Development of synaptic fidelity and action potential robustness at an inhibitory sound localization circuit: effects of otoferlin‐related deafness

Sound localization involves information analysis in the lateral superior olive (LSO), a conspicuous nucleus in the mammalian auditory brainstem. LSO neurons weigh interaural level differences (ILDs) through precise integration of glutamatergic excitation from the cochlear nucleus (CN) and glycinergic inhibition from the medial nucleus of the trapezoid body (MNTB). Sound sources can be localized even during sustained perception, an accomplishment that requires robust neurotransmission. Virtually nothing is known about the sustained performance and the temporal precision of MNTB–LSO inputs after postnatal day (P)12 (time of hearing onset) and whether acoustic experience guides development. Here we performed whole‐cell patch‐clamp recordings to investigate neurotransmission of single MNTB‐LSO fibres upon sustained electrical stimulation (1–200 Hz/60 s) at P11 and P38 in wild‐type (WT) and deaf otoferlin (Otof) knock‐out (KO) mice. At P11, WT and KO inputs performed remarkably similarly. In WTs, the performance increased drastically between P11 and P38, e.g. manifested by an 8 to 11‐fold higher replenishment rate (RR) of synaptic vesicles and action potential robustness. Together, these changes resulted in reliable and highly precise neurotransmission at frequencies ≤100 Hz. In contrast, KO inputs performed similarly at both ages, implying impaired synaptic maturation. Computational modelling confirmed the empirical observations and established a reduced RR per release site for P38 KOs. In conclusion, acoustic experience appears to contribute massively to the development of reliable neurotransmission, thereby forming the basis for effective ILD detection. Collectively, our results provide novel insights into experience‐dependent maturation of inhibitory neurotransmission and auditory circuits at the synaptic level.

MNTB-LSO synapses are formed by axonal boutons that are larger than classical bouton terminals (Gjoni, Aguet et al., 2018;Helfert & Schwartz, 1986) yet still much smaller than endbulb or calyx of Held synapses. They are also inconspicuous compared with the highly specialized morphology of ribbon synapses at inner hair cells (Yu & Goodrich, 2014). The facts that MNTB-mediated inhibition is crucial for proper ILD analysis (Beiderbeck et al., 2018;Grothe & Pecka, 2014;Masterton et al., 1967) and that sound sources are reliably localized even against noisy backgrounds (Kerber & Seeber, 2012) argue for extraordinarily robust neurotransmission, despite the bouton-type terminals. Indeed, MNTB-LSO synapses of P11 mice perform more reliably upon sustained high-frequency stimulation in vitro than other auditory and non-auditory bouton-type synapses (Brill et al., 2019;Krächan et al., 2017). Virtually nothing is known about the sustained neurotransmission behaviour after P11, including the question of whether this parameter matures further and, if so, which mechanisms determine maturation.
Here we analysed the robustness and temporal precision of MNTB-LSO inputs in developing WTs and Otof KOs. Using Otof KOs, we evaluated the impact of disrupted spontaneous activity and acoustic experience on the maturation of neurotransmission. To do so, we electrically stimulated single MNTB fibres in minimal stimulation experiments (1-200 Hz/60 s) and recorded evoked inhibitory postsynaptic currents (eIPSCs) from LSO principal neurons in acute slices from P11 and P38 WTs and age-matched Otof KOs.
Sustained MNTB fibre stimulation at P11 resulted in similar synaptic attenuation, fidelity, replenishment rate (RR) and temporal precision in WTs and Otof KOs. Moreover, action potential (AP) failures occurred at a rate of >66% in each genotype during sustained high-frequency stimulation. In P38 WTs, single MNTB fibres performed virtually failure-free, and neurotransmission was highly precise during 60 s challenge periods at ≤100 Hz. Furthermore, high fidelity and precision were achieved by a drastic increase of the RR between P11 and P38 (8 to 11-fold). In contrast, single fibre fidelity in P38 Otof KOs remained as low as at P11. Both genotypes displayed similar maturation of AP fidelity between P11 and P38. The reduced neurotransmission fidelity of P38 Otof KOs was not due to a smaller readily releasable pool (RRP), reduced MNTB fibre excitability, altered eIPSC-to-noise ratio in failure detection, or insufficient recovery between high-frequency trains. Instead, the reduced fidelity can be explained by a virtual lack of an age-dependent RR increase. We further validated our empirical results by computational modelling, which established a reduced RR per release site in P38 Otof KOs.
Taken together, we demonstrate an age-dependent increase of neurotransmission performance in WTs that is due to a drastic increase in SV replenishment. Such an increase is largely absent in deaf Otof KOs. In contrast, AP fidelity appears to develop normally in Otof KOs. Collectively, our findings demonstrate that the maturation of reliable neurotransmission in the central auditory system (MNTB-LSO inputs) during ongoing high-frequency stimulation requires acoustically driven synaptic transmission in the cochlea which may thus ensure proper ILD coding.

Animals and ethical approval
Animal breeding and experiments were approved by the regional councils of the Land Rhineland-Palatinate according to the German Animal Protection Law (TSchG §4/3) and followed the guidelines for the welfare of laboratory animals. The authors understand and conform to the principles and regulations described in The Journal of Physiology (Grundy, 2015). Experiments were performed on Otof deaf5/deaf5 mice (Longo-Guess et al., 2007) and age-matched control (Otof +/+ ) mice of both sexes, which were bred in the animal facility of the University of Kaiserslautern. The mice had a mixed C57BL/6J, CH3 background and were genotyped as described previously (Longo-Guess et al., 2007). Otof +/+ and Otof deaf5/deaf5 mice were typically littermates. Throughout the paper, Otof deaf5/deaf5 mice are designated 'Otof KOs' or 'KOs' . Otof +/+ mice are designated 'WTs' . n corresponds to the number of recorded cells.

Electrical stimulation of MNTB fibres: orthodromic activation
A stimulation electrode (glass capillary, ∼10 μm tip diameter) was connected to a stimulus isolator (STG 4002,Reutlingen,Germany) and placed at the lateral edge of the MNTB. To stimulate MNTB fibres in a stepwise manner, the stimulus amplitude was increased in small increments of 5 μA, repeated 10 times. Monopolar stimulus pulses of 100 μs were applied. Synaptic depression was avoided by stimulating at 0.5 Hz. Single MNTB fibre recruitment (minimal stimulation) was achieved when stimulation resulted in eIPSCs at the LSO principal neuron (see Fig. 1Ca and b). Subsequent stimulus amplitudes were also tested to determine a range of suitable stimulus amplitudes for minimal stimulation (see Fig. 1D). If 5 μA steps were too big to resolve a suitable stimulus amplitude for consistent single fibre stimulation (e.g. by recruitment of an additional fibre), manual adjustments of the stimulus amplitude were made within the 5 μA increment until a reliable single fibre stimulation was achieved (tested with ∼10 sweeps). Inconsistent stimulation of MNTB fibres was obvious in 'jumps' of eIPSC amplitude and was typically noticed online. In such cases, the attempt to find a suitable stimulus amplitude for single fibre stimulation was repeated. If an inconsistent fibre recruitment was unnoticed online, jumps of eIPSC peak amplitudes were assessed offline. Recordings with an inconsistent fibre number were excluded from the analysis. The minimal stimulation paradigm at P38 was largely performed on the same cells as in the previous study .
The minimal stimulation paradigm at P11, however, was performed independently from this previous study.
Sustained stimulation of MNTB fibres started with a 60 s baseline normalization period and stimulation at 0.2 Hz. Subsequently, a 60 s challenge was followed by 60 s recovery. The stimulus frequency during challenge was increased (1,2,5,10,20,50,100,200 Hz), whereas it was constantly 1 Hz during recovery. A total of 23,760 stimuli were given throughout such a protocol. A 10 s pause was introduced between the last stimulus of a recovery period and the first stimulus of a subsequent challenge period in order to store data. The sustained stimulation protocol was used for minimal stimulation at P11 and P38 (Figs 2,3,5,6,8 to 15) and for maximal stimulation at P38 (Fig. 7). Maximal stimulation at 100 Hz for 1 s (20 repeats, 20 s pause in between) was performed in WTs and Otof KOs at P11 and P38 (Fig. 8, traces are averages of 20 repeats).

Assessment of AP fidelity of MNTB fibres: antidromic stimulation experiments
We assessed the capability of MNTB neurons to reliably generate and conduct APs along their axons during sustained high-frequency challenge. For that purpose, we obtained somatic current-clamp recordings from MNTB neurons and electrically stimulated their axons at the medial edge of the LSO to elicit antidromic APs. Experimental conditions (electrophysiology set-up, stimulus isolator, stimulation pipettes) were the same as for orthodromic stimulation, except for the sample frequency (50 kHz). The stimulus amplitude was set to a twofold threshold. Each 60 s challenge period (1, 10, 50, 100, 200 Hz) was followed by a 1 Hz 60 s recovery period.

