Synaptic reliability and temporal precision are achieved via high quantal content and effective replenishment: auditory brainstem versus hippocampus

Abstract

Key points

• Auditory brainstem neurons involved in sound source localization are equipped with several morphological and molecular features that enable them to compute interaural level and time differences.

• As sound source localization works continually, synaptic transmission between these neurons should be reliable and temporally precise, even during sustained periods of high‐frequency activity.

• Using patch‐clamp recordings in acute brain slices, we compared synaptic reliability and temporal precision in the seconds–minute range between auditory and two types of hippocampal synapses; the latter are less confronted with temporally precise high‐frequency transmission than the auditory ones.

• We found striking differences in synaptic properties (e.g. continually high quantal content) that allow auditory synapses to reliably release vesicles at much higher rate than their hippocampal counterparts.

• Thus, they are indefatigable and also in a position to transfer information with exquisite temporal precision and their performance appears to be supported by very efficient replenishment mechanisms.

Abstract

At early stations of the auditory pathway, information is encoded by precise signal timing and rate. Auditory synapses must maintain the relative timing of events with submillisecond precision even during sustained and high‐frequency stimulation. In non‐auditory brain regions, e.g. telencephalic ones, synapses are activated at considerably lower frequencies. Central to understanding the heterogeneity of synaptic systems is the elucidation of the physical, chemical and biological factors that determine synapse performance. In this study, we used slice recordings from three synapse types in the mouse auditory brainstem and hippocampus. Whereas the auditory brainstem nuclei experience high‐frequency activity in vivo, the hippocampal circuits are activated at much lower frequencies. We challenged the synapses with sustained high‐frequency stimulation (up to 200 Hz for 60 s) and found significant performance differences. Our results show that auditory brainstem synapses differ considerably from their hippocampal counterparts in several aspects, namely resistance to synaptic fatigue, low failure rate and exquisite temporal precision. Their high‐fidelity performance supports the functional demands and appears to be due to the large size of the readily releasable pool and a high release probability, which together result in a high quantal content. In conjunction with very efficient vesicle replenishment mechanisms, these properties provide extremely rapid and temporally precise signalling required for neuronal communication at early stations of the auditory system, even during sustained activation in the minute range.

Key points

• Auditory brainstem neurons involved in sound source localization are equipped with several morphological and molecular features that enable them to compute interaural level and time differences.

• As sound source localization works continually, synaptic transmission between these neurons should be reliable and temporally precise, even during sustained periods of high‐frequency activity.

• Using patch‐clamp recordings in acute brain slices, we compared synaptic reliability and temporal precision in the seconds–minute range between auditory and two types of hippocampal synapses; the latter are less confronted with temporally precise high‐frequency transmission than the auditory ones.

• We found striking differences in synaptic properties (e.g. continually high quantal content) that allow auditory synapses to reliably release vesicles at much higher rate than their hippocampal counterparts.

• Thus, they are indefatigable and also in a position to transfer information with exquisite temporal precision and their performance appears to be supported by very efficient replenishment mechanisms.

Abbreviations

AMPA

α‐amino‐3‐hydroxy‐5‐methyl‐4‐isoxazolepropionic acid

CA

cornu ammonis

CGP

2,6‐di‐tert‐butyl‐4‐(3‐hydroxy‐2,2‐dimethyl‐propyl)‐phenol

CNQX

6‐cyano‐7‐nitroquinoxaline‐2,3‐dione

CTZ

cyclothiazide

d‐AP5

d(−)‐2‐amino‐5‐phosphonopentanoic acid

DG

dentate gyrus

EC

entorhinal cortex

ePSC

evoked postsynaptic current

f_50Amp

frequency of half‐maximal depression

f_50Fid

frequency of half‐maximal fidelity

GABA

γ‐aminobutyric acid

ILD

interaural level differences

I_RRP

current amplitude evoked upon complete release of RRP

K_V channel

voltage‐gated potassium channel

LSO

lateral superior olive

m

quantal content

MNTB

medial nucleus of the trapezoid body

n.d.

not determined

NMDA

N‐methyl‐d‐aspartate

N_RRP

number of vesicles in RRP

P_v

release probability of one vesicle

q

quantal size

RRP

ready releasable pool

sPSC

spontaneous postsynaptic current

STD

short‐term depression

STF

short‐term facilitation

Introduction

Binaural cues to sound localization include interaural level differences (ILD) and interaural time differences. Due to their small head size and their ability to hear ultrasound frequencies, small mammals heavily rely on ILD analysis (review: Grothe & Pecka, 2014). At the centre of ILD analysis is the lateral superior olive (LSO), a medullary brainstem nucleus whose binaural neurons receive excitatory and inhibitory input from the ipsilateral and contralateral ear, respectively. Within the contralateral inhibitory ILD branch, three conspicuous morphological features are manifested in the presynaptic terminals: (1) ultrafast and temporally precise ribbon synapses of inner hair cells (Li et al. 2014; Wichmann & Moser, 2015); (2) endbulbs of Held, i.e. giant presynaptic terminals connecting auditory nerve fibres with globular bushy cells in the cochlear nucleus; (3) calyces of Held connecting these globular bushy cells with principal neurons in the contralateral medial nucleus of the trapezoid body (MNTB; Oertel, 1999). MNTB neurons in turn provide glycinergic input to LSO neurons (Glendenning et al. 1985; Spangler et al. 1985; Sommer et al. 1993). The ribbon synapses, as well as the endbulb and calyceal synapses, enable faithful neurotransmission. Ultimately, however, the performance of MNTB–LSO synapses, which are non‐calyceal boutons and appear morphologically inconspicuous, is crucial for proper function of the ILD circuit. As sound localization works continually (the position of a noisy waterfall can be determined at any instance over a period of hours), MNTB–LSO synapses should be able to function persistently without getting completely exhausted. They should produce few if any failures in information flow, even at stimulation frequencies > 100 Hz, a rate that is common to the central auditory system. Although short‐term depression (STD) does occur at MNTB–LSO synapses, i.e. the synaptic strength declines in an activity‐dependent manner, it is moderate and occurs over tens of seconds (Giugovaz‐Tropper et al. 2011; Walcher et al. 2011; Kramer et al. 2014). It is an open question whether other synaptic systems, particularly non‐auditory ones, perform equally well in this time domain. Recently, STD behaviour in the millisecond‐second range has been assessed in a meta‐analysis across 61 studies for auditory and non‐auditory synapses (Friauf et al. 2015). This meta‐analysis revealed 50% STD at a > 10‐fold higher frequency for conventional bouton‐type auditory synapses than for the non‐auditory group (74 Hz vs. 7 Hz).

In the present study, we investigated synaptic reliability as well as temporal precision during sustained high‐frequency activity extending into the minute range to mimic the continuous activation common in natural environments (Lalor et al. 2009). We compared glycinergic MNTB–LSO synapses to two hippocampal ones, namely glutamatergic CA3–CA1 synapses (Szirmai et al. 2012) within the Cornu ammonis (CA) and GABAergic synapses between the entorhinal cortex (EC) and dentate granule cells (DG; Germroth et al. 1989; Liu et al. 1998). All three types display conventional presynaptic boutons but differ considerably regarding the stimulus‐driven and spontaneous spike activity levels. Contrary to the auditory brainstem, the hippocampal formation is not confronted with ultrafast and temporally precise high‐frequency information processing. Rather, its function is to generate cognitive maps and to mediate episodic‐like memory processes, which requires computations at very different timescales (Knierim et al. 2006; Smith & Mizumori, 2006). For example, the well‐documented spatial firing patterns (place fields) of rodent hippocampal neurons comprise discharge rates of only ∼1 spike s^–1 (McHugh et al. 1996).

We performed whole‐cell voltage‐clamp recordings in acute brain slices of juvenile mice and assessed in detail synaptic plasticity in response to sustained stimulation. We also quantified the fidelity of synaptic transmission and analysed recovery from depression as well as the temporal precision. We hypothesized that auditory synapses of the ILD pathway, in which synaptic reliability is of uttermost importance, are more indefatigable and temporally more precise and display a higher fidelity than hippocampal ones during sustained stimulation. Our results show that the auditory synapses indeed differ from their hippocampal counterparts in many aspects, namely resistance to synaptic fatigue, low failure rate, rapid and efficient recovery from fatigue and superior temporal precision. The differences are parallelled by a high rate of vesicle release which, in turn, requires effective replenishment mechanisms of so far unknown nature.

Methods

Ethical approval

Animal breeding and experiments were approved by the regional council according to the German animal protection act (TSchG §4, Absatz 3) and the guidelines for the welfare of experimental animals released by the European Communities Council Directive (2010/63/EEC).

Animals

Experiments were performed on juvenile C57BL/6N mice of both sexes and a weight of 4–10 g at postnatal day 11 ± 1. As assessed by various means, onset of hearing occurs between postnatal day 9–14 in mice (Mikaelian & Ruben, 1965; Ehret, 1976, 1983; Ehret & Romand, 1992) and synaptic transmission in the MNTB–LSO pathway is reportedly quite mature by postnatal day 11 (Kim & Kandler, 2003; Sonntag et al. 2011). However, structural refinement still proceeds under the influence of auditory experience (Kandler & Gillespie, 2005) and STD behavior changes at least until the 3rd postnatal week (Kramer et al. 2014). In the hippocampus, mechanisms necessary for the expression of adult‐like synaptic plasticity become available around postnatal day 11, when animals first open their eyes and begin to explore their environment (Dekay et al. 2006). But presynaptic plasticity mechanisms persist to influence information processing until the 3rd postnatal week (Dumas, 2005).

