New insights into cochlear sound encoding

Introduction

In the mammalian cochlea, inner hair cells (IHCs)—the genuine sensory cells of the cochlea transform sound-induced mechanical signals into a neural code at their ribbon synapses. Upon hair bundle deflection, mechanotransducer channels, located in the stereociliar tips, provide hair cell depolarization. This process leads to presynaptic glutamate release from IHCs onto spiral ganglion neurons (SGNs), and ultimately activates the auditory pathway. Coding of sound at IHC ribbon synapses achieves impressive performance: each glutamatergic presynaptic active zone (AZ) of an IHC provides the sole excitatory input to a postsynaptic SGN. Yet, each single AZ drives SGN spike rates at sound onset in the kilohertz range and supports firing at hundreds of hertz during ongoing stimulation. Moreover, these synapses are capable of transmitting information on the timing of the stimulus with submillisecond precision.

The underlying mechanisms that mediate this performance have remained enigmatic but likely relate to the molecular and structural specializations of the IHC ribbon-type AZ. Here, we briefly review the latest progress on the molecular anatomy and physiology of the IHC ribbon synapse, with a focus on the presynaptic AZ (for recent reviews of the postsynaptic SGN, see 1,2.) Dysfunction or loss of IHC synapses causes a specific form of sensorineural hearing impairment: auditory synaptopathy (recently reviewed in 3). We will then summarize recent experimental and theoretical work that has corroborated the Ca²⁺ nanodomain hypothesis of Ca²⁺ influx-exocytosis coupling at IHC ribbon synapses. Finally, we will provide a brief overview of the current state of the debate on the mode of exocytosis at the hair cell AZ, which remains a hot topic of current research.

Unconventional presynaptic molecular composition

The synaptic ribbon represents the most prominent structural deviation from “conventional” glutamatergic synapses of the vertebrate central nervous system (Figure 1). Depending on the cell type, developmental stage, and animal species under investigation, the ribbon can assume various shapes and sizes^4–6. The main structural component of synaptic ribbons is RIBEYE⁷, a protein assembled from an aggregation-prone A domain and an enzymatically active B domain, both of which are transcribed from the CtBP2 gene⁸. The synaptic ribbons help cluster large complements of Ca²⁺ channels and readily releasable vesicles at the IHC AZ, thereby enabling synchronous auditory signaling and also promoting continuous vesicle replenishment^9–11. Ribbons are also critical for sensory processing in the retina, where they serve similar functions^12–14, and, in addition, seem to play a role in coupling Ca²⁺ channels to release sites¹⁵.

Figure 1. Inner hair cells drive sound encoding in several spiral ganglion neurons.

Schematic drawing of an inner hair cell (gray) and its synapses with spiral ganglion neurons (black). Inset shows super-resolution (4Pi) images of an immunolabeled inner hair cell synapse with the synaptic ribbon (red) placed opposite to the center of the postsynaptic AMPA receptor cluster (green). Each spiral ganglion neuron is thought to receive input from one ribbon-type inner hair cell active zone at the postsynaptic swelling of its peripheral neurite.

In addition to unexpectedly finding that IHC ribbon synapses appear to operate without neuronal SNAREs¹⁶ and the classic neuronal Ca²⁺ sensors synaptotagmin 1 and 2^17,18, we have recently come to realize that SNARE regulators such as complexins^19,20, as well as priming factors of the Munc13 and CAPS families²¹ which are critical for transmission at many synapses, also seem to be missing from IHCs. Instead, hair cells employ the multi-C₂-domain protein otoferlin²², a member of the ferlin family of membrane fusion-related proteins (reviewed in 23,24), which is a tail-anchored protein and requires the TRC40 pathway for efficient targeting to the endoplasmic reticulum²⁵. Otoferlin clusters below the synaptic ribbon²¹ and seems to assume multiple roles in hair cell exocytosis. For example, otoferlin has been suggested (i) to act as the putative Ca²⁺ sensor in IHCs^22,26, (ii) to facilitate vesicular priming and replenishment^21,27, and (iii) to participate in exocytosis-endocytosis coupling through direct interaction with the adaptor protein 2 (AP-2) complex^28,29. It is tempting to speculate that IHCs evolved this unconventional molecular machinery in order to achieve the utmost performance. One possible requirement could be a rapid and low-affinity engagement of synaptic vesicles with release sites with molecular links to a nearby Ca²⁺ channel, followed by rapid clearance of vesicular lipid and proteins from that site for it to be quickly reloaded. Clearly, more work is required to elucidate the molecular fusion machinery of IHCs.