Data analysis
Peak amplitudes of eIPSCs were analysed with custom-written routines (Dr Alexander Fischer, University of Kaiserslautern) for IGOR Pro (Wavemetrics, Lake Oswego, OR, USA) running Patcher's Power Tools (Max Planck-Institute for Membrane Biophysics). The mean eIPSC peak amplitude during the 0.2 Hz for 60 s normalization period was set to 100% and used for normalization. eIPSC peak amplitudes ≤1.5 times the quantal size (q) were considered as synaptic failures. Such failures corresponded to a fidelity of 0, and eIPSC amplitudes ≥1.5 times q to a fidelity of 1. For the fidelity analysis, we globally set q to 22 pA as determined previously (Krächan et al., 2017;Müller et al., 2019). Notice that our '1.5 × q' criterion is conservative and results in an underestimated number of released SVs.
To quantify the frequency dependence of eIPSC amplitudes (Figs 2Cc and 3Cc), we fitted the data by Hill functions. Hill functions were also used for the fidelity rate (Figs 2Dc and 3DC) and the AP failure rate (Fig. 4C).
Voltage traces from antidromic stimulation experiments were analysed for AP fidelity. MNTB neurons displaying V rest <-60 mV were included in the analysis. APs were detected by a threshold-based search (peak amplitude ≥-50 mV).
The RRP-related current (I RRP ) was determined using the method of Elmqvist & Quastel (1965). The number of SVs in the RRP (N RRP ) was determined by dividing the I RRP by the respective q of each cell. We derived q from distributions of spontaneous IPSC amplitudes (Brill et al., 2019;Krächan et al., 2017;Müller et al., 2019) and Gaussian fitting incorporated into Excel spreadsheets (Kemmer & Keller, 2010). The initial release probability (P v ) was determined by dividing the number of released SVs at the first eIPSC by the N RRP . The replenishment was determined by cumulative summation of all eIPSC amplitudes during a 100 Hz for 60 s train. The slope of a linear fit during seconds 50 to 60 (s 50-60 ) was used to calculate RR in absolute current (Körber et al., 2015;Mendoza Schulz et al., 2014;Parthier et al., 2018;Wen et al., 2016). Division by q yielded the RR in SVs/stimulus (interstimulus interval at 100 Hz = 0.01 s), which was converted to SVs/s (cf. Fig. 9B).
Latencies were determined as the time from the peak of the stimulation artefact to the peak of the eIPSC using custom-written IGOR routines. Synaptic failures were excluded from the latency analysis. The standard deviation of the latency (latency SD) served as a measure of synaptic precision (Brill et al., 2019;Krächan et al., 2017). In the sustained stimulation protocol, the latency SD was determined during s 50-60 for individual recordings. Latency and precision analyses were omitted at P11 for frequencies >50 Hz due to the high number of failures during s 50-60 (cf. Fig. 2D). All experiments and analyses were performed blind to the genotype.

Computational modelling
We applied a deterministic model that considers M release sites (subsequently named 'sites'), where each site is either occupied by an SV or empty. An empty site is reoccupied at a replenishment rate RR i between the i th and (i+1) th stimulus. N RRPi denotes the number of occupied sites just before the arrival of the i th AP, and P v is the probability of SV fusion upon an AP. We assume that all sites are occupied at rest (Neher, 2017), i.e. there is no empty site at the start of the challenge (M = N RRP1 ). Further, where m i is the number of SVs released in response to the i th stimulus (= quantal content) and f is the stimulus frequency (100 Hz). The term N RRPi+1 = N RRPi × (1 − P v ) represents N RRP depletion in response to the i th stimulus, and the term f ) depicting the decrease of empty sites through replenishment.
We assume that RR builds up initially (Weingarten, 2018) such that RR 1 , RR 2 and RR 3 differ. To obtain basic parameters and starting with the fourth stimulus, we assume that RR remains constant during the first second, namely RR i = RR 4 . The model parameters M, P v , RR 1 , RR 2 , RR 3 and RR 4 are estimated by performing least-square fitting with the averaged empirical data for the first second. To predict long-term SV release, we assume that RR decreases monotonically as per a double exponential decay via if stimulus i > Delay. τ 1 and τ 2 are decay time constants for RR, g is a constant, and RR min is the minimum replenishment rate after a very large number of stimuli (going to infinity). If i < Delay, RR is assumed to be constant and equal to RR 4 . τ 1 , τ 2 , g, Delay and RR min are determined by fitting the model to the number of SVs released for the entire challenge period (all 6000 stimuli).
To model replenishment during recovery periods, we again use equations (1) and (2), but with F = 1 Hz. RR is assumed to increase monotonically during each recovery period as per the following equation: RR recmax , RR recmin and τ rec are estimated by fitting the model to the empirically obtained recovery data.
To consider AP failures in MNTB neurons and their stochastic nature, we slightly modified equation (2) by changing the stimulus frequency f to f i , i.e. to an instantaneous frequency. In the presence of one failure between two successful events and with f = 100 Hz, f i = 50 Hz and in the presence of two failures, f i = 33 Hz, etc. We applied the modified model to generate Fig. 11. Note that several processes addressed in our models, such as quantal content, SV release, or the RR, are discrete processes in reality. Therefore, our discontinuous model variables ought to be discrete integer values in sensu stricto. However, for simplicity, we replaced them with continuous parameters (see also Neher, 2017).

Statistics
Sample data are presented as means ± SD. Statistical analysis was performed with Origin Pro 8.6 (OriginLab, Northampton, MA, USA). Normally distributed samples (Kolmogorov-Smirnov) were compared in paired or unpaired two-tailed t tests. A homo-or heteroskedastic t test was performed based on the equality of variances determined by an F test. If data were not normally distributed, a Mann-Whitney U test was applied. In case of multiple comparison, critical α values were post hoc Šidák corrected (Abdi, 2007).
Significance levels were then:

Results
Neurotransmission between auditory brainstem neurons is highly resistant to fatigue. Synaptic inputs show remarkable fidelity and high temporal precision, even when activated hundreds of times per second (Brill et al., 2019;Friauf et al., 2015;Krächan et al., 2017;Sonntag et al., 2011;Taschenberger & von Gersdorff, 2000). In the present study, we investigated whether the neurotransmission fidelity of MNTB-LSO inputs matures after hearing onset and if so, whether the maturation requires acoustic experience.

No developmental fidelity increases for sustained neurotransmission in Otof KOs
In a previous paper , we described impaired circuit refinement in Otof KOs, evidenced by more − yet weaker − MNTB fibres converging on a single LSO principal neuron (cf. Fig. 1Aa in the present study). To ensure comparability between WTs and Otof KOs, we here conducted single MNTB fibre stimulation (= minimal stimulation; Fig. 1Ab). Very low stimulus amplitudes failed to elicit an eIPSC, yet amplitude increases in small increments ultimately resulted in activation of a single MNTB fibre and reliable eIPSCs ( Fig. 1B-D). For further neurotransmission analysis, we chose a suitable stimulus amplitude from the range of stimulus amplitudes at which stable single fibre activation was achieved (Fig. 1D). Under these conditions, mean eIPSC peak amplitudes were: P11 WTs: 253 ± 32 pA; P11 Otof KOs: 156 ± 26 pA; P38 WTs: 470 ± 56 pA; P38 Otof KOs: 160 ± 36 pA (Fig. 1E and F;n = 11,12,13,11). The results are consistent with those presented earlier .
Having determined the suitable stimulus amplitude, we assessed the neurotransmission behaviour of single MNTB fibres during sustained minimal stimulation (1-200 Hz/60 s). For the two genotypes, current traces from a P11 MNTB-LSO input at 1, 10 and 100 Hz are shown in Fig. 2A and B. With increasing frequency, synaptic attenuation (for definition, see Friauf et al., 2015) increased similarly in WTs and Otof KOs (s 50-60 ; Table 1). Steady-state attenuation for WTs and Otof KOs reached ∼80% of the normalized eIPSC amplitude at 1 Hz, ∼50% at 10 Hz, and ∼10% at 100 Hz, resulting in 50% depression frequencies (f 50Amp ) of 8 and 10 Hz, respectively ( Fig. 2C; Table 1). At each frequency, steady-state attenuation did not differ between WTs and Otof KOs ( Fig. 2C; Table 1). Neurotransmission failures occurred similarly often in WTs and Otof KOs at the end of 60 s challenge periods ( Fig. 2Ac and Bc), revealing similar fidelity rates for a given frequency ( Fig. 2D; Table 2).
To assess developmental changes in the fidelity of sustained neurotransmission, we next performed minimal stimulation experiments at P38, an adult-like age so far unexplored in either genotype. Example current traces at 1, 10 and 100 Hz are shown in Fig. 3A and B. As at P11, synaptic attenuation increased in a frequency-dependent manner ( Fig. 3Ca and b; Table 1). Steady-state amplitudes for WTs and Otof KOs at s 50-60 were ∼80% at 1 Hz, ∼50% at 10 Hz and ∼20% at 100 Hz, resulting in similar f 50Amp values of 9 and 10 Hz, respectively (Fig. 3Cc). In each genotype, the majority of steady-state amplitudes was statistically indistinguishable at both P11 and P38 (12/16 frequencies, Table 1). At P38, neurotransmission in WTs was virtually failure-free up to 50 Hz (Fig. 3Ac, Da and c; Tables 1 and 2). In contrast, neurotransmission failures already occurred at 5 Hz in Otof KOs ( Fig. 3Db and c; Table 2). Accordingly, significantly more failures took place in P38 Otof KOs at ≥10 Hz, resulting in a 50% fidelity frequency (f 50TotFid ) of 150 Hz for WTs and 41 Hz for Otof KOs ( Fig. 3Dc; Table 2). Between P11 and P38, neurotransmission fidelity ≥20 Hz increased considerably in and 100 Hz (Aa, Ba) and respective close-ups at the trains' start (s 0-0.4 ; Ab, Bb) or end (s 59.6-60 ; Ac, Bc). Triangles above traces mark synaptic failures. Ca, Cb, time course of mean normalized eIPSC amplitudes in P11 WTs (Ca) and Otof KOs (Cb). Cc, frequency dependence of mean normalized amplitude during s 50-60 . Da-c, as C, but for fidelity. Time courses in Ca-Cb and Da-Db are parabolic weighted moving averages of three (1-2 Hz), five (5-20 Hz), or nine data points . Fits in Cc and Dc are sigmoidal regressions. Minimum and maximum values were set to 0 and 1, respectively. Shaded areas depict 95% confidence intervals. f 50Amp and f 50TotFid values (stippled lines) were determined from sigmoidal regressions. Notice that f 50TotFid reflects the fidelity of both action potential conduction and synaptic transmission. Numbers above plots are number of cells. Statistical comparison of normalized amplitude and fidelity between WTs and Otof KOs was done via an unpaired two-tailed t test, except for fidelity at 1-5 Hz (U test). Details in Tables 1 and 2. eIPSC: evoked inhibitory post-synaptic currents; KO: knock-out; MNTB: medial nucleus of the trapezoid body; WT: wild type. WTs (Table 2). By contrast, such a developmental increase was largely absent from Otof KOs, resulting in similar fidelity values for both ages (6/8 frequencies, Table 2). Likewise, f 50TotFid increased less than 1.7-fold with age in Otof KOs, compared with >4-fold in WTs (Figs 2Dc and 3Dc; Table 2). Taken together, neurotransmission fidelity does not differ between WT and Otof KOs at hearing onset. Fidelity increases strongly in WTs with acoustic experience as evidenced by highly reliable neurotransmission at P38, even during sustained high-frequency challenge. In contrast, the fidelity remains immature in Otof KOs, implying that acoustic experience is essential for developing robust and failure-free neurotransmission.

