Responses of neurons in the feline inferior colliculus to modulated electrical stimuli applied on and within the ventral cochlear nucleus; Implications for an advanced auditory brainstem implant

Abstract

Auditory brainstem implants (ABIs) can restore useful hearing to persons with deafness who cannot benefit from cochlear implants. However, the quality of hearing restored by ABIs rarely is comparable to that provided by cochlear implants in persons for whom those are appropriate. In an animal model, we evaluated elements of a prototype of an ABI in which the functions of macroelectrodes on the surface of the dorsal cochlear nucleus would be integrated with the function of multiple penetrating microelectrodes implanted into the ventral cochlear nucleus. The surface electrodes would convey most of the range of loudness percepts while the intranuclear microelectrodes would sharpen and focus pitch percepts. In the present study, stimulating electrodes were implanted chronically on the surface of the animal’s dorsal cochlear nucleus (DCN) and also within their ventral cochlear nucleus (VCN). Recording microelectrodes were implanted into the central nucleus of the inferior colliculus (ICC). The electrical stimuli were sinusoidally modulated stimulus pulse trains applied on the DCN and within the VCN. Temporal encoding of neuronal responses was quantified as vector strength (VS) and as full-cycle rate of neuronal activity in the ICC. VS and full-cycle AP rate were measured for 4 stimulation modes; continuous and transient amplitude modulation of the stimulus pulse trains, each delivered via the macroelectrode on the surface of the DCN and then by the intranuclear penetrating microelectrodes. In the proposed clinical device the functions of the surface and intranuclear microelectrodes could best be integrated if there is minimal variation in the neuronal responses across the range of modulation depth, modulation frequencies, and across the four stimulation modes. In this study VS did vary as much as 34% across modulation frequency and modulation depth within a stimulation mode, and up to 40% between modulation modes. However, these intra- and inter-mode variances differed for different stimulation rates, and at 500 Hz the inter-mode differences in VS and across the range of modulation frequencies and modulation depths was <= 24% and the intra-modal differences were <= 15%. The findings were generally similar for rate encoding of modulation depth, although the depth of transient amplitude modulation delivered by the surface electrode was weakly encoded as full-cycle rate. Overall, our findings support the concept of a clinical ABI that employs surface stimulation and intranuclear microstimulation in an integrated manner.

1. INTRODUCTION

When disabling hearing loss results from loss of the inner hair cells of the cochlea, good hearing usually can be restored by a cochlear implant. However, some persons with deafness cannot benefit from cochlear implants due to traumatic injury to the auditory nerve, to auditory nerve tumors, or to ossification or malformation of the cochlea. When both cochlea or auditory nerves are so afflicted, useful hearing may be restored by stimulating electrodes implanted onto the surface of the cochlear nucleus (Bento et al. 2008; Colletti et al. 2004; Colletti et al. 2005; Colletti et al. 2002; Colletti et al. 2009b; Edgerton et al. 1982; Grayeli et al. 2007; House and Hitselberger 2001; Lenarz et al. 2006; Maini et al. 2009; McCreery 2008; Otto et al. 2002; Schwartz et al. 2003; Seki et al. 2007; Vincenti et al. 2008; Xiao et al. 2007). The quality of the hearing provided by these Auditory Brainstem Implants (ABIs) differs across users and for the etiology of the hearing loss but in most cases is sufficient to assist with lip reading and for detection and differentiation of many environmental sounds. For some ABI users in suitable listening environments, some comprehension of speech without the aid of lip readings (“open-set speech recognition”) has been achieved (Colletti and Shannon 2005). However, the ABIs now in clinical use rarely restore hearing and speech comprehension that is comparable to what can be achieved by most users of a cochlear implants (Behr et al. 2007; Colletti et al. 2005; Colletti et al. 2009a; Grayeli et al. 2007; Kuchta 2007; Lenarz et al. 2002; Lenarz et al. 2001; Otto et al. 2002; Vincenti et al. 2008).