Electrophysiology

Whole‐cell recordings were performed from LSO principal neurons, CA1 pyramidal neurons and DG granule cells in acute brain slices at nearly physiological temperature (36.5 ± 0.5°C). Hippocampal slices were cut on a vibratome (VT 1200S, Leica) in ice‐cold solution containing (in mm) 2.5 KCl, 1.25 NaH₂PO₄, 2.0 sodium pyruvate, 3.0 myo‐inositol, 26 NaHCO₃, 260 glucose (H₂O), 6 MgCl₂, 0.25 CaCl₂. 300‐μm‐thick coronal and 400‐μm‐thick horizontal slices were generated for CA1 and DG recordings, respectively. Brainstem slices containing the LSO were generated as described previously (Hirtz et al. 2011). Slices were stored at room temperature in carbogenated artificial cerebrospinal fluid (ACSF) containing (in mm) 125 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 2 sodium pyruvate, 3 myo‐inositol, 0.44 l‐(+)‐ascorbic acid, 25 NaHCO₃, 10 glucose (H₂O), 1 MgCl₂, 2 CaCl₂ (pH = 7.3). One slice at a time was transferred into a recording chamber and superfused with carbogenated ACSF. The recording chamber was mounted on an upright microscope equipped with infra‐red and differential interference contrast optics, a 4x‐objective and a 60x‐objective. Evoked excitatory or inhibitory postsynaptic currents (eEPSCs or eIPSCs) were recorded from LSO principal neurons, CA1 pyramidal neurons (Liu et al. 1998), or DG granule cells (Germroth et al. 1989) while electrically stimulating MNTB fibres, Schaffer collaterals, or the perforant path, respectively. LSO principal neurons were selected by their spindle shape and their electrophysiological properties (Sterenborg et al. 2010). CA1 principal neurons were identified by their location within the pyramidal layer and their shape and DG granule cells were selected by their location and their globular soma with a diameter of ∼10 μm. Furthermore, only granule cells with an input resistance between 250–400 MΩ were accepted (Liu et al. 1998). Patch pipettes (GB150(F)‐8P, Science Products) with a final resistance of 2.5–6 MΩ and theta‐glass stimulation pipettes with an opening diameter of 15–25 μm (theta‐glass 1.5 mm 6IN, World Precision Instruments) were generated with a horizontal puller (Brown Flaming P‐87, Sutter Instruments). The internal solution used for LSO and CA1 recordings contained (in mm) 140 potassium‐6‐gluconate, 10 Hepes, 5 EGTA, 1 MgCl₂, 2 Na₂ATP, 0.3 Na₂GTP (both from Serva Electrophoresis). For eIPSC recordings from DG neurons, the pipette solution contained (in mm) 10 Hepes, 5 EGTA, 1 MgCl₂, 130 KCl, 2 ATP‐Na₂ and 0.3 GTP‐Na₂. The patch pipette was connected to an amplifier (EPC10, HEKA). Data were recorded and visualized with PatchMaster software (HEKA). All recordings were sampled at 20 or 50 kHz and low–pass filtered at 7.6 kHz. Liquid junction potentials (15.4 mV at low chloride, –3.5 mV at high chloride) were compensated online by the amplifier. The series resistance amounted to 10–25 MΩ and was compensated by 10–30%.

Bipolar theta‐glass stimulation pipettes filled with ASCF were placed at the lateral edge of the MNTB (to stimulate MNTB fibres), in the molecular layer of the CA2 region (to stimulate Schaffer collaterals), or at the border between presubiculum and DG (to stimulate the perforant pathway). 100‐μs‐long electrical current pulses were applied through a pulse generator (STG 4002 or STG 4004, Multi‐Channel Systems). The stimulus amplitude was set to a value at which ePSC peak amplitudes were stable and maximal (2–6 mA). ePSC onset latencies were ∼1 ms for MNTB–LSO synapses and ∼2–3 ms at both hippocampal synapse types, suggestive of monosynaptic connections. In addition, blockade of glutamatergic transmission in the GABAergic EC–DG pathway ruled out a disynaptic circuit via the involvement of interneurons. Each analysis started with a 120 s stimulation epoch at 0.2 Hz to establish a baseline for the amplitudes (set to 100%). Baseline ePSC peak amplitudes amounted to ∼140 pA for CA3–CA1, ∼110 pA for EC–DG and ∼720 pA for MNTB–LSO synapses. Noise amplitudes were similar across recordings (CA3–CA1: 14 pA; EC–DG: 13 pA; MNTB–LSO: 18 pA). After having established a baseline, synapses were challenged for 60 s at various frequencies of 1–200 Hz in ascending order (in the following referred to as ‘challenge period’). Each challenge period was followed by a 60 s recovery period, during which hippocampal synapses were stimulated at 0.2 Hz, a frequency at which no short‐term plasticity occurs (Dekay et al. 2006), whereas MNTB–LSO synapses were stimulated at 1 Hz.

Pharmacological compounds

If not stated otherwise, all chemicals were obtained from AppliChem (AppliChem GmbH, Darmstadt, Germany). To pharmacologically isolate eEPSCs in CA1 neurons, γ‐aminobutyric acid (GABA)_A receptor‐mediated and glycine receptor‐mediated inhibitory transmission were simultaneously blocked with 5 μm strychnine (Jonas et al. 1998). Likewise, eIPSCs in DG neurons were pharmacologically isolated with 50 μm 6‐cyano‐7‐nitroquinoxaline‐2,3‐dione (CNQX), 20 μm d‐(–)‐2‐amino‐5‐phosphonopentanoic acid (d‐AP5) and 5 μm 2,6‐di‐tert‐butyl‐4‐3‐hydroxy‐2,2‐dimethyl‐propyl‐phenol (CGP55845) to block α‐amino‐3‐hydroxy‐5‐methyl‐4‐isoxazolepropionic acid (AMPA), N‐methyl‐d‐aspartate (NMDA) and metabotropic GABA_B receptors, respectively. Postsynaptic AMPA receptor desensitization of CA3–CA1 synapses was assessed with 50 μm cyclothiazide (CTZ).

Data analysis

The foot‐to‐peak amplitude of each PSC was assessed automatically with custom‐written software (squid), implemented in IGOR Pro 6.34A (WaveMetrics). Further analysis was done with Excel 2010 (Microsoft) and Origin 9 (OriginLab) and statistical analysis was done with the Excel plug‐in WinSTAT (R. Fitch Software). Peak PSC amplitudes had to exceed the 95% peak‐to‐peak noise level 2‐fold to be considered a response. If, for example, the 95% peak‐to‐peak noise level was 20 pA, outward‐directed peak ePSC amplitudes currents needed to surpass +40 pA or were otherwise considered failures. The instantaneous fidelity rate was determined as the mean of the neuronal sample. Consequently, the fidelity rate was 60% when six out of ten neurons responded with a PSC. To determine the mean ePSC peak latency and standard deviation (SD), average values were calculated at the single cell level. Each ePSC plus the following nine events (skipping failures) were taken into account. The obtained average and SD values were binned in 1 s intervals and population values were determined across single cells for each interval.

Statistics

Outliers (>4 times standard SD above or below mean) were excluded from further analysis. All data sets were normally distributed (Kolmogorov–Smirnov) and, therefore, Student's paired or unpaired two‐tailed t tests were performed. In case of multiple comparisons, a post hoc Šidák correction was applied (Abdi, 2007). Data fitting as well as moving averaging was performed with Origin 9. If not stated otherwise, values from individual neurons are depicted by dots and the arithmetic mean and the standard error of the mean (SEM) are illustrated with bars and whiskers, respectively. Numbers of recorded neurons are provided in the tables or figure legends. Significance is indicated as ^* P < 0.05, ^** P < 0.01 and ^*** P < 0.001.

Modelling

We adapted the synaptic plasticity model from earlier work (Dittman et al. 2000; Yang & Xu‐Friedman, 2008) and implemented the routines in IGOR Pro (WaveMetrics). Our model incorporates synaptic vesicle depletion, calcium‐dependent vesicle replenishment, calcium‐dependent facilitation and postsynaptic receptor desensitization. To determine synaptic parameters for CA3–CA1 synapses, the average ePSC course across all cells was calculated. The model was simultaneously applied to the averaged 1, 2, 20 and/or 50 Hz traces and the χ² value was minimized. As we were interested in sustained activity, the model had to describe the last 10–30 s of the amplitude course. Furthermore, it was crucial to describe the initial phase of depression or facilitation (first 5–10 pulses), when vesicles are exclusively drawn from the readily releasable pool (RRP) and, therefore, the release probability of one vesicle (P _v) dominates the temporal course. As control experiments with CTZ did not reveal significant postsynaptic receptor desensitization in the glutamatergic CA3–CA1 projection (cf. Fig. 5 B), we omitted receptor desensitization in our model.

Figure 5. Postsynaptic receptor desensitization contributes only marginally to synaptic depression.

A, effect of 50 μm CTZ on sEPSCs. Exemplary current traces in ACSF (black) and CTZ (red). Amplitude: 29.8 ± 2.3 pA in ACSF, 37.6 ± 2.2 pA in CTZ, P = 2.2 × 10^–3; decay time: 9.4 ± 0.8 ms in ACSF, 20.2 ± 2.0 ms in CTZ, P = 1.4 × 10^–4. Ba, current traces of a representative CA1 neuron in ACSF and CTZ, depicting the first 5 s (upper row) and last 5 s (lower row) of a challenge period. Bb, time course of normalized mean amplitudes during a 20 Hz challenge period (n = 8 neurons). Peak amplitudes are illustrated as the weighted moving average of five data points. Bc, expanded time course of the first and last 5 s shown in panel Bb and statistics (paired t test). Normalized amplitudes during first 5 s: 138.5 ± 15.9% in ACSF, 161.7 ± 21.8% in CTZ, P = 0.176; during last 5 s: 24.5 ± 7.1% in ACSF, 29.4 ± 6.8% in CTZ, P = 0.438. C, quantal size (q) determined during low‐frequency stimulation (black) and challenge periods at high frequency (purple). Upper traces, exemplary current traces from CA1, DG and LSO neurons (Ca–c). Each trace depicts five ePSCs plus several sPSCs (arrowheads) during high‐frequency stimulation (20, 20 and 50 Hz, respectively). Low‐frequency stimulation was done at 0.2, 0.2 and 1 Hz, respectively (not shown). Lower left graph, distribution plots of the sPSC amplitudes from the three representative neurons (bin width = 5 pA). From these distributions, q was determined through a Gaussian fit from 0 pA to the bin after the first local maximum (stippled curves). Insets show resulting Gauss curves and q values for the representative cases (n = number of analysed sPSCs). Lower right graph, population data and statistics for each synapse type (paired t test).