AP fidelity of MNTB neurons
Neurotransmission failures may have two reasons. First, axon terminals may be unable to release transmitter-filled SVs, which results in synaptic failures. Second, presynaptic neurons may be unable to elicit APs and/or to reliably conduct them into the axon terminals; for example, because of reduced excitability caused by Na v channel inactivation (Ulbricht, 2005). To address the second reason, namely, reduced AP fidelity, we performed antidromic stimulation experiments in which we recorded from individual MNTB somata while electrically stimulating their axon at the medial border of the LSO, akin to a previous study (Kramer et al., 2014). Challenge periods lasted 60 s and comprised several frequencies . . Although this implies considerable limitations of AP generation and/or propagation at the end of 100 Hz challenge, the value is significantly lower than the total failure rate of 93 ± 7% (Table 2 and Fig. 4D; see also Fig. 2Da and Dc). Neurotransmission comprises AP propagation and SV release (total failures = AP failures + synaptic failures). Consequently, the difference 93% -66% suggests that limitations in the terminal axon boutons contribute by 27% to the high total failure rate of 93%, i.e. by more than a quarter (Fig. 4E). We reason that such limitations comprise insufficient SV replenishment and/or SV exocytosis.
Results from P11 Otof KOs did not differ much from those of age-matched WTs (AP failure rate = 75 ± 38%; range: 0-100%; n = 11, two of them failure-free; Fig. 4Ab-Cb and D). The value was statistically indistinguishable from the total failure rate of 96 ± 5% (Fig. 4D), but the substantial difference of 21% supports the idea of some limitations in the axon terminals (Fig. 4E). Collectively, the results further support our conclusion that MNTB-LSO inputs of WTs and Otof KOs perform similarly at the time of hearing onset. The two cohorts did not differ significantly in the total failure rate nor the AP failure rate ( Fig. 4D; Table 2; P values 0.54 and 0.28). The results also demonstrate that a low AP fidelity contributes to a major extent to their relatively low performance at P11 (>3/4).
34 ± 10 (14) 12 ± 6 (14) P11 OtofKO  The situation in young adults was quite different. P38 WT neurons displayed an AP failure rate of 20 ± 5% during s 50-60 of 100 Hz challenge (range: 0-97%; n = 13, nine of them failure-free; Fig. 4Ac-Cc and D). The value was >3-fold lower than the 66% in P11 WTs ( Fig. 4D; Table 2), implying considerable development of the factors determining AP propagation during the first postnatal month (e.g. myelin sheet and nodes of Ranvier). The AP failure rate was statistically indistinguishable from the total failure rate of 24 ± 21%, indicating that synaptic failures play no major role at young adult MNTB-LSO connections. As the synaptic failure rate of P38 WTs appears to be ∼7-fold lower than at P11 (Fig. 4E, 4% vs. 27%), we conclude that AP as well as synaptic fidelity improve considerably after hearing onset in WTs, and that the release and replenishment machineries develop to a level that provides high robustness and indefatigability to neurotransmission.
In deaf P38 Otof KOs, a mean AP failure rate of 29 ± 7% occurred during s 50-60 of 100 Hz challenge (range: 0-98%; n = 16, eight of them failure-free; Fig. 4Ad-Cd and D). The value is >2.5-fold lower than the 75% at P11 (Fig. 4D), implying considerable maturation of AP fidelity despite deafness. Given the high total failure rate of 86 ± 12% in P38 Otof KOs (Fig. 4E), 2/3 of the failures (57% of 86%) appear to be due to deficiencies in the axon terminals. Moreover, the almost 15-fold higher value of 57% compared with 4% in WTs (Fig. 4E) points to the crucial importance of functional otoferlin and acoustic experience for the maturation and Otof KOs (Cb). Cc, frequency dependence of mean normalized eIPSC amplitudes during s 50-60 . Da-c, as C, but for fidelity. Time courses in Ca-Cb and Da-Db are parabolic weighted moving averages of three (1-2 Hz), five (5-20 Hz), or nine data points . Fits in Cc and Dc are sigmoidal regressions. Minimum and maximum values were set to 0 and 1, respectively. Shaded areas depict 95% confidence intervals. f 50Amp and f 50TotFid values (stippled lines) were determined from sigmoidal regressions. Numbers above plots are number of cells. Statistical comparison of normalized amplitude and fidelity between WTs and Otof KOs was done via an unpaired two-tailed t test, except for fidelity at 1-5 Hz (U test). Details in Tables 1 and 2. eIPSC: evoked inhibitory post-synaptic currents; KO: knock-out; MNTB: medial nucleus of the trapezoid body; WT: wild type.

Figure 4. No differences in AP fidelity between P38 WTs and Otof KOs
A, original voltage traces from four MNTB neurons of WT and Otof KO at P11 (Aa, Ab) and P38 (Ac, Ad). The axon of MNTB neurons was stimulated at the medial border of the LSO to elicit antidromically propagated APs (100 Hz for 60 s). Traces depict 100 ms close-ups during s 59.6-60 . Triangles mark AP failures. B, dot plots from the neurons in A for five stimulation frequencies (1-200 Hz), illustrating AP fidelity to the last 40 stimulus pulses. Red frames in 100 Hz plots mark the time window shown in A. Each dot represents a single AP, and numbers to the right of the plots depict the number of successful events. Notice increasing number of failures at frequencies ≥50 Hz. C, frequency dependence of AP failure rate. Failure-free AP propagation during s 50-60 (e.g. 1000 APs at 100 Hz) will result in a failure rate of 0%. Individual single cell data curves are colour-coded, and fits are sigmoidal regressions. Minimum and maximum values were set to 0 and 1, respectively. Shaded areas depict 95% confidence intervals. Stippled lines mark the failure rate at 100 Hz; exact values are provided in circles. D, statistical analysis of failure rates during s 50-60 of 100 Hz challenge (three-way ANOVA and Šidák's correction). E, synaptic failure rates calculated from the data depicted in D (Synaptic failures = Total failures -AP failures; e.g. 27% = 93% -66%). F, frequency dependence of AP fidelity and total fidelity (cf. panel Ca-d and Fig. 2Cc, Dc). of robust release and replenishment machineries after hearing onset. Figure 4D illustrates another interesting aspect: the AP failure rate declines considerably from P11 to P38 in each genotype (from 66% to 20% in WTs, from 75% to 29% in KOs), implying that acoustic experience is not essential to achieving high AP fidelity. Moreover, at each age the AP failure rate is statistically indistinguishable between cohorts, indicating no functional impairment in MNTB axons upon otoferlin loss. Most important for the present study, however, are the differences in the fidelity of SV replenishment and/or SV release. Whereas the participating machineries become normally very resilient and indefatigable after hearing onset, as evidenced by the very low synaptic failure rate in P38 WTs, the improvement is lacking in deaf Otof KOs (Fig. 4E).
We finally assessed AP fidelity by determining f 50APFid values. At P11, f 50APFid amounted to 78 Hz in WTs and 53 Hz in Otof KOs (Fig. 4Fa,b). f 50TotFid was about 2-fold lower, namely 37 and 24 Hz (Fig. 4Fa,b and G; see also Fig. 2Da,c), demonstrating a substantial contribution of AP failures to the total transmission behaviour in juveniles. At P38, f 50APFid was 195 Hz in WTs vs. 150 Hz in Otof KOs (Fig. 4Fc, d and G). Both values were ∼2.5-fold higher than at P11 (Fig. 4G), again pointing to hearing-independent improvement of AP generation and conduction properties. Finally, the small 1.3-fold difference between f 50APFid and f 50TotFid in P38 WTs, as opposed to a 3.7-fold difference in Otof KOs, again supports our conclusion that synaptic failures play no major role in young adults, but they do so upon otoferlin loss-induced deafness (Fig. 4G).

