Abstract
The mammalian auditory system encodes sounds with subtypes of spiral ganglion neurons (SGNs) that differ in sound level sensitivity, permitting discrimination across a wide range of levels. Recent work suggests the physiologically defined SGN subtypes correspond to at least three molecular subtypes. It is not known how information from the different subtypes converges within the cochlear nucleus. We examined this issue using transgenic mice of both sexes that express Cre recombinase in SGNs that are positive for markers of two subtypes: CALB2 (calretinin) in type 1a SGNs and LYPD1 in type 1c SGNs, which correspond to high- and low-sensitivity subtypes, respectively. We crossed these with mice expressing floxed channelrhodopsin, which allowed specific activation of axons from type 1a or 1c SGNs using optogenetics. We made voltage-clamp recordings from bushy cells in the anteroventral cochlear nucleus (AVCN) and found that the synapses formed by CALB2- and LYPD1-positive SGNs had similar EPSC amplitudes and short-term plasticity. Immunohistochemistry revealed that individual bushy cells receive a mix of 1a, 1b, and 1c synapses with VGluT1-positive puncta of similar sizes. We used optogenetic stimulation during in vivo recordings to classify chopper and primary-like units as receiving versus nonreceiving 1a- or 1c-type inputs. These groups showed no significant difference in threshold or spontaneous rate, suggesting the subtypes do not segregate into distinct processing streams in the AVCN. Our results indicate that principal cells in the AVCN integrate information from all SGN subtypes with extensive convergence, which could optimize sound encoding across a large dynamic range.
Significance Statement
Sound information is carried to the brain by auditory afferents that differ in their sensitivity to sound. Auditory afferents fall into three subtypes differing in their spontaneous and sound-evoked activity. These subtypes can be distinguished by anatomical, physiological, and molecular features. However, it is not well understood how the three subtypes relay information to the brain. We addressed this question using optogenetic stimulation to study the properties of two afferent subtypes in the anteroventral part of the cochlear nucleus. We found that the properties of synapses are indistinguishable between subtypes, and they show extensive convergence, suggesting that combining inputs with different sound level sensitivity is important for processing sound.
Introduction
The auditory nerve encodes sound stimuli over a wide range of intensities, on the order of 140 dB, while preserving sensitivity to small changes in intensity. An important adaptation supporting this capability is that spiral ganglion neurons (SGNs) vary in their sensitivity to sound intensity, and each SGN is sensitive to small changes in sound intensity over a relatively limited dynamic range (∼20 dB). SGN sensitivity correlates with spontaneous firing rate, such that low, medium, and high “spont” SGNs have relatively low, medium, and high sensitivity, respectively. In some species (cat, Liberman, 1978; gerbil, Huet et al., 2016), the range of spontaneous rates appears distinct for each class, which suggests that each class supports specialized functions.
The sound sensitivities of SGNs are correlated with multiple anatomical features. Low-spont and high-spont SGNs differ in which side of the hair cell they contact, the sizes of their presynaptic ribbons and postsynaptic AMPA receptor patches, and their axon diameters (Liberman, 1982; Liberman et al., 2011). Sound level sensitivity appears to depend on the properties of presynaptic calcium channels (Ohn et al., 2016; Özçete and Moser, 2021).
In addition, SGNs have three molecular subtypes. One subtype, called 1a, expresses calretinin (CALB2) and appears to correspond to high-spont SGNs (Petitpré et al., 2018; Shrestha et al., 2018; Sun et al., 2018). The 1c subtype expresses LYPD1 and Pou4f1 and appears to correspond to the low-spont subtype. The 1b subtype expresses calbindin (CALB1) and could correspond to medium-spont fibers. The existence of molecular specializations supports the view that SGN subtypes carry out distinct functions, requiring the expression of particular proteins. Electrophysiological recordings in the cochlea from postsynaptic processes of CALB2- and LYPD1-positive SGNs show spontaneous activity levels similar to those expected from subtypes 1a and 1c (Siebald et al., 2023). These extensive differences “suggest that the separate [spontaneous rate] groups may play fundamentally different roles in the process of auditory perception” (Ryugo, 1992), but how they do that is not clear. An important issue to resolve is how the different SGN subtypes influence their targets in the brain.
Two major targets of SGNs in the anteroventral cochlear nucleus (AVCN) are bushy cells and T-stellate cells. Bushy cells in vivo respond to sounds similarly to SGN afferents and are therefore called “primary-like,” while T-stellate cells respond with regularly spaced spikes, and are called “choppers” (Osen, 1969; Cant and Morest, 1979; Young et al., 1988; Young and Oertel, 2004). SGNs form large axosomatic synapses onto bushy cells called “endbulbs of Held” (Lorente de Nó, 1981). Endbulb morphology in cats correlates with spontaneous rate, and bushy cells appear to receive multiple endbulbs with similar morphology, which suggests that bushy cells receive synapses from SGNs of the same spontaneous rate class (Sento and Ryugo, 1989; Ryugo and Sento, 1991). However, calretinin immunohistochemistry indicates that bushy cells in mice receive varying ratios of type 1a versus non-1a endbulbs (Wang et al., 2021). Thus, it is not clear how the diverse level sensitivity of SGNs carries forward into bushy cells.
To address these issues, we used transgenic mice that expressed channelrhodopsin in CALB2+ and LYPD1+ neurons to specifically study type 1a and 1c SGN inputs to AVCN principal cells. We were unable to identify a transgenic mouse line for specifically labeling 1b SGNs. Optogenetic stimulation revealed that EPSCs of CALB2+ and LYPD1+ endbulbs had similar physiological properties, including short-term depression. In addition, we identified units in the AVCN in vivo that received input from CALB2+ or LYPD1+ SGNs, and they showed similar sensitivity to sound. Overall, our results indicate that the different subtypes of SGNs converge extensively on principal cell types in the AVCN.
Materials and Methods
All experiments were performed with the approval of the University at Buffalo's Institutional Animal Care and Use Committee. Experiments were conducted on mice of both sexes. Strains were (1) CBA/CaJ mice (The Jackson Laboratory RRID:IMSR_JAX:000654), (2) mice expressing a floxed channelrhodopsin-eYFP construct (Ai32, The Jackson Laboratory RRID:IMSR_JAX:024109; Madisen et al., 2012), which we refer to as “ChR2” mice, (3) mice with a CreERT2 construct at the CALB2 locus (The Jackson Laboratory RRID:IMSR_JAX:013730; Taniguchi et al., 2011), which we refer to as “CALB2-Cre” mice, (4) mice with a CreERT2 construct at the LYPD1 locus (Siebald et al., 2023), called “LYPD1-Cre” mice. Each line was backcrossed with CBA/CaJ mice for at least five generations. Experimental mice were generated by crossing CALB2- and LYPD1-Cre mice with ChR2 mice, which we refer to for simplicity as “CALB2” and “LYPD1” mice. Genotypes were confirmed by PCR using primers listed in Table 1.
