Introduction

Perception depends on the ability of neurons to encode discrete features of complex external stimuli. These computations are determined by the type of information received from the periphery, the nature and position of the synapses, and the biophysical properties of the target neuron. For instance, in the auditory system, spiral ganglion neurons (SGNs) encode sound information from hair cells in the cochlea and distribute it to specialized target neurons in the cochlear nucleus complex (CNC) that extract individual elements of the original sound, such as frequency, phase, intensity, and timing. The outcome of each dedicated computation continues to be processed through parallel ascending pathways and ultimately to the cortex, where the auditory features are reassembled to generate a percept. Thus, understanding how these first computations are made is a key step towards deciphering the basis of perception.

To generate accurate percepts of complex auditory stimuli, neurons must compute both what frequencies are present and when those frequencies occur across multiple time scales. For instance, overlapping sound stimuli, such as two competing speakers in a noisy room, are perceptually distinguished by correctly binding together frequencies with coherent onsets that then continue to change together over time^1,2. Such computations require coincidence detection that can accurately encode co-occurring frequencies with submillisecond precision. Frequency information is communicated by SGNs, whose auditory nerve fibers (ANFs) project through the eighth nerve, bifurcate, and spread tonotopically to fill each division of the CNC (Fig. 1A). In addition, SGNs fall into physiologically distinct subtypes that are recruited at different intensities, thereby allowing sounds to be detected across a wide dynamic range and in the presence of background noise. Many target neurons receive SGN inputs from a limited range of frequencies and fire at the onset of the sound, effectively breaking it up into its frequency components. This presents a challenge for perception as auditory circuits must ultimately bind together co-occurring frequencies while also retaining information about their sequence to locate and recognize sounds.

Figure 1.

In the mammalian auditory system, precise encoding of broadband timing information begins with the octopus cells of the CNC. Octopus cells are excitatory neurons that bind together co-occurring frequency information on a submillisecond timescale and send this information along one of the parallel ascending pathways in the auditory brainstem. Octopus cells are named for their large-diameter tentacle-like dendrites^3,4, which are oriented unidirectionally across a tonotopic array of SGNs such that each neuron integrates inputs from a wide range of frequencies^5–8. SGNs provide the major excitatory inputs onto octopus cells. Biophysically, octopus cells have low input resistances near rest (∼4MΩ), fast time constants (∼200µs), and large low-voltage-activated potassium (∼40nS at rest) and hyperpolarization activated (∼62nS at rest) conductances. Together these properties give these cells impressively narrow windows of coincidence detection on the order of 1 millisecond^9–16. This combination of receiving SGN innervation across broad frequencies and their biophysical specializations establish octopus cells as spectrotemporal coincidence detectors that can reliably encode the timing of complex stimuli, such as the broadband transients found in speech and other natural sounds^13,17,18. Fittingly, in vivo recordings from octopus cells demonstrate their ability to phase lock to broadband transients at rates up to 1kHz^19–21. Moreover, computational models of octopus cells demonstrate that onset responses are governed by the cell’s biophysical specializations and are, in large part, the result of temporal summation of excitation^22–28. The simplicity of its connectivity combined with the precision of its temporal computations makes the octopus cell an attractive model for understanding how specialized anatomical and electrophysiological properties contribute to neuronal computations.

Although the octopus cell’s integration of SGN inputs within a very narrow time frame enables canonical coincidence detection, such a model does not explain how other temporal features of sound stimuli are encoded. Indeed, octopus cells encode spectrotemporal sequences within their broadly-tuned response areas, like frequency modulated sounds, that likely require further circuit specializations²⁹. Although threshold somatic depolarization can be sufficient to activate an octopus cell²³, the vast majority of synapses are found on dendrites. Further, SGN inputs are organized tonotopically along octopus cell dendrites, with inputs from high frequency regions located more distal than those from low frequency regions. Dendritic morphology, passive cable properties, active resting membrane properties, and the spatial and temporal relationship between synaptic inputs can all impact excitatory post synaptic potential (EPSP) summation as excitation sweeps across the dendritic arbor and towards the soma. This raises the possibility that computations made in the dendrites influence the effective window of coincidence detection by the octopus cell. Such mechanisms could enable flexible processing that is adaptive to the dynamics of the environment without compromising the fidelity of high-fidelity coincidence computations.

Here, we sought to define the circuit mechanisms that allow octopus cells to act as coincidence detectors across time scales. We generated a comprehensive anatomical and physiological map of excitatory and synaptic inputs onto octopus cell somas and dendrites and examined how this circuit organization influences octopus cell activation. Through a combination of in vitro experiments and computational modeling, we show that the somatic summation of excitation is shaped by dendritic inhibition. Thus, octopus cells depend on compartmentalized computations that enable preservation of timing information both at the moment of stimulus onset and within an extended window for evidence accumulation, which is fundamental for the spectrotemporal integration of natural auditory stimuli.

