IPSC during the 40-s-period is shown on the right (#400 and #4000, respectively). (B 1 ) Time course of IPSC peak amplitudes from two LSO neurons at stimulation frequencies ranging from 1-200 Hz (for color code, see inset). Peak values to the first 40 pulses (pulse #1-40) are plotted. (B 2 ), Like (B 1 ), but depicting the time course of IPSC peak amplitudes during the complete 40-s-periods. For each stimulation frequency, peak amplitudes were sampled at 1-s-intervals. Notice the stronger amount of STD than at 37 • C, particularly at higher frequencies.
P11, because the latter displayed a statistically significant decline already at 50 Hz. Like at P11, mean peak IPSC 30−40 s amplitudes declined in a frequency-dependent manner in P19/37 • C LSO neurons, but they remained above 25% of the control up to 100 Hz (Figures 7B 2 ,C, Table 1). Noticeably, this level was about 4-fold higher than at P11/37 • C. The moderate difference between P19/37 • C and P11/37 • C in STD behavior was also reflected by the failures. Interestingly, at virtually every stimulus frequency, both the percentage of neurons with failures and the failure rate were lower at P11 ( Table 2). For example, at 100 Hz, 27% of the P11/37 • C neurons displayed failures during the first 40 pulses (0.8% mean failure rate), whereas 40% of the P19/37 • C neurons did so (3.5% mean failure rate). Thus, the performance of the MNTB-LSO connections shortly after hearing onset was surprisingly better in the millisecond range than that observed 1 week later. However, the opposite finding was achieved in the 40 s range, because P11 synapses displayed failure rates >20% at 50 Hz, whereas P19 synapses remained virtually failure-free up to 100 Hz ( Table 2).
Noticeably, the absolute amplitudes should be treated carefully. First, they may have been reduced by the usage of kynurenic acid (Mok et al., 2009). Second, the driving force for inward-directed Cl-flow may have been slightly artificial because of the chloride concentration of 2 mM in the whole-cell patch pipettes.

TEMPERATURE-DEPENDENCY OF SYNAPTIC ATTENUATION AND RECOVERY
In a next series of experiments, we assessed the capacity of P11 LSO neurons to recover from synaptic attenuation. To do so, we again obtained an initial baseline for normalization (1 Hz, 40 s) and then stimulated uniformly with 50 Hz, now extending the stimulus train to 60 s (Figure 1B 2 ). Immediately thereafter, the stimulus frequency was reduced to 1 Hz, and IPSCs were recorded for another 60 s. In the following, we refer to these two 60-s epochs as challenge period and recovery period, respectively. Recordings were performed at 37 • C (n = 22) and 26 • C (n = 17). Exemplary results from a neuron at 26 • C and another at 37 • C are illustrated in Figures 9A 1 ,B 1 , respectively, and the ensemble data are shown in Figures 9A 2 ,B 2 . As expected, synapses displayed synaptic attenuation during the first 40 s as shown before (cf. Figures 3B 2 , 5B 2 ). By the very end of the challenge period, peak IPSC amplitudes (IPSC 60 s ) had completely collapsed to Notice the virtual collapse to <10% at frequencies ≥50 Hz. Black frame depicts the last 10 data points in each trace that underwent statistical analysis shown in panel (C). (C) IPSC peak amplitudes decreased with increasing stimulation frequency more profoundly than at P11/37 • C. N numbers in the inset of (B 1 ) correspond to all panels. * * p < 0.01; * * * p < 0.001.
0.0 ± 0.0% at 26 • C, yet they remained at 20.6 ± 3.3% at 37 • C (Figures 9A 2 ,B 2 ). The latter value did not differ significantly from that obtained after 40 s (27.7 ± 4.8%; p = 0.226), corroborating that 37 • C responses are quite stable in the secondto-minute range. In contrast, IPSCs declined further between 40 and 60 s at 26 • C (IPSC 40 s : 2.9 ± 1.4%; p = 0.049). Together, this demonstrates that glycinergic MNTB-LSO neurotransmission is sustained at a steady-state level in the minute range only at physiological temperature. The better performance in the minute range at 37 • C was also evident in drastically higher peak amplitudes during the last 10 s of the challenge period: at 37 • C, the mean IPSC 50−60 s value was 23.3 ± 3.5%, but it was almost 40-fold lower at 26 • C (0.6 ± 0.2%; p = 1 × 10 −6 ; Figures 9C,D).
At both temperatures, IPSC 110−120 s peak amplitudes of the P11 MNTB-LSO connections were significantly higher than IPSC 50−60 s amplitudes ( Table 3; estimated recovery: 37 • C: 3.6fold; 26 • C: 36.0-fold). IPSC 110−120 s values amounted to 107.7 ± 5.0% of the control value at 26 • C and thus were overshooting (cf. Dinkelacker et al., 2000). At 37 • C, the corresponding value was significantly lower (83.8 ± 6.4%, p = 0.007; Table 3; Figure 9D). As the glycinergic neurotransmission recovered robustly after a 60-s-stimulus train (cf. Figure 9), we decided to challenge the synapses even more with the aim to determine their limits. We extended the stimulation time and applied the 60 s/50 Hz challenge and 60 s/1 Hz recovery protocol for a total of ten times (Trial 1-10, Figure 1B 2 ), resulting in 30,600 pulses over a period of 20 min. The results from such "marathon experiments" are depicted in Figure 10. On average, the glycinergic MNTB-LSO neurotransmission was resistant to very prolonged stimulation, as evidenced by the fact that mean peak amplitudes did not decline below 15% (Figures 10A,B). Nevertheless, synaptic augmentation progressed, albeit only slightly, from trial to trial ( Figure 10B, Table 3). For example, the mean IPSC 50−60 s in trial 10 was significantly lower than that in trial 1 (18.3 ± 3.6% vs. 23.3 ± 3.5%, p = 0.020). Likewise, the IPSC 1 amplitudes declined significantly, albeit only moderately, from trial 1 to trial 10 ( Table 3). Similar effects were seen for IPSC 2 , whereas IPSC 10 and IPSC 1 s did not decline during 20-min stimulation.