Primers for genotyping transgenic (TG) mouse lines
Expression of ChR2 was induced in LYPD1 and CALB2 mice by tamoxifen injection (TI). Tamoxifen (Sigma-Aldrich T5648) was dissolved in 100% ethanol (10 mg/ml), added to equal volume corn oil (Sigma-Aldrich C8267), spun in a SpeedVac until the ethanol evaporated (∼40 min), and aliquoted and frozen until ready to be used. Mice were injected beginning at P28 with 0.075 mg/g for 4 d and then maintained until P42–P52 for brain-slice experiments or 2–8 months for in vivo electrophysiology. Mice not injected with tamoxifen were used as controls, and are referred to as “LYPD1TI−” and “CALB2TI−.”
Brain-slice electrophysiology
LYPD1 or CALB2 mice were anesthetized with 200 mg/kg ketamine plus 10 mg/kg xylazine (K/X) and perfused with ice-cold sucrose solution [containing the following (in mM): 76 NaCl, 75 sucrose, 25 NaHCO3, 25 glucose, 2.5 KCl, 1.25 NaH2PO4, 7 MgCl2, 0.5 CaCl2]. Then sagittal brain slices were cut using a Leica VT1200 (142 µm) and incubated at 34°C for 20 min in artificial cerebrospinal fluid [ACSF; containing the following (in mM): 125 NaCl, 26 NaHCO3, 20 glucose, 2.5 KCl, 1.25 NaH2PO4, 1 MgCl2, 1.5 CaCl2, 4 Na ʟ-lactate, 2 Na-pyruvate, 0.4 Na ʟ-ascorbate, bubbled with 95% O2–5% CO2]. Slices were then kept at room temperature until recordings. Slices were mounted on coverslips coated in poly-ʟ-lysine (Sigma-Aldrich P4832), placed under an Olympus BX51WI microscope, and perfused with ACSF containing 1 µM strychnine (Sigma-Aldrich 58793) at 3–4 ml/min using a pump (403U/VM2; Watson Marlow) running through an inline heater to maintain the temperature at 34°C (SH-27B with TC-324B controller; Warner Instruments). In some experiments, asynchronous (delayed) release was enhanced by replacing divalents in the ACSF with 3 mM SrCl2 and 0.1 mM MgCl2.
Whole-cell patch-clamp recordings were made from BCs in the AVCN using borosilicate patch pipettes with resistance of 1.3–2.3 MΩ and voltage-clamped with a MultiClamp 700B (Molecular Devices) controlled by an ITC-18 interface (Instrutech), driven by custom-written software (mafPC) running in Igor (WaveMetrics). Pipettes were filled with internal solution containing the following (in mM): 35 CsF, 100 CsCl, 10 EGTA, 10 HEPES, and 1 QX-314, pH 7.3, 300 mOsm. For assessing AMPA currents, BCs were held at –70 mV with access resistance of 5–15 MΩ compensated to 70%. BCs were identified by EPSCs with fast decay kinetics (τ < 0.2 ms) and half-widths <0.5 ms. For assessing NMDA currents, BCs were held at +40 mV. The liquid junction potential was calculated to be 6.6 mV for voltage-clamp and 8.8 mV for current-clamp internal solutions with respect to normal ACSF. Recordings were not corrected for the junction potential. SGN axons were stimulated electrically with a glass microelectrode placed 30–50 µm away from the soma with currents of 4–20 µA through a stimulus isolator (World Precision Instruments, A360). Individual synaptic inputs were isolated by adjusting the electrode position and pulse amplitude. Optogenetic stimulation was done with a 0.2 ms light pulse produced by a CoolLED pE-300ultra attached to the fluorescence port of the microscope. The diameter of the illuminated area was minimized using the field diaphragm to restrict stimulation to a single presynaptic SGN axon. PPRs were evoked with an interval of 15 ms, which was the minimum reliable interval for optogenetic stimulation.
For measuring quantal amplitudes, we used spontaneous EPSCs (sEPSCs), which in bushy cells are commonly 1–10% of the amplitude of electrically evoked EPSCs. EPSCs with large amplitudes never occur in bushy cells without electrical or optogenetic stimulation. Therefore, sEPSCs evidently represent the release of individual quanta, and the use of tetrodotoxin is not necessary.
Spiral ganglion immunohistochemistry
LYPD1 or CALB2 mice were anesthetized with K/X, then perfused with 0.9% saline plus 0.25 g/L heparin, followed by 4% buffered paraformaldehyde (PFA, Sigma-Aldrich 158127). Cochleas were postfixed in 4% PFA overnight and then decalcified in 100 mM EDTA (pH adjusted to 7.2 with HCl; Sigma-Aldrich E6758) for 3 d. Decalcified cochleas were embedded in 20% gelatin, fixed overnight in 4% PFA, and transferred to 20% sucrose in phosphate-buffered saline (PBS: 0.9% NaCl, 0.02 M phosphate, pH = 7.4). Frozen sections were cut on a sliding microtome (American Optical) at 50 µm thickness. Sections were washed in PBS, permeabilized with 0.5% Triton X-100 (Sigma T8787) in PBS (PBST) for 10 min, washed in PBS, blocked in 1% normal donkey serum in PBST for 1 h, and incubated overnight at 4°C in primary antibody solution [anti-calretinin (αCR) for CALB2 mice, anti-Pou4f1 (αPou4F1) for LYPD1 mice; Table 2] in PBST. To enhance the visibility of eYFP expression, an antibody against GFP (αGFP) was included. Sections were then washed in PBS and incubated for 1 h with secondary antibodies (Table 2). Sections were then washed in PBS, mounted on slides, and coverslipped with ProLong Diamond Antifade (Invitrogen, P36961). A portion of the middle third of the SGN region in Rosenthal's canal in each cochlea was imaged on a confocal microscope (FV1000, Olympus). Between 64 and 103 (86 ± 11, mean ± standard deviation) SGNs in each image series (one or two cochleas per mouse, three mice, 26–46 optical sections per series) were scored for immunoreactivity to subtype-specific markers (αCR for subtype 1a and αPou4F1 for subtype 1c) and overlap with eYFP immunoreactivity.