Results

The balance of excitatory and inhibitory synapses is different in somatic and dendritic domains

To determine the wiring pattern that drives octopus cell computations, we generated a detailed map of excitatory and inhibitory synaptic inputs (Fig. 1). Overall, octopus cells receive abundant excitatory VGLUT1+ innervation from SGNs^30,31 and sparse inhibitory innervation from glycinergic neurons, as visualized using the glycinergic Cre driver Glyt2^Cre and the Ai34 synaptophysin-tdTomato (syp/tdT) fusion protein reporter (Fig. 1B). In addition, sparse inhibitory inputs nestle between SGN inputs, especially on octopus cell dendrites (Fig. 1C).

Quantification of the number and distribution of presynaptic puncta onto octopus cells revealed marked differences in the ratio of excitation and inhibition in the somatic and dendritic compartments. Since innervation patterns have never been systematically analyzed, we made three-dimensional reconstructions of 16 octopus cells and their excitatory SGN (n = 8 cells, 4 mice) and inhibitory (n = 8 cells, 3 mice) inputs. Octopus cells were visualized using a Thy1 reporter and presynaptic puncta were labeled with the syp/tdT reporter driven either by Foxg1^Cre (Fig. 1D) or Glyt2^Cre (Fig. 1E). Consistent with qualitative assessment, the density of SGN inputs was higher (10.7 ± 3.0 SGN puncta/100µm²) than that of inhibitory inputs (4.2 ± 0.8 puncta/100µm², Fig. 1F). Moreover, the relative proportions of excitatory and inhibitory inputs differed in the soma and dendrites (Fig. 1G). On somas, SGNs provided dense innervation that continued on the proximal dendrite, then gradually declined with distance from the soma. By contrast, somas received very few inhibitory inputs. On dendrites, inhibitory puncta were evenly distributed. As a result, octopus cells have a strikingly different average ratio of excitatory and inhibitory puncta on the soma (7:1) and on the dendrite (5:2), suggesting that each compartment contributes differentially to the final computation made by the octopus cell (Fig. 1H).

The majority of excitatory synapses on octopus cells are from type Ia SGNs

Although uniformly glutamatergic, SGNs exhibit stereotyped physiological differences in response thresholds that could affect the nature of their inputs onto octopus cells and influence octopus cell coincidence detection^{32–35,35,36}. It was recently discovered that there are three molecularly distinct SGN subtypes, referred to as Ia, Ib, and Ic SGNs, which correlate with previously shown physiological groups^37–43 (Fig. 2A). Therefore, we further categorized excitatory inputs based on SGN subtype identity. These can be identified with the presence of Ntng1-dependent reporter expression⁴⁴ in Ib and Ic SGNs (Ib/c) and its absence in Ia SGNs^40,41,43, coupled with very low to undetectable levels of calretinin (CR-) in Ic SGNs, and moderate to high levels of calretinin (CR+, CR++) in Ib and Ia SGNs (Fig. 2B). Ntng1^Cre-labeled Ib/c SGNs accounted for 60.1 ± 2.6% of the entire population, with 28.5 ± 12.2% Ib SGNs, 31.6% Ic SGNs, and 39.9 ± 2.6% Ia SGNs (Fig. 2C: n = 1599 neurons, 4 mice; mean ± SD). Further, proportions of SGN subtypes matched scRNA-seq estimates (Fig. 2C), indicating that this approach provides full coverage. SGN subtype identity was further confirmed by examining the spatial organization of SGN peripheral processes en route to the IHCs in the cochlea (Fig 2A, Supp. Fig. 1A-C).

Figure 2.

Within the ventral cochlear nucleus (VCN), where octopus cells reside, Ntng1^Cre labeling was restricted to SGNs (Supp. Fig. 1D). In the VCN, expression of Ntng1-tdT and CR in SGN central axons was consistent with the moderate to undetectable levels of CR in Ntng1-tdT SGN somas in the periphery (Supp. Fig. 2E). Thus, Ntng1^Cre-driven expression of syp/tdT is an appropriate tool for mapping subtype-specific connectivity onto octopus cells.

Reconstruction of Ntng1^Cre-labeled Ib/c puncta (Fig. 2D) demonstrated that octopus cells are dominated by inputs from Ia SGN fibers, which are the fibers with the lowest response thresholds and highest rates of spontaneous activity. Octopus cell dendrites received 4.1 ± 1.0 puncta/100µm² from Ib/c SGNs (Fig. 2E, magenta: n = 9 cells, 5 mice; mean ± SD), accounting for 38% of the total SGN density. Given that Ntng1-tdT+ cells account for 60.1% of the SGN population (Fig. 2C, magenta), Ib/c inputs were underrepresented on octopus cells. Octopus cells receive similarly low innervation from Ic inputs (Supp. Fig. 2G-I: n = 6 cells, 2 mice), as estimated from the degree of sparse labeling of Ic axons achieved by Myo15^iCre reporter expression (Supp. Fig. 2A-F) and the expected proportion of Ic SGNs in the ganglion (Supp. Fig. 2F). By contrast, Ia SGNs, which comprise only ∼40% of the total population (Fig. 2C, yellow), accounted for 62 ± 9.7% of SGN synapses on octopus cells (Fig. 2E, yellow: 6.6 ± 1.0 puncta/100µm²). All three subtypes showed the same overall distribution from the soma to the distal dendrite (Fig. 2F). Together, excitatory and inhibitory puncta densities in the innervation maps indicate the average octopus cell receives ∼1035 SGN synapses (642 Ia SGN, 393 Ib/c SGN) and ∼354 inhibitory synapses. Additionally, the majority of synapses on the average octopus cell (83%) are found on dendrites, highlighting their critical role in the octopus cell computation.