Stable stimulus artefacts and missing correlation between number of preceding AP failures and subsequent eIPSC amplitude
From the above AP fidelity analysis, we exclude differences in excitability as a serious aspect of different transmission behaviour. We further analysed stimulation stability during high-frequency challenge, as unstable stimulus amplitudes may compromise AP generation. We found no evidence for unreliable stimulation conditions, because peak-to-peak stimulus artefacts did not fluctuate over time (not shown). Moreover, there was no (negative) correlation between peak-to-peak stimulus artefacts and the AP failure rate.
A further possibility to distinguish between synaptic depression and reduced excitability is to determine whether there is a (positive) correlation between the number of preceding failures and the amplitude of the subsequent non-failure eIPSC. In the case of synaptic depression, and in contrast to an excitability scenario, such an amplitude should not be especially high. We determined the distributions of non-failure eIPSC amplitudes during the steady-state phase (s 50-60 ) of 100 Hz challenge at P38 (see Figs 3Ac, Bc and 9Ab for raw traces). WTs as well as Otof KOs displayed right-skewed (left-leaning) distributions with the maximum in the 40-80 pA bin for each genotype ( Fig. 5Aa and Ac). However, the highest eIPSC amplitudes differed considerably ( Fig. 5Ab and  We also checked for reduced excitability by analysing the difference and the ratio of amplitudes from subsequent eIPSCs (i and i+1) with and without failures in between. Again, this was done at P38 during s 50-60 of 100 Hz challenge ( Fig. 5Ba-d). In both genotypes, eIPSC i+1 did not increase systematically when the number of failures became higher. Even with 10 failures (110 ms gap), eIPSC i+1 was similar to eIPSC i . This implies that within-challenge recovery gaps of 110 ms are too short to enable effective replenishment of empty sites and recovery from depression. A notable difference between genotypes became obvious at ≥11 failures (≥120 ms gaps). Here, eIPSC i+1 in WTs was ∼60 pA or ∼2-fold higher than eIPSC i (Fig. 5Ba-d), implying effective replenishment if within-challenge recovery gaps are ≥120 ms. This time window relates to an effective stimulation frequency of ∼8Hz. In contrast to WTs, within-challenge recovery was absent in Otof KOs ( Fig. 5Bc and d).
Together, the results argue against excitability differences between WTs and Otof KOs as an important factor in explaining the difference in the ability to fire APs at high rate.
Since pooled sample data fail to reveal the characteristics of individual measurements, we next analysed the performance of individual neurons by generating cumulative plots of eIPSC amplitudes obtained at various numbers of preceding failures (Fig. 5C). At P38, when failure rates differ drastically between WTs and Otof KOs (cf. Fig. 4D and E), these plots displayed no systematic right-shift when the number of preceding failures increased and more time was available for replenishment, neither during the 100 Hz nor the 50 Hz train (Fig. 5Ca-d). Instead they consistently displayed mean eIPSC amplitudes of ∼80 pA in the WT and ∼50 pA in the KO, confirming weak replenishment in the tens of milliseconds range. In contrast, the two inputs recuperated substantially during the first 2 s of the 1 Hz recovery period (eIPSC in WT: 219 pA; in KO: 116 pA), J Physiol 600.10 demonstrating robust replenishment capacity in the second range. Recovery will be addressed in detail in the next section.

Normal recovery from synaptic attenuation in Otof KOs
The above results, including those from the antidromic stimulation experiments, imply that the low total fidelity in P38 Otof KOs is caused predominantly by synaptic failures (Fig. 4D and F; see also Fig. 3D). To assess whether KOs may display impaired recovery from synaptic attenuation, we analysed single fibre eIPSCs during a 1 Hz for 60 s recovery period after each challenge period. Example current traces for P11 WTs and Otof KOs after 100 Hz challenge are shown in Fig. 6A. Mean recovery time courses were unaltered between genotypes, reaching steady-state levels of ∼80%, independent of the preceding challenge frequency (s 110-120 ; Fig. 6Ba and b, Table 3). At P38, steady-state recovery was also indistinguishable between genotypes (Fig. 6C and D, Table 3). Steady-state levels amounted again to ∼80%, independent of challenge frequency, genotype or age (Fig. 6E, Table 3). eIPSCs did not recover to 100%, because 1 Hz stimulation already led to a slight synaptic depression (Figs 2C and 3C;Krächan et al., 2017;Kramer et al., 2014). Notably, introducing a 10 s pause between recovery and the subsequent challenge period did result in ∼100% recovery (cf. starting values of challenge periods in Figs 2C and 3C). Taken together, Otof KOs display normal recovery behaviour at both ages. Therefore, we conclude that the reduced fidelity at P38 cannot be explained by insufficient recovery capability per se. However, rapid within-challenge recovery, i.e. recovery within the short 10 ms windows during 100 Hz challenge,

Figure 6. Normal recovery from synaptic attenuation in Otof KOs
A, eIPSCs of a P11 WT (Aa) and an Otof KO neuron (Ab) during 1 Hz for 60 s recovery following 100 Hz challenge. Right side in Aa and Ab shows eIPSC overlays at higher temporal resolution. B, time course of mean normalized eIPSC amplitudes during recovery periods in P11 WTs (Ba) and Otof KOs (Bb). C, D, as A-B, but at P38. E, mean normalized amplitudes during s 110-120 of the recovery periods. Statistical comparison of normalized amplitudes between WTs and Otof KOs at both ages was done via an unpaired t test. Details in Table 3. eIPSC: evoked inhibitory post-synaptic current; KO: knock-out; WT: wild type. Normalized amplitude values (means ± SD) were determined during s 110-120 of the recovery period (cf. Fig. 6). 100% represents the baseline value. Values in brackets depict number of cells. P values were determined by a t test. eIPSC: evoked inhibitory post-synaptic current; KO: knock-out; WT: wild type.
appears to be drastically impaired upon hearing loss (Fig. 4D).
Fidelity analysis during single fibre stimulations is not biased by low eIPSC-to-noise ratios eIPSC peak amplitudes ≤1.5 times q (33 pA with q = 22 pA) were considered a failure in this study.
Our threshold was ∼2-fold higher than the peak-to-peak noise level of ∼15 pA. Low eIPSC-to-noise ratios may cause detection limits and thus compromise fidelity assessment in that small eIPSCs remain undetected. To check the robustness of our fidelity analysis, we performed maximal stimulation experiments at P38 and determined the fidelity with the same criteria used for minimal stimulation. Despite pronounced synaptic attenuation during s 50-60 , eIPSC peak amplitudes were large enough to be clearly distinguishable from noise ( Fig. 7Aa and  b). Consequently, maximal stimulation yielded higher eIPSC-to-noise ratios than minimal stimulation. Even at higher eIPSC-to-noise ratios, Otof KOs showed more synaptic failures than WTs, resulting in 40% and 75% lower fidelity at 100 and 200 Hz, respectively ( t tests). In accordance with these findings, f 50TotFid was ∼40% lower in Otof KOs (112 vs. 194 Hz; Fig. 7C). Taken together, the reduced fidelity in Otof KOs during both minimal and maximal stimulation demonstrates the robustness of our threshold-based fidelity analysis, even when eIPSC-to-noise ratios are low. Furthermore, the higher number of converging fibres in Otof KOs (6 vs. 4; cf. Fig. 1Aa,  does not rescue the fidelity.
The initial P v was normal in Otof KOs at both ages ( Fig. 8Db; P11 WT: 0.13 ± 0.05, n = 15; P11 Otof KO: 0.12 ± 0.04, n = 7; P = 0.88; P38 WT: 0.17 ± 0.06, n = 12; P38 Otof KO: 0.18 ± 0.07, n = 8; P = 0.89, unpaired t test). Moreover, neither genotype displayed an age-related change (WT: P = 0.04; Otof KO: P = 0.08, unpaired t tests, Šidák's correction). Notably, P v is the combined release probability of several fibres here because of maximal stimulation. Of course, P v values of individual fibres fluctuate around this mean value. Taken together, because of the normal initial RRP and normal initial P v in Otof KOs, both parameters do not contribute to the reduced fidelity in P38 mutants. We reason that the reduced fidelity may be due to release deficits which become manifest during sustained stimulation. A lower replenishment capacity may be one cause.