Antibodies used for immunohistochemistry
AVCN immunohistochemistry
LYPD1 and CALB2 mice were anesthetized with K/X and then perfused with saline and PFA, as above. Brains were postfixed at room temperature for 1 h and then placed in 20% sucrose overnight. Sagittal sections were cut frozen at 25–50 µm thickness on a sliding microtome (American Optical) and treated similarly to cochlear sections. Slices were incubated in primary antibody solution (αVGluT1, αGFP, αcalretinin; Table 2) in PBST, washed in PBS, and incubated in secondary antibody solution. Image series were collected on a Leica TCS SP8 confocal. Endbulbs were identified as rings of VGluT1+ immunoreactivity around unlabeled somata (i.e., bushy cells) in the core of AVCN. VGluT1+ portions of endbulbs were traced in each optical section and reconstructed across adjacent optical sections using a custom-written program running in Igor (Wavemetrics), and the total volume was calculated by summing areas of traced puncta. Subtype identity for each VGluT1+ area was determined by overlapping expression with calretinin immunoreactivity (subtype 1a) or eYFP (subtype 1c) or neither (subtype 1b). Three to five image series were collected in three mice; each series had 28–51 optical sections, spaced every 0.5 µm. Only cells that appeared to be fully enclosed within the image series were quantified.
Auditory brainstem responses (ABRs)
ABRs were measured in a sound-attenuating booth (Med Associates) that was electrically shielded with a grounded steel lining and acoustically shielded with Sonex sound-attenuating foam (Acoustical Solutions). K/X-anesthetized mice were placed on a platform warmed by a water recirculating pump (T/Pump, Gaymar). The recording electrode (27 gauge, 13 mm, Rochester Electro-Medical S83018-R9) was placed subdermally near the pinna, with a reference electrode placed along the midline and a ground electrode in the hindleg, connected to a Medusa4Z preamplifier. A free-field speaker (MF1, Tucker-Davis Technologies) was placed 10 cm in front of the mouse's head, connected to a TDT ABR rig powered by a WS computer with an R26 processor running BioSig software (Tucker-Davis Technologies). The speaker was calibrated using a PCB Piezotronics microphone and signal conditioner (part nos. 378C01 and 480C02). Responses to 0.1 ms broadband clicks were bandpass filtered at 300–3,000 Hz and averaged over 512 presentations at each sound level, stepping down from 90 to 10 dB SPL in 5 dB steps. ABR waveforms and thresholds were analyzed in WaveMetrics Igor, identifying the threshold by visual inspection.
AVCN in vivo electrophysiology
Anesthesia was induced in mice with K/X. Anesthesia was confirmed every 30 min using toe pinch and supplemented with 20:1 mg/kg ketamine/xylazine if a response was observed. Mice were administered 100–150 µl 0.9% saline SQ every 30 min. The mouse was placed on a platform maintained at 38°C by a water recirculating pump (T/Pump, Gaymar) in a chamber lined with grounded steel sheeting for electrical shielding and Sonex sound-attenuating foam (Acoustical Solutions). Stimuli were generated in WaveMetrics Igor with a sampling rate of 200 kHz and relayed by the data acquisition system to a Crown XLi 800 amplifier, which drove a Peerless XT25BG60-04 (Tymphany) speaker placed 20 cm from the mouse. Tone levels were calibrated using a Larson Davis Casella Cel sound level meter.
Hair was removed from the scalp using Nair, an incision was made along the midline, and the scalp was removed to expose the posterior skull. The bone was etched with 30–35% hydrogen peroxide, and a head holder was glued to it using OptiBond (Kerr Dental, 36519) and Nexus universal adhesive (Kerr Dental, 36991) cured with a UV light (Henry Schein, Maxima RU1200, #1126227). A small hole was drilled over the left cochlear nucleus and widened to a diameter of 2 mm. The paraflocculus and lateral cerebellum were aspirated until the cochlear nucleus could be visualized. An optical fiber (500 µm, 0.63 NA, Prizmatix) was positioned over the cochlear nucleus and attached to an LED light source (Optogenetics-LED-Blue Prizmatix).
Sharp electrodes were pulled from borosilicate glass (1.5 mm OD, 0.86 ID, with filament, Sutter Instrument) to a tip length of ∼4 mm. Electrodes were filled with 3 M KCl, yielding resistances typically near 15 MΩ. The electrode was directed at a 45° angle into the cochlear nucleus and lowered using Scientifica IVM mounted on a PatchStar manipulator. A 15 ms noise burst was played repeatedly as a search stimulus (∼92 dB SPL, random number generator running at ∼10 kHz), while advancing the electrode into the AVCN. Observing any action potentials within a 50 ms window was taken as evidence of a nearby unit.
Once a unit was detected, its tuning curve was collected to identify its characteristic frequency (CF). A rate-level function (RLF) was measured for 40 ms tones (5 ms cosine rise–fall times) at CF. The spontaneous rate was measured from periods between stimuli. Optogenetic stimuli were delivered through the optical fiber. Most experiments used light pulses of 5.5–7.5 ms duration. Tones or optogenetic stimuli were presented every ∼0.2 s. In many cases, the recording electrode detected a voltage pulse immediately after optical stimulation, which was either an artifact from the LED stimulation or a field potential driven by optogenetic stimulation. We referred to this pulse as the “neuroptic,” by analogy with the neurophonic triggered by sound stimuli. The neuroptic was large enough to interfere with spike detection. Therefore, we recorded the neuroptic in the same region after losing the cell, or after moving the electrode slightly, and subtracted it from traces to improve spike detection during analysis.
The absolute threshold of each unit was assessed from the rate-level function at CF, where the spike rate was significantly greater than spontaneous (Krishnamoorthy and Thomson, 2004). For a better comparison between units of different CF, we converted absolute thresholds to relative thresholds following Suthakar and Liberman (2021). That is, we assessed ABR thresholds to tone stimuli (5 ms duration, 2 ms gate, 10–55 dB SPL, 6–32 kHz) for six ears from CBA/CaJ mice and estimated the ABR threshold at each unit's CF from the linear interpolation. The relative threshold was calculated as the difference between each unit's absolute threshold and the estimated ABR threshold.