Octopus cell reconstructions showed the same basic wiring patterns regardless of where each cell was positioned in the octopus cell area. The tonotopic position of all reconstructed octopus cell somas was estimated in 3D reconstructions aligned to a normalized CNC model of tonotopy⁴⁵. Octopus cells had similar morphologies (Supp. Fig. 3E-G) and patterns of synaptic innervation (Supp. Fig. 3H-M) regardless of where they were positioned along the tonotopic axis. Thus, we have established a wiring diagram that finds low threshold, Ia SGN synapses to be the primary input to both the soma and dendrites of octopus cells. Additionally, the whole-neuron wiring diagram identifies a dendritic domain where inhibitory synapses are approximately equal in number to the Ib/c excitatory synapses from the periphery.

Figure 3.

SGN inputs to octopus cells facilitate at high stimulation frequencies

Whether or not an octopus cell responds to its inputs depends on when and how EPSPs travel to and then summate in the soma. To determine if SGN subtypes transmit information differently to their central targets, we performed in vitro whole-cell current clamp recordings of octopus cells (Fig. 3A) while using Channelrhodopsin-2 (ChR2) to stimulate either all SGNs (Foxg1-ChR2) or only Ib/c SGNs (Ntng1-ChR2). Trains of ChR2-evoked SGN stimulation, ranging from 5 to 50Hz, in both the total SGN population (Fig. 3B, black: n = 8 cells, 5 mice) and the Ib/c SGN population (Fig. 3B, magenta: n = 7 cells, 6 mice) exhibited no differences in paired-pulse plasticity at any frequency of stimulation (p > 0.35 at all interstimulus intervals, Tukey’s HSD), although the Ib/c population exhibited higher variability than the total SGN population (at 20ms: SD = 0.11, SD = 0.24, respectively).

ChR2-evoked synaptic responses are known to artificially undergo synaptic depression^46,47. To determine if the paired pulse depression measured in ChR2-stimulated experiments was physiological (Fig. 3B), we used electrical stimulation to evoke EPSPs from SGNs. Electrically-evoked SGN EPSPs had higher paired-pulse ratios than ChR2-evoked EPSPs and were mildly facilitating at short (20ms) intervals (Fig. 3B-C, open circles: n = 5 cells, 3 mice), consistent with an octopus cell’s ability to respond reliably to click trains in vivo^5,7,20,48. In contrast, previous results using electrical stimulation demonstrated short-term depression of SGN inputs to octopus cells^49,50. However, these experiments were carried out in the presence of higher, non-physiological levels of extracellular calcium. We repeated paired pulse plasticity experiments with non-physiological calcium concentrations (2.4mM) and similarly found that electrically-evoked EP-SPs from SGNs resulted in short-term depression at 50Hz of electrical stimulation (Fig. 3C, grey: n = 3 cells, 2 mice), though not to the degree observed when using full-field, ChR2-evoked inputs.

Glycine evokes inhibitory post synaptic potentials that are occluded by a low input resistance

Given the increased density of inhibitory synapses on octopus cell dendrites, we considered the possibility that somatic and dendritic compartments contribute differently to the temporal computation made by octopus cells. A role for inhibition has never been incorporated into octopus cell models as previous efforts failed to reveal physiological evidence of functional inhibitory synapses onto octopus cells either in vitro^12,51,52 or in vivo²⁹. Similarly, we did not observe light-evoked (Glyt2-ChR2) inhibitory post synaptic potentials (IPSPs) in octopus cell somas during whole-cell current clamp recordings from P30-45 mice. Since inhibitory synapses are located primarily on octopus cell dendrites, we posited that their voltage spread to the soma is limited given the extremely low input resistance of octopus cell somas.

To decrease electrotonic isolation of the dendrites and increase input resistance, we pharmacologically blocked voltagegated potassium (K_v) and hyperpolarization-activated cyclic nucleotide-gated (HCN) channels using 100µM 4-Aminopyridine (4-AP) and 50µM ZD 7288 (ZD). This cocktail increased octopus cell membrane resistance (Fig. 4A). To isolate inhibition, 15µM 2,3-dioxo-6-nitro-7-sulfamoyl-benzo[f]quinoxaline (NBQX) was added to block AMPA receptor activation. Consistent with our hypothesis, the increase in input resistance unveiled light-evoked IPSPs in recordings from octopus cell somas (Fig. 4B). Bath application of 500nM strychnine (STN) fully abolished IPSPs (Fig. 4C), confirming functional glycinergic inhibitory synaptic transmission onto octopus cells.

Figure 4.