A large part of the difference in the quality of the hearing restored by cochlear implants and ABIs can be attributed to the unique ways in which each access key features of the auditory system. The mammalian cochlear nucleus complex is organized “tonotopically” whereby the spectrum of acoustic frequencies is represented along an ordered spatial gradient within each of its subdivisions, thus preserving the ordered decomposition of acoustic frequencies implemented by the biomechanics of the cochlea (Brawer et al. 1974; Cant and Benson 2003; Frisina 2001; Frisina et al. 1990; Leake and Snyder 1989; Malmierca et al. 2002; Moore 1987; Osen 1969; Shivdasani et al. 2008). This tonotopic organization of the auditory system constitutes much of the basis of “place pitch” and the hardware and software of multi-channel cochlear implants is configured to access this organization to the maximum extent possible. However, the anatomy of the cochlear nucleus is less favorable for a similar approach with an ABI. A clinical ABI’s array of stimulating electrode is implanted onto the free surface of the dorsal cochlear nucleus where it can be mechanically stabilized by the lateral recess of the 4^th ventricle (Fayad et al. 2006; Schwartz et al. 2008) but much of the tonotopic ordering of acoustic frequencies within the dorsal cochlear nucleus is along its shallow-to-deep axis and this ordering cannot be accessed optimally by stimulating electrodes distributed across the free surface of the nucleus (el-Kashlan et al. 1991). In one series of 60 ABI users, only 60% were able to pitch rank the percepts elicited by even a subset of their surface electrodes with any degree of certainty and consistency (Otto et al. 2002). Even for those users who demonstrated some ability to pitch rank their electrodes, there was much ambiguity and uncertainly in the ranking of adjacent or nearby electrodes. By contrast, most users of cochlear implants do, with practice, develop considerable ability to discriminate and rank the pitch of even complex sounds (Moore 1993; Oxenham 2008; Sucher and McDermott 2007; Wilson and Dorman 2008). It has been our goal to lay the groundwork for an analogous cochlear nucleus auditory prosthesis that could be implemented with the sound processors currently now used in ABIs and with minimal modifications of their software. The proposed device is based on ongoing work to develop high-density arrays of penetrating microelectrodes and on the findings from a clinical trial in which patients received an ABI that included an array of surface electrodes and also an array of penetrating electrodes into their ventral cochlear nucleus.

In that clinical trial in which each of 10 patient received an array of 12 electrodes on the surface of their CN and also received an array of 10 penetrating microelectrodes into their ventral cochlear nucleus (VCN) the penetrating “microwire” microelectrodes in the VCN did elicit percepts which the users described as tones of various pitches. This demonstrated the ability of these penetrating microelectrodes to access a restricted portion of the tonotopic organization of the human ventral cochlear nucleus (Otto et al. 2008) and as predicted by the animal studies described below. Conversely, the percepts elicited by the macroelectrodes on the surface of the patient’s CN were not described as distinctly tonal but were described as much louder than the percepts elicited by the penetrating microelectrodes which were able to convey a very limited range of loudness that the users described as “faint”. Those descriptions correspond well to findings from our animal studies. The dorsal and ventral cochlear nuclei of the cat and other mammals send separate tonotopically–ordered projections to the contralateral central nucleus of the inferior colliculus (ICC) and also send more limited projections to the ipsilateral inferior colliculus (Cant and Benson 2003; Friauf and Ostwald 1988; Malmierca et al. 2005; Powell and Cowen 1962; Warr 1966). Our previous animal studies have characterized how microelectrodes in the feline ventral cochlear nucleus are able to activate neurons in a restricted portion of the tonotopic gradient of the contralateral ICC (McCreery et al. 2013; McCreery et al. 1998; McCreery et al. 1997, 1992; McCreery et al. 2000; McCreery et al. 2001). There remains some uncertainty as to how perceived sound intensity (“loudness”) is related, at various levels of the auditory system, to the number of neurons activated the sound, by the volume of neuronal tissue activated, by the activation of specific “loudness coding” neurons, or by some combinations thereof. However there is good evidence that “loudness” is not coded simply as neuronal discharge rate at the levels of the auditory nerve, the cochlear nucleus, and the ICC (Ehert and Merzenich 1988; Moore 1993; Relkin and Doucete 1997; Rohl and Uppenkamp 2012). In the Inferior colliculus, in the medial geniculate nucleus of the thalamus and in the primary auditory cortex the volume of activated neuronal tissue (as determined in human subjects by functional MRI) increases exponentially with perceived sound intensity (Rohl and Uppenkamp 2012). In our animal studies we observed that each the macroelectrodes on the surface of the dorsal cochlear nucleus activated neurons over a greater span of depths in the ICC than did the penetrating microelectrodes in the ventral cochlear nucleus (McCreery et al. 2010). Thus the differences in the loudness elicited by the patient’s surface and penetrating electrodes is consistent with present models of how loudness is coded in the auditory system and with our findings from our animal studies.