Results

Synaptic plasticity during sustained activity

To compare auditory MNTB–LSO synapses with hippocampal CA3–CA1 and EC–DG synapses regarding neurotransmission during sustained activity, we challenged the synaptic systems for 60 s with increasing frequencies of 1–200 Hz. Representative current traces at 1 Hz, 10 Hz and 50 Hz are illustrated in Fig. 1 Aa–c. In the early phase of the stimulus trains, CA3–CA1 synapses consistently displayed facilitation (Fig. 1 Aa), whereas the other two systems showed STD in a frequency‐dependent manner (Fig. 1 Ab and c). Further into sustained stimulation, CA3–CA1 synapses also depressed when frequencies were ≥5 Hz. In each system, ePSC peak amplitudes generally declined rapidly during the first few seconds and reached steady‐state values after ∼20 s (Fig. 1 Ba–c). We quantified the capability of the synaptic systems to transmit signals over several tens of seconds by comparing the last ten ePSC peak amplitudes obtained from each challenge period (Fig. 1 Ca–c). Even at a frequency as low as 1 Hz, sustained stimulation led to significant depression at EC–DG synapses, with amplitudes declining to ∼75% (Fig. 1 Cb; Table 1). In contrast, CA3–CA1 and MNTB–LSO synapses showed no statistically significant plasticity at the end of such a 1 Hz train (Fig. 1 Ca and c). With stimulation at 2 Hz, synaptic depression was significant in EC–DG and MNTB–LSO synapses, leading to steady‐state levels of 65% and 87%, respectively. By contrast, CA3–CA1 synapses did not differ from baseline (Fig. 1 Ca).

Figure 1. MNTB–LSO synapses depress less than hippocampal synapses during sustained stimulation.

Aa–c, exemplar current traces of the first 20 ePSCs recorded from CA1, DG and LSO neurons during presynaptic fibre stimulation with 1, 10 and 50 Hz (top to bottom in each panel). Ba–c, time course of normalized ePSC amplitudes during challenge periods of 1–200 Hz. Peak amplitudes are illustrated as the weighted moving average of three (1–2 Hz), five (5–20 Hz) or nine (50–200 Hz) data points. Ca–c, mean amplitudes ± SEM of the last ten ePSCs of each challenge period. For statistics, values were compared to the baseline. Da–c, correlation between amplitude and stimulation frequency (data from Ca–c). Sigmoidal regression reveals the frequency of half‐maximal depression (f _50Amp). Grey areas depict the 95% confidence band of the sigmoidal regressions. ^* P < 0.05, ^*** P < 0.001, Student's paired t test. Details and statistics in Table 1. n.d., not determined.

Table 1.

Synaptic depression during challenge periods and recovery from depression

Challenge periods
Frequency (Hz)		1	2	5	10	20	50	100	200
CA3–CA1	Last 10 ePSCs	116.4 ± 13.0	98.9 ± 11.1	72.7 ± 9.9	42.5 ± 8.6	17.8 ± 2.5	14.2 ± 3.2	7.5 ± 1.4	n.d.
	P value	0.241	0.927	0.023	6.5 × 10–5	1.2 × 10–7	2.7 × 10–11	1.3 × 10–17	n.d.
	n	10	10	10	10	10	10	10	n.d.
EC–DG	Last 10 ePSCs	75.4 ± 3.8	64.6 ± 4.8	44.7 ± 4.6	29.9 ± 4.6	17.6 ± 2.8	9.6 ± 1.8	n.d.	n.d.
	P value	8.8 × 10–6	2.0 × 10–6	4.0 × 10–9	1.6 × 10–10	1.8 × 10–15	2.7 × 10–16	n.d.	n.d.
	n	15	15	14	14	12	7	n.d.	n.d.
MNTB–LSO	Last 10 ePSCs	105.2 ± 6.9	86.7 ± 6.1	71.0 ± 6.5	62.3 ± 5.7	n.d.	44.4 ± 5.6	11.0 ± 2.1	7.3 ± 1.4
	P value	0.459	0.044	2.6 × 10–4	2.6 × 10–6	n.d.	5.0 × 10–8	1.7 × 10–14	9.0 × 10–10
	n	19	20	20	20	n.d.	16	13	7
Recovery periods
Frequency (Hz)		1	2	5	10	20	50	100	200
CA3–CA1	Last 5 ePSCs	96.2 ± 14.2	74.3 ± 8.9	74.8 ± 16.6	49.0 ± 9.4	35.8 ± 9.6	23.1 ± 6.8	23.0 ± 7.9	n.d.
	P value	0.796	0.018	0.001	4.1 × 10–4	8.9 × 10–5	1.3 × 10–6	1.0 × 10–5	n.d.
	n	10	10	10	10	10	10	10	n.d.
EC–DG	Last 5 ePSCs	85.3 ± 4.5	90.7 ± 5.9	79.7 ± 10.3	64.9 ± 12.1	71.7 ± 15.1	73.2 ± 17.0	n.d.	n.d.
	P value	0.058	0.137	0.071	0.012	0.088	0.165	n.d.	n.d.
	n	15	15	14	14	12	7	n.d.	n.d.
MNTB–LSO	Last 5 ePSCs	101.8 ± 6.4	95.4 ± 5.9	88.5 ± 6.8	87.6 ± 6.5	n.d.	93.1 ± 8.3	91.3 ± 9.7	82.3 ± 9.7
	P value	0.781	0.447	0.103	0.069	n.d.	0.420	0.387	0.118
	n	19	20	20	20	n.d.	16	13	7

Depicted are the normalized mean values (% ± SEM) of the last 10 pulses of each challenge period (cf. Fig. 1), the last five pulses of each recovery period (cf. Fig. 2), the corresponding P values (tested against baseline) and the number of recorded cells (n). n.d., not determined. Values in bold indicate significance.

Stimulation at ≥5 Hz caused synaptic depression in all synaptic systems, but steady‐state levels varied strongly between auditory and hippocampal synapses. At 10 Hz, MNTB–LSO synapses were capable of transmitting signals with stable amplitudes of ∼60% (Table 1), whereas both types of hippocampal synapses depressed to considerably lower steady‐state levels of ∼40% (CA3–CA1) and ∼30% (EC–DG) of the baseline. At 50 Hz, MNTB–LSO synapses still transmitted in a fairly robust manner, with amplitudes amounting to ∼45% of the baseline. By contrast, hippocampal synapses displayed significantly lower steady‐state amplitudes of ∼15% and ∼10% (cf. Table 1).

Steady‐state values of ePSC peak amplitudes as a function of the stimulation frequency are illustrated in Fig. 1 Da–c. Assuming that an infinitely low frequency does not lead to synaptic plasticity, whereas an infinitely high frequency results in 100% depression, frequency‐dependent depression can be described by a sigmoidal regression. Based on such a regression, we determined the frequency of half‐maximal depression (f _50Amp) for each synaptic system (also described as ‘critical frequency’; Kraushaar & Jonas, 2000). We obtained f _50Amp values of 9 Hz and 4 Hz for CA3–CA1 and EC–DG, respectively. As the f _50Amp value at MNTB–LSO synapses amounted to 17 Hz, the hippocampal synapses could be distinguished by their ∼2‐fold and > 4‐fold higher upper cut–off frequency.

In summary, STD is present and occurs in a frequency‐dependent manner in all three synaptic systems. Above 5 Hz, it is more evident at hippocampal synapses than at the MNTB–LSO synapses. The latter are capable of transmitting signals during sustained stimulation up to 50 Hz with stable amplitudes > 40%, whereas both hippocampal synapse types fail to do so at > 10 Hz. Together, these data corroborate the idea that the auditory brainstem synapses are less fatigable than the hippocampal ones.

Recovery from synaptic depression

We next turned to analysis of recovery behaviour from synaptic depression. To do so, both types of hippocampal synapses were stimulated at a low frequency of 0.2 Hz after each challenge period, whereas MNTB–LSO synapses were stimulated at 1 Hz. The change of ePSC peak amplitudes was analysed over a period of 60 s (Fig. 2 Aa–c). The rationale for a lower stimulation frequency in the hippocampus was provided by our finding that CA3–CA1 and EC–DG synapses showed short‐term plasticity at 1 Hz (cf. Fig. 1 Ca and b), which could have compromised the recovery process.

Figure 2. EC–DG and MNTB–LSO synapses recover efficiently from attenuation.

Aa–c, time course of ePSC recovery following challenge periods obtained from CA1, DG and LSO neurons. Mean amplitudes during the last 10 s of the challenge periods are also plotted for clarity (–10–0 s). During recovery, CA3–CA1 and EC–DG synapses were stimulated at 0.2 Hz and MNTB–LSO synapses at 1 Hz. Insets show exemplary traces of the last ePSC of the 50 Hz challenge period as well as the 1st and last ePSC of the recovery period (from left to right). Ba–c, mean amplitudes of the last five ePSCs of the recovery period. For statistics, values were compared to the baseline (^* P < 0.05, ^** P < 0.01, ^*** P < 0.001, paired t test). Details and statistics in Table 1. n.d., not determined.

CA3–CA1 synapses failed to fully recover from STD elicited at frequencies ≥5 Hz (Fig. 2 Ba; Table 1). Even after a 5 Hz challenge period, amplitudes did not reach 75% of the baseline. In contrast, EC–DG synapses showed robust recovery after each challenge period except 10 Hz (Fig. 2 Bb). MNTB–LSO synapses recovered most effectively, with amplitudes exceeding 80% after each challenge period and becoming statistically indistinguishable from baseline (Fig. 2 Bc; Table 1). To further assess the ineffective recovery behaviour of CA3–CA1 synapses, we extended the recovery period to 120 s. Still, amplitudes did not increase further (not shown).

Fidelity of sustained synaptic transmission

The decline of ePSC amplitudes during challenge periods (cf. Fig. 1) may occur for several reasons. Mean peak amplitudes of, for example, 20% may be due to the fact that all individual ePSC amplitudes have declined to this value. Likewise, half of the stimulus pulses might elicit PSCs with peak amplitudes of 40% of the baseline, whereas the other half might result in transmission failures. Other combinations are also possible. In order to address the issue of transmission failures, we analysed the fidelity of synaptic transmission during the challenge periods. For this purpose, we determined the rate of successes, i.e. when an ePSC followed a stimulus pulse. Results for 1, 10 and 50 Hz are depicted in Fig. 3 and details for all frequencies are shown in Table 2. At 1 Hz, all synapse types were virtually failure‐free (Fig. 3 Aa–c). Nevertheless, EC–DG synapses showed a slightly higher failure rate, namely altogether 28 failures in response to a total of 900 pulses (n = 15 neurons), resulting in a 96.9% fidelity. The corresponding percentages for CA3–CA1 synapses and MNTB–LSO synapses were 99.7% and 100%, respectively. At 10 Hz, MNTB–LSO synapses were still able to transmit with a very high fidelity of 99.8% (Fig. 3 Bc). In contrast, both hippocampal synapse types featured a considerable number of failures, which resulted in average fidelity values of 83.1% and 60.9% at CA3–CA1 and EC–DG synapses, respectively (Fig. 3 Ba and b). Even at 50 Hz, transmission at MNTB–LSO synapses occurred with extraordinarily high fidelity of 99.6% (Fig. 3 Cc), yet the average fidelity at this frequency became very low for both hippocampal synapse types (CA3–CA1: 23.3%; EC–DG: 5.6%; Fig. 3 Ca and b). Together, these results demonstrate that the fidelity of synaptic transmission during sustained high‐frequency stimulation is drastically higher at the auditory synapses than at the hippocampal ones.