Statistical analysis
We tested data for normality using Shapiro–Wilk tests (detailed results are not shown, for clarity). Normally distributed data are compared using Student's t test or ANOVA and summarized as mean ± standard error of the mean (SEM). Post hoc tests were corrected for multiple comparisons using Holm's correction. Non-normally distributed data are compared using Mann–Whitney rank tests and summarized as median ± median absolute deviation.
Results
Mouse models for studying SGN subtypes
We backcrossed all transgenic alleles into the CBA/CaJ line to avoid hearing problems associated with aging in the background C57 strain (Hunter and Willott, 1987). Previous reports have indicated high specificity of expression in the CALB2-Cre and LYPD1-Cre lines for 1a and 1c SGNs, respectively (Siebald et al., 2023). We were unable to identify a comparable mouse line that would restrict labeling to subtype 1b SGNs. After backcrossing, we verified the expression of the ChR2-eYFP construct in sections throughout the spiral ganglion. Fluorescence was particularly strong in the cell membrane in both mouse strains (Fig. 1A,B), likely reflecting the expression of the ChR2 in the membrane, which is ideal for optogenetic stimulation. We verified that ChR2 was restricted to the subtype of interest by colabelling with appropriate antibodies. We applied anti-calretinin (αCR) antibodies in CALB2 mice and found that there was substantial overlap with eYFP expression (Fig. 1A). Similarly, we applied anti-Pou4f1 (αPou4f1) antibodies in LYPD1 mice and found there was substantial overlap with eYFP expression (Fig. 1B). In both strains, the majority of neurons with either label were double-labeled (CALB2, 83 ± 3% in three mice, five cochleas; LYPD1, 89 ± 2% in three mice, six cochleas; Fig. 1C). In the CALB2 mice, 16 ± 3% of neurons were αCR-positive, but not eYFP-positive, while eYFP expression without αCR immunoreactivity was rare (0.4 ± 0.3% of labeled SGNs). In LYPD1 mice, few SGNs were only αPou4f1-positive (3.8 ± 1.5%) or eYFP-positive (6.9 ± 0.3%). Thus, it seemed likely that optogenetic stimulation would largely activate the SGN subtypes expected.
ChR2-eYFP expression in the spiral ganglion and cochlear nucleus. A, B, Confocal images of the spiral ganglion from (A) CALB2 and (B) LYPD1 mice. Images show eYFP fluorescence (yellow) overlaid with immunoreactivity to antibodies against (A, blue) calretinin or (B, red) Pou4f1. The white arrow in B indicates a cell that shows eYFP fluorescence, but no Pou4f1 immunoreactivity. Scale bar in A also applies to B. C, Quantification of SGNs that did or did not express eYFP and subtype-specific markers. Markers are values from individual cochleas (1 to 2 cochleas per mouse, 3 mice). eYFP expression is closely linked to marker expression in both 1a and 1c mouse lines. D–G, Confocal stacks showing overviews of the cochlear nucleus (Di, Fi) and closeups where the auditory nerve enters (Dii, E, Fii, G). eYFP (yellow) is highly expressed in tamoxifen-injected mice (D, CALB2; F, LYPD1) and sparse in mice not injected with tamoxifen (E, CALB2TI−; G, LYPD1TI−). VGluT1-immunolabelling (magenta) labels endbulbs in the AVCN, for context. Scale bar in E applies to close-ups in D–G. The dotted boxes in Di and Fi show locations of closeups in Dii and Fii.
We also confirmed ChR2-eYFP expression in the AVCN. CALB2 and LYPD1 are broadly expressed among SGNs during development, raising the possibility that “leaky” expression of Cre at early ages could drive eYFP-ChR2 expression in SGNs before marker expression refines to the mature subtype. We tested this by examining the AVCN from mice that were injected with tamoxifen, which showed large numbers of eYFP-expressing fibers in CALB2 (Fig. 1D) and LYPD1 (Fig. 1F) mice. In contrast, with no tamoxifen injection (TI−), there were no or few eYFP-positive fibers in CALB2TI− (Fig. 1E) or LYPD1TI− mice (Fig. 1G). This suggests that “leaky” Cre expression is minimal. We saw similar results for three replicates of each genotype. Thus, optogenetic stimulation is unlikely to activate significant numbers of misexpressing SGNs.
We evaluated auditory function in these mouse lines by measuring ABRs. It was important to confirm auditory function because calretinin is expressed throughout the nervous system including in the auditory system. The CALB2-Cre construct disrupts one CALB2 allele (Taniguchi et al., 2011), raising the possibility that auditory function could be affected. The LYPD1-Cre and floxed ChR2 constructs are not known to influence function more broadly. Thresholds for 0.1 ms clicks were near 30 or 35 dB SPL for CBA/CaJ, CALB2, LYPD1, CALB2TI−, and LYPD1TI− mice (Fig. 2A,B; ANOVA F(4,28) = 1.44, p = 0.25). We quantified the amplitude of the first positive-going ABR peak with respect to baseline (Wave 1), which is thought to reflect activity in the cochlea/auditory nerve 1(Melcher and Kiang, 1996). We found a similar dependence of wave 1 amplitude with sound level across the non–ChR2-expressing strains. We pooled these controls and compared ABR-level functions against the ChR2-expressing strains. There was no significant difference between control, CALB2, and LYPD1 strains (Fig. 2C; two-factor ANOVA F(2,560) = 1.25, p = 0.29). We observed small differences in the amplitudes of Waves 2 and 3 of the ABR in CALB2 and LYPD1 mice compared with the control (data not shown). This suggests that ChR2 expression in SGNs did not significantly reduce hearing sensitivity or responses in the periphery, though there could be some minor central effects. Therefore, it seemed appropriate to use these transgenic mice to investigate the SGN subtypes further.
ABR assessment of transgenic lines showing relatively normal auditory activity. A, Representative ABR waveforms from a CBA/CaJ mouse (left, black), a CALB2 mouse (middle, red), and a LYPD1 mouse (right, blue), in response to 0.1 ms click stimuli. B, ABR thresholds for ears from CBA/CaJ mice (“CBA”), CALB2 mice not injected with tamoxifen (“CTI−”), LYPD1 not injected with tamoxifen (“LTI−”), CALB2 mice injected with tamoxifen (“C”), and LYPD1 mice injected with tamoxifen (“L”). C, ABR-level functions for wave 1 of the ABR for control (black), CALB2 (red), and LYPD1 (blue) mice. Markers show mean and standard error of the mean (SEM) for each genotype.