Frontiers in Neural
We also investigated the development of failures across trials during the challenge period. In a first step, we analyzed all IPSCs during the first 40 pulses. No failure was seen during trial 1, hardly any in trial 5 (0.4%), yet almost 10% in trial 10 ( Table 4), indicative of some worsening with time. In a second step, we analyzed the failures during the last second of the challenge period (59-60 s). Whereas the percentage of neurons with failures increased only moderately (45.5% in trial 1, 57.1% in trial 5 and 10), the failure rate increased more than 2-fold (from 14.3% in trial 1 to 29.3% in trial 10; Table 4), again demonstrating some worsening performance with time.

SYNAPTIC ATTENUATION DURING VERY PROLONGED STIMULATION AT P11/26 • C (10 TRIALS, 20 MIN, 50 Hz)
We next performed the "marathon experiments" at 26 • C to assess temperature dependency (Figure 11). In contrast to 37 • C, the mean IPSC 1 amplitude decreased drastically and highly significantly from trial 1-10, becoming more than 3-fold reduced ( Table 3). Likewise, IPSC 2 , IPSC 10 , and IPSC 1 s decreased significantly with increasing trial number, and the effects were much more pronounced than at 37 • C ( Table 3). The time course of depression was similar across trials, resulting in almost completely collapsed amplitudes after about 35 s in each case. Mean IPSC 50−60 s peak amplitudes were <5% of the control, with a significant difference between trial 1 and 5 only (Table 3, Figure 11B), probably because amplitudes jittered considerably. This result was due to the almost complete collapse obvious already in trial 1.
Trial-to-trial development of the failures also displayed major differences between 26 and 37 • C. During the first 40 pulses, the failure rate increased steadily from trial to trial, and much more rapidly than at 37 • C. This was evidenced by a considerable increase from 14.3 to 22.3% between trial 1 and 5 at 26 • C, yet a negligible increase from 0 to 0.4% at 37 • C ( Table 4). During the last second of the challenge period, however, there was no major difference in the failure rate between trials. Again, this can be explained by the fact that depression was already very robust in the first trial. Together, the failure rate was considerably higher than at 37 • C.