Figure 3. MNTB–LSO synapses perform with superior fidelity.

Fidelity rate depicted during challenge periods at 1 Hz (A), 10 Hz (B) and 50 Hz (C). Time course of the fidelity rate of CA3–CA1, EC–DG and MNTB–LSO synapses for each stimulus pulse (dots) and as weighted moving average of three (A), five (B) and nine (C) data points. Current traces show the first five and last five responses of a representative neuron (upper and lower row, respectively). Failures are marked by arrowheads. Details in Table 2.

Table 2.

Fidelity rates during challenge periods

Frequency (Hz)		1	2	5	10	20	50	100	200
CA3–CA1	Fidelity last 10 s	100.0 ± 0.0	99.6 ± 0.4	98.2 ± 0.9	74.7 ± 4.7	28.1 ± 8.1	14.7 ± 9.1	1.2 ± 0.8	n.d.
	P value	—	0.339	0.076	3.0 × 10–4	1.5 × 10–5	9.4 × 10–6	2.1 × 10–15	n.d.
	n	11	11	11	11	11	10	10	n.d.
EC–DG	Fidelity last 10 s	94.6 ± 3.7	95.7 ± 1.7	85.1 ± 1.7	48.6 ± 4.6	18.4 ± 4.5	4.2 ± 1.1	n.d.	n.d.
	P value	0.178	0.002	0.007	3.5 × 10–5	1.7 × 10–9	1.7 × 10–10	n.d.	n.d.
	n	15	15	14	14	12	7	n.d.	n.d.
MNTB–LSO	Fidelity last 10 s	100.0 ± 0.0	100.0 ± 0.0	100.0 ± 0.0	99.8 ± 0.3	n.d.	99.0 ± 0.6	60.2 ± 8.4	44.7 ± 10.1
	P value	—	—	—	—	n.d.	0.144	5.0 × 10–4	0.002
	n	12	12	9	9	n.d.	15	13	7

Depicted are the mean values (% ± SEM) of the fidelity rate (cf. Fig. 4 B), the corresponding P values (tested against baseline) and the number of recorded cells (n). n.d., not determined. Values in bold indicate significance.

The fidelity rate did not stay constant throughout the challenge periods (e.g. Fig. 3 B). Rather, transmission failures became more numerous, both with time and stimulation frequency (Fig. 4 Aa–c). During the last 10 s of the 1 Hz challenge period, each synapse type performed with unimpaired fidelity (Fig. 4 Ba–c; Table 2). At 2 and 5 Hz, only the EC–DG synapses became slightly unreliable (96% and 85%). When the stimulation frequency was increased to 10 Hz, their fidelity rate declined considerably to ∼50%. The fidelity of CA3–CA1 synapses was also impaired at this frequency (∼75%), whereas it was virtually unaffected at the MNTB–LSO synapses (99.8%). Even at 50 Hz, the latter synapse type performed with nearly perfect fidelity (99%), whereas the two hippocampal synapse types became drastically unreliable (CA3–CA1: 15%; EC–DG: 4%). Such a low fidelity never occurred at the auditory synapses, where the rate declined to 60% at 100 Hz and to 45% at 200 Hz stimulation, the highest frequency applied.

Figure 4. MNTB–LSO synapses display a > 10‐fold higher frequency of half‐maximal fidelity (f _50Fid).

Aa–c, time course of the mean fidelity rate during presynaptic fibre stimulation at 1–200 Hz. Time courses are illustrated as weighted moving average of three (1–2 Hz), five (5–20 Hz), or nine (50–200 Hz) data points. Ba–c, mean fidelity rate during the last 10 s of each challenge period. For statistics, values were tested against the baseline (paired t test). Ca–c, correlation between fidelity rate and stimulation frequency (data from Ba–c). Sigmoidal regression reveals the frequency of half‐maximal fidelity (f _50Fid). Grey areas depict the 95% confidence band of the sigmoidal regressions. Details and statistics in Table 2. ^* P < 0.05, ^** P < 0.01, ^*** P < 0.001, paired t test; n.d., not determined.

In a next step, we determined f _50Fid, the frequency at which the fidelity of transmission is half‐maximal. To do so, we employed the data obtained at all stimulation frequencies (1–200 Hz) and assumed that the fidelity reaches 100% at infinitely low and 0% at infinitely high stimulation frequencies. Sigmoidal regression revealed an f _50Fid of 167 Hz at the MNTB–LSO synapses (Fig. 4 Cc), which was more than an order of magnitude higher than at the hippocampal counterparts (∼15 Hz for CA3–CA1, ∼10 Hz for EC–DG; Fig. 4 Ca and b). Thus, it appears that faithful synaptic transmission occurs at drastically higher frequencies in the auditory brainstem. Together, each synapse type showed a higher f _50Fid than f _50Amp, indicative for a mixed contribution of synaptic failures and declined ePSC amplitudes onto synaptic depression. However, failures have a drastically higher impact at hippocampal synapses, as evidenced by a smaller difference between f _50Fid and f _50Amp (CA3–CA1: 15 Hz vs. 9 Hz; EC–DG: 10 Hz vs. 4 Hz; MNTB–LSO: 166 Hz vs. 17 Hz).

Postsynatic receptor desensitization contributes only marginally to the decline of ePSC amplitudes

Synaptic depression may arise from presynaptic or postsynaptic mechanisms, such as vesicle depletion or receptor desensitization. AMPA receptors of the CA1 region may desensitize to ∼60% within a few milliseconds (Colquhoun et al. 1992). Nevertheless, at stimulation frequencies of 5–25 Hz, AMPA receptor desensitization contributes only little to STD in CA1 neurons (Arai & Lynch, 1998; Hagler & Goda, 2001). To assess desensitization during sustained activity (60 s), we analysed EPSC amplitudes at CA3–CA1 synapses in the absence and presence of CTZ, a drug known to block AMPA receptor desensitization (Yamada & Tang, 1993; Kessler et al. 2000). Consistent with previous results (Wang & Kaczmarek, 1998; Hjelmstad et al. 1999), application of 50 μm CTZ resulted in increased amplitudes and prolonged decay times of spontaneous excitatory postsynaptic currents (sEPSCs, Fig. 5 A), providing a positive control for the pharmacological effectiveness of the drug. Nevertheless, initial and late eEPSC amplitudes were unchanged during 20 Hz challenge periods (Fig. 5 Ba–c; P _First100 = 0.176, P _Last100 = 0.438). We therefore conclude that AMPA receptor desensitization does not contribute to synaptic depression during sustained activity of CA3–CA1 synapses.

To further assess whether receptor desensitization may have contributed to synaptic depression, we determined the quantal size q, i.e. the current amplitude evoked by the release of a single vesicle. We did so for all three synapse types during low‐frequency and high‐frequency stimulation (Fig. 5 C). A reduction of q during high‐frequency stimulation would indicate postsynaptic receptor desensitization. At both types of hippocampal synapses, q remained virtually unchanged during high‐frequency stimulation (Fig. 5 Ca and b; CA3–CA1: q _0.2Hz = 17.3 ± 1.2 pA, q _20Hz = 18.3 ± 1.2 pA, P = 0.126; EC–DG: q _0.2Hz = 20.5 ± 1.8 pA, q _20Hz = 22.7 ± 2.1 pA, P = 0.051), suggesting that receptor desensitization can be ruled out. A small reduction of ∼10% occurred at the MNTB–LSO synapses (Fig. 5 Cc; q _1Hz = 21.3 ± 0.9 pA, q _50Hz = 19.1 ± 1.0 pA, P = 0.001). Based on these findings, we conclude that postsynaptic receptor desensitization contributes, if at all, only marginally to synaptic depression.

Efficient counteraction against vesicle depletion

The amplitude of a postsynaptic current response (ePSC) is the mathematical product of three parameters:

$$e\mathit{PSC}=N_{\mathrm{RRP}}{P}_{\mathrm{v}}q,$$

with N _RRP representing the number of readily releasable vesicles (equivalent to the size of the readily releasable pool [RRP]), P _v the release probability of a single vesicle and q the quantal size. In order to determine I _RRP, i.e. the current amplitude evoked upon completely releasing the RRP, we evaluated the recordings from EC–DG synapses (20 Hz) and MNTB–LSO synapses (100 Hz) through the method of Elmqvist and Quastel (Fig. 6 Ba and b, Ca and b; Elmqvist & Quastel, 1965; Neher, 2015). Because of the short‐term facilitation (STF) behaviour of CA3–CA1 synapses (Fig. 6 Aa), we could not apply the method at this synapse type. Instead, we here utilized the synaptic plasticity model introduced by Dittman and colleagues (Dittman et al. 2000) to derive P _v and subsequently calculate I _RRP  = ePSC ₁ /P _v (Fig. 6 Ab). Based on spontaneous PSCs, we estimated q for all three synapse types (Fig. 6 Ac, Bc, Cc and Dd) and calculated N _RRP as the ratio I _RRP /q. We obtained a 5‐fold higher N _RRP at MNTB–LSO synapses than at EC–DG synapses (202 vs. 40 vesicles; Fig. 7 A). By contrast, P _v was in the range of ∼20% for both MNTB–LSO and EC–DG synapses and did not differ significantly between the two synapse types (16% vs. 25%; P = 0.31; Figs 6 Dc and 7 A). P _v was drastically lower at CA3–CA1 synapses (1%), whose N _RRP comprised 228 vesicles. Estimation of N _RRP by means of the Elmqvist and Quastel method, which relies on extrapolation of the depression curve, has its limitations (Neher, 2015). We therefore also employed the Schneggenburger–Meyer–Neher method (Schneggenburger et al. 1999), by which the cumulative release plot is back‐extrapolated. This estimation method revealed smaller N _RRP values (69 for MNTB‐LSO synapses, 19 for EC‐DG synapses) than the Elmqvist and Quastel method and respective P _v values of ∼40% and ∼30%. Thus, N _RRP was 3.7‐fold higher at the auditory synapse type, confirming our overall conclusion that this type displays a large size of the readily releasable pool and a high release probability.