Synaptic physiology of SGN subtypes
To study the synaptic properties of 1a and 1c SGNs, we prepared acute slices of the AVCN and made voltage-clamp recordings from bushy cells. We compared EPSCs evoked by optogenetic and electrical stimulation (oEPSCs and eEPSCs, respectively) in slices from LYPD1 mice (Fig. 3A). We verified optical and electrical stimulation activated different synapses using paired stimulation at short intervals. A lack of short-term depression in the second EPSC indicated the two stimulated synapses were distinct. The oEPSCs appeared similar to eEPSCs, with no significant difference in amplitudes (eEPSC, 1.91 ± 0.71 nA; oEPSC, 1.43 ± 0.34 nA; p = 0.32, paired Mann–Whitney rank test, Fig. 3B) or kinetics (eEPSC width at half-height, 0.30 ± 0.02 ms; oEPSC, 0.31 ± 0.02 ms, p = 0.27, Student's t test, Fig. 3C). This suggests that optogenetic activation triggered reasonably normal neurotransmitter release and is an appropriate method to examine subtype-specific synaptic properties.
Properties of SGN synapses onto bushy cells in the AVCN. A, Representative EPSCs recorded in a bushy cell following paired electrical (left) or optogenetic (right) stimulation in a slice from a LYPD1 mouse (interpulse interval of 15 ms). Stimulus artifact after electrical stimulation is blanked. Optogenetic EPSCs were aligned by their rising phase before averaging. B, C, Comparison of EPSC amplitudes (B) and half-widths (C) evoked by electrical or optogenetic stimulation in the same bushy cell. Lines connect data recorded from the same bushy cells. Open markers are (B) median ± median deviation or (C) mean ± standard error from 14 cells in four LYPD1 mice. D, Amplitudes of optogenetically evoked EPSCs from LYPD1 and CALB2 mice. E, Paired-pulse ratio (PPR, 15 ms) in response to electrical or optogenetic stimulation. Lines connect data recorded from the same bushy cells (11 cells, 4 mice). F, Optogenetically evoked PPRs from LYPD1 and CALB2 mice. G, EPSCs evoked optogenetically at +40 mV from LYPD1 (left) and CALB2 (right) mice, showing a short-latency AMPA-receptor–mediated peak, followed by a NMDA-receptor–mediated slow component, which was quantified at 2 ms after the AMPA peak. H, NMDA:AMPA ratios from traces similar to panel G, from LYPD1 (8 cells, 4 mice) and CALB2 (9 cells, 4 mice). No comparisons show significant differences.
We measured oEPSCs in slices from LYPD1 and CALB2 mice. We saw a wide distribution of oEPSC amplitudes in both genotypes. oEPSC amplitudes showed no significant difference between LYPD1 (1.6 ± 1.0 nA, 31 cells, 11 mice) and CALB2 mice (2.2 ± 1.4 nA, 19 cells, 9 mice; p = 0.77, unpaired Mann–Whitney rank test, Fig. 3D). Thus, putative 1a and 1c SGN subtypes have similar EPSC amplitudes on bushy cells.
We also assessed short-term synaptic plasticity using the paired-pulse ratio (PPR), which is sensitive to the presynaptic probability of release at short intervals. Endbulbs typically show strong short-term depression with pairs of stimuli (Isaacson and Walmsley, 1996; Oleskevich and Walmsley, 2002; Yang and Xu-Friedman, 2008; Cao and Oertel, 2010; Chanda and Xu-Friedman, 2010). The shortest interval that was reliable for optogenetic stimulation across all preparations was 15 ms. In some cells, we elicited pairs of pulses using both electrical and optogenetic stimulation (ePPR and oPPR, respectively), which showed no significant differences (Student's t test, 11 cells, p = 0.67, Fig. 3E). This suggests that the oPPR is an appropriate measure for characterizing SGN subtype synapses. oPPRs did not differ significantly between LYPD1 (0.80 ± 0.04, 28 cells, 10 mice) and CALB2 mice (0.87 ± 0.04, 19 cells, 9 mice; p = 0.18, Student's t test, Fig. 3F). Thus, PPR and presynaptic release probability are similar for 1a and 1c SGN subtypes.
We also compared NMDA:AMPA ratios in LYPD1- and CALB2-positive synapses, by holding bushy cells at +40 mV during optogenetic stimulation (Fig. 3G). The AMPA receptor–mediated component of the oEPSC was straightforwardly the short-latency peak. The NMDA-receptor–mediated component did not typically show a distinct peak, so we quantified the NMDA component as the current at 2 ms after the AMPA peak. The NMDA:AMPA ratio in LYPD1 mice (0.037 ± 0.014, eight cells, four mice) and CALB2 mice (0.017 ± 0.003, nine cells, four mice) did not differ significantly (p = 0.12, Mann–Whitney test, Fig. 3H).
We examined quantal sizes made by LYPD1- and CALB2-positive synapses. First, we measured the amplitudes of spontaneous EPSCs (sEPSCs) in cells from LYPD1 and CALB2 mice (Fig. 4A), which most likely reflect the release of single quanta from endbulbs of all types. sEPSC amplitudes were not statistically different between the two strains (LYPD1, 159 ± 13 pA, 16 cells, 8 mice; CALB2, 153 ± 12 pA, 18 cells, 10 mice; p = 0.76, Student's t test, Fig. 4Ci). In addition, we quantified the kinetics of sEPSCs by measuring the width at half-height, which did not differ significantly between the two strains (LYPD1, 0.21 ± 0.01 ms, 15 cells, eight mice; CALB2, 0.22 ± 0.01, 14 cells, nine mice; p = 0.50, Fig. 4Di), suggesting the two strains had similar AMPA receptor expression.
Quantal properties in subtypes. A, Representative recording from a LYPD1 mouse (left) with individual sEPSCs overlaid (middle) and average sEPSC (right). B, Optogenetically evoked EPSCs in the same cell as panel A with 3 mM Sr2+ in the ACSF to enhance delayed release (oEPSCDR). Left panel, Representative trace in response to optogenetic stimulation (markers). The middle panel shows overlaid oEPSCDR, with average at right. C, Amplitudes of sEPSCs (i) and oEPSCDR (ii) recorded in LYPD1 (“L”; i, 16 cells, 8 mice; ii, 17 cells, 9 mice) and CALB2 (“C”; i, 18 cells, 10 mice; ii, 16 cells, 9 mice). oEPSCDR amplitude differed significantly between the two groups (p = 0.01), but sEPSC amplitude did not (p = 0.61). D, Width at half-maximum (half-width) of sEPSCs (i) and oEPSCDR (ii). There were no significant differences between LYPD1 and CALB2 mice.