To determine the types of glycinergic receptors contributing to IPSPs, we pharmacologically blocked subsets of glycine receptors (Fig. 4C). IPSPs were reduced upon addition of 20µM picrotoxin (PTX), which blocks homomeric glycine receptors^53–55. Sequential addition of 100µM cyclothiazide (CTZ), which blocks α2-containing homomeric and heteromeric glycine receptors^56,57, nearly abolished the remaining IPSPs, and responses were fully abolished with further application of 500nM STN. Collectively, these data demonstrate that glycinergic synaptic contacts onto octopus cell dendrites are functional. Additionally, glycine receptor subunit composition (Fig. 4C) implicates a role for both large conductance extrasynaptic β-subunit lacking homomeric receptors and synaptically localized α2β receptors with slower kinetics^58–60.

To confirm whether the confinement of IPSPs to the dendrites is consistent with our understanding of octopus cell biophysics, we developed an improved biophysically and anatomically accurate model of octopus cells based on our findings (Supp. Fig. 4)^18,61. This model performed as predicted based on our experimental results. As in our current-clamp recordings (Fig. 4A), removal of K^v and HCN conductances in the model changed the input resistance and current-voltage relationship of the neuron, resulting in reduced electronic isolation (Fig. 4D, blue). In control conditions, stimulation of dendritic glycinergic conductances induced negligible hyperpolarizing voltage changes (Fig. 4E, black). With increased input resistance, dendritic IPSPs measured at the soma were similar to in vitro recordings (Fig. 4E-F). Thus, this model confirms that dendritic IPSPs can elicit somatic hyperpolarization in octopus cells when electrotonic dendritic isolation is reduced.

While blocking K_v and HCN allowed us to reveal IPSPs at the soma, 4-AP increases the duration of the already unphysiological ChR2-evoked presynaptic action potential⁴⁷, resulting in altered release probabilities and synaptic properties, amongst other caveats⁶². To directly confirm that the increase in somatically-measured IPSP amplitude can be explained by changes in input resistance alone, we used the octopus cell model to simulate dendritic glycinergic conductances and measure changes in dendritic current and somatic potential amplitude in the presence of blocked K_v and HCN channels. K_v and HCN block and the resulting change in input resistance increased the magnitude of soma-measured IPSPs for all glycine conductances (Fig. 4G, blue). Dendrite-measured currents did not change as a result of the hyperpolarization induced by K_v and HCN block regardless of synaptic location on the dendrites or the magnitude of the glycine conductance (Fig 4G, dark orange, light orange). Together, results from the model replicate those collected in vitro and provide evidence of functional glycinergic synaptic transmission that is difficult to detect with in vitro somatic recordings.

Inhibition decreases the magnitude and advances the timing of dendritic SGN inputs

SGN synapses onto octopus cell dendrites are arranged tonotopically, with higher frequency SGNs from the base of the cochlea terminating on the distal dendrites and lower frequency SGNs from more apical positions terminating more proximally (Fig. 1A). This organization has been proposed to re-synchronize coincidentally firing SGNs that are activated at slightly different times due to the time it takes for the sound stimulus to travel from the base to the apex of the cochlea¹⁸. To test how inhibition in the dendrites shapes coincidence detection, we first used our model to explore the influence of simultaneous activation of inhibitory and excitatory synapses at varying locations along the dendritic tree. By placing inhibitory synapses on proximal or distal dendrites and moving the relative location of excitation, we modelled the effect of on-path and off-path inhibition⁶³ on somatically recorded EPSPs (Fig. 5A, Supp. Fig. 5). For both inhibition proximal to excitation (Fig. 5B-C: on-path) and inhibition distal to excitation (Fig. 5D-E: off-path), the model predicted that inhibition reduces EPSP amplitude and accelerates EPSP peak timing at the soma. Thus, the presence of inhibition appears to enable modulation of EPSP timing in dendritic compartments during continuous auditory stimuli when inhibition can be recruited beyond the onset of a sound and thus allow for adaptable temporal processing during the duration of a stimulus.

Figure 5.

To directly test if the model’s prediction that temporally coincident excitation and inhibition affects the timing and amplitude of excitatory SGN inputs as they travel towards the octopus cell soma, we coincidently activated SGNs and glycinergic inputs in vitro. In these experiments, the octopus cell properties were not altered pharmacologically and inhibition was undetectable or only visible with averaging over many sweeps (Fig. 5F, blue). When synaptic inhibition was evoked together with excitation (Fig. 5F, green), the soma-recorded EPSPs were smaller than when excitation was evoked alone (Fig. 5F, black: n = 8 cells, 6 mice). ChR2-evoked inhibition decreased EPSP heights by 25.2 ± 9.0% (Fig. 5G, green) and shifted the peak of EPSPs forward 57.5 ± 26µs (Fig. 5H, green). This effect was mimicked by bath application of 25µM glycine (Fig. 5G-H, blue: n = 4 cells, 3 mice). Further, bath application of 1µM STN had the opposite effect, resulting in larger EPSPs, delayed peak times, and increased half-widths (Fig. 5F-H, orange: n = 5 cells, 4 mice). Thus, the timing of EPSP arrival may be shaped both by the release of synaptic glycine and by tonically active glycine channels. Of note, many SGNs also terminate on the octopus cell soma, where inhibition is minimal. This suggests that the octopus cell’s ability to act as a coincidence detector depends on two stages of compartmentalized computations, one in the dendrite that combines excitation and inhibition to provide important information about which frequencies co-occur in a complex sound stimulus and one in the soma that is restricted by the rigid temporal summation window for coincidence detection. Together with the electrotonic properties of the octopus cell and the dominance of low threshold, low jitter Ia SGN inputs, these combined computations can enable reliable coincidence detection and proper cross-frequency binding needed for perception of sound.