Frontiers in Neural Circuits
www.frontiersin.org

TEMPERATURE DEPENDENCY OF RECOVERY DURING VERY PROLONGED STIMULATION (10 TRIALS, 20 MIN, 50 Hz)
At both 37 and 26 • C, the course of recovery from synaptic attenuation was quite variable across neurons, particularly in later trials (Figures 10A, 11A). As assessed by the IPSC 50−60 s /IPSC 110−120 s ratios, peak amplitudes recuperated significantly in all trials and at both temperatures ( Table 3), but at 37 • C, the recovery courses were much more reproducible (Figures 10B, 11B). At 37 • C, mean IPSC 110−120 s amplitudes amounted to about 76.9-83.8% of the control value, with no significant differences between trial 1-5 and trial 1-10 ( Figure 10B, Table 3). In contrast, mean IPSC 110−120 s amplitudes at 26 • C became significantly reduced more than 2.5-fold ( Figure 11B, Table 3, 107.7 ± 5.0% in trial 1, 43.9% ± 5.5 in trial 10). For a direct illustration of the temperature-dependent differences and for further quantification, we superimposed the curves obtained at both temperatures during trial 1, 5, and 10 ( Figure 12A). At the end of each challenge period, IPSC 50−60 s values at 26 • C were consistently lower than at 37 • C (Figures 12A,B; p = 1 × 10 −6 , p = 2 × 10 −4 , p = 1 × 10 −4 for trials 1, 5, 10). At the end of the recovery period, however, the situation was more diverse: at 37 • C, IPSC 110−120 s values were significantly lower than those obtained at 26 • C in trial 1, yet they were higher in trial 10. No difference occurred in trial 5 (Figures 12A,C; p = 0.007, p = 0.188, p = 0.005 for trials 1, 5, 10). The temperature dependency of STD and amplitude recovery across trials was reflected by the time course of the failure rates. During the first 40 pulses at 37 • C, failure rates of about 10% occurred in trial 10 only, whereas this level was reached already in trial 5 at 26 • C (Table 4, Figure 12D). At the end of the challenge period (59-60 s), the temperature effect on the failure rate was even more pronounced, because values at 26 • C exceeded 60% in all trials (Table 4, Figure 12D). In contrast, the failure rate during 59-60 s was drastically lower at 37 • C (<30%), increasing moderately with trial number (Table 4, Figure 12D). Concerning the recovery at 37 • C during the first 10 s (60-70 s), 9.1% of the neurons displayed failures in trial 1, whereas 14.3% did so both in trial 5 and trial 10, demonstrating a mild increase (Table 4). At 26 • C, however, the percentage of neurons with failures was high already in trial 1 (52.9%) and remained high until the end (Table 4). Regarding the 60-70 s failure rate, the recovery of P11/37 • C MNTB-LSO synapses was very good initially, yet it deteriorated from trial to trial and became 8.5-fold worse (trial 1: 1.4%; trial 10: 11.9%, Table 4, Figure 12E). In contrast, at 26 • C the failure rate was much higher already from the beginning (trial 1: 16.5%) and remained at this high level during the "marathon experiment" (Table 4, Figure 12E). Thus, the performance at 37 • C in the 20-min range was manifold superior to that at 26 • C, but it also became less reliable with time. At the end of the recovery period (110-120 s) in trial 1, the failure rate was negligible at both temperatures (Table 4, Figure 12E). By the time trial 10 was finished, however, it had increased considerably at both temperatures, with a similar rate of about 10% in both cohorts. Still, the percentage of neurons displaying failures at 26 • C was almost twice as high as at 37 • C (17.6 vs. 9.5%). Together, these data demonstrate a relatively minor trial-to-trial deterioration of the glycinergic MNTB-LSO transmission at physiological temperature, whereas at lower temperature, the performance becomes impaired quickly.