Figure 6. MNTB–LSO synapses display a substantially larger initial quantal content (m ₁) than hippocampal synapses.

Aa, Ba and Ca, representative current traces, depicting the first 20 ePSCs from CA3–CA1 (20 Hz), EC–DG (20 Hz) and MNTB–LSO (100 Hz) synapses. Ab, Bb and Cb, because of facilitation, the initial release probability (P _v) of CA3–CA1 synapses was modelled as in Dittman et al. (2000) and the size of the initial readily releasable pool (I _RRP) was determined as the quotient of the amplitude of the first evoked postsynaptic current in a train (ePSC₁) and P _v. For EC–DG and MNTB–LSO synapses, I _RRP was determined via the method of Elmqvist & Quastel (1965), with a linear fit over the first four ePSCs. Ac, Bc and Cc, exemplar sPSCs from the neurons depicted in Aa, Ba and Ca; q was determined as in Fig. 5. Da–e, mean values of ePSC₁, RRP,P _v,q and m ₁. Statistics were applied for experimental data sets, but not for the model (unpaired t test).

Figure 7. Vesicle replenishment during sustained synaptic activity is up to 16‐fold more efficient at MNTB–LSO synapses than at hippocampal synapses.

A, functional properties of CA3–CA1, EC–DG and MNTB–LSO synapses: ePSC₁, initial I _RRP, initial P _v, q, m ₁ and number of vesicles in the initial RRP (N _RRP). ^#Values determined through modelling. For CA3–CA1 synapses, the model incorporated facilitation and calcium‐dependent replenishment. Slight discrepancies between numbers (e.g. m ₁ = 5.1 and m ₁ = ePSC ₁ /q = 95.5/20.0 = 4.8 at EC–DG synapses) are explained by different averaging procedures (single cells vs. means). B, recorded and modelled time course of depression at the three synapse types during a 35‐pulse stimulus train at 50 Hz. The recorded time course of m is depicted by open dots (weighted moving average of nine data points). The modelled time course of m is depicted by dashed lines (calcium‐dependent vesicle replenishment was neglected). Coloured numbers at the X‐axis show the pulse number at which m becomes < 1, i.e. when vesicle release would ebb away. Upward‐directed arrows illustrate the efficiency of counteracting activity against vesicle pool exhaustion. Numbers at arrowheads depict the quantal content that was measured at the respective pulse number. C, quantal content during steady‐state periods as a function of stimulation frequency (last ten ePSCs in a challenge period; m _last10). Da–c, number of vesicles per second as a function of stimulation frequency. Dashed lines indicate the maximal replenishment rate.

Next, we calculated the quantal content m ₁, i.e. the number of vesicles released in response to the first presynaptic action potential during a stimulus train, as m ₁  = ePSC ₁ /q (Fig. 6 De). Both types of hippocampal synapses displayed low values (CA3–CA1: 2.5 vesicles; EC–DG: 5 vesicles; Figs 6 De and 7 A). They are brought about by a very low P _v at CA3–CA1 synapses on the one hand and a very low N _RRP at EC–DG synapses on the other hand (Fig. 7 A). Remarkably, m ₁ at MNTB–LSO synapses was drastically higher (25 vesicles). Thus, it seems that MNTB–LSO synapses generously expend their transmitter resources. In order to further address this issue, we performed a thought experiment and modelled vesicle depletion employing N _RRP and P _v, thereby assuming that no vesicle replenishment occurs between stimulus pulses during a challenge period. Under these conditions, transmission at MNTB–LSO synapses would decline below 1 vesicle at the 21st pulse (Fig. 7 B). The value of m of CA3–CA1 and EC–DG synapses declines below 1 vesicle after 30 and 8 pulses, respectively. In reality, the three synapse types released an average of 12.9, 1.7 and 2.3 vesicles at the above‐mentioned pulse numbers (Fig. 7 B). The higher numbers are most likely due to vesicle replenishment mechanisms and our results imply that replenishment is by far most effective at MNTB–LSO synapses, thus efficiently preventing depletion of the RRP, despite a very high m.

The analysis documented in Fig. 7 B was restricted to the first 35 pulses during challenge periods lasting 60 s. To assess vesicle exocytosis at the end of such extended periods, when ePSCs were in steady state (cf. Fig. 1 Ba–c), we determined the average number of vesicles released during the last ten pulses (m _last10) and plotted the values in a frequency‐dependent fashion. Again, MNTB–LSO synapses displayed drastically higher values than their hippocampal counterparts (Fig. 7 C), pointing to considerably higher exocytosis and concomitant endocytosis activities. From the data depicted in Fig. 7 C, we estimated the number of vesicles that are released per second at the end of the challenge periods as a function of frequency. Such an estimate illustrates the ‘power’ (work/time) that a synaptic system can execute (Abbott et al. 1997; Dekay et al. 2006). We found a maximal exocytosis rate of ∼560 vesicles s^–1 at MNTB–LSO synapses, more than 15‐fold higher than at the hippocampal synapses for which we obtained similar values (Fig. 7 Da–c).

Taken together, N _RRP and P _v at MNTB–LSO synapses are high, resulting in a large quantal content m (high number of exocytosed vesicles). Despite this fact, these synapses can maintain neurotransmission during sustained high‐frequency stimulation and at a high power. These features appear to be closely coupled to efficient vesicle replenishment mechanisms counteracting vesicle depletion.

Benefits of a large quantal content at MNTB–LSO synapses

The data above imply that sustained neurotransmission at MNTB–LSO synapses, in parallel with a large quantal content, requires very efficient replenishment mechanisms that are associated with high metabolic energy costs. What might be the benefit of such a scenario with high energy costs?

As outlined in the Introduction, MNTB–LSO synapses participate in sound localization for which temporal precision is a crucial prerequisite. Synaptic transmission is precise when vesicle fusion is constrained to a short time window and the time of a single vesicle fusion event is well described by a positively skewed distribution (Minneci et al. 2012). In a theoretical approach, we accounted for this relation with a gamma distribution (Fig. 8 A). Parameters were chosen to yield an expectation of 1 ms, with a standard error of 0.3 ms. If a single vesicle undergoes stimulus‐triggered exocytosis, the latency is very likely to deviate from expectation. The randomness of the process results in relatively large jitter from trial to trial. However, with increasing quantal content, the average latency of a multivesicular release event converges to expectation (Fig. 8 B). As a result, the trial‐to‐trial differences will be smaller (when more vesicles are released during a single fusion event, because $\sigma /\sqrt{m}$ will become smaller (Fig. 8 C). We reason that the benefit of a large quantal content, although it is associated with high metabolic energy costs, is high temporal precision in synaptic transmission.

Figure 8. A large quantal content results in temporal precision.

A, the latency of a single spike–triggered fusion event was assumed to follow a positively skewed gamma distribution (Minneci et al. 2012). Parameters were chosen to yield an expectation E(X) of 1 ms and a standard deviation sigma(X) of 0.3 ms. Main graph shows the cumulative probability and inset shows the distribution of the latency. B, simulation of the average latency upon release of m vesicles (1–100), based on the distribution shown in A. Ten independent simulations were superimposed, of which two are highlighted with open or filled black dots; remainders are shown by open grey dots. C, latency jitter assessed as standard deviation (SD_Latency) of the simulations shown in B. Following the central limit theorem, the relationship between m and SD_Latency is described by σ/$\sqrt{m}$ (black fit line).

In order to test this assumption experimentally, we analysed the ePSC peak latency for the three synapse types and compared the temporal precision during sustained activation within the challenge periods. In particular, we determined the time dependence and frequency dependence. Exemplar colour plots illustrating the variability of ePSCs as a function of time and frequency are depicted in Fig. 9 A–C. Even at the lowest stimulation frequency of 1 Hz, the temporal precision at the MNTB–LSO synapses was strikingly more precise than at the hippocampal counterparts (Fig. 9 A). The profound performance difference became even more evident during 10 and 50 Hz trains, when the precision at the MNTB–LSO synapses remained exquisite in time but degraded considerably in the hippocampus, thereby becoming quite chaotic (Fig. 9 B and C). Moreover, whereas the time course of the ePSC latency appeared to be rather independent of the stimulation frequency at the MNTB–LSO synapses (Fig. 9 Dc), this was not the case at the hippocampal synapses (Fig. 9 Da and b). The mean latency values from the population data also demonstrated a remarkable stability in the auditory brainstem, both in terms of time dependence and frequency dependence (Fig. 10 Ac; Table 3) and a significant latency increase became only apparent upon stimulation at ≥50 Hz (Fig. 10 Bc). By contrast, CA3–CA1 synapses and EC–DG synapses showed much less stability and displayed significant latency shifts already during challenge periods at ≥5 Hz (Fig. 10 Aa and b and Ba and b). To assess the temporal variability of the ePSC peak amplitudes over a stimulus train, we quantified the jitter, i.e. the SD of the latency (SD_Latency). Again, there was a striking difference between the auditory synapses on the one side and the two hippocampal ones on the other. SD_Latency values during the challenge periods (1, 10, 50 Hz) ranged from 0.08 to 0.18 ms at MNTB–LSO synapses and were manifoldly lower than at the CA3–CA1 and EC–DG synapses (0.49–3.96 ms; 1.00–5.09 ms; Fig. 10 C). MNTB–LSO synapses showed a significant increase in SD_Latency only at 100 Hz (Fig. 10 Dc; Table 3), whereas CA3–CA1 synapses did so above 5 Hz (Fig. 10 Da). Interestingly, there was no change in SD_Latency at EC–DG synapses, but they were temporally imprecise already at 1 Hz and stayed so at higher frequencies (Fig. 10 Db). To directly compare the temporal precision across the three synapse types, we finally calculated the coefficient of variation (CV_Latency), defined as the ratio SD_Latency/mean latency. We obtained the same results as shown for the SD_Latency alone (Table 3).

Figure 9. MNTB–LSO synapses display shorter and temporally more precise ePSC peak latencies than hippocampal synapses.