Discussion

Coincidence detection plays a critical role in many cognitive and perceptual processes, from the ability to localize sound, to the binding of auditory and visual features of a common stimulus. Depending on the computation, the temporal window for integration can range widely, thereby requiring circuitry with distinct anatomical and physiological properties. Here, we describe a two-domain mechanism for coincidence detection that can detect co-occurring frequencies with different degrees of precision. By mapping and selectively activating synaptic inputs onto octopus cells both in vitro and in a computational model, we revealed that compartmentalized dendritic nonlinearities impact the temporal integration window under which somatic coincidence detection computations are made. The arrival of many small, reliable excitatory inputs (Fig. 3) from low-threshold SGNs (Fig. 2) is continuous throughout an ongoing stimulus. We demonstrate that glycinergic inhibition to octopus cell dendrites (Fig. 1) can shift the magnitude and timing of SGN EPSPs as they summate in the soma (Fig. 4,5). The narrow window for coincidence detection computations allows the octopus cell to respond with temporal precision using momentary evidence provided by SGNs at the onset of the stimulus. We propose that, as a stimulus persists, inhibition onto octopus cell dendrites can adjust the timing of excitation before arriving at the soma for the final input-output computation. This allows the cell to make an additional computation with a slightly longer window for evidence accumulation without compromising the accuracy of the somatic onset computation (Fig. 6).

Figure 6.

As coincidence detectors in the auditory system, octopus cells are faced with the challenge of recognizing complex sounds that include many frequencies that co-occur from the beginning to the end of the stimulus. As shown by in vivo recordings^6,29, octopus cells respond well to cues that include complex spectrotemporal patterns, including frequency modulations beyond the onset of the stimulus²⁹. Given that the auditory environment is filled with overlapping sound stimuli, such responses presumably allow the octopus cell to encode which frequencies belong to which sound. Our data thus support the role of octopus cells beyond simple onset coincidence detectors that rely solely on the temporal summation of excitation. The results suggest that, despite high K_v and HCN conductances at rest, the addition of dendritic inhibition transforms the magnitude and timing of SGN signals as they arrive in the cell body, which may expand their response selectivity and allow them to become a slightly leakier integrator and thus accumulate evidence beyond onsets. Although, this inhibition is difficult to detect because of shunting, our data demonstrate that it is both present and impactful.

As well as needing to work beyond onsets, an effective coincidence detector in the auditory system must also function reliably across a range of sound intensities. Intensity information is encoded by the number and types of SGNs that are activated in the cochlea. The Ia, Ib, and Ic molecular subtypes defined in mouse^40,41,43 broadly correspond to the anatomically and physiologically defined subtypes described across species^42,64. We find that the majority of inputs onto octopus cells come from Ia SGNs, which most closely correspond to the low-threshold, high-spontaneous rate population. Consistent with this result, single-unit SGN recordings in cats demonstrated a bias towards low-threshold, high-spontaneous rate axon collaterals in the octopus cell area³³. Low-threshold SGNs are also characterized by short first spike latencies and low temporal jitter^65–68. A hallmark of the octopus cell is the fact that it only fires action potentials when many SGN inputs are activated within a narrow period of time⁶⁹. The presence of many low-threshold and temporally precise inputs on the octopus cell may help ensure that coincidence detection still works reliably for quiet sounds. Further, Ia inputs onto octopus cells do not exhibit paired-pulse depression, not unlike low levels of depression seen in Ia inputs to bushy cells⁷⁰. The presence of SGN inputs without paired-pulse depression could be beneficial for encoding sustained auditory signals. Finally, although Ia SGNs are over-represented, Ib and Ic inputs are also present. Since precise, low-threshold SGN responses can be saturated by background noise such that responses to relevant stimuli are masked^38,65,68,71, recruitment of higher threshold SGNs at higher sound intensities may compensate for this tradeoff.

The presence of inhibitory inputs onto dendrites is a fundamental feature of the nervous system and, in other systems, contributes to a neuron’s final computation. For example, direction selectivity computations in dendrites of retinal cells require excitation-inhibition interactions in dendritic compartments^72,73. In pyramidal cells of the cortex and hippocampus, the spatial distribution of inhibition impacts dendritic non-linearities in a branch selective manner^74–79. However, octopus cells do not share all mechanisms for dendritic computation as neurons in the cortex and hippocampus. The lack of backpropagating action potentials and dendritic calcium spikes in octopus cells causes submillisecond timing differences in dendritic summation of coincident excitation and local inhibition to be consequential in subthreshold computation before somatic action potential generation.