POSSIBLE CONTRIBUTION OF MNTB AXON CONDUCTION PROBLEMS TO IPSC FAILURES
In a final series of experiments, we addressed the question whether possible AP conduction problems (Bucher and Goaillard, 2011) in axons of MNTB neurons may lead to the absence of neurotransmission and, therefore, to the absence of IPSCs. To do so, we compared our results obtained upon MNTB stimulation (orthodromic, synaptic) while recording from P11/37 • C LSO neurons with those obtained via antidromic, nonsynaptic stimulation of MNTB axons in the LSO while recording from MNTB neurons in the current-clamp mode. Orthodromic stimulation resulted in high-fidelity synaptic transmission during the first 40 pulses at stimulus frequencies from 1 to 200 Hz (Figures 13A 1 ,B 1 ). The great majority of neurons (6 of 8) displayed virtually no failures up to 100 Hz, but the fidelity declined to 98 ± 2% at 200 Hz and 84 ± 0.13% at 333 Hz (Figure 13C 1 ). During the last 40 pulses of the 40-s train, however, many more failures were apparent, and 100% fidelity was maintained only up to 10 Hz (Figures 13A 2 -C 2 ). Above 10 Hz, the fidelity rate gradually declined to 83 ± 12% at 50 Hz and to 17 ± 10% at 333 Hz (Figure 13C 2 ). 90% fidelity (4 failures in 40 events) was computed at 280 Hz for the first 40 pulses and at 28 Hz for the last 40 pulses.
In the antidromic stimulation experiments, the pattern of successful responses (APs) and failures was quite similar to that obtained via orthodromic stimulation (Figures 13D-F). The performance declined with increasing stimulus frequency, such that APs occurred in response to the first 40 pulses in 93 ± 5% of cases at 100 Hz and 83 ± 11% at 200 Hz (Figure 13F 1 ). No failure occurred at frequencies ≤50 Hz. During the last 40 pulses, the performance was worse, the respective values being 45 ± 13% at 100 Hz and 6 ± 4% at 200 Hz (Figure 13F 2 ). Ninety percent fidelity occurred at 160 Hz during the first 40 pulses and at 55 Hz during the last 40 pulses. Thus, axons of P11/37 • C MNTB neurons can conduct APs absolutely reliably up to 50 Hz, even upon very prolonged stimulation. Collectively, the results from orthodromic and antidromic stimulation show that some fidelity decrease in synaptic transmission of the MNTB-LSO connection can be attributed to impaired AP conduction properties of MNTB neurons, rather than to deficits of the synaptic transmission machinery. However, the observed STD behavior is unlikely due to AP conduction problems and is rather a presynaptic effect (as further addressed in the Discussion).