A–C , time series of colour‐coded ePSCs from representative CA1, DG and LSO neurons during challenge periods at 1 Hz (A), 10 Hz (B) and 50 Hz (C). Black and grey current traces show the first five and last five ePSCs, respectively, whose peak latency range is depicted by double‐sided arrows. Each colour‐coded line represents a single current trace with current amplitudes ranging from grey to red and black dots highlighting the peak latency. Lines with failures were blanked. Arrowheads mark the stimulus events. D, time course of the peak latency from representative CA3–CA1 (Da), EC–DG (Db) and MNTB–LSO synapses (Dc) at 1, 10 and 50 Hz, illustrating the time and frequency dependence of the temporal jitter.

Figure 10. MNTB–LSO synapses maintain temporal precision at much higher stimulation frequencies than hippocampal synapses.

Aa–c, time course of peak latencies recorded from CA1, DG and LSO neurons during challenge periods at 1, 10 and 50 Hz. Ba–c, frequency dependence of latency. Values of the first five (black dots) and last five ePSCs (grey dots) were averaged. Ca–c, time course of SD_Latency, same cells as in A. Da–c, frequency dependence of SD_Latency, same responses as in B. Due to the high failure rate above 50 Hz (CA3–CA1) and 10 Hz (EC–DG), the statistical analysis was restricted to a smaller frequency range. Paired t test, details and statistics in Table 3. Note different scaling of Y‐axes in Ba and b vs. Bc and in Da and b vs. Dc.

Table 3.

Latency and temporal jitter during challenge periods

Latency during sustained stimulation
Frequency (Hz)		1	2	5	10	20	50	100	200
CA3–CA1	First 10 ePSCs	6.06 ± 0.44	6.10 ± 0.41	5.98 ± 0.45	5.90 ± 0.37	5.80 ± 0.31	6.03 ± 0.35	n.d.	n.d.
	Last 10 ePSCs	5.99 ± 0.43	6.04 ± 0.37	6.45 ± 0.44	7.42 ± 0.53	8.40 ± 0.57	8.58 ± 0.48	n.d.	n.d.
	P value	0.644	0.711	0.019	0.012	0.001	0.001	n.d.	n.d.
	n	10	10	10	10	10	10	n.d.	n.d.
EC–DG	First 10 ePSCs	8.05 ± 0.53	8.06 ± 0.56	8.44 ± 0.76	8.27 ± 0.74	n.d.	n.d.	n.d.	n.d.
	Last 10 ePSCs	7.85 ± 0.61	7.97 ± 0.55	7.28 ± 0.52	6.83 ± 0.68	n.d.	n.d.	n.d.	n.d.
	P value	0.502	0.729	0.030	0.004	n.d.	n.d.	n.d.	n.d.
	n	15	15	14	13	n.d.	n.d.	n.d.	n.d.
MNTB–LSO	First 10 ePSCs	1.65 ± 0.08	1.68 ± 0.09	1.73 ± 0.10	1.76 ± 0.11	n.d.	1.74 ± 0.14	1.78 ± 0.11	1.8 ± 0.17
	Last 10 ePSCs	1.68 ± 0.08	1.72 ± 0.10	1.78 ± 0.11	1.79 ± 0.11	n.d.	2.01 ± 0.11	2.74 ± 0.11	2.45 ± 0.18
	P value	0.182	0.239	0.205	0.233	n.d.	0.006	4.0 × 10–7	8.4 × 10–4
	n	19	19	20	20	n.d.	16	13	6
Temporal jitter (SDLatency) during sustained stimulation
Frequency (Hz)		1	2	5	10	20	50	100	200
CA3–CA1	First 10 ePSCs	0.76 ± 0.11	0.63 ± 0.07	0.72 ± 0.13	0.74 ± 0.13	0.78 ± 0.17	0.77 ± 0.13	n.d.	n.d.
	Last 10 ePSCs	0.79 ± 0.15	1.16 ± 0.43	0.71 ± 0.16	1.64 ± 0.35	2.09 ± 0.31	2.87 ± 0.54	n.d.	n.d.
	P value	0.849	0.260	0.964	0.012	0.009	0.004	n.d.	n.d.
	n	10	10	10	10	10	10	n.d.	n.d.
EC–DG	First 10 ePSCs	1.68 ± 0.16	1.71 ± 0.25	2.02 ± 0.35	2.20 ± 0.38	n.d.	n.d.	n.d.	n.d.
	Last 10 ePSCs	1.65 ± 0.31	1.22 ± 0.11	1.91 ± 0.31	2.54 ± 0.52	n.d.	n.d.	n.d.	n.d.
	P value	0.923	0.070	0.782	0.539	n.d.	n.d.	n.d.	n.d.
	n	15	15	14	13	n.d.	n.d.	n.d.	n.d.
MNTB–LSO	First 10 ePSCs	0.08 ± 0.01	0.09 ± 0.02	0.11 ± 0.02	0.12 ± 0.03	n.d.	0.12 ± 0.03	0.10 ± 0.02	0.15 ± 0.04
	Last 10 ePSCs	0.11 ± 0.03	0.11 ± 0.03	0.15 ± 0.04	0.15 ± 0.03	n.d.	0.17 ± 0.03	0.33 ± 0.04	0.36 ± 0.10
	P value	0.297	0.364	0.185	0.300	n.d.	0.143	5.6 × 10–5	0.091
	n	19	19	20	20	n.d.	16	13	6
Temporal jitter (CVLatency) during sustained stimulation
Frequency (Hz)		1	2	5	10	20	50	100	200
CA3–CA1	First 10 ePSCs	0.13 ± 0.02	0.11 ± 0.01	0.12 ± 0.02	0.12 ± 0.02	0.13 ± 0.02	0.13 ± 0.02	n.d.	n.d.
	Last 10 ePSCs	0.12 ± 0.02	0.12 ± 0.07	0.11 ± 0.02	0.22 ± 0.04	0.24 ± 0.03	0.33 ± 0.06	n.d.	n.d.
	P value	0.922	0.291	0.715	0.009	0.024	4.5 × 10−3	n.d.	n.d.
	n	10	10	10	10	10	10	n.d.	n.d.
EC–DG	First 10 ePSCs	0.21 ± 0.02	0.21 ± 0.03	0.22 ± 0.03	0.25 ± 0.03	n.d.	n.d.	n.d.	n.d.
	Last 10 ePSCs	0.20 ± 0.03	0.16 ± 0.02	0.25 ± 0.03	0.33 ± 0.04	n.d.	n.d.	n.d.	n.d.
	P value	0.714	0.132	0.491	0.172	n.d.	n.d.	n.d.	n.d.
	n	15	15	14	13	n.d.	n.d.	n.d.	n.d.
MNTB–LSO	First 10 ePSCs	0.05 ± 0.01	0.05 ± 0.01	0.06 ± 0.01	0.06 ± 0.01	n.d.	0.06 ± 0.01	0.05 ± 0.01	0.08 ± 0.01
	Last 10 ePSCs	0.06 ± 0.01	0.05 ± 0.01	0.07 ± 0.01	0.08 ± 0.01	n.d.	0.08 ± 0.01	0.12 ± 0.01	0.14 ± 0.04
	P value	0.494	0.497	0.300	0.184	n.d.	0.165	1.4 × 10−4	0.183
	n	19	19	20	20	n.d.	16	13	6

Depicted are the mean latencies (ms), SD_Latency ± SEM (ms) and the CV_Latency ± SEM of the first ten and last ten ePSCs of each challenge period (cf. Fig. 10), the corresponding P values and the number of recorded cells (n). n.d., not determined.

The observed latency increase during challenge periods at higher frequencies prompted us to perform an additional statistical analysis that comprised all latency/time data points. We considered these data points separately for each cell and each stimulation frequency and fitted a simple linear regression model to each of them:

$$L{}_j=\alpha +\beta t_j+R_j,j=1,...,N,$$

where t_j is the time of the stimulus, L_j is the peak latency and R ₁,…, R_N are independent and identically distributed (i.i.d.) mean–zero residuals. Although the number of stimuli is 60 times the stimulation frequency, the actual sample size was considerably smaller when many failures occurred, particularly at higher frequencies. When plotting the data, the distribution of the L_j was not Gaussian, but considerably heavy‐tailed. Therefore, we did not estimate the intercept α and the slope β by least squares, but by the robust iteratively reweighted least squares approach (Holland & Welsch, 1977), using the MATLAB routine robustfit with default weights and tuning parameter. The resulting slope estimates for each cell can be approximately considered as a sample of i.i.d. normal random variables with mean μ_β by the asymptotic theory of M‐estimates (cf. Huber & Ronchetti, 2009). Hence, one may apply a t test for testing the null hypothesis (μ_β = 0) against the alternative μ_β ≠ 0, where the latter corresponds to a tendency of a change in latency with time for the whole cell population. The estimates of the population means for the three synapse types at all stimulation frequencies are depicted in Table 4. MNTB–LSO and CA3–CA1 synapses displayed a significant latency increase with stimulation time for higher frequencies, starting at 50 Hz and 10 Hz, respectively. The same effect was seen for individual cells where the large majority of slope estimates are significantly positive for higher stimulation frequencies. At EC–DG synapses, the analysis confirmed a significant latency decrease with stimulation time at 5 and 10 Hz. However, at 50 Hz the situation reversed to a clearly positive slope (P = 0.115), demonstrating a tendency for a positive correlation between latency and time from 50 Hz onwards. As there were only five neurons with enough ePSCs left, the data at 50 Hz are insufficient to reject the null hypothesis. These results imply an intermediate optimal frequency range of 5–10 Hz, where EC–DG synapses work fastest.

Table 4.