No developmental increase in replenishment capacity in Otof KOs
To determine the replenishment capacity, we calculated the cumulative number of released SVs for each genotype (minimal stimulation, 100 Hz for 60 s). eIPSC analysis at P11 confirmed that SVs are primarily released at the start of the train in both genotypes (Fig. 9Aa, left half). At the train's end, the number of SVs released per stimulus was low in either genotype, as evidenced by an excessively high number of failures ( Fig. 9Aa right half; Figs 2 and 3). The same scenario was evident in P38 Otof KOs (Fig. 9Ab). In contrast, SV release persisted in P38 WTs, and failures were therefore rare, even at the train's end (Figs 9Ab and  3Ac). Consequently, the cumulative number of released SVs amounted to ∼24,000 SVs in P38 WTs, a ∼6-fold higher value than in P11 WTs (4000 SVs), ∼8-fold higher than in P11 Otof KOs (3000 SVs), and ∼4-fold higher in P38 Otof KOs (6000 SVs; Fig. 9B). To determine the replenishment capacity during s 50-60 , we linearly fitted the last 10 s of the cumulative SV release and determined the slope as a proxy for the RR (Fig. 9Ba and b). RR s50-60 was similar between WTs and Otof KOs at P11 (Fig. 9Ba and C; WT: 44 ± 26 SVs/s, n = 7; Otof KO: 33 ± 27 SVs/s, n = 9; P = 0.45, unpaired t test). WTs displayed an ∼8-fold higher RR at P38 than at P11 ( Fig. 9C; 348 ± 160 SVs/s, n = 12; P = 8.8E-5, unpaired t test). By contrast, RR of Otof KOs did not change significantly with age ( Fig. 9C; P38 Otof KOs: 70 ± 34 SVs/s; n = 9; P = 0.03, unpaired t test). In line with this, the RR in P38 Otof KOs was 80% lower than in age-matched WTs (P = 0.0002, unpaired t test).
There is a caveat to the analysis performed in Fig. 9A-C because the obtained RR values do not consider the contribution of AP failures and detection failures. Such failures amounted to 75% in P11 Otof KOs and to 29% in P38 Otof KOs during s 50-60 of 100 Hz challenge (cf. Fig. 4D), thereby resulting in lower effective stimulation frequencies and thus potentially in erroneous, overestimated RR values. In the section 'Computational modelling reveals reduced RR per site in Otof KOs' , below, we will elaborate this concern and refute it.
Previously, we described a difference in single fibre quantal content between WTs and Otof KOs . To control for this difference, we also compared the normalized time courses of cumulatively released SVs (maximum = 100%; Fig. 9Da and b). WT and Otof KO time courses overlapped considerably at P11, confirming the similarity in incremental SV release. By contrast, P38 Otof KOs revealed a reduced slope during s 50-60 , thus manifesting the reduced incremental SV release compared with WTs (Fig. 9Db). Remarkably, the reduced slope became manifested already after ∼20 s. As RR and fidelity were high in P38 WTs, we checked for a potential correlation between the two parameters. To do so, we plotted RR against fidelity for all four cohorts (Fig. 9Ea). We found a sigmoidal relationship, revealing that 50% fidelity is reached at a RR of ∼165 SVs/s, whereas 100% fidelity is acquired at RRs >500 SVs/s (Fig. 9Eb).
Taken together, the RR of single MNTB-LSO inputs increase ∼8-fold from P11 until P38, which explains the high neurotransmission fidelity during sustained J Physiol 600.10 high-frequency stimulation at P38. Because deaf Otof KOs lack a high RR, the formation of a robust replenishment capacity appears to depend on acoustic experience.

Computational modelling reveals reduced RR per site in P38 Otof KOs
To further validate our empirical data, we applied computational modelling and determined RR throughout challenge and recovery (100 Hz for 60 s and 1 Hz for 60 s, respectively). Moreover, we assessed RR for single sites (Methods, equation 3). The model captured the experimental data very well for each cohort (Fig. 10A). In P38 WTs, RR peaked at ∼620 SVs/s (Fig. 10Ba). Corresponding values in the other three cohorts were less than half, yet similar for inter-cohort comparison (P11 WT: ∼260 SVs/s; P11 Otof KO: ∼280 SVs/s; P38 Otof KO: ∼270 SVs/s). Each cohort reached a steady-state RR after

Figure 9. No developmental increase in replenishment capacity in Otof KOs
Aa, single fibre eIPSCs of a P11 WT (black) and a P11 Otof KO neuron (red) at the start and end of a 100 Hz for 60 s train. Numbers above eIPSCs indicate the cumulative number of released SVs. Ab, same as Aa, but at P38. B, time course of cumulatively released SVs in WTs and Otof KOs at P11 (Ba) and P38 (Bb). Thin lines depict single recordings, thicker lines are mean values, and thickest lines (s 50-60 ) mark period in which RR was quantified. C, statistics for RR. Numbers in plot are number of cells. Statistical comparison between WTs and Otof KOs was done via an unpaired t test. D, normalized time course of cumulatively released SVs (maximum = 100%). E, total fidelity as a function of the RR (period analysed: s 50-60 ). Data points were colour-coded for groups (Ea) and described with a sigmoidal fit (Eb, red line). Shaded area is 95% confidence interval. eIPSC: evoked inhibitory post-synaptic currents; KO: knock-out; RR: replenishment rate; SV: synaptic vesicle; WT: wild type. J Physiol 600.10 ∼20 s that amounted to 33 SVs/s in P11 WTs, 30 SVs/s in P11 Otof KOs, 356 SVs/s in P38 WTs, and 66 SVs/s in P38 Otof KOs (s 50-60 ; Fig. 10Ba). Thus, steady-state RRs increased ∼11-fold in WTs, but only ∼2-fold in Otof KOs between P11 and P38. These model-based numbers are close to the proxies from the experimental data (8-fold and 2-fold, respectively; cf. Fig. 9C). When normalized to the maximum rate, RR during s 50-60 dropped to ∼10% in P11 WTs and P11 Otof KOs (Fig. 10Bb). RR in P38 Otof KOs declined to ∼25%, whereas P38 WTs maintained an RR of ∼60% throughout the train. As the quantal content differed up to ∼3-fold across cohorts (Fig. 10A), we normalized the RR to the initial RRP for each cohort (initial RRP size determined by the model). In P11 WTs and Otof KOs, the RRP became replenished 1-fold/s during s 50-60 (Fig. 10Ca). Corresponding values were ∼1.5-fold/s in P38 Otof KOs and ∼4-fold/s in P38 WTs. Pooling challenge and recovery trains, the cumulative replenishment of the RRP was ∼70-fold in P11 WTs, ∼61-fold in P11 Otof KOs, ∼240-fold in P38 WTs, and ∼130-fold in P38 Otof KOs (Fig. 10Cb). When related to a single site, RR amounted to ∼0.8 SVs/s in P11 WTs, ∼0.7 SVs/s in P11 Otof KOs, ∼4.5 SVs/s in P38 WTs, and ∼1.5 SVs/s in P38 Otof KOs (s 50-60 ; Fig. 10Da). These values correspond to ∼10% of the maximal RR per site in P11 WTs, ∼10% in P11 Otof KOs, ∼50% in P38 WTs, and ∼20% in P38 Otof KOs (Fig. 10Db). About 95% of the sites were empty during steady-state in P11 WTs, ∼95% in P11 Otof KOs, ∼80% in P38 WTs, and ∼90% in P38 Otof KOs (Fig. 10Ea). Although these values appear to be quite similar, the P38 WT cohort is clearly distinct. Finally, the reoccupation time for a site was ∼1.23 s in P11 WTs, ∼1.37 s in P11 Otof KOs, ∼0.23 s in P38 WTs, and 0.67 s in P38 Otof KOs (Fig. 10Eb).