DISCUSSION
Mature LSO neurons receive powerful inhibitory, glycinergic input from MNTB neurons, which are able to respond reliably to high frequency stimulation supplied via the excitatory calyx of Held, up to 800 Hz for brief trains lasting ca. 20 ms (Taschenberger and von Gersdorff, 2000;Mc Laughlin et al., 2008). Our results provide evidence that the subsequent glycinergic MNTB-LSO connections are less astounding in this time range, as they sustain faithful synaptic transmission merely to considerably lower frequencies (ca. 100 Hz at P11/37 • C, cf. Figure 3). Although featuring robust STD and synaptic attenuation, the P11/37 • C connections are nevertheless continuously functional in the 10 s-to-1 min range and, if recovery periods are introduced, neurotransmission is reliably sustained for at least 20 min upon 50 Hz stimulation. The performance crucially depends on temperature, as at 26 • C, faithful neurotransmission falls close at >10 Hz, and collapses almost completely during prolonged stimulation with 50 Hz. Our data also demonstrate some moderate maturational changes from P11 to P19. The unusually short glycinergic IPSC decay time likely reflects specifically fast   Figure 1B 2 ). Ten trials of 50 and 1 Hz/60 s episodes were applied (Trial 1-10), amounting to a total of 30,600 stimuli. (A) Time course of the IPSC peak amplitudes (one color per neuron, n = 21-22). Mean values ± s.e.m. are shown in black (simple moving average of five data points). Due to the smoothing, failures are no longer discernable. (B) Superposition of the ten averaged time courses, illustrating the similarities across trials (traces for trials 1, 5, and 10 are highlighted by a thick black, turquoise, and magenta line, respectively). Statistical comparison was done between trials 1 and 5 and trails 1 and 10. * p < 0.05. deactivation kinetics of GlyRs in LSO neurons, and thus rapid channel closure, which optimizes the integration of interaural intensity differences. On the other hand, the relatively normal recovery course from STD rules out a specific, ultrafast replenishment machinery in glycinergic MNTB-LSO synapses (Cho et al., 2011). STD in glycinergic synapses is much less studied than in GABAergic synapses, and analyses are surprisingly limited to the auditory system, where they are only recent and rudimentary (Couchman et al., 2010;Kuo and Trussell, 2011). The present Frontiers in Neural Circuits www.frontiersin.org study provides the most extensive analysis of glycinergic shortterm plasticity so far, and it also provides a fresh look into neurotransmission within a prolonged time window. In contrast to most studies, yet similar to two previous reports (Galarreta and Hestrin, 1998;Klyachko and Stevens, 2006), our analysis comprised normalization of the peak amplitude of IPSC 1 and all subsequent IPSCs to a control value (average from 40 IPSCs obtained at 1 Hz stimulus frequency prior to high frequency stimulation; Figures 1B 1 ,B 2 ). Under these conditions, mean peak amplitudes of IPSC 1 , which was always evoked after a 3-min pause, exceeded 110% at every stimulus frequency when recordings were performed at 26 • C (Table 1, Figures 5B 1 ,B 2 ). We propose that this initial overshoot may be due to disturbed and imprecise regulation of docking and/or priming processes, resulting in an increased number of fusion-competent vesicles after long periods of synaptic silence (cf. Hermann et al., 2007;Klug, 2011;Klug et al., 2012), for very similar treatises on the lack of chronic background activity on short-term plasticity). As such overshooting IPSCs were rare at 37 • C, the disturbed regulation appears to be mainly a temperature artifact. It may also explain the overshooting recovery course in trial 1 at 26 • C (Table 3, Figures 9C,D). Furthermore, our results add a general caveat to room temperature studies, because they overestimate the amplitude of the first event and, consequently, the amount of STD.

CHARACTERISTICS OF STD
In virtually all of our recordings, IPSC peak amplitudes declined rapidly within the first 10 events, such that the IPSC 10 amplitudes Frontiers in Neural Circuits www.frontiersin.org Mean values ± s.e.m. are shown in black (simple moving average of five data points). (B) Superposition of the ten averaged time courses, illustrating the differences between trials, particularly during the recovery phases (traces for trials 1, 5, and 10 are highlighted by a thick black, turquoise, and magenta line, respectively). Statistical comparison was done between trials 1 and 5 and trails 1 and 10. * * p < 0.01; * * * p < 0.001.
reached a fraction of the control value at stimulus frequencies ≥5 Hz, regardless of age and temperature (cf. Figures 3B 1 ,  5B 1 , 7B 1 ). Our results further show a prominent frequency dependency. When comparing the amount of STD observed in the present study with data obtained from various synapse types within and outside the auditory system, STD in the glycinergic MNTB-LSO connections is neither particularly high nor low, but tends to be on the low side ( Table 5, cf. values for 100 Hz stimulus frequency). We found no indication for short-term facilitation, consistent with the idea that STD is key to maintaining temporal precision (Kuba et al., 2002;Cook et al., 2003). The absence of short-term facilitation at P11 and P19 in the MNTB-LSO synapses makes a developmental regulation of short-term plasticity unlikely within this time period, namely depression in younger and facilitation in mature synapses, as described for glutamatergic EPSPs in the rat neocortex between P14 and P18 (Reyes and Sakmann, 1999) and for GABAergic IPSCs in the gerbil auditory cortex after hearing onset (Takesian et al., 2010). Interestingly, preliminary results obtained from P30 MNTB-LSO connections also revealed no short-term facilitation, but STD, consistent with findings from mature endbulb of Held synapses (Wang and Manis, 2008).

PLATEAU PHASE DURING PROLONGED STIMULATION
An interesting observation in the course of STD was a plateau phase, which was centered at 3-4 s and lasted several seconds. This plateau phase was particularly evident at P11/37 • C during 50 Hz stimulation ( Figure 9C). A similar phenomenon, named delayed response enhancement, has been described in glutamatergic hippocampal synapses, where it depends on presynaptic synapsins I and II, is temperature-dependent, and most prominent in adults "marathon experiments." The underlying mechanism of this plateau phase in the glycinergic MNTB-LSO connections needs to be characterized in future studies.