Although this work uncovers a role for inhibition, understanding of octopus cell computations is limited by the fact that it remains unclear what kind of information is carried by inhibitory inputs. Although the presence of presynaptic glycinergic puncta in the octopus cell area^80–83 and glycinergic receptor expression in octopus cells^84–88 is well established, it is unknown where this glycinergic innervation originates. Of the local neurons within the CNC that provide inhibition to the VCN, there is no evidence of connections to octopus cells from D-Stellate⁵², L-Stellate⁸⁹, or tuberculoventral cells⁹⁰. Outside of the CNC, terminal degeneration experiments in cats suggested the superior periolivary nucleus (SPON) and the ventral division of the lateral lemniscus (VNLL) as potential sources of descending inhibition to the octopus cell area⁹¹. Octopus cells provide excitatory input to both the SPON^92–96 and the VNLL^20,97–101, raising the possibility of feedback inhibition from the auditory brainstem as a circuit mechanism for elongated temporal summation windows during ongoing stimuli. Although feedback inhibition is not rapid enough to prevent or alter the onset response that octopus cells are well-known for, it could limit the duration of a response or change the effective coincidence detection window as the stimulus continues. Future studies will be required to identify the source of inhibition and its organization along the dendrites. If inhibitory inputs tonotopically match the local, narrowly-tuned dendritic SGN inputs, it is possible that frequency matched inhibition could influence spectral selectivity or feature extraction. On the other hand, broadly tuned inhibition could reduce depolarization block or serve as a temporal milestone that signals gaps or offsets. Further characterization of in vivo octopus cell responses in complex sound environments may clarify the effect of noise on signal detection and could reveal additional features of this cell’s contributions to perception of the auditory world.

Acknowledgements

We thank Dr. Nace Golding (University of Texas at Austin), Dr. Matthew McGinley (Baylor College of Medicine), Dr. Phillip Joris (KU Leuven), and Dr. Bernardo Sabatini for helpful discussions and feedback. Sadie Quinn, Lucy Lee, and Ryan Merrow provided valuable technical assistance. Dr. Bruce Bean (Harvard Medical School) generously provided access to electrophysiological equipment. Slc6a5^{tm1.1(Cre)Ksak} mice were kindly provided by Dr. Wade Regehr (Harvard Medical School). Ntng1^em1(Cre)Kfra mice were made and kindly provided by Dr. Fan Wang (Massachusetts Institute of Technology). Myo15a^{tm1.1(Cre)Ugds} mice were kindly provided by Dr. Stefan Heller (Stanford). We thank Rigoberto Ramirez, Tenzin Paljorwa, and Edgar Ramirez for animal care support. We are grateful to the Neurobiology Imaging Facility (NIF) for software availability and to the HMS Research Instrumentation Core for the design and fabrication of temperature regulation equipment. This work was supported by grants from the BRAIN Initiative 1R01NS118402 to L.V.G., the National Institute on Deafness and Other Communication Disorders 5R01DC009223 to L.V.G. and 1F32DC020070 to L.J.K, the William Randolph Hearst Fund to L.J.K, and the Broad Institute’s Stanley Center for Psychiatric Research to S.H. and G.F.

Author Contributions

Conceptualization, L.J.K. and L.V.G.; Methodology, S.H.; Formal Analysis, L.J.K and S.H.; Investigation, L.J.K, S.M., and S.S.; Resources, G.F.; Writing, L.J.K. and L.V.G.; Visualization, L.J.K. and S.H.; Supervision, G.F. and L.V.G.; Funding Acquisition, L.J.K, G.F., and L.V.G. Declaration of Interests

Declaration of Interests

The authors declare no competing interests.

Materials and Methods

Animal Use and Transgenic Mouse Lines

All procedures were approved by and conducted in accordance with Harvard Medical School Institutional Animal Care and Use Committee. Male and female mice (Mus musculus) were bred on a C57BL/6 background at the Harvard Center for Comparative Medicine or obtained from Jackson Laboratories (Bar Harbor, ME). Mice were housed in groups of up to five animals and maintained on a 12hr light/dark cycle. Transgenic alleles were heterozygous for each transgene for all experimental animals. Descriptions of allele combinations for all experiments can be found in Supplemental Table 1.

Spiral ganglion neurons (SGNs) and their central auditory nerve fibers (ANFs) were targeted using either Foxg1^{tm1.1(Cre)Ddmo} (Foxg1^Cre)¹⁰² or Foxg1^Flp, both of which drive robust reporter expression neurons in the auditory and vestibular ganglion^103,104 and the neocortex^105,106, but not in brainstem or midbrain neurons. Foxg1^Flp mice were generated by crossing the Foxg1^tm1.1Fsh mouse line¹⁰⁷ with the Tg(EIIa-Cre)C5379Lmgd mouse line¹⁰⁸, then backcrossing to isolate the flp transgene and remove the Cre transgene.

Inhibitory inputs to octopus cells were targeted with Slc6a5^{tm1.1(Cre)Ksak} mice (Glyt2^Cre)¹⁰⁹.

Octopus cells were sparsely labeled with the Tg(Thy1-YFP)HJrs (Thy1) mouse line¹¹⁰. This line labels ∼0-15 octopus cells amongst other neurons throughout the brain.

