Audiotactile interactions in the mouse cochlear nucleus

Multisensory integration of auditory and tactile information occurs already at the level of the cochlear nucleus. Rodents use their whiskers for tactile perception to guide them in their exploration of the world. As nocturnal animals with relatively poor vision, audiotactile interactions are of great importance for this species. Here, the influence of whisker deflections on sound-evoked spiking in the cochlear nucleus was investigated in vivo in anesthetized mice. Multichannel, silicon-probe electrophysiological recordings were obtained from both the dorsal and ventral cochlear nucleus. Whisker deflections evoked an increased spiking activity in fusiform cells of the dorsal cochlear nucleus and t-stellate cells in ventral cochlear nucleus, whereas bushy cells in the ventral cochlear nucleus showed a more variable response. The response to broadband noise stimulation increased in fusiform cells and primary-like bushy cells when the sound stimulation was preceded (~ 20 ms) by whisker stimulation. Multi-sensory integration of auditory and whisker input can thus occur already in this early brainstem nucleus, emphasizing the importance of early integration of auditory and somatosensory information.

In the auditory system, integration of information from the cochlea with information from other sensory modalities begins at the earliest processing stages in the cochlear nucleus [1][2][3][4][5][6] . Anatomical studies demonstrate that several somatosensory structures in the brainstem provide inputs to the cochlear nucleus (CN). These regions include the dorsal column nuclei, consisting of the gracile 7 and cuneate 8,9 nuclei that receive proprioceptive and somatosensory inputs from the lower and upper body, respectively. Projections to the cochlear nucleus also arise from the spinal trigeminal nucleus 10-14 that conveys touch sensation from the face. Projections from these regions form mossy fibre inputs that synapse onto granule cells in the cochlear nucleus granule-cell domain, as well as onto dendrites of ventral cochlear nucleus (VCN) bushy cells 15 and D-stellate cells 16 . The D-stellate cells inhibit fusiform cells in dorsal cochlear nucleus (DCN) [17][18][19] . The parallel fibre axons of granule cells in turn provide excitatory synaptic input to DCN cartwheel and fusiform cells. The presence of somatosensory inputs to a primary auditory nucleus such as the cochlear nucleus is intriguing in the context of the role of the pinnae and neck in generating sound localization cues 16 , as well as the suppression of self-generated signals 20 . Correct interpretation of this information requires integration of auditory signals with somatosensory and proprioceptive signals conveying information about head and pinna position. Studies in cats and guinea pigs 6,[21][22][23] have demonstrated that direct electrical stimulation of brainstem somatosensory nuclei evokes neuronal responses in the DCN and VCN 15,21,24 . These data suggest that activation of CN granule cells by these somatosensory inputs excites fusiform cells and provides feedforward inhibition to fusiform cells through the inhibitory interneurons, the cartwheel cells. The association between sound and whisker stimulation may be a consequence of similar encoding mechanisms: both senses process information that produces mechanical displacements of tissue (i.e., the basilar membrane for auditory and the skin for somatosensory) and are processed in frequency based codes in the cerebral cortex 25 . Rodents use a set of roughly 30 whiskers on each side of the snout, palpating surfaces through a 5-20 Hz forward-backward motion known as "whisking" 26,27 . Whisker-mediated object identification can thus be used as a model to learn more about the mechanisms of multisensory processing and the transformation of this processing to a behavioural output. In the present in vivo electrophysiology study, spiking activity was measured in spike-sorted single units, and audiotactile interaction in the mouse cochlear nucleus was investigated using whisker stimulation in combination with sound. The results suggest that whisker stimulation can modify the sound-evoked spiking activity in the cochlear nucleus.

Whisker-evoked responses in fusiform cells in DCN and bushy cells in VCN.
The whiskers on the mouse whisker pad were deflected (1000 deflections at 5 Hz) using a magnetic stimulation system, and spiking activity was measured in the CN using multi-channel silicon probes. In DCN fusiform cells whisker stimulation evoked increased spiking activity above the spontaneous rate. The increase in spiking started 6.0 ± 1.8 ms after whisker movement. Calculated over the full spike-sorted cell population (n = 45 cells; 8 animals) whisker stimulation increased median spiking from 1.09 spikes/s [IQR: 0.55-3.40] to 4.56 spikes/s with whisker stimulation [IQR: 2.00-6.90] (p < 0.0001, Wilcoxon matched-pairs signed rank test; time window for calculating spikes was 50 ms/sweep; 1000 sweeps were recorded; Fig. 2a, b).
In primary-like bushy cells in VCN the spontaneous activity (n = 49 cells; 5 animals) did not change with whisker stimulation (Fig. 2c, d) It should be noted, however, that although when averaging over the entire sample there was no effect of whisker stimulation, there were individual PL cells where small effects were observed (Fig. 2d). The number of recorded t-stellate cells in VCN was comparatively low (n = 10 cells, from one animal, two different recording sites, 2 shank electrode), but there was an increase in spiking from 0.30 spikes/s [IQR: 0.13-0.57] to 1.20 spikes/s [IQR: 0.45-1.8] when whiskers were deflected (Wilcoxon matched-pairs signed rank, p = 0.0020; IQR: 0.32-1.35) effect size: 0.63 (Fig. 2e, f). This increase could be detected 7.2 ± 2.3 ms after whisker movement.