RELIABILITY DURING VERY PROLONGED STIMULATION
The present study is quite novel in that the vast majority of studies so far have assessed synaptic depression with very few tetanic pulses (ca. 20), whereas we have stimulated over a period of 60 s and in 10 trials, resulting in 30,600 stimuli total (marathon experiments). We did so because the IPSC peak amplitudes had not reached a steady state after 20 pulses. Rather, they further declined considerably, e.g., reaching values of <30% after 40 s at 50Hz/P11/37 • C (cf. Table 1, Figure 3B 2 ). These results are consistent with the rare results described elsewhere for such IPSCs, namely in the hippocampus (Klyachko and Stevens, 2006). EPSCs in this intermediate range have been analyzed more extensively, for example in the calyx-MNTB synapse (deLange et al., 2003), primary hippocampal neurons (Sara et al., 2002), autaptic hippocampal neurons (Lambert et al., 2010), and the neuromuscular junction (Richards et al., 2003). Together the results, like ours, show that STD does not reach a plateau level during the first 1-2 s of tetanic stimulation. In our case, steady state was not reached until about 30 s into the experiment. Glycinergic IPSC peak amplitudes of the MNTB-LSO connection were not depressed below 20%, and the failure rate was remarkably low, even when stimulation lasted ≥1 min (P11/37 • C/50 Hz; cf. Figure 10B, Table 4). As our stimulus paradigm included a 1-min-long recovery period between subsequent 1-min challenge epochs, we cannot rule out that IPSC peak amplitudes will diminish further when the stimulus train is extended beyond 1 min. Preliminary results (Bakker and Friauf, unpublished) indeed demonstrate a further decline when stimulation lasts 10 min, yet the transmission stays functional, with mean peak amplitudes amounting to about 15% during the last minute.

PARTICIPATION OF GABA B RECEPTORS?
As shown in gerbils, LSO neurons release GABA during spiking activity which acts as a retrograde transmitter at presynaptic GABA B receptors, leading to an adjustment of the synaptic strength (Magnusson et al., 2008) and rendering the system more dynamic . Such a scenario can be ruled out in our experiments which were done under voltage-clamp conditions, thus preventing spike generation. Moreover, in current-clamp conditions, the glycinergic MNTB-LSO neurotransmission would have been hyperpolarizing because of the chloride concentrations used in the bath and pipette solutions. Therefore, we conclude that retrograde signaling via presynaptic GABA B receptors was not confounded by our pharmacological blockade of these receptors with CGP55845. Still, GABA may have been co-released together with glycine from the axon terminals of MNTB neurons (Kotak et al., 1998;Nabekura et al., 2004;Wojcik et al., 2006)   Numbers to the right are the numbers of successful spike propagation. (E 2 ) Dot plots from the same MNTB neuron, again stimulated at seven frequencies (1-200 Hz), but now illustrating fidelity behavior to stimulus number 3961-4000. Frame at 100 Hz depicts the scenario shown in (D 2 ). Notice the occurrence of an increasing number of failures at frequencies ≥100 Hz. (F 1 ,F 2 ) Fidelity data of the population of LSO neurons (n = 7) as a function of stimulus frequency. 100% fidelity means 40 successful responses to 40 stimuli (40/40). Mean values ± s.e.m. are shown by black squares. Notice that the course is similar to that seen upon orthodromic stimulation.