To isolate quanta associated with synapses of individual subtypes, we used optogenetic stimulation in trains in the presence of ACSF containing 3 mM SrCl2 and 0 mM CaCl2, to enhance delayed release (also called “asynchronous release”; Fig. 4B; Miledi, 1966; Dodge et al., 1969; Goda and Stevens, 1994; Chen and Regehr, 2000; Xu-Friedman and Regehr, 2000). Then we quantified the amplitudes of quanta that occurred after the synchronous EPSC. The EPSCs from optogenetically evoked delayed release in Sr (oEPSCDR) were significantly larger in LYPD1 mice compared with CALB2 (LYPD1, 165 ± 9 pA, 17 cells, 10 mice; CALB2, 137 ± 6 pA, 16 cells, 9 mice; p = 0.013, Student's t test, Fig. 4Cii). We tested if oEPSCDR's were larger in LYPD1 mice as a result of multiple overlapping quantal events by comparing their kinetics. The half-widths of oEPSCDR's were not significantly slower in LYPD1 mice (LYPD1, 0.22 ± 0.01 ms, 17 cells, nine mice; CALB2, 0.24 ± 0.01 ms, 16 cells, nine mice; p = 0.97, Student's one-tailed t test, Fig. 4Dii). This suggests that quanta released from LYPD1-positive endbulbs are slightly larger than quanta released from CALB2-positive endbulbs.
Subtype synaptic structure
We also compared the morphology of synapses made by the different SGN subtypes onto bushy cells. Our strategy was to label all synapses onto bushy cells using anti-VGluT1 antibodies (αVGluT1; Fig. 5Ai), type 1a SGNs using anti-calretinin (αCR) antibodies (Fig. 5Aii), and type 1c SGNs by using eYFP expression in LYPD1 mice (Fig. 5Aiii). We identified type 1a-like synapses as immunoreactive to both αVGluT1 and αCR, type 1c-like as immunoreactive to αVGluT1 and expressing eYFP, and type 1b-like as labeled only with αVGluT1. We took confocal stacks and found bushy cells surrounded by a mixture of type 1a-, 1b-, and 1c-like synapses (Fig. 5Aiv). αVGluT1 immunoreactivity marks areas of SGN fibers with high vesicle density and therefore likely represents regions where synaptic transmission takes place. We traced αVGluT1-positive puncta through multiple sections and quantified their volumes (Fig. 5B). Puncta volumes were similar across subtypes (1a-like, 1.26 ± 0.99 µm3 median ± median absolute deviation, 20 puncta; 1b-like, 1.03 ± 0.41 µm3, 16 puncta; 1c-like, 1.39 ± 0.47 µm3, 15 puncta; 11 cells). There was no significant difference in puncta volumes of the different subtypes (p = 0.56, Kruskal–Wallis test).
Convergence of subtypes. A, Representative confocal section of the AVCN, with synaptic areas around bushy cells labeled with antibodies against VGluT1 (i, αVGluT1) and calretinin (ii, αCR) in a LYPD1 mouse (iii, eYFP). iv, Merged images from panels i–iii. Puncta labeled with eYFP and immunopositive for αVGluT1 (1c-like) appear magenta, puncta labeled with eYFP and immunopositive for αCR (1a-like) appear yellow, and puncta only immunopositive for αVGluT1 (1b-like) are red. B, Volumes of reconstructed αVGluT1+ puncta immunopositive for αCR (1a-like, 20 puncta), labeled with eYFP (1c-like, 15 puncta), and neither (1b-like, 16 puncta). Solid markers are measures of individual puncta (51 puncta, 11 cells, 3 mice), and open markers are median ± median absolute deviation. C, Percent of αVGluT1+ volume onto each cell that is immunopositive for αCR (1a-like), labeled with eYFP (1c-like) and neither (1b-like). Solid markers are individual measures for 11 cells from 3 mice, and open markers are mean ± standard error of the mean.
We also quantified the percentage of each subtype's contribution to the total αVGluT1-positive volume around each cell. Some cells received αVGluT1-positive synaptic input predominantly from one of the SGN subtypes (1a-, 1b-, or 1c-like) and received little from the other subtypes (Fig. 5C). Some cells received αVGluT1-positive inputs with similar volumes across SGN subtypes. On average, cells received 40.7 ± 6.2% (mean ± SEM) 1a-like, 30.1 ± 7.2% 1b-like, and 29.2 ± 7.1% 1c-like synaptic volume, which is similar to the proportion of SGN subtypes (Petitpré et al., 2018; Shrestha et al., 2018; Sun et al., 2018).
In vivo electrophysiology
We examined how the information carried by the different SGN subtypes is processed in the AVCN by quantifying the responses of their targets. We predicted that the activity of units in AVCN would resemble their auditory nerve inputs. Our strategy was to assess a unit's characteristic frequency, threshold, and spontaneous rate and then use optogenetic stimulation to identify if it received excitation from fibers of subtype 1a (“1a-receiving” vs “1a-nonreceiving”) in CALB2 mice or of subtype 1c (“1c-receiving” vs “1c-nonreceiving”) in LYPD1 mice. We predicted that 1a-receiving units would have higher spontaneous rates and lower thresholds compared with 1a-nonreceiving units and that 1c-receiving units would have lower spontaneous rates and higher thresholds compared with 1c-nonreceiving units.
Figure 6 shows representative recordings of a chopper and a primary-like unit. Choppers show simple biphasic action potentials, and their responses to tones show regular interspike intervals (Smith and Rhode, 1987; Taberner and Liberman, 2005; Fig. 6A). Primary-like units respond to sounds with a rapid onset of spiking that adapts (Fig. 6B). In these recordings, observing a prepotential was uncommon, so we relied on the position and depth of the recording electrode and the adaptation of the response to identify primary-like units. We collected tuning curves from each unit to identify CF and measured a rate-level function at CF for each unit to identify the threshold (Fig. 6C). Then we delivered brief (∼5 ms) pulses of blue light to determine if the unit received optogenetically activatable inputs. In the example chopper unit in Figure 6D, optogenetic stimulation led to short-latency action potentials, typically on a subset of sweeps. For units with significant spontaneous firing, we verified the spikes were elicited by optogenetic stimulation by varying the timing of the light pulse during the sweep. Because the auditory nerve is the sole excitatory input to bushy cells and stellate cells, it is most likely that cells that respond to optogenetic stimulation in CALB2 mice receive input from CALB2-positive SGNs, so we call those units 1a-receiving. Similarly, units that respond to optogenetic stimulation in LYPD1 mice likely receive input from LYPD1-positive SGNs, so we call them 1c-receiving.