AP failures affect SV release only mildly during ongoing high-frequency stimulation
The above results on RR values (Figs 9 and 10B-D) neglect the contribution of AP failures. Such failures became particularly frequent at P11 during ongoing 100 Hz challenge (Fig. 4). They result in an increased replenishment time (multiples of 10 ms), which in turn increases the N RRP and ultimately leads to a higher quantal content. To address the issue and to quantify the effect on the quantal content, we added a probabilistic component to equation (2). This component implements random AP failures, therewith creating a realistic scenario. We focused on s 50-60 of the 100 Hz trains and modelled the time course of SV release for each cohort (Fig. 11A-C). Insets  66, 75, 20, 29%; P v 14.5, 13, 19, 15.5%; RR per site 0.8, 0.7, 4.5, 1.6 SVs/s. Note different y-axes for P38 WTs. Horizontal lines depict mean quantal content. Shaded areas are s 57-57.5 periods magnified in insets. B, cumulative fraction of quantal content, based on the computational modelling (10 repeats) in panel A (compare with experimental data in Fig. 5C). Numbers to the left of colour-coded columns depict number of preceding AP failures, whereas numbers to the right depict number of events. The mean 'recovery gain' from 0 failures to the maximal number of failures is provided in the panels (e.g. 0.5 SVs in A, upon occurrence of 13 subsequent AP failures (relating to a replenishment time of 140 ms). C, computational modelling of the mean quantal content during s 50-60 . The AP failure rate was chosen arbitrarily in a broad range from 0 to 90%. The empirical AP failure rate (mean ± SD) is shown for orientation (cf. Fig. 4D). To demonstrate the effect of P v , each panel depicts two curves. Normalized values are depicted by the y-axis on the left, absolute values by the y-axis on the right. A supplementary document 'Interactive modelling of quantal content.xlsm' allows an interactive view of the scenario and adjustment of the parameters AP failure rate, P v and RR per site. AP: action potential; KO: knock-out; RR: replenishment rate; SV: synaptic vesicle; WT: wild type. show 500 ms close-ups of s 57-57.5 and illustrate exponential depression as well as replenishment for various numbers of AP failures (best seen at P38 WT). Analysis of the s 50-60 period revealed steady-state levels for the quantal content in each cohort (average for P11 WT: 0.9 SVs/success; P11 Otof KO: 1.1; P38 WT: 4.2; P38 KO: 0.9; mean in each case) and relatively little variability after the occurrence of AP failures. We next modelled the performance of each cohort by generating cumulative plots of the quantal content, like for the analysis of experimental data in Fig. 5C. These plots contain curves obtained after various numbers of preceding AP failures (Fig. 11B). At P11, the curves revealed merely a small right-shift of the mean quantal content, mounting to ≤0.5 SVs,even with 13 AP failures, which relate to 140 ms replenishment time (Fig. 11Ca,b). This small 'recovery gain' was also seen at P38 for the KOs (Fig. 11Cc). Only the P38 WTs displayed a right-shift of 2.3 SVs with 4 AP failures (= 50 ms replenishment time; Fig. 11Cd). Whereas the P38 WTs showed a systematic recovery gain (∼0.6 SVs for each 10 ms period), the other three cohorts did not.
We further modelled the dependence of SV release on the AP failure rate by changing this value from 0% to 90% (Fig. 11Ca-d). The computational results revealed relatively flat curves for the quantal content over a broad AP failure range and a >4-fold increase only for failure rates >80%. In a final step, we increased P v 2-fold and found almost no effect (Fig. 11Ca-d). Taken together, 50-60 s into ongoing high-frequency stimulation, SV release appears to be quite independent of the number of AP failures and the release probability. SV release in the steady-state phase is therefore determined merely by the RR, in accordance with previous modelling observations (Neher, 2017).
Collectively, computational modelling confirmed our empirical data and revealed a multi-fold higher RR per site in P38 WTs than in the three other cohorts. The high RR in P38 WTs appears to form the basis for robust and high-fidelity neurotransmission during sustained high-frequency stimulation. Our modelling results further suggest an increase in RR for Otof KOs between P11 and P38, potentially arguing for an activity-independent, hard-wired mechanism or a compensatory mechanism. Nevertheless, even with such a mechanism, acoustic experience is the main contributor for the increase in RR and for the normal development of synaptic performance.
Taken together, the neurotransmission latency (assessed at low-frequency stimulation) shortens during development at MNTB-LSO inputs, as expected from higher release synchrony and faster receptor kinetics. At each age, the latency was normal in Otof KOs, suggesting that maturation depends neither on peripheral spontaneous activity nor on acoustic experience.

Impaired latency stability during sustained high-frequency stimulation in P38 Otof KOs
In a next step, we assessed latency stability during sustained high-frequency challenge. A stable latency is a temporal measure for high precision, robustness and fidelity (Brill et al., 2019;Krächan et al., 2017). Example current traces of a P11 WT and a P11 Otof KO neuron are amplitude colour-coded overlays of eIPSCs from the early and late phase of challenge trains (Fig. 13A). At 1-10 Hz, the latency remained stable throughout the train in both genotypes ( Fig. 13B and C; check black circles below current traces in panels Aa and b). At 20-200 Hz, however, there was a trend of a latency increase during the challenge period, and the increase was frequency dependent (Fig. 13G; Table 4).
For P38, example current traces are shown in Fig. 13Da  and b. Again, the latency remained quite stable up to 10 Hz challenge in each genotype ( Fig. 13E and F; Table 4). At ≥20 Hz, however, it increased during challenge, and the increase was more pronounced in Otof KOs. It also increased with increasing stimulation frequency Normalized latency values (means ± SD) were determined during s 50-60 of the challenge period. 100% represents the latency at the start of each train (cf. Fig. 12). Values in brackets depict number of cells. P values were determined by a t test. n.d, not determined due to a high number of synaptic failures. KO: knock-out; WT: wild type.

Reduced temporal precision of eIPSCs in P38 Otof KOs
A large quantal content forms the basis for temporally precise neurotransmission at MNTB-LSO inputs (Krächan et al., 2017). As a measure of precision, we determined the latency jitter during s 50-60 of stimulation (1-200 Hz). The jitter (= latency SD) increased similarly with the stimulus frequency in P11 WTs and Otof KOs Figure 12. Normal developmental shortening of neurotransmission latency in Otof KOs A, example peak-scaled eIPSCs (graphical mean of 12 events) obtained at 0.2 Hz stimulation. Bars above eIPSCs indicate synaptic latency (from artefact peak to eIPSC peak). B, artefact-aligned and peak-scaled eIPSCs for direct latency comparison (stimulation artefacts blanked for clarity). C, statistics for latency. Numbers in plot are cell numbers. Statistical comparison between WTs and Otof KOs was done via an unpaired t test. eIPSC: evoked inhibitory post-synaptic currents; KO: knock-out; WT: wild type.

Figure 13. Increased latency shift during high-frequency challenge in P38 Otof KOs
A, overlay of example current traces from a P11 WT (Aa) and Otof KO (Ab) at the start and end of 0.2, 1, 10 and 100 Hz trains. Traces are colour-coded with a gradient from baseline to peak (grey to red). Colour-coded line plots below current traces belong to the traces above and depict artefacts at 0 ms and eIPSC peak amplitudes (black circles). B, time course of absolute total latency (NB: total latency = AP conduction time + synaptic latency) for P11 WTs (Ba) and P11 Otof KOs (Bb). C, as B, but for normalized latency. D-F, as A-C, but for P38. G, mean normalized latency during s 50-60 of the challenge period. Statistical comparison between WTs and Otof KOs was done via an unpaired t test. n.d., not determined due to a high number of synaptic failures. Time courses are simple moving averages over three (1-2 Hz), five (5-20 Hz) or nine data points . Broken lines are due to many failures, prohibiting a statistical analysis. Details in Table 4. eIPSC: evoked inhibitory post-synaptic current; KO: knock-out; WT: wild type.
( Fig. 14Aa and b; black circles mark peak amplitudes). P38 inputs also showed a frequency-dependent increase in latency SD, but it was less pronounced in the WTs at >10 Hz (Fig. 14B). Sample results confirmed the above observations and revealed statistically distinguishable latency SD values at P11 for only one of five frequencies tested (Table 5). In contrast, P38 Otof KOs had a significantly higher latency SD than WTs for seven of eight frequencies tested ( Fig. 14C; Table 5).

Figure 14. Lower precision of latency during sustained high-frequency challenge in P38 Otof KOs
A, example current traces from a WT (Aa) and an Otof KO neuron (Ab) at P11, depicting eIPSCs during s 50-60 of 0.2, 1, 10 and 100 Hz trains. Traces are colour-coded with a gradient from baseline (grey) to peak (red). Black circles around red dots highlight the eIPSC peak. B, as A, but at P38. C, mean temporal precision during s 50-60 of challenge. Statistical comparison was done via an unpaired t test. n.d., not determined, due to a large number of failures in most cells, which prohibited a continuous latency SD analysis. Details in Table 5. eIPSC: evoked inhibitory post-synaptic current; KO: knock-out; WT: wild type. Latency SD values to eIPSC peak (means ± SD) were determined during s 50-60 of the challenge period (cf. Fig. 13). Values in brackets depict number of cells. P values were determined by an unpaired two-tailed t test. n.d., not determined due to the high number of synaptic failures that permitted latency SD analysis. eIPSC: evoked inhibitory post-synaptic current; KO: knock-out; WT: wild type.
Taken together, a high RR enables P38 WT MNTB-LSO inputs to maintain a high quantal content during sustained high-frequency stimulation, thus achieving temporally precise synaptic transmission. Maturation of synaptic fidelity is largely absent in Otof KOs, as shown by a lower RR, more failures and temporal imprecision. Collectively, we conclude that acoustic experience is essential for proper development of resilient synaptic performance, which is likely a prerequisite for proper sound localization. In the following chapter, we will provide further evidence that the synaptic release and replenishment mechanisms appear to become impaired upon otoferlin loss, in contrast to AP robustness.