Rate of latency change during challenge periods, determined by linear regression

Frequency (Hz)		1	2	5	10	20	50	100	200
CA3–CA1	Slope	1.9 ± 2.2	–2.5 ± 3.7	5.1 ± 2.6	21.5 ± 3.6	25.8 ± 6.5	33.0 ± 11.1	n.d.	n.d.
	P value	0.404	0.213	0.081	2.1 × 10–4	0.003	0.016	n.d.	n.d.
	n	10	10	10	10	10	10	n.d.	n.d.
EC–DG	Slope	–1.7 ± 5.3	–4.8 ± 3.7	–19.0 ± 6.4	–21.8 ± 8.8	8.6 ± 14.6	22.1 ± 11.0	n.d.	n.d.
	P value	0.750	0.213	0.011	0.030	0.569	0.115	n.d.	n.d.
	n	15	15	14	13	11	5	n.d.	n.d.
MNTB–LSO	Slope	0.8 ± 0.5	0.8 ± 0.6	0.3 ± 0.3	–0.3 ± 0.4	n.d.	3.1 ± 1.3	15.0 ± 2.2	9.3 ± 2.0
	P value	0.145	0.213	0.362	0.426	n.d.	0.025	1.8 × 10–5	0.006
	n	19	19	20	10	n.d.	16	13	6

Depicted is the change of the peak latency (slope ± SEM; μs s⁻¹), determined by a linear regression model, the corresponding P values (tested against 0) and the number of recorded cells (n). For EC–DG and CA3–CA1 and 100 Hz, there were too many failures at 100 Hz for a meaningful analysis. We also included only experiments with at least 10 responses. Thus, two cells from EC–DG were excluded at 50 Hz. n.d., not determined. Values in bold indicate significance.

Finally, we found that the latency increase over time was not homogeneous by fitting piecewise linear regression models to the data obtained from MNTB–LSO synapses at 50 Hz. The number of segments was chosen by the BIC model selection criterion (Akaike, 1978). For most cells, two or three segments were found, demonstrating that the rate of latency increase is not constant over time. As with our initial comparison of the first and last ten responses (cf. Fig. 10 B) we were unable to address the intermediate conditions, this analysis sheds new light onto the dynamics of temporal precision. It suggests that more than one factor is involved in the timing of vesicle release during sustained high‐frequency stimulation.

In summary, our results on indefatigability and temporal precision of ePSCs demonstrate a striking difference between auditory brainstem and hippocampal synapses. MNTB–LSO synapses are much less fatigable than their hippocampal counterparts and, at the same time, display the highest and most constant temporal precision. The latter appears to be achieved via a very high quantal content, which in turn is most likely paralleled by most efficient vesicle replenishment mechanisms in order to guarantee superior synaptic fidelity.

Discussion

The present study emphasizes on several aspects of synaptic transmission at auditory and hippocampal synapses of juvenile mice, particularly on indefatigability and temporal precision. As continuous stimulation is virtually ubiquitous in natural environments (Lalor et al. 2009), experiments employing sustained stimulation (60 s) formed the centre of our study. We revealed striking performance differences between auditory MNTB–LSO synapses and hippocampal synapses (EC–DG and CA3–CA1). First, frequency‐dependent synaptic depression, as assessed in the tens of seconds‐to‐minute range, is least profound at the MNTB–LSO synapses (Fig. 1). Second, recovery from synaptic fatigue is efficient at both the MNTB–LSO and EC–DG synapses, yet appears to be quite ineffective at CA3–CA1 synapses (Fig. 2). Third, faithful transmission without failures is a hallmark of MNTB–LSO synapses when compared to both hippocampal synapse types; it is evidenced by a ∼10‐fold higher f _50Fid value (Figs 3 and 4). Fourth, quantal content at MNTB–LSO synapses is considerably higher (Fig. 6). Fifth, vesicle replenishment mechanisms of MNTB–LSO synapses are most efficient, enabling them to continually release ∼560 vesicles s^–1 (Fig. 7). Finally, the temporal precision of synaptic responses during activity in the seconds‐to‐minute range is drastically higher in the auditory brainstem (Figs 9 and 10) and probably caused by a high quantal content, i.e. multivesicular release properties (Fig. 8).

One important aspect of our study is that it combines high‐frequency stimulation with sustained stimulation, resulting in up to 12,000 stimuli per challenge period. Such long‐lasting stimulus paradigms have been used previously (Forsythe et al. 1998; Galarreta & Hestrin, 1998; Kraushaar & Jonas, 2000; Fernandez‐Alfonso & Ryan, 2004; Kim & Ryan, 2010). Likewise, there are comparative studies across various synapse types in the millisecond‐to‐second range (Dittman et al. 2000; Blitz et al. 2004). Nonetheless, to our knowledge the present study is the first which compares synapse types with a combined high‐frequency and sustained stimulation approach. The extensive analysis of temporal precision appears to be another unique feature of our study.

Spike activity levels and temporal filtering in vivo

Both stimulus‐driven and spontaneous action potential firing drive postsynaptic neurons at rates that vary considerably between neuronal systems. Upon sound‐driven stimulation, auditory neurons can sustain high‐frequency synaptic activity during 100 ms stimulation episodes with median values of > 300 spikes s^–1 and the rate may increase to > 700 and > 600 spikes s^–1 (data of MNTB neurons from mice and rats; Kopp‐Scheinpflug et al. 2008). In response to ipsilateral acoustic stimulation lasting 10 s, LSO neurons can fire action potentials at a steady‐state rate of ∼300 Hz (Tsuchitani & Boudreau, 1966). In contrast, hippocampal neurons of freely moving and exploring rodents fire an average of only 20–35 spikes s^–1 and only for brief periods of ∼500 ms (McHugh et al. 1996; Klyachko & Stevens, 2006). Similar spike rates were reported in vivo for MNTB neurons during spontaneous discharge activity, even for tens of seconds (Hermann et al. 2007; Kopp‐Scheinpflug et al. 2008; Sonntag et al. 2009). Spontaneous discharge activity in hippocampal neurons occurs at much lower frequencies in vivo, namely ∼1.5 Hz in CA3 pyramidal neurons and ∼6 Hz in layer III EC neurons (Fernandez‐Ruiz et al. 2012; Hahn et al. 2012). In line with these differences, our in vitro results demonstrate temporal filtering at MNTB–LSO synapses at a drastically higher frequency than at hippocampal synapses (f _50Fid ∼170 Hz vs. ∼13 Hz; Fig. 4). Nevertheless, each synapse type appears to be well adapted to its functional demands as evidenced by the close match between the in vivo spike activity levels and the corresponding f _50Fid values.

Indefatigability of MNTB–LSO synapses

The indefatigability that we determined for MNTB–LSO synapses during sustained activation (Figs 1, 2, 3, 4), together with their high quantal content (Fig. 7), implies robust replenishment mechanisms of synaptic vesicles. As the replenishment occurs within tens of seconds, it is most likely obtained via efficient endocytosis, an energy‐expensive task that goes along with high metabolic costs. Interestingly, in the rodent neocortex, an increase in the mean firing frequency by only 1 spike s^–1 elevates energy consumption by as much as 6.5 μmol ATP per gram grey matter per minute (Attwell & Laughlin, 2001). Moreover, ∼30% of the ATP may be required for pumping out the Na⁺ ions that generate the spikes, whereas the remaining 70% reflects ATP used on calcium pumping or on vesicle exocytosis and endocytosis (Rangaraju et al. 2014). Based on this, it was calculated that the use of ATP per synaptic bouton per spike is 244 ATP molecules (Engl & Attwell, 2015). Consequently, high firing rates exceeding > 300 spikes s^–1, in combination with a high quantal content, imply a distinctly high energy budget in MNTB neurons. There is no metabolic data on these neurons, but the superior olivary complex and other auditory brainstem regions are known to consume the highest amount of energy among many brain regions (Sokoloff, 1977, 1981) and are abundantly equipped with proteins assigned to metabolic processes (Kaltwaßer et al. 2013; Moritz et al. 2015).

The maintenance of high‐fidelity synaptic transmission at high rates is associated with effective vesicle recycling which, in turn, prevents RRP depletion. Sustained high bandwidth transmission is correlated with fast vesicle reloading and a large size of the vesicle pool (Saviane & Silver, 2006). The rate of exocytosis compared to endocytosis and the reuse of retrieved vesicles, determine how slowly the available synaptic vesicle pool is depleted (Fernandez‐Alfonso & Ryan, 2004). As long as the stimulation frequency is below a critical value, depletion of available synaptic vesicles is almost absent, even during sustained stimulation. The observed decrease of amplitudes and fidelity (Figs 1 and 3) may be caused by imbalanced exocytosis and endocytosis of synaptic vesicles, thus resulting in gradual accumulation of vesicle membrane on the surface of the presynaptic terminals and increasingly limiting the fidelity of synaptic transmission. In line with this, faster and less saturable endocytosis makes maturating synapses capable of maintaining prolonged high‐frequency transmitter release (Renden & von Gersdorff, 2007). Interestingly, endocytosis accelerates upon sustained stimulation in a Ca²⁺‐dependent manner and [Ca²⁺]_i and calmodulin speed up the refilling of the fast releasable vesicle pool (Wu et al. 2005; Lee et al. 2010). It is conceivable that this process is distinctly different across synapse types.

Size of the readily releasable pool: unconventional vs. conventional synapses

In cochlear inner hair cells, the total RRP comprises ∼600 synaptic vesicles (Khimich et al. 2005), at the endbulb of Held, numbers amount to ∼1000 vesicles (Lin et al. 2011) and > 1500 vesicles appear to form the RRP in the calyx of Held (Neher & Sakaba, 2008; Neher, 2010). At the mouse neuromuscular junction, another unconventional type, the RRP size averages ∼1700 vesicles (Ruiz et al. 2011). Values for goldfish retinal bipolar cells range from 1000 to 6000 (Heidelberger et al. 1994; Mennerick & Matthews, 1996; von Gersdorff, 2001). In contrast, values are drastically smaller for single hippocampal synapses, which are of conventional anatomy (∼5 vesicles; Dobrunz & Stevens, 1997). Reported numbers for other single conventional synapses are also in the same low range: the vesicle numbers amount to ∼2 (cerebellar molecular interneurons; Trigo et al. 2012), ∼7–8 (cerebellar climbing fibre and parallel fibre synapses; Xu‐Friedman et al. 2001), ∼10 (superior colliculus cultures; Kirischuk & Grantyn, 2000), ∼12 (hippocampal cultures; Stevens & Tsujimoto, 1995), ∼12 (retinal amacrine cells; Borges et al. 1995) and ∼16–27 (piriform cortex; Schikorski & Stevens, 1999). Our data indicate that the initial RRP of conventional, bouton‐type MNTB–LSO synapses is composed of ∼200 vesicles (Fig. 7). As will be discussed in the following paragraph, this large RRP is likely due to convergent inputs and heavily branching axons of MNTB neurons, whereas the number of vesicles per release site is probably very small.