Besides the presence of otoferlin, IHC AZs are characterized by large Ca²⁺ channel clusters, which localize underneath the presynaptic density³⁰ and consist predominantly of pore-forming L-type Ca_V1.3 subunits^31,32, auxiliary Ca_Vβ2³³, and likely yet-to-be-identified Ca_Vα2δ subunits³⁴. Ca²⁺ channel clustering depends on multiple molecular scaffolds, such as Bassoon or the ribbon (or both)^10,35 as well as RIM2α and β³⁶. The seamless interplay and correct localization of these proteins is not only required for establishing a normal Ca²⁺ channel complement^10,36 but also critical to stabilize a large readily releasable pool of vesicles at the AZ^10,11,36. Moreover, IHC AZs contain additional scaffolds such as Piccolino, a short splice variant of Piccolo³⁷, and the Usher protein harmonin that directly interacts with presynaptic Ca²⁺ channels, regulates their gating, and likely promotes their proteasomal degradation^38,39. Although the endocytic machinery is still largely uncharted, it was recently shown to include AP-2^28,29, dynamins^40,41, amphiphysin, and clathrin heavy chain⁴¹.

Tight spatial coupling of Ca²⁺ channels and vesicular Ca²⁺ sensors

The manner in which Ca²⁺ influx couples to vesicle fusion critically determines the properties of synaptic transmission. Two limiting cases can be distinguished (reviewed recently in 42–44): (i) “pure” Ca²⁺ nanodomain control, in which the Ca²⁺ driving the fusion of a given vesicle is contributed by an individual voltage-gated Ca²⁺ channel, and (ii) “pure” Ca²⁺ microdomain control, where the amount of Ca²⁺ at the Ca²⁺ sensor is governed by a population of Ca²⁺ channels, with negligible impact of individual channels. Aside from the precise topography of the channels with respect to the vesicular Ca²⁺ sensors and their numbers and open probabilities, the Ca²⁺-binding properties of the vesicular Ca²⁺ sensor and the cytosolic Ca²⁺ buffering at the AZ govern the coupling⁴⁵. Previous work has examined the Ca²⁺-binding properties of the Ca²⁺ sensor of fusion by combining whole-cell Ca²⁺ uncaging and membrane capacitance measurements in IHCs of mice right after hearing onset⁴⁶. In these experiments, a requirement for four to five Ca²⁺ ions to bind cooperatively prior to fusion was indicated, and an overall low Ca²⁺ affinity of the sensor can be assumed. Note however, that this approach yielded massive exocytosis (added membrane equivalent to 15% of the cell’s surface). Hence, it is unlikely to be entirely mediated by exocytosis at IHC AZs, but probably also involved extrasynaptic exocytosis. This calls for revisiting the intrinsic Ca²⁺ dependence of synaptic vesicle fusion by using refined approaches to exocytosis at AZs, ideally of more mature IHCs.

Classic⁴⁷ and more recent^30,31,48 work indicates that hair cell AZs harbor tens of Ca²⁺ channels on average. However, within a given IHC, regardless of its tonotopic position, the number of Ca²⁺ channels per AZ varies dramatically. This presynaptic heterogeneity is thought to be related to the requirements of wide dynamic range sound encoding^30,49–51. Interestingly, IHCs display opposing gradients of their AZs for Ca²⁺-channel complement (higher at the modiolar side, facing the ganglion) and voltage-dependence of activation (voltage for half-maximal activation more hyperpolarized at the pillar side, facing away from the ganglion)⁵².

Moreover, depending on the experimental strategy, estimates for the maximal Ca²⁺ channel open probability at IHC AZs vary between 0.2⁴⁸ and 0.4³⁰. To date, the exact topography of individual Ca²⁺ channels within the observed stripe-like clusters beneath the ribbon and their putative molecular linkers to vesicular release sites remains to be experimentally determined. Here, biophysically constrained modeling has proven to be a useful tool in exploring the consequences and feasibility of various scenarios (see below, 30,53). Moreover, another interesting aspect in this context will be the detailed identification of the molecular processes governing AZ maturation, in particular, in regard to the progressive tightening of Ca²⁺ channel-synaptic vesicle coupling during early postnatal development^30,48,54.