Whisker stimulation increased the sound-evoked response in DCN fusiform cells and VCN bushy cells.
Stimulation of the ipsilateral whiskers (ipsilateral to recording site in the CN) during acoustic stimulation increased the sound-driven responses of fusiform cells and bushy cells. The magnitude of the effect was dependent on the delay between whisker stimulation and sound stimulation onset. The whisker stimulation onset shifted from 50 ms before sound (+ 50), to 20 ms after onset of sound (−20); the protocols were performed in a random order (Supplementary Table S2).
In fusiform cells, the number of sound-evoked spikes increased for the + 20, + 10, + 5 and − 10 protocol (Fig. 3, Table 1 and Supplementary Fig. S1). The protocols where whisker stimulation preceded sound stimulation increased spiking with approx. 4 spikes/s (27 to 31 spikes/s) (average from the + 20, + 10, + 5 protocols). In the other protocols (+ 50, − 5, − 20) the bimodal stimulation did not significantly change the absolute evoked spikes compared to sound only stimulation. It remains to be investigated if small differences in timing, e.g. between the − 5 and − 10 protocol, relate to non-linear effects on encoding sensory information in the DCN. The bimodal response (calculated as BI; Fig. 6) was on average 9% smaller than the linear summation of the sound and whisker-evoked response (averaged over the + 20 and + 10, protocols; p < 0.001, Wilcoxon signed rank test, theoretical median zero).
In VCN, cells were classified based on the PSTH pattern as either primary-like (PL) or primary-like-withnotch bushy cells (PLN). In PL bushy cells, the + 20 and + 10 bimodal protocols significantly increased the number of sound evoked responses during whisker stimulation (Fig. 4, Table 2 and Supplementary Fig. S1) with approx. 5 spikes/s (48 to 53 spikes/s; average from the + 20, + 10 protocols). The bimodal response of the PL bushy cells was ~ 11% larger than the linear summation of the sound and whisker-evoked response (averaged over the + 20, + 10 protocols; p < 0.001, Wilcoxon signed rank test, theoretical median zero; Fig. 6). The + 50 protocol induced a response that was ~ 5% smaller than the linear summation of the sound and whisker-evoked response, indicating that although the bimodal response was slightly larger than only sound, it was smaller than the unimodal sum.
In t-stellate cells (n = 10), the whisker alone stimulation induced a short-lasting (~ 20 ms) increased spiking (Fig. 2f), hence, when the whisker stimulation preceded the tone, the whisker stimulation did not influence the sound evoked response (Fig. 5, Table 2). Although the bimodal response was not significantly larger, when averaging over the sample, in some cells the brief burst of whisker-evoked activity could increase the bimodal response. This effect is corroborated by the relatively large BI (for the − 20 and − 10 protocols; Fig. 6).