Normal maturation of AP latency in Otof KOs
The above analyses (Figs 12-14) assessed neurotransmission latency (= total latency) and thus did not differentiate between AP conduction time and synaptic latency. In order to disentangle one from the other, we addressed temporal aspects of AP conduction from the above antidromic experiments (cf. Fig. 4). The time course of AP latencies of P11 MNTB fibres during 60 s trains at frequencies of 1-200 Hz is illustrated in Fig. 15Aa and b (cells for which we obtained <100 APs during s 50-60 at ≥10 Hz were excluded from the analysis). WTs as well as Otof KOs displayed stable values throughout the 1 and 10 Hz trains, but increasing temporal instability towards higher values at 50, 100 and 200 Hz. For example, WT values at 10 Hz were 0.82 ms at the train's start and 0.88 ms during s 50-60 (Fig. 15Ca left; Table 6, increase of 7%, n = 13), whereas corresponding Otof KO values were 1.05 and 1.10 ms (Fig. 15Cb left, increase of 5%, n = 10). At 100 Hz, we observed 0.95 and 1.54 ms for WTs and 1.10 and 1.53 ms for Otof KOs ( Fig. 15Ca and b right; Table 6, increases of 62% and 39%, n = 10 and 6). Thus, there appears to be no abnormality in AP conduction behaviour upon otoferlin loss around hearing onset.
At P38, AP latencies were not only considerably shorter than at P11, but also more stable during the trains ( Fig. 15Ba and b). Values for 10 Hz were: WT 0.53 ms at the start, 0.55 ms during s 50-60 ; KO 0.54 ms at the start, 0.56 ms during s 50-60 ( Fig. 15Da and b left panels; Table 6, 4% increase in each case, n = 13 and 16). Corresponding values for 100 Hz were: WT 0.53 ms at the start, 0.73 ms during s 50-60 ; KO 0.51 ms at the start, 0.82 ms during s 50-60 ( Fig. 15Da and b, right panels; Table 6, 38% and 61% increase; n = 11 and 15). As at P11, no difference was obvious between genotypes, indicating that AP conduction properties develop normally despite deafness.
Our results on AP conduction properties allowed us to estimate the synaptic latency at MNTB-LSO inputs (synaptic latency = total latency -AP latency). We did this for 10 Hz and 100 Hz trains and compared between frequency, genotype and age (Fig. 15C-D). For each condition, the synaptic latency was ∼2-fold longer than the AP latency (range 1.5-2.7 ms). During 10 Hz trains, the synaptic latency was quite stable, regardless of genotype and age ( Fig. 15Ca and b, Da and b, left panels). In contrast, the 100 Hz challenge was accompanied by considerable instability. At this frequency, only the P38 WT MNTB-LSO inputs transmitted steadily and robustly,  Fig. 12). B, as A, but at P38. Note different y-axis for P38 WT. C, time course of total latency, AP latency and estimated synaptic latency for P11 WT (Ca) and P11 Otof KO (Cb) for 10 Hz (left) and 100 Hz (right). D, as C, but at P38. (NB: total latency = AP latency + synaptic latency). Time courses are simple moving averages over three (1 Hz), five (10 Hz) or nine data points . Broken lines are due to many failures. n in C and D depict cell numbers for total latency, AP latency. AP: action potential; KO: knock-out; WT: wild type. J Physiol 600.10 with a synaptic latency of 1.16 ms at the train's start and 1.30 ms during s 50-60 , whereas the other three cohorts displayed tremendous latency instability ( Fig. 15Ca and b, Da and b, right panels). We conclude from these results that a stable AP latency appears to develop independently of acoustic experience, in contrast to the synaptic latency.

Discussion
Previously, we described impaired synaptic refinement in the MNTB-LSO circuit of Otof KOs at P11, around the onset of hearing . In the present study, we assessed the performance of this neurocircuit during sustained high-frequency stimulation in juvenile and young adult mice. Our main findings from WTs and Otof KOs are: (1) at P11, MNTB-LSO inputs of Otof KOs perform normally; (2) between P11 and P38, increases of AP robustness, synaptic performance, RR of SVs and temporal precision of MNTB-LSO inputs occur in WTs; (3) except for AP robustness, these parameters remain largely immature in Otof KOs; (4) more MNTB fibres converging on a given LSO neuron do not rescue the low performance of P38 Otof KOs; (5) computational modelling reveals RR deficits at single release sites in P38 Otof KOs; (6) a high RR and high AP robustness act together in enabling reliable and temporally precise neurotransmission in P38 WTs, whereas P38 Otof KOs suffer particularly from a reduced RR. Collectively, we provide novel insights into the maturation of an inhibitory neurocircuit and its dependence on experience. We establish acoustic experience as the main contributor to the developmental increase of its synaptic performance (Fig. 16).

P11 synapses perform normally in Otof KOs
Before hearing onset (∼P12 in mice; Ehret, 1976Ehret, , 1983, spontaneous spike activity propagates through the immature auditory system (Tritsch et al., 2007(Tritsch et al., , 2010. Glycinergic MNTB-LSO synapses are strengthened between P4 and P11 (Bach & Kandler, 2020;Kim & Kandler, 2003). They become more resilient to synaptic depression during short trains (20 pulses), most likely due to an increased RRP (Alamilla & Gillespie, 2013). Spontaneous prehearing activity is drastically reduced in Otof KOs, leading to weaker MNTB-LSO inputs by P11 . Although the basic synaptic strength is lower, we found normal neurotransmission and an unaltered RR in Otof KOs at P11 (Figs 2, 9 and 10). These observations let us conclude that the neurotransmission performance develops independently from spontaneous prehearing activity. Consequently, its development appears to be uncoupled from the prehearing strengthening mechanisms. Furthermore, the similarity between WTs and Otof KOs at P11 provides a valuable basis to study the impact of acoustic experience on this crucial parameter.

Acoustic experience is required to increase RR, but not the AP fidelity
Our P38 cohorts comprised the range of P31-49 and, therefore, WTs have encountered ∼3-5 weeks of acoustic experience. By contrast, Otof KOs have not (Longo-Guess et al., 2007). We previously verified the absence of MNTB spiking upon acoustic stimulation . Thus, Otof KOs allow the differentiation of experience-dependent maturation from experience-independent maturation. In our previous study and the present one, we report normal acceleration of receptor kinetics, normal increase of SV release synchrony, normal latency shortening and normal maturation of AP fidelity in Otof KOs. In conclusion, these parameters are not shaped by acoustic experience, a result that we did not expect a priori.
Collectively, there are differential effects of deafness on the functional maturation of auditory brainstem circuitry.
Acoustic experience impacts structural refinement, the maturation of biophysical properties, and the development of temporal integration within SOC circuits (Leake et al., 2006;Leao et al., 2004;Leao et al., 2005;Pilati et al., 2016;Walmsley et al., 2006;Werthat et al., 2008). A major novel finding of the present study is the considerable maturation of neurotransmission performance after hearing onset. This is particularly manifested by an ∼8-fold increase in the steady-state RR between P11 and P38 and an ∼6-fold increase when RR is normalized to a single site (Figs 9C, 10Da and 16). In contrast, RR increases only 2-fold in Otof KOs and this mild increase is not statistically significant (Figs 9C  and 10Da). Nevertheless, we do not exclude a minor experience-independent or a compensatory mechanism Figure 16. Proposed scenarios for neurotransmission during sustained high-frequency stimulation A, SV replenishment at the start of 100 Hz challenge for WTs and Otof KOs at P11 and P38. SV replenishment is provided by the recycling pool (RecP), the reserve pool (ResP), and partly by the endocytic pathway. B, SV replenishment during s 50-60 of 100 Hz challenge. RecP and ResP are likely depleted. Stippled lines indicate replenishment that is insufficient to maintain reliable neurotransmission. Between P11 and P38, replenishment capacity increases ∼11-fold in WTs (from 33 to 356 SVs/s), yet only ∼2-fold in Otof KOs (from 30 to 66 SVs/s). The extraordinarily high RR of 356 SVs/s in P38 WTs forms the basis for reliable neurotransmission during sustained high-frequency activity. eIPSCs are shown below schemes, triangles mark synaptic failures. RR values in panels are derived from the model (e.g. top left: 263 SVs/s). Notice that the replenishment capacity is most likely underestimated because of our strict identification criterion for release events. The issue is further addressed in the Discussion. eIPSC: evoked inhibitory post-synaptic current; KO: knock-out; RR: replenishment rate; SV: synaptic vesicle; WT: wild type. J Physiol 600.10 in Otof KOs that leads to the slight RR increase. Even so, we conclude that the increase of RR mainly depends on acoustic experience.