Using optogenetics to identify 1a- and 1c-receiving units in the AVCN in vivo. A, B, Responses to sound for representative chopper (A) and primary-like (B) units. Top traces, CF tone stimulus (A, 17 kHz, 42.4 dB SPL; B, 16 kHz, 88.8 dB SPL). Middle, Rasters of responses over 15 trials. Bottom, Histogram from all trials. Insets, Individual (gray) and average (black) spikes. Scale bars in insets: A, 0.1 mV, 1 ms; B, 0.2 mV, 1 ms. C, Representative tuning curve and rate-level function for a primary-like unit in a LYPD1 mouse. D, E, Rasters showing action potentials (dots) and optogenetic stimulation (blue rectangles) of representative units in a LYPD1 mouse. The occurrence of spikes during optogenetic stimulation indicates that the chopper unit in D is 1c-receiving, whereas the lack of spikes indicates that the primary-like unit in E is 1c-nonreceiving. There was a reduction in spiking shortly after the light flash in E, most likely because excitatory inputs to unrecorded inhibitory cells were activated.
In addition, we recorded from units in which optogenetic stimulation did not appear to trigger any response. The example in Figure 6E was a primary-like unit. Lack of response was unlikely to be a result of poor light penetration into the tissue, because responsive and nonresponsive cells could be recorded in the same electrode penetration. Furthermore, some nonresponding units with high spontaneous rates had lower spike rates after optogenetic stimulation (e.g., Fig. 6E), most likely because auditory nerve stimulation led to the activation of inhibitory inputs to the cell being recorded. Thus, units that did not respond to optogenetic stimulation most likely received no or subthreshold excitatory input from the corresponding SGN subtype. We therefore refer to those units as 1a-nonreceiving in CALB2 mice and 1c-nonreceiving in LYPD1 mice. During our recordings, we encountered 23 1a-receiving and 28 1a-nonreceiving units in CALB2 mice and 35 1c-receiving and 43 1c-nonreceiving units in LYPD1 mice.
We quantified the threshold from the rate-level function at CF and converted it to a threshold relative to the ABR threshold at that frequency. We also measured the spontaneous rate for each unit during periods when no stimulus was presented, typically for at least 10 s. We examined the relationship between relative threshold versus spontaneous rate, which is highly correlated in SGNs from mice of similar ages (Suthakar and Liberman, 2021). We saw no correlation in either primary-like units (Fig. 7A,C) or choppers (Fig. 7B,D). Furthermore, we saw no significant difference between the spontaneous rate or threshold of 1c-receiving (13) and 1c-nonreceiving (20) primary-like units (spontaneous rate p = 0.52; threshold p = 0.44; 13 1c-receiving, 20 1c-nonreceiving units; Fig. 7A) or choppers (spontaneous rate p = 0.61; threshold p = 0.84; 22 1c-receiving, 23 1c-nonreceiving units; Fig. 7B), nor between 1a-receiving and 1a-nonreceiving primary-like units (spontaneous rate p = 0.98; threshold p = 0.30; 13 1a-receiving, 12 1a-nonreceiving units; Fig. 7C) or choppers (spontaneous rate p = 0.38; threshold p = 0.56; 10 1a-receiving, 16 1a-nonreceiving units; Fig. 7D; Mann–Whitney U test, all comparisons). Thus, 1a and 1c fibers do not appear to contribute to physiological specialization among primary-like or chopper units of the AVCN, suggesting there is extensive convergence of these subtypes onto principal cells of the AVCN.
Comparing features of 1a- and 1c-receiving units in AVCN. Spontaneous rate and sound level thresholds are indistinguishable between 1c-receiving versus 1c-nonreceiving units (A, B) or between 1a-receiving versus 1a-nonreceiving units (C, D), for both primary-like (A, C) and chopper (B, D) units. Small markers indicate spontaneous rate and relative threshold of individual units, and large markers indicate median ± median absolute deviation.
Discussion
We compared the synapses made by two subtypes of SGNs on their postsynaptic target principal cells in the core of the AVCN. We used transgenic mouse lines to target type 1a versus type 1c SGNs, which are thought to correspond to neurons with high versus low spontaneous rate and sound level sensitivity (Petitpré et al., 2018; Shrestha et al., 2018; Sun et al., 2018). Stimulation of specific subtypes in vitro using optogenetics indicated that 1a and 1c synapses had similar EPSC amplitude, short-term depression, and NMDA:AMPA ratio. Furthermore, we used optogenetics to identify the targets of these subtypes in the AVCN and found they had similar spontaneous rates and sound thresholds.
This raises important questions about the function of the different SGN subtypes. A picture has formed about a suite of characteristics that correlate with the SGN subtype. SGNs with high sensitivity (i.e., low threshold) tend to receive synapses on the pillar (outer) pole of the hair cell, while SGNs with low sensitivity (i.e., high threshold) tend to receive synapses on the modiolar (inner) pole (Liberman, 1982). Calcium channels at pillar synapses tend to have lower voltage thresholds than modiolar synapses (Özçete and Moser, 2021). The presynaptic ribbons that organize releasable vesicles tend to be larger on the pillar side and smaller on the modiolar side, while the apposing patches of postsynaptic AMPA receptors tend to be smaller on the pillar side and larger on the modiolar side (Liberman et al., 2011). The endbulbs formed by SGNs with a lower spontaneous rate tend to have more complex morphologies than those a with higher spontaneous rate (Sento and Ryugo, 1989). These various characteristics have been linked to molecular classes that can be distinguished on the basis of the transcriptome (Petitpré et al., 2018; Shrestha et al., 2018; Sun et al., 2018). Targetted recordings in cochlear explants found that subtype 1a SGNs have predominantly higher spontaneous activity than 1c (Siebald et al., 2023).