Discussion
In the dorsal and ventral cochlear nucleus, the sensory-evoked spiking activity to sound, whisker deflections and bimodal (whisker and sound) stimulation was investigated using in vivo electrophysiology in anaesthetized mice.
Whisker deflections evoked increased spiking activity above spontaneous rates in the cochlear nucleus. The effect was most prominent in DCN fusiform cells, with a smaller, transient response in VCN t-stellate cells. www.nature.com/scientificreports/ www.nature.com/scientificreports/ Fusiform cells receive multisensory information via granule cells, whereas for t-stellate cells a multisensory input pathway remains to be investigated. In VCN bushy cells, however, the sample average did not increase significantly, although in some cells ( Fig. 2d) there was a clear increase in spiking with whisker deflections. The cochlear nucleus is primarily involved in sound processing, thus the pairing of whisker stimulation with sound was investigated. Pairing sound stimulation with whisker deflections increased the sound-evoked responses in fusiform cells. The effect was dependent on the temporal relationship between the whisker defection and the sound stimulation. An optimal range was found to be when the whisker deflection began 20 ms before the sound onset. The average bimodal increase was, however, smaller than the sum of the unimodal activity (sound plus whisker) (mean BI ≈ − 9 (average from the + 20 and + 10 protocols; Fig. 6). The variability in the bimodal effect could indicate that multisensory integration is more prominent in certain areas and/or that some cells were less activated by the stimulation parameters used. In VCN, cells were classified based on the PSTH pattern as either primary-like (PL) or primary-like-with-notch bushy cells (PLN), the former thus putatively being spherical bushy cells and the latter globular bushy cells 30 . Globular bushy cells are important for encoding interaural level differences, whereas the spherical bushy cells play a role in encoding interaural time and level differences 31 . Based on a limited sample (n = 7) in contrast to PL cells, the sound evoked response of PLN cells is not modulated by whisker deflections. Whereas in PL cells there was a bimodal enhancement (average bimodal increase, that was larger than the sum of the unimodal activity; mean BI ≈ 11 (average from + 20 & + 10 protocols); Fig. 6). In the t-stellate population, it's noteworthy that the whisker deflection alone caused a transient initial burst (Fig. 2f), and when the whisker deflection preceded the tone, the bimodal activity was not significantly increased compared to only sound. Delaying the whisker deflection relative to the tone increased bimodal activity, because the whisker evoked burst was preserved in the bimodal response (Fig. 6b).
Anatomical data show that DCN receives input from the somatosensory system via dorsal column nuclei and the spinal trigeminal nuclei that project to the granule cell domain 9,10,13,32 . The DCN thus plays a central role in early multi-sensory integration of sound and tactile stimuli. The spinal trigeminal nuclei also mediate sensory information from the whiskers 33 . In rodents audiotactile interactions in the cochlear nucleus have been less examined 34 . The DCN is involved in auditory spatial perception in the vertical plane and suppression of selfgenerated signals 20 . The interactions of sound with the head creates different acoustic spectra depending on the angle of the incoming sound. These difference in the acoustic spectra enable the DCN to determine the sound location with respect to the ear 35 . Knowing the position of the ears also necessities integrating head position. Head position is determined by proprioceptive inputs from the neck. Breathing and sniffing, activities that selfgenerate sound 21 , are both linked to whisking 36 . It is thus possible that whisker deflection per se is not the most important input to DCN, but rather the information on head position inherent in the sensory signal evoked by whisker movements. Furthermore, sound produced when the whisker touches an object is another source of information that could be used for multisensory integration. Hence, whisker input might provide information on subtracting self-generated sounds 20 . In guinea pigs, fusiform cells have been shown to integrate auditory with somatosensory inputs evoked by stimulation of face and neck muscles 3,16 . This type of somatosensory stimulation induced D-stellate cell inhibition, with the degree of inhibition increasing or decreasing depending on the temporal relationship between the somatosensory and the auditory signal. In the present experiments, when whisker activation preceded the sound stimulation, the bimodal response was larger compared to only sound stimulation. Under this condition it is therefore possible, that the degree of D-stellate inhibition to fusiform cells was very small.
Interactions between the somatosensory and auditory systems has been shown in humans [37][38][39] , cats 22 and guinea pigs 15,40 . It remains to be investigated to what degree similar multisensory integration is important in mice and rats for tactile information processing based on whisker inputs. One hypothesis is that, similar to that found in humans 39 , depending on the frequency band, sound can enhance or decrease the perception of surface textures. Depending on the texture discrimination task the performance might thus be better or worse when it is accompanied with sounds of different frequencies. www.nature.com/scientificreports/

Materials and methods
The study was carried out in compliance with the ARRIVE guidelines and were approved by the Institutional Animal Care and Use Committee at the University of Michigan (IACUC; Protocol #009,202). We confirmed that all experiments were performed in accordance with IACUC guidelines and regulations. 25 mm lateral to midline; depth ~ 5250 µm (from skull)) using the RZ2 multichannel acquisition system from Tucker-Davis Technologies (FL, USA). The raw signal (acquired at 25 kHz) was bandpass filtered between 300-5000 Hz and signals that exceeded the background noise with at least 4 SDs were detected as spikes. Recorded units were sorted offline into single-units with a customised, semi-automatic MATLAB algorithm via k-means clustering of n principal components (k and n were user-specified) of peak-amplitude aligned waveforms. Furthermore, spike-sorted units were analysed (Neuroexplorer, Nex Technologies, USA) with crossand autocorrelation to ensure that they were separate units ( Supplementary Fig. S2). Cross-correlation was done with spike-sorted units in adjacent channels. The refractory period in the autocorrelogram was estimated as the peak from the hazard function. The refractory period for fusiform cells in DCN was 9.8 ms ± 1. Auditory Stimulation. Experiments were performed in a double-walled sound shielded chamber and acoustic signals were generated by the RX8 DSP hardware from Tucker-Davis Technologies (TDT). Sound stimulation was 50 ms broadband noise bursts (200 Hz-20 kHz) with 2 ms rise/fall times presented unilaterally through a closed, calibrated earphone to the left ears. 500 repetitions were presented at 80 dB SPL. The response latency was calculated as the timepoint when the spike rate in a 1 ms bin after stimulus onset was higher than the mean ± 2 SD calculated from 50 ms preceding the sound stimulation.