Ib/c SGNs were targeted using the Ntng1^em1(Cre)Kfra (Ntng1^Cre) mouse line, which drives expression in neurons throughout the nervous system (Supp. Fig. 1F) and disrupts expression of the endogenous allele⁴⁴. Auditory brainstem responses in adult Ntng1^Cre/+ mice are normal. Ic SGNs were sparsely targeted with the Myo15a^{tm1.1(Cre)Ugds} (Myo15^iCre) mouse line¹¹¹.

Fluorescent reporters included Gt(ROSA)26Sor^{tm14(CAG-tdTomato)} (Ai14, tdT)¹¹², Gt(Rosa)26Sor^{tm34.1(CAG-Syp/tdTomato)} (Ai34, syp/tdT), and Gt(Rosa)26Sor^{tm1.2(CAG-EGFP)Fsh} (RCE:FRT, EYFP)¹¹³. We also used Gt(Rosa)26Sor^{tm32(CAG-COP4*H134R.EYFP)} (Ai32, ChR2)¹¹⁴ to drive synaptic activity in in vitro slice experiments.

Histology and Reconstructions

For immunohistochemical labeling, mice were deeply anesthetized with isoflurane and transcardially perfused with 15mL of 4% paraformaldehyde (PFA) in 0.1M phosphate-buffered saline (PBS) using a peristaltic pump (Gilson). Whole skulls containing brain and cochlea were immediately transferred to 20mL of 4% PFA and post-fixed overnight at 4°C. Fixed brains and cochlea were removed from the skulls and washed with 0.1M PBS.

Brains were collected from mice of both sexes, aged 28-38 days, and embedded in gelatin-albumin hardened with 5% glutaraldehyde and 37% PFA¹¹⁵. Sections were cut at 35, 65, or 100μm with a vibrating microtome (Leica VT1000S) and free-floating tissue was collected in 0.1M PBS. For sections less than 65μm, tissue was permeabilized and nonspecific staining was blocked in a solution of 0.2% Triton X-100 and 5% normal donkey serum (NDS, RRID: AB_2337258) in 0.1M PBS for 1 hour. After blocking, tissue was treated with primary antibody in a solution containing 0.2% Triton X-100 and 5% NDS in PBS for 1-2 nights at room temperature. Primary antibodies used were: chicken anti-GFP (1:1000, RRID:AB_10000240), rabbit anti-RFP (1:1000, RRID:AB_2209751), goat anti-calretinin (1:1000, RRID:AB_1000034), and guinea pig anti-VGLUT1 (1:500, RRID:AB_887878). Sections were washed in 0.1M PBS then incubated in a secondary antibody solution (1:1000) containing 0.2% Triton X-100 and 5% NDS for 2-3hrs at room temperature. Tissue sections were mounted on charged slides and coverslipped (Vectashield Hardset Antifade Mounting Medium with DAPI), and imaged using a Zeiss Observer.Z1 confocal microscope.

For 100μm sections, tissue was washed in CUBIC-1A solution for 1hr for strong permeabilization and delipidization^116,117. Tissue was then further permeabilized and nonspecific staining was blocked in a solution of 0.2% Triton X-100 and 5% NDS in 0.1M PBS for 1hr. After blocking, tissue was treated with primary antibody in a solution containing 0.2% Triton X-100 and 5% NDS in PBS for 4 nights at 37°C. Primary antibodies used were: chicken anti-GFP (1:1000, RRID:AB_10000240), and rabbit anti-RFP (1:1000, RRID:AB_2209751). Sections were then incubated in a secondary antibody solution (1:400) containing 0.2% Triton X-100 and 5% NDS for 4 nights at 37°C. Tissue sections were pre-incubated in CUBIC2 solution, then temporarily mounted on uncharged slides with CUBIC2 solution for immediate imaging using a Zeiss Observer.Z1 confocal microscope.

Octopus cells and synaptic puncta were reconstructed in Imaris (Oxford Instruments). YFP signal from the target octopus cell was used to generate a surface reconstruction and mask syp/tdT signal. Dendrites were reconstructed using the masked YFP signal and separated into 10μm increments. Masked syp/tdT puncta were marked and localized to a 10μm increment of the dendritic tree. Synapse counts, dendrite metrics, and masked channels were exported to Excel (Microsoft) for further analysis.

Cochlea were collected from mice of both sexes, aged 28-42 days. The bony labyrinth of the inner ear was decalcified in 0.5M ethylenediamine tetraacetic acid (EDTA) for 3 nights at 4°C and embedded in gelatin-albumin hardened with 5% glutaraldehyde and 37% PFA. Sections were cut at 65μm with a vibrating microtome (Leica VT1000S) and free-floating tissue was collected in 0.1M PBS. Sections were washed in CUBIC-1A solution for 1hr for strong permeabilization and delipidization. Tissue was further permeabilized and nonspecific staining was blocked in a solution of 0.2% Triton X-100 and 5% NDS in 0.1M PBS for 1hr. After blocking, tissue was treated with primary antibody in a solution containing 0.2% Triton X-100 and 5% NDS in PBS for 2 nights at room temperature. Primary antibodies used were: chicken anti-GFP (1:1000, RRID:AB_10000240), rabbit anti-RFP (1:1000, RRID:AB_2209751), goat anti-calretinin (1:1000, RRID:AB_1000034).