Endogenous Ca²⁺ buffering has been studied in hair cells of various species^53,55–58 typically revealing substantial concentrations of Ca²⁺-binding sites (up to a few millimolar). Recently, measurements of exocytic membrane capacitance changes in mutant IHCs lacking the three major cytosolic EF-hand Ca²⁺-binding proteins (that is, calbindin-28k, calretinin, and parvalbumin) were combined with Ca²⁺ buffer substitution—using different concentrations of synthetic Ca²⁺ chelators with either slow (EGTA) or fast (BAPTA) kinetics—and computational modeling of stimulus-secretion coupling⁵³. With this combinatorial approach, the effective average coupling distance between the Ca²⁺ sensor of the release site and the nearest Ca²⁺ channels was estimated to equate to approximately 17 nm in mature IHCs and this is well in line with a Ca²⁺ nanodomain-like control of exocytosis and similar to previous estimates (approximately 23 nm;⁵⁹).

The notion of a Ca²⁺ nanodomain-like control of exocytosis is supported by observations of a lower apparent Ca²⁺ cooperativity (approximately 1.4) of IHC exocytosis upon changes in the channel open probability, when compared with that found with changes in single-channel Ca²⁺ current (3–4;^30,31,59). The interpretation of these discrepant apparent cooperativity estimates as evidence for Ca²⁺ nanodomain-like control of exocytosis was further substantiated by modeling³⁰. There, 50 Ca²⁺ channels were distributed over an area of 80 × 400 nm, aiming to match the Ca²⁺-channel clusters, assuming different topographies of the Ca²⁺ channels to a dozen release sites that were placed at the rim of the Ca²⁺ channel cluster. Modeling of Ca²⁺-triggered exocytosis was constrained by experimental observations as much as possible. When channels were randomly positioned, the apparent Ca²⁺ cooperativity of exocytosis during changes in the number of Ca²⁺ channels was close to two, and hence higher than the experimentally observed value of 1.4. The physiological Ca²⁺ cooperativity was best matched when allocating one molecularly coupled channel to each release site, while the other channels were distributed randomly but respected an exclusion zone of 10 nm around the coupled channels. This was taken to support the notion of Ca²⁺ nanodomain-like control of exocytosis at the IHC AZs of mice after hearing onset, likely realized by molecular coupling of a “private” channel to the release site.

During development, Ca²⁺ influx-exocytosis coupling tightens from Ca²⁺ microdomain-like control to Ca²⁺ nanodomain-like control³⁰, as also found for other synapses⁶⁰. At this point, Ca²⁺ microdomain control of exocytosis seems less likely for mature IHC synapses. Initially, this mechanism had been considered as an explanation for the low apparent Ca²⁺ cooperativity of exocytosis in mature IHCs, which was thought to employ a linear Ca²⁺ sensor (one Ca²⁺-binding step by synaptotagmin 4) after the onset of hearing⁶¹. Alternatively, a lower cooperativity, even in the presence of a sensor with several Ca²⁺-binding steps, was suggested to result in a linear apparent Ca²⁺ dependence in whole-cell membrane capacitance measurements, due to summing exocytosis from heterogeneous, but Ca²⁺ microdomain-governed AZs in IHC capacitance measurements of exocytosis⁶². However, whereas the former hypothesis seems hard to reconcile with the comparable and supralinear intrinsic Ca²⁺ dependence of fusion prior and after hearing onset³⁰, the latter appears incompatible with the indications for Ca²⁺ nanodomain control of exocytosis at the single-synapse level⁵⁹.

Future experiments should elucidate the topography and mobility of individual Ca²⁺ channels within the cluster and dissect putative linker proteins connecting release sites to the nearest Ca²⁺ channel(s). Moreover, further characterization of the molecular composition of the Ca_V1.3 Ca²⁺ channel complexes, also testing the presence and role of splice variants of the pore-forming Ca_V1.3α subunit, will assist in understanding (i) the molecular mechanisms that govern the heterogeneity of Ca²⁺ channel expression, (ii) the developmental tightening of excitation-secretion coupling, and (iii) the respective contributions of individual Ca²⁺ channel variants to shaping the efficiency of presynaptic Ca²⁺ influx at individual IHC AZs.

Large and variable excitatory postsynaptic currents: the uniquantal versus multiquantal release debate

We will now focus on the mode of vesicular release at auditory ribbon synapses, a mechanism that remains only partially understood. The phenomena that raised uncertainty about this fundamental process of synaptic transmission are (i) a remarkable heterogeneity of AMPA receptor-mediated excitatory postsynaptic current (EPSC) amplitudes, that can range from about 20 to more than 800 pA at postsynaptic SGN terminals of rodents and (ii) the differences in release kinetics, as reflected in variable rise times and waveforms of the EPSCs⁶³.