Somatosensory Stimulation.
To stimulate the whiskers ipsilateral to the CN recording side a custombuilt magnetic whisker stimulation system was used. This enabled whisker deflection in the absence of sound. The whiskers were lightly covered with magnetic paint (k03151000, Krylon) and an electromagnet coupled to an isolated electromagnetic driver was placed close to the whiskers. Signals to drive the magnet were generated using the TDT Synapse software. The whiskers were deflected (ramp and hold) 1000 times at 5 Hz with a 100 ms www.nature.com/scientificreports/ duration pulse. The whiskers were actively moved using the magnet, in the rostral direction (i.e. protraction).
To ensure that sound-driven responses in the CN were caused by whisker movements, and not by other tactile or auditory cues, a control experiment with magnetic stimulation was performed by omitting the magnetic paint, and thus not eliciting movement of the whiskers when the magnet was activated. In this control experiment no effect of magnetic stimulation was observed on the spontaneous or sound-driven responses in fusiform cells (Supplementary Table S1). The time window used to calculate the spontaneous activity was 50 ms before the stimulus onset. The time window to calculate the whisker response was from stimulus onset to 50 ms (for the t-stellate cells it was only 25 ms; the t-stellate response was only on initial burst) after stimulus onset. The whiskers ipsilateral to the CN recording site (left side) were deflected. The response latency was calculated as the timepoint when the spike rate in a 1 ms bin after stimulus onset was higher than the mean ± 2 SD calculated from 50 ms preceding the whisker stimulation. The delay from magnet onset to whisker movement (~ 10 ms) was subtracted.

Bimodal Stimulation.
To study multisensory integration in DCN and VCN the acoustic stimulation was combined with the whisker stimulation. A bimodal stimulation protocol was used based on previous in vitro 41 and in vivo 42 studies on spike timing dependent plasticity in DCN. Seven different protocols with varying delays between the sound and whisker stimulation were used: + 50 ms, + 20 ms, + 10 ms, + 5 ms, − 5 ms, − 10 ms and − 20 ms. The prefix " + " means that the whisker stimulation starts before the auditory stimulation and the "−" that the whisker stimulation starts after the auditory stimulation. The different protocols were repeated 500 times and the spiking output compared to 500-times acoustic stimulation alone. The order of the protocol was randomized between experiments (Supplementary Table S2). The time window used to calculate the bimodal response was ~ 50 ms for protocols + 50, + 20, + 10 and + 5. When the whisker stimulation started after the sound the time window was reduced. For the − 5 protocol the time window was ~ 45 ms, for the − 10 protocol the time window was ~ 40 ms and for the − 20 protocol the time window was ~ 30 ms (the exact time varied 1-2 ms depending on the stimulus artefact). To compare the bimodal response to the unimodal responses (only sound or only whisker deflections) the percentage change was calculated as a bimodal integration index BI ≡ Bi−S−W S+W * 100, where Bi is bimodal, S is sound, and W is whisker response in spikes/s 21,43 . BI > zero, is called a bimodal enhancement, meaning that the bimodal response was larger than the sum of the unimodal response.
Statistics. The raw data (spikes/s) did not pass normality tests (neither the D' Agostino & Pearson omnibus normality test, nor the Shapiro-Wilk normality or the Kolmogorov-Smirnov normality test); the data was righthand skewed-relatively few data points with a high spike rate and many with low spike rate. Furthermore, the sample variances in each group were not equal. Thus, the groups were compared with individual non-parametric tests (Wilcoxon matched-pairs signed rank test, exact p-value), with p-values corrected for multiple testing using www.nature.com/scientificreports/ Bonferroni correction. There is one comparison per row/condition (sound vs. whisker deflection + sound); thus, the number of comparisons per family was seven, one for each protocol (+ 50, + 20 etc.). In the tables (Tables 1  and 2) the p-values from the individual Wilcoxon matched-pairs signed rank test were multiplied by seven. Unless stated otherwise the median and interquartile range (IQR) is plotted. The interquartile range is the difference between the third and first quartiles. Due to the expected variability in the response of different cells to bimodal stimulations, the experimental unit was each individual cell, rather than averaging cells from each animal. To reduce the Type 1 error rate, taking into account the large sample size when using the individual cells as the experimental units, the significance level was set at p < 0.01. The correlation coefficient r was calculated as a measure of the effect size: r = z √ 2 * n . Z is the z-score, n = number of single units, 2n = the total number of observations, including the cases where the difference is zero.