PRE-OR POSTSYNAPTIC MECHANISM OF DEPRESSION?
STD can be caused by depletion of transmitter release (von Gersdorff and Matthews, 1997). Likewise, postsynaptic receptors can be desensitized (Jones and Westbrook, 1996;Overstreet et al., 2000) or saturated (Kirischuk et al., 2002) by repetitive exposure to neurotransmitter. However, in contrast to GABA A receptors (Jones and Westbrook, 1995), GlyRs do not desensitize heavily, i.e., they do not enter long-lived closed states in the prolonged presence of glycine (Akaike and Kaneda, 1989;Lewis et al., 1991;Melnick and Baev, 1993;Harty and Manis, 1996;Breitinger et al., 2004;Mørkve and Hartveit, 2009). Therefore, we conclude that the observed STD is a presynaptic phenomenon. STD is often ascribed to depletion of the readily releasable vesicle pool (Rizzoli and Betz, 2005).
Inactivation of presynaptic Ca 2+ channels (Forsythe et al., 1998;Xu and Wu, 2005;Hennig et al., 2008;Mochida et al., 2008) or the refractoriness of release sites during repetitive activation (Dittman et al., 2000) are also feasible (see González-Inchauspe et al., 2007, for counterarguments). In contrast, feedback activation of presynaptic autoreceptors, as shown for GABA, glutamate, and adenosine Zucker and Regehr, 2002), appears to be unlikely, because high-affinity metabotropic GlyRs are unknown. Quite to the contrary, activation of presynaptic ionotropic GlyRs opens presynaptic Cl − channels, which in turn depolarizes the terminal, resulting in opening of voltage-operated Ca 2+ channels, elevated presynaptic [Ca 2+ ] i , and, ultimately, enhanced neurotransmitter release (Turecek and Trussell, 2001;Zucker and Regehr, 2002  results of increased STD at lower temperature are also in line with a presynaptic effect, because physiological temperatures accelerate vesicle recruitment and reduce STD (Kushmerick et al., 2006). Finally, we can rule out a collapse of the chloride driving force in the LSO neurons during repetitive stimulation (Ehrlich et al., 1999), and thus a postsynaptic effect, because we performed our experiments in the whole-cell mode, thereby clamping E Cl to −112 mV and generating a constant driving force. Our results from antidromic stimulation experiments imply that AP conduction failures, which occurred only at stimulus frequencies ≥100 Hz, do not at all contribute to impaired transmission at frequencies ≤50 Hz. Rather, the presynaptic release machinery cannot faithfully follow such frequencies, resulting in failures. Furthermore, as the fidelity of AP conduction at 100 Hz was 45% (Figure 13F 2 ), and as about eight MNTB neurons converge onto a single LSO neuron (Noh et al., 2010;Hirtz et al., 2012), the probability that all of these eight neurons simultaneously display a conduction failure calculates to 1% (55% 8 ). This low value does not explain the 64% failures in IPSCs (cf. Figure 13C 2 ). Only at 200 Hz, AP conduction failures appear to contribute to IPSC failures, because the probability of an AP conduction failure in all eight converging MNTB neurons (94% 8 = 61%) comes close to the observed IPSC failure rate of 71%. Finally, also the course of recovery cannot be explained by spike failures. In sum, although glycinergic MNTB-LSO connections undergo STD and synaptic attenuation, they can sustain high frequency transmission at hearing onset over long periods (minutes) and thousands of stimuli if stimulus frequency does not exceed 50 Hz. They also recover robustly (τ = 6.3 s). At P11, connections appear to function quite reliably, yet sustained transmission at 100 Hz is reached only by P19. If the physiological temperature is reduced by 11 • C, the performance worsens drastically, in that 10 Hz is now the upper frequency for reliable transmission. Depletion of transmitter release is the most likely cause of the depression behavior. It is important to assess the impact of glycine recycling in replenishing the transmitter pool. One aspect is neuronal and glial re-uptake from the synaptic cleft, which is mediated by glycine transporters 1 and 2, respectively. The role of these transporters may be addressed in pharmacological (Jeong et al., 2010;Jiménez et al., 2011) and genetic knock-out experiments (Gomeza et al., 2003a,b), employing the same battery of stimuli as used in the present study.

ACKNOWLEDGMENTS
This work was supported by the Deutsche Forschungsgemeinschaft (DFG Grant FR 1784/17-1 to Eckhard Friauf) and the DFG Research Training Group "Molecular, physiological and pharmacological analysis of cellular membrane transport" (GRK 845, Florian Kramer and Eckhard Friauf). Further support was provided by the Center for Mathematical and Computational Modeling, Kaiserslautern. We thank Jennifer Winkelhoff for excellent technical support and Dr. Marco Rust, Dr. Jonathan Stephan, and Alex Fischer for valuable comments on an earlier version of the manuscript.