Conventionally, quanta of neurotransmitter are released spontaneously (“uniquantal release”), thereby producing so-called miniature EPSCs (mEPSCs). These mEPSCs are uniform in size and have characteristic monoexponentially decaying waveforms. Such mEPSCs are thought to correspond to spontaneous and statistically independent fusion of individual vesicles, constituting the basis of the quantal hypothesis of transmitter release⁶⁴. At IHC synapses, large variance of EPSC amplitudes and waveforms is found in individual postsynaptic boutons even in complete absence of stimulation. Synchronized (statistically dependent) release of multiple vesicles (“synchronized multiquantal release”) was offered as a plausible mechanism explaining such EPSC heterogeneity^59,63,65. Synchronized multiquantal release has also been indicated for hair cell synapses of turtle and bullfrog papillae^66–69, and for other sensory synapses, such as those in retinal bipolar cells^70–72.

Heterogeneity of the EPSC shape might result from varying degrees of synchronicity of multiquantal release^63,73. Potential mechanisms mediating synchronized multiquantal release include release site synchronization and compound or cumulative fusion of vesicles (reviewed in 2,4,74). The synaptic ribbon might contribute to synchronizing multiquantal release by clustering presynaptic Ca²⁺ channels and tethering a large complement of release-ready synaptic vesicles^{10,21,36,71,75}. One thought is that Ca²⁺ influx through an individual Ca²⁺ channel could synchronously trigger the fusion of several nearby vesicles underneath the synaptic ribbon⁶⁶. Alternatively, vesicles might pre-fuse to each other to form larger quanta prior to fusion to the plasma membrane (compound exocytosis), or fuse to a vesicle that is in the process of release (cumulative exocytosis).

Recently, uniquantal release through a dynamic fusion pore has been suggested as an alternative hypothesis for IHC synapses of rodents⁷⁶. The motivation came from considering the spike rates of hundreds of hertz over prolonged periods of time, which, according to the multiquantal hypothesis (on average, six vesicles per EPSC⁵⁹), would require at least sixfold-higher vesicle release rates, which seem very high considering measured rates of sustained exocytosis of about 700 vesicles per second²⁷. Moreover, in conditions omitting the synchronizing effect of presynaptic Ca²⁺ entry, large monophasic, as well as multiphasic, EPSCs persisted⁷⁶, seemingly arguing against a Ca²⁺-synchronized multiquantal release scenario at the IHC ribbon synapse. In addition, biophysically constrained mathematical modeling of compound exocytosis suggested the presence of large vesicles near the AZ membrane, which was not found in electron microscopy of stimulated samples. These latter findings are difficult to reconcile with a synchronized multiquantal release mode to take place at mammalian IHC ribbon synapses. So, could a uniquantal release model offer an explanation for the large heterogeneity of EPSC amplitudes?

Uniquantal release through a dynamic fusion pore has also been proposed for other synapses^77,78; however, for this scenario to be plausible at IHC synapses, two major questions arise: (i) could the glutamate content of a single synaptic vesicle elicit the observed large EPSCs (on average, about 300 pA) and (ii) could transmitter release be governed by a fusion pore that might regulate the extent and timing of release events? Using mathematical modeling—constrained by morphological estimates of the postsynaptic AMPA receptor clusters (Figure 1) and assumptions on glutamate content and AMPA receptor density and function—the study concluded that single-vesicle release might suffice to trigger large-amplitude EPSCs⁷⁶. Moreover, model prediction and deconvolution of EPSCs suggested that fusion pore regulation could account for the observed variable EPSC shapes. In detail, the authors suggested a model in which there is either immediate and full collapse upon vesicle fusion (potentially explaining monophasic EPSCs) or alternatively the formation of a transitory instable fusion pore prior to collapse, potentially flickering open and closed—a mechanism permitting progressive glutamate unloading that may explain the observation of multiphasic EPSCs. In addition, variable vesicle size that occurs also in the absence of homotypic vesicle-to-vesicle fusion and different filling states of vesicles⁷⁹ might contribute to the EPSC heterogeneity at IHC synapses. Finally, other mechanisms such as spill-over of glutamate from neighboring synapses and postsynaptic receptor properties need to be considered when relying on postsynaptic recordings.

These different hypotheses might not be mutually exclusive, and appear to strongly depend on the chosen experimental model system. Further experiments, such as (i) membrane capacitance recordings of individual fusion events, (ii) electron tomography of synapses immobilized within milliseconds after stimulation, or (iii) super-resolution fluorescence live-cell imaging of vesicular exocytosis, will help in the future to pinpoint the mode of exocytosis at hair cell ribbon synapses.