All these lines of evidence support that SGNs are made up of multiple subtypes that are specialized to serve distinct functions, most likely related to encoding different sound level ranges. However, we saw little difference between subtype 1a and 1c in synaptic physiology or in the sound responses of their targets in the AVCN. Our synaptic physiology results are consistent with those of Zhang et al. (2022), where the SGN subtype was determined after electrophysiological recordings using calretinin immunohistochemistry. Synaptic inputs showed similar depression whether the recorded bushy cell was primarily surrounded by calretinin-immunopositive synapses (i.e., likely type 1a) or not (i.e., likely not type 1a). This suggests that any differences in the computational function of type 1a and type 1c SGNs are not supported by different synaptic properties.
The overall population of SGN synapses is highly variable in terms of synaptic depression, which can have major consequences for response fidelity (Yang and Xu-Friedman, 2009). Remarkably, SGN synapses that converge on the same bushy cell have similar levels of synaptic depression (Yang and Xu-Friedman, 2012). Furthermore, levels of depression at SGN synapses onto bushy cells change in response to increased ambient sound levels and reduced peripheral activity (Ngodup et al., 2015; Zhuang et al., 2017, 2020; Wong and Xu-Friedman, 2022b), and this plasticity appears to occur through a pathway that depends on nitric oxide (Wong and Xu-Friedman, 2022a). Thus, depression levels are tightly regulated at SGN synapses formed onto AVCN neurons, suggesting they play an important role in processing sound dynamics. Yet our present results indicate that levels of depression do not correlate with the SGN subtype. Type 1a SGNs are expected to have higher levels of spontaneous and stimulus-evoked activity in vivo than type 1c SGNs, but our evidence does not support that type 1a synapses have lower depression than type 1c synapses. Therefore, the high variability in endbulb synaptic depression previously found with normal acoustic experience is not likely to be related to the SGN subtype.
The diversity of intensity sensitivity observed among SGNs is often thought of as a way of apportioning the wide range of intensities the auditory system has to deal with. The different sensitivity classes of SGNs have long been considered an adaptation for breaking down the enormous range of intensities into more limited dynamic ranges. However, we observed that bushy cells received a mix of 1a, 1b, and 1c inputs, and there was no particular differentiation between bushy cells, for example, where some bushy cells received only subtype 1a synapses, while others received only 1b or 1c. Indeed, the synaptic volumes around bushy cells in Figure 5 appeared similar to the reported proportions of SGN somata of the different subtypes (Petitpré et al., 2018; Shrestha et al., 2018; Sun et al., 2018). This contrasts with the preservation of the tonotopic map into the AVCN, where bushy and T-stellate cells receive inputs from a limited frequency region of the cochlea. Indeed, the auditory system preserves tonotopy at all stages from the cochlea to the cortex, reflecting the importance of frequency coding in sound analysis. Intensity information does not appear to be treated in the same way. Rather, the different intensity classes mix at their targets in the AVCN, suggesting that sound intensity plays a less central role in sound processing. One reason could be that communication signals or environmental sounds can arrive at the ear at extremely wide-ranging levels, depending on the distance to the origin and intervening obstacles, and in the presence of masking noise. Therefore, processing must rely on integrating spectral and amplitude fluctuations over time, rather than absolute values. Maintaining separate streams of intensity processing may not optimally support such analysis.
The lack of segregation of SGN subtypes may reflect the special function of units in the AVCN. That is, the mixing of 1a and 1c SGN synapses onto the same bushy cells may optimally support the function of bushy cells, which are commonly understood to relay highly precise temporal information to brain areas involved in sound localization. Bushy cell responses in vivo are clearly influenced by sound intensity (Gibson et al., 1985; Joris et al., 1994), but the encoding of temporal characteristics, such as sound onset or phase-locking, may benefit from the integration of multiple SGN inputs regardless of their level sensitivity. One could evaluate this possibility experimentally by functionally silencing SGN subtypes using opto- or chemogenetic manipulation and assessing changes in the temporal fidelity of bushy cells.
We observed high convergence from subtypes 1a, 1b, and 1c onto bushy cells (Fig. 5). All bushy cells we analyzed received endbulbs from at least two subtypes, and most (8/11 = 73%) received all three. However, optogenetic stimulation in either LYPD1 or CALB2 mice drove spiking in ∼45% of units. This difference may arise from nonuniformity in synaptic strength. It is possible that small-volume endbulbs produce EPSPs too weak to trigger postsynaptic action potentials after optogenetic stimulation, and we would expect such SGNs to have little influence over the activity of their postsynaptic bushy cell, including their intensity sensitivity. It will be important to evaluate the diversity of endbulbs further, to understand if a subset dominates bushy cell output, and how that might change with acoustic experience and aging.
Unlike bushy cells, the responses of chopper units do not preserve the temporal fine structure of sounds but do encode spectral cues and sound envelope, so the lack of difference in threshold or spontaneous rate between choppers that receive 1a or 1c input seems unexpected. We predict that anatomical study of the synapses onto T-stellate cells would show convergence of SGN subtypes similar to what we observed with endbulbs in Figure 5. However, addressing that question may be difficult because of the complex dendritic morphology of T-stellate cells and lack of specific cell markers. It will also be valuable to compare synaptic targets of SGN subtypes outside of the AVCN core.
It is also possible that the strong convergence of SGN subtypes we observed is an adaptation that specifically supports mouse hearing. For mice, which have poor spatial acuity to broadband and narrowband sounds (Lauer et al., 2011), localizing sound sources may be less critical than sound detection and near-field discrimination. Such tasks may not require segregated streams for processing sound amplitude. It may instead be advantageous to sum activity from SGNs with different threshold sensitivity to ensure detection. SGNs of mice vary in level sensitivity (Taberner and Liberman, 2005; Suthakar and Liberman, 2021), but do not show the clear category boundaries that occur in some other species (Liberman, 1978; Huet et al., 2016). It would be valuable to examine how SGN subtypes sort onto their targets in other species that specialize in directional hearing.
Footnotes
We thank James Engel, Jr, and Kimberly Nguyen for establishing the animal colony, Wade Sigurdson for his help with imaging, and Alexander Gennaro for his support and help throughout the project. This work was supported by National Institutes of Health Grants R01 DC015508 (M.A.X.-F.) and R01 DC019514 (U.M.) and the David M. Rubenstein Fund for Hearing Research (A.M.L., U.M.).
The authors declare no competing financial interest.
- Correspondence should be addressed to Matthew A. Xu-Friedman at mx{at}buffalo.edu.