The axonal arbor of a single MNTB neuron gives rise to ∼150 presynaptic boutons (Hirtz et al. 2012), but it branches widely such that a single MNTB neuron most likely innervates more than one LSO neuron, possibly 3–5. Therefore, we estimate that the synaptic contact between an individual MNTB neuron and an individual LSO neuron is achieved via approximately 30–50 axonal boutons. About 8 MNTB neurons converge onto a single LSO neuron (Noh et al. 2010; Hirtz et al. 2012). As we stimulated our MNTB–LSO synapses at submaximal intensities, we estimate that only 2–3 converging MNTB neurons were activated. Based on this value, the initial RRP of ∼200 vesicles appears to be spread over roughly 100 boutons (40 boutons per MNTB neuron × 2.5 converging MNTB neurons), which relates to ∼2 readily releasable vesicles per presynaptic bouton, i.e. per functional release site. If the above calculation is performed with ∼300 boutons per MNTB axon, a finding reported by Kandler and coworkers (Clause et al. 2014), the number of readily releasable vesicles per presynaptic bouton shrinks to ∼1. Therefore, single MNTB–LSO synapses appear to be similar to other conventional synapses. Notably, even in unconventional synapses, the number of readily releasable vesicles per active zone is as low as 1–3 (calyx of Held: Schneggenburger et al. 1999, endbulb of Held: Ryugo et al. 1996, neuromuscular junctions: Ruiz et al. 2011; Wen et al. 2016). Only in goldfish bipolar cells, which feature ribbon synapses, it appears to be one magnitude higher (∼17; Heidelberger et al. 1994). Thus, it appears that MNTB–LSO synapses achieve their relatively large RRP by two mechanisms. First, each MNTB neuron displays a highly bifurcating axonal arbor in the LSO. Second, several MNTB neurons converge onto a single LSO neuron. Combined, both mechanisms result in numerous synaptic contacts and a high number of release sites, although each synapse is of conventional, bouton‐type anatomy.

Efficient vesicle replenishment during sustained activity

The fact that MNTB–LSO synapses can maintain a maximal release rate of ∼560 vesicles s^–1, which is more than 15‐fold higher than at hippocampal synapses (Fig. 7), points to very efficient replenishment mechanisms. As the replenishment takes place in the order of tens of seconds, it most certainly involves efficient endocytosis of released vesicles. Endocytosis correlates with the quantal content (Kuromi & Kidokoro, 1999). Therefore, our data showing a drastically higher quantal content at MNTB–LSO synapses further support an extraordinarily effective endocytosis machinery at this synapse type, despite its conventional synapse morphology. Zebrafish motoneuron–muscle cell synapses achieve sustained swimming behaviour through fast release sites at which vesicle replenishment occurs at a > 60‐fold higher rate than at slow release sites, which are only active below 1 Hz (Wen et al. 2016). When being stimulated for 300 ms at 100 Hz, the replenishment rate amounts to 0.37 vesicles ms^–1. Not only can MNTB–LSO synapses maintain an even higher rate of ∼0.56 vesicles ms^–1, but they can sustain this for tens of seconds, a > 100‐fold longer timescale. From the vantage point of replenishment time, a minimum of ∼1.8 ms is thus required to replenish a single vesicle in the MNTB axon terminals, whereas it takes at least ∼2.7 ms and ∼28 ms to accomplish this at zebrafish motoneuron synapses and hippocampal synapses, respectively. Taken together, the data on vesicle replenishment rate and replenishment time further support our conclusion that MNTB–LSO synapses can release transmitter with exquisitely high power and, remarkably, even during sustained activation in the minute range.

High temporal precision of synapses in the auditory system

The exquisitely high temporal precision at MNTB–LSO synapses, even during sustained stimulation (Figs 9 and 10), is possibly the most striking feature when compared to hippocampal synapses. It enables them to provide adequate contralateral inhibitory input to the ILD detector, in concert with the ribbon synapses of inner hair cells as well as the endbulbs and calyces of Held. All of these unconventional synapses are located upstream in the neuronal ILD chain. Our modelling as well as experimental data imply a close association of a large quantal content, equivalent to multivesicular release, with exquisite temporal precision (Figs. 6, 7, 8, 9, 10). Evidence for highly synchronous multivesicular release has previously been accumulated in auditory (Glowatzki & Fuchs, 2002; Grant et al. 2010; Graydon et al. 2011; Li et al. 2014) and vestibular hair cells, where it triggers the millisecond‐range vestibule‐ocular reflex (Zhao & Klein, 2004; Rennie & Streeter, 2006; Vincent et al. 2014). The multivesicular release from both types of hair cells produces large EPSPs that instantaneously surpass the firing threshold. At the MNTB–LSO synapses, it ensures large and temporally precise IPSPs of short duration that are integrated with brief EPSPs transmitted from the ipsilateral side. Our data thus add to the knowledge of temporal precision by including novel findings about inhibitory synapses.

The low temporal jitter at the MNTB–LSO synapses (Fig. 10; Table 3) is in agreement with previous reports demonstrating an activity‐dependent degradation of temporal precision of only minor extent for the calyx of Held–MNTB synapse (Fedchyshyn & Wang, 2007; Lorteije et al. 2009; Grande & Wang, 2011). Another striking example of temporal precision in the auditory system is the precisely timed first spike in auditory nerve fibres in response to sound onset. In cats, the fibres’ minimum latencies range from 1–3 ms and the jitter is as low as 0.1 ms (Heil & Irvine, 1997). In frog auditory nerve fibres, the jitter is as low as 0.3 ms (Feng, 1982). These values are ∼10‐fold lower than in monkey retinal ganglion cells (Uzzell & Chichilnisky, 2004). The basis for precise first spike latency is unclear. Several possibilities have been suggested, remarkably an increased RRP size (Wittig & Parsons, 2008), an elaborate presynaptic release machinery of specialized proteins within the active zone (i.e. SNAREs; Hibino et al. 2002), or an increase of the local Ca²⁺ level by restricting Ca²⁺ diffusion (Roberts, 1994). Auditory nerve fibres of mice lacking synaptic ribbons in cochlear hair cells display degraded onset coding (Buran et al. 2010), suggesting that the ability to hold docked vesicles at the active zone accounts for the temporal precision of a synapse.

One important aspect concerning the superior temporal precision of the auditory synapse type pertains to different axon properties in the three neural systems, including myelination differences. Schaffer collaterals of CA3 pyramidal neurons are unmyelinated and have conduction velocities of ∼0.3 m s⁻¹ (Andersen et al. 1978; Soleng et al. 2003 b). By contrast, axons of MNTB principal neurons are myelinated, large‐to‐medium diameter fibres that have a conduction velocity of ∼33 m s⁻¹ (Tsuchitani & Johnson, 1991). The drastic difference may well contribute to the substantial differences in latency and temporal jitter observed in the present study (cf. Table 3). Interestingly, a recent study in dysmyelinated brainstems found prolonged spike latencies at the calyx of Held nerve terminal which were indicative of a 3‐fold reduction in action potential conduction velocity (Kim et al. 2013). Likewise, the temporal precision was impaired and the spike failure rate increased. Finally, stimulus trains applied to CA3 pyramidal neurons result in a cumulative increase of the axonal conduction time which may influence transmitter release (Soleng et al. 2003 a). Together, the results of these studies demonstrate the importance of myelination. They also point to a caveat in the interpretation of our data, namely that the superior temporal precision of the MNTB–LSO connection may not solely be due to synaptic differences (e.g. specialized release machineries), but also to differences in the information processing of axons (Debanne, 2004). Such axonal differences may also contribute to the superior fidelity of the MNTB–LSO connection (cf. Fig. 3), because myelin greatly reduces the effective membrane capacitance of axons (Barrett & Barrett, 1982), thus leading to more secure and reliable excitability.

Morphofunctional and molecular factors for indefatigability and temporal precision

What might be the morphofunctional and/or molecular basis for the differences in indefatigability and temporal precision displayed by the three synapse types analysed in the present study? Most likely, one needs to search for specializations in the presynaptic machinery (for treatises on postsynaptic aspects and the synaptic cleft geometry that can modulate synaptic strength, see Xu‐Friedman & Regehr, 2004 and Freche et al. 2011). At the calyx of Held–MNTB synapses, a greater number of small active zones with few docked vesicles, a larger pool of releasable vesicles and a higher efficiency of release conspire to optimize sustained high‐frequency transmission (Taschenberger et al. 2002). A higher efficiency of release may be achieved by an increased Ca²⁺ sensitivity which will result in a higher P _v. When Ca²⁺ influx into the presynaptic terminal is reduced, this is accompanied by increased fluctuations in both PSC amplitude and latency (Isaacson & Walmsley, 1995).

In a recent review (Kopp‐Scheinpflug & Tempel, 2015), decreased temporal precision of neuronal signalling was described upon mutations in genes encoding for potassium channels (Kv1.1, Kv3.3) or regulatory proteins involved in synaptic release (e.g. complexins). Other candidate proteins for temporal fidelity are synaptojanin 1 (Trapani et al. 2009), Munc13/CAPS family priming proteins (Brose et al. 1995; Liu et al. 2008; Lipstein et al. 2013; Imig et al. 2014) and the Ca²⁺ sensors synaptotagmin‐1 and synaptotagmin‐2 (Xu et al. 2007).

Unfortunately, electron microscopic data on presynaptic terminals contacting LSO neurons are rare (Helfert & Schwartz, 1986; Korada & Schwartz, 1999; Schwartz & Eager, 1999) and we are not aware of any ultrastructural study that has explicitly addressed the vesicular architecture in presynaptic terminals originating from axons of glycinergic MNTB neurons. It will be important to elucidate the molecular, structural and functional mechanisms that contribute to the distinct performance differences across various synapse types. Revealing their peculiarities will be essential to our understanding of synaptic physiology and plasticity, which more and more come into the spotlight of neuropsychiatric disorders (Siegert et al. 2015; Tang et al. 2015).

Additional information

Competing interests

The authors declare no competing financial interests.

Author contributions

All authors were involved in the design of the study, approved the final version of the manuscript and agree to be accountable for all aspects of the work. All persons designated as authors qualify for authorship and all those who qualify for authorship are listed. Experiments were performed at the University of Kaiserslautern by E.G.K. and A.U.F. Data were analysed by E.G.K., A.U.F. and J.F. and the manuscript was written by E.F., E.G.K. and A.U.F.