Sections were then incubated in a secondary antibody solution (1:500) containing 0.2% Triton X-100 and 5% normal goat serum for 2-3hrs at room temperature. Tissue sections were mounted on charged slides, coverslipped (Vectashield Hardset Antifade Mounting Medium with DAPI), and imaged using a Zeiss Observer.Z1 confocal microscope.

Acute Slice Electrophysiology

Data were obtained from mice of both sexes, aged 24-47 days. Mice were deeply anesthetized with isoflurane and perfused transcardially with 3mL of 35°C artificial cerebral spinal fluid (ACSF; 125mM NaCl, 25mM glucose, 25mM NaHCO₃, 2.5mM KCl, 1.25mM NaH₂PO₄, 1.4mM CaCl₂, and 1.6mM MgSO₄, pH adjusted to 7.45 with NaOH). For high calcium concentration experiments presented in Fig. 3C, ACSF contained 125mM NaCl, 25mM glucose, 25mM NaHCO₃, 2.5mM KCl, 1.25mM NaH₂PO₄, 2.4mM CaCl₂, and 1.3mM MgSO₄. Mice were rapidly decapitated and the brain was removed and immediately submerged in ACSF. Brains were bisected and 250μm slices were prepared in the sagittal plane with a vibrating microtome (Leica VT1200S; Leica Systems). Prepared slices were incubated for 30min at 35°C, then allowed to recover at room temperature for at least 30min. ACSF was continuously bubbled with 95% O₂/5% CO₂.

Whole-cell recordings were conducted at 35°C using a Multiclamp 700B (Molecular Devices) in current-clamp mode with experimenter adjusted and maintained bridge balance and capacitance compensation. Data were filtered at 12kHz, digitized at 83–100kHz, and acquired using pClamp9 (Molecular Devices). Neurons were visualized using infrared Dodt gradient contrast (Zeiss Examiner.D1; Zeiss Axiocam 305 mono). Glass recording electrodes (3–7MΩ) were wrapped in parafilm to reduce capacitance and filled with an intracellular solution containing 115mM K-gluconate, 4.42mM KCl, 0.5mM EGTA, 10mM HEPES, 10mM Na₂Phosphocreatine, 4mM MgATP, 0.3mM NaGTP, and 0.1% biocytin, osmolality adjusted to 300mmol/kg with sucrose, pH adjusted to 7.30 with KOH. All membrane potentials are corrected for a 11mV junction potential.

For optogenetic activation, full-field 470nm blue light was presented through a 20x immersion objective (Zeiss Examiner.D1). Onset, duration, and intensity of light was controlled by a Colibri5 LED Light Source (Zeiss). For electrical stimulation, glass stimulating electrodes were placed in the auditory nerve root and 20µs current pulses were generated with a DS3 current stimulator (Digitimer).

Analysis and Statistical Tests

Cell counts and habenula measurements were performed in ImageJ/FIJI software (National Institutes of Health). Electrophysiology data were analyzed using custom scripts and NeuroMatic analysis routines¹¹⁸ in Igor Pro (Wavemetrics).

For data with equal variance (Levene’s test), one-way ANOVAs with Tukey’s HSD post hoc test were used where appropriate to determine statistical significance. For data with non-homogenous variances, one-way ANOVAs with a Welch F test were used with a Tukey’s HSD post hoc test. Errors and error bars report standard deviation (SD) or standard error of the mean (SEM) as noted in figure legends and throughout the text.

Computational Modelling

Computer simulations were performed using the NEURON 8.2 simulation environment¹¹⁹, with an integration time constant of 25µs. The morphology of the octopus neuron was obtained from McGinley et al., 2012. The active and passive properties of the model were optimized to match the experimental recordings. We set the passive parameters as follows: internal or axial resistance (R_i or R_a) to 150Ω.cm, membrane resistance (R_m) to 5KΩ.cm², capacitance (C_m) to 0.9µF/Cm² and resting membrane potential (V_m) to -65mV. We included the following ion-channel conductances in our morphologically realistic octopus neuron model: Fast Na⁺ (ḡ_NA), low-voltage activated K⁺ (ḡ_KLT), high threshold K⁺ (ḡ_KHT), fast transient K⁺ (ḡ_KA), hyperpolarization-activated cyclic nucleotide-gated HCN (ḡ_h), leak K⁺ (ḡ_leak). The kinetics of all the active ion-channel conductances were obtained from Manis and Campagnola, 2018 and the maximal conductance was optimized to match experimental data. We introduced a scaling factor (scl) to scale the maximal conductance to match the sag and input resistance of the experimental recordings (Supp. Fig. 4). Reversal potentials for HCN, Na⁺ and K⁺ respectively were (in mV), E_h= -38, E_Na= 50 and E_K= -70. Excitatory AMPA synaptic conductance and Inhibitory glycine synaptic conductance were introduced in the proximal and distal dendrites to test the impact of dendritic inhibition on the EPSP height and peak time. The rise and decay time of AMPA and glycine conductance were set to 0.3ms and 3ms respectively, to mimic the fast synaptic transients observed in octopus cells. The reversal potential of AMPA and glycine conductance was set to 0 and -80mV respectively.