No correlation between instantaneous AP failure rate and subsequent SV release
Our results of negligible recovery in the tens of milliseconds range (cf. Figs 5 and 11) are in good agreement with a classical study at calyx of Held synapses (7% after 50 ms, 53% after 1000 ms; Wu & Borst, 1999). They are also consistent with a recent SOC study which characterized inputs from the ventral CN to medial olivo-cochlear neurons (Romero & Trussell, 2021). The authors evoked a test postsynaptic current at various time intervals after a high-frequency pulse train (20 or 50 Hz) and found virtually no recovery from short-term depression at intervals <200 ms. Ultimately, they fitted the recovery behaviour by a mono-exponential function with a time constant of ∼3 s. Our group has observed very similar recovery from short-term depression at MNTB-LSO inputs, namely only very minor fast recovery in the tens of milliseconds range (Weingarten, 2018).
In several auditory brainstem nuclei and at cerebellar climbing fibres, synapses show accelerated recovery when presynaptic Ca 2+ that enters during the AP is elevated (Blitz et al., 2004;Foster et al., 2002;Wang & Kaczmarek, 1998;Yang & Xu-Friedman, 2008). In line with this, chelating presynaptic Ca 2+ with extrinsic buffers decreases fast recovery. This has also been demonstrated in a 'skipped-stimulus approach' in which one stimulus in a high-frequency train was omitted (Yang & Xu-Friedman, 2015). Recovery of the eEPSC amplitude was affected for 60 ms in a Ca 2+ -dependent manner. We reason that MNTB fibres are effectively protected from presynaptic Ca 2+ build-up, particularly during the steady-state phase of ongoing high-frequency stimulation. Their complex repertoire of cytosolic Ca 2+ buffers (Schwaller, 2010) is well described (Bazwinsky-Wutschke et al., 2016;Friauf, 1994;Lohmann & Friauf, 1996). Activity-related build-up of Ca v channel inactivation or occluded clearance of release sites may also contribute to our results (Forsythe et al., 1998;Neher, 2010).
Considerable amplitude variation for eIPSCs to subsequent stimuli is not unique to the present study (cf. Figs 3Bb, 7Ab and 9A). Rather, substantial stimulus-to-stimulus variability has been reported by several groups at single MNTB-LSO fibres, even with low-frequency stimulation when the RRP likely remains large. eIPSC amplitudes can vary by several hundreds of picoampere and they depend heavily on the Cl − driving force (Noh et al., 2010: 400 pA;Michalski et al., 2013Michalski et al., : 1000Gjoni, Zenke et al., 2018: 150 pA;Müller et al., 2019: 500 pA). It therefore appears that the variability reflects normal biological variance and not excitability problems.
Taken together, the mechanisms for the extraordinarily robust performance of auditory (and other) synapses during sustained stimulation are enigmatic. It will be compelling to investigate MNTB-LSO synapses using the 'flash-and-freeze' technique (Watanabe, Liu et al., 2013;Watanabe, Rost et al., 2013). Interestingly, this technique has recently been adopted for sustained high-frequency stimulation in acute brain slices (Borges-Merjane et al., 2020).

Comparison of RR between MNTB-LSO synapses and other synapse types
Comparing RRs across various synapse types and neurocircuits is not straightforward, because regions and experimental conditions differ substantially (Hallermann & Silver, 2013). The numbers of stimulated fibres|boutons|release sites differ considerably, making a comparison of the absolute RR difficult, if not impossible. We used computational modelling to determine the RR at single sites, which can serve as a reliable normalization parameter for cross-comparisons. For P38 WT MNTB-LSO synapses, we obtained an RR of ∼4.5 SVs/s per site (Fig. 10Da). In the following meta-analysis, we compare this value with other synapse types (Table 7). Lucas et al. (2018) challenged calyx of Held-MNTB synapses at 100 Hz for 30 s; a steady-state RR of J Physiol 600.10 ∼4.4 SVs/s can be deduced from their results (Table 7). For 100 Hz for 60 s stimulation of cerebellar mossy fibre bouton-granule cell synapses (Saviane & Silver, 2006), the same challenge conditions as in the present study, we estimate a steady-state RR of ∼4.5 SVs/s per site (Table 7). In 10 Hz for 10 s stimulation experiments at the Drosophila neuromuscular junction (NMJ), Delgado et al. (2000) calculated an RR of ∼2 SVs/s per site. At the zebrafish NMJ, 100 Hz for 20 s challenge causes mainly asynchronous SV release (Wen et al., 2016), indicating strong SV depletion and an RR too low to supply SVs for failure-free neurotransmission over tens of seconds. Recently, we employed 100 Hz for 60 s stimulation to compare the performance of MNTB-LSO inputs with two hippocampal and two auditory synapse types (CA3-CA1, entorhinal cortex-dentate gyrus (EC-DG), CN-LSO, lateral lemniscus-inferior colliculus (LL-IC); Brill et al., 2019;Krächan et al., 2017). The RR per site at CA3-CA1 and EC-DG synapses amounted to ∼0.02 and ∼0.1 SVs/s, respectively. These values are ∼200-fold and ∼50-fold lower than at MNTB-LSO synapses. CN-LSO and LL-IC synapses displayed an RR per site of 0.6 and 0.1 SVs/s (Table 7). These values are ∼8-fold and ∼50-fold lower than at MNTB-LSO synapses.
Taken together, the meta-analysis indicates an extraordinarily high RR per site at MNTB-LSO synapses, imparting more robustness and reliability to synaptic transmission than at NMJ, LL-IC, CN-LSO, EC-DG and CA3-CA1 synapses. Furthermore, the high RR at the inhibitory MNTB-LSO synapses appears to be on a par with that at excitatory calyx of Held and excitatory cerebellar mossy fibre terminals (both being ultra-high performers).

Technical considerations: detection failures
Our analysis of empirical eIPSC amplitudes as a function of preceding failures (Fig. 5) showed virtually no relationship up to 120 ms. Computing amplitudes via a probabilistic model explained the experimental findings ( Fig. 11 and Suppl. document 'Interactive modelling of quantal content.xlsm'). AP failures as well as synaptic failures contribute to transmission failures, and detection failures form a third source. Indeed, because of our strict threshold (>1.5-fold q, see Methods), we most likely overestimated the failures. Moreover, during sustained high-frequency stimulation, minimally filled SVs may be exocytosed due to insufficient presynaptic control of q, resulting in metaphorical firing of blank cartridges. Even if the molecular processes of SV replenishment (filling of empty sites) and SV exocytosis are effective, the strength of synaptic transmission will be drastically reduced. For example, if q decreases to 11 pA (half of the normal q of 22 pA), a simultaneous exocytosis of at least three SVs would be required to surpass the threshold criterion. Our analysis therefore fails to detect such small 'subthreshold' responses. Nevertheless, we want to point out that the problem of detection failures affects only the results demonstrated in Figs 3, 5 and 7. The results demonstrated in Figs 9 and 10, however, are not affected, i.e. our comparison of replenishment rates across cohorts needs no correction.

Implications for human deafness
Among the >120 genes involved in hearing impairment (http://hereditaryhearingloss.org), mutations of the human OTOF gene occur at a frequency of ∼2-7%, i.e. much higher than average (Duman et al., 2011;Iwasa et al., 2013;Iwasa et al., 2019;Rodriguez-Ballesteros et al., 2008). Might the results of our study have any implications for the relatively large cohort of deaf patients carrying OTOF mutations? Deaf patients carrying bilateral cochlear implants (CIs) have only very limited spatial hearing ability, and their horizontal-plane sound localization strongly relies on ILD cues (Aronoff et al., 2010;Mayo & Goupell, 2020). They can coarsely localize a single source in quiet, but their performance declines rapidly in the presence of other sound (Williges et al., 2018). Due to the limited dynamic range of hearing aids, ILD information is reduced (Kerber & Seeber, 2012) and it becomes difficult to create an auditory scene or to segregate target speech from spatially separated noise.
As shown in this study, MNTB-LSO inputs, which participate in ILD analysis, display impaired performance in Otof KO mice, particularly during ongoing stimulation. It is compelling to investigate whether the lower performance, a result of lacking acoustic experience, is irreversible. Supplying Otof KOs with CIs or restoring hearing via gene therapy are possible approaches (Akil & Lustig, 2019;Al-Moyed et al., 2019;Tertrais et al., 2019). Both approaches may need to consider the various (critical) periods during which synaptic function develops, both through activity-dependent and activity-independent mechanisms. Considering the characteristic specificities of these periods will likely help to optimize CI strategies and to minimize the impairments of deaf patients.

Conclusions
The present study and our previous one  shine light on activity-dependent and activity-independent maturation of an inhibitory brainstem circuit involved in sound localization (Friauf & Lohmann, 1999;Kandler et al., 2009). During the prehearing period, elimination and strengthening of MNTB-LSO synapses depend on spontaneous activity.
After hearing onset, sound-evoked activity shapes the robustness and temporal precision of neurotransmission. Nevertheless, several parameters develop normally in Otof KOs, implying genetic determination and, therefore, some independence of acoustic experience. Overall, an interplay of activity-dependent and activity-independent processes orchestrates the maturation of this inhibitory brainstem circuit. We are still at the beginning of better comprehending the complexity of these processes.