The concordances between animal studies and the findings from the clinical study of patients with ABI that include surface and penetrating electrodes implanted on and in their cochlear nucleus helps to validate our cat model as a platform for the pre-clinical development of improved central auditory prostheses. Our goal is to use this platform to lay the groundwork for an ABI that will allow greater integration and synergy between the demonstrated ability of stimulating electrodes on the surface of the dorsal cochlear nucleus to convey loudness and the ability of the intranuclear microelectrodes to convey place pitch. One finding from the clinical trial described above suggests a need for devices that can place a large number of microelectrodes into the user’s ventral cochlear nucleus while minimizing the tissue damage associated with their implantations and long-term residence. As a group, the 10 patients that received penetrating microelectrodes in their ventral cochlear nuclei did not perform better on speech recognition tests than did patients who received only the surface electrodes (Otto et al. 2008). However, only two of the 10 patients detected percepts from more than 2 of their penetrating microelectrodes, and those two patients did demonstrate improved performance on standard tests of speech perception (recognition of NU- Chip words). This suggests the value of an ABI that places a large number of microstimulating electrodes into the user’s ventral cochlear nuclei so many microelectrodes will be positioned and distributed to elicit auditory percepts spanning a wide range of pitch. There has been significant progress in developing arrays of penetrating microelectrodes in which multiple independent microelectrodes sites are distributed along the length of each penetrating shank (Han et al. 2011). A clinical device incorporating microelectrode arrays of this design would allow many microelectrodes to be distributed along the tonotopic gradient(s) within the ventral cochlear nucleus while minimizing the number of penetrating shanks and thus minimizing the risk of tissue injury.

The prospects for a clinical trial of such a novel ABI would be greatly enhanced by its compatibility with the hardware and software now used in the sound processors of cochlear implants and ABIs. Certainly the device would need to be configured so as not to compromise the demonstrated benefits provided by the ABI now in clinical use. Virtually all sound processors of modern cochlear implant and in ABIs employ variants and enhancements of the Continuous Interleaved Sampling (CIS) strategy. After being partitioned into frequency bands by a bank of contiguous bandpass filters, the wide dynamic range of environmental sound (up to ~ 100 dB) is mapped into the much smaller range of electrical hearing (10-15 dB) and the resulting continuous time-varying signal from each channel is parsed into trains of amplitude-modulated charge-balanced stimulus pulses. To reduce channel interactions, the stimulus pulses are displaced temporally (interleaved) when applied to adjacent or proximal electrodes within the scala tympani (Wilson and Dorman 2008; Wilson et al. 1993). We envision an analogous central auditory prosthesis in which the frequency bands (channels) assigned to the surface electrodes would be pitch-ranked to the extent possible, but their sound processor channels would be relatively wide and would overlap and encompass the narrower pass-bands of the pitch-ranked channels serving the penetrating electrodes. The users’ pitch ranking of the penetrating microelectrodes should be relatively unambiguous due to the tonal quality of their percepts (Otto et al. 2008).

In acoustic hearing, pitch is of central importance for speech and music perception, for differentiation of environmental sounds and for localization and separation of sound sources. Data from normal-hearing listeners suggest that the low-frequency, low-numbered harmonics within complex sounds are of prime importance in the perceptual segregation of competing sounds from different sources (Oxenham 2008). Also, modulation frequency and modulation depth are important features of sounds which allows them to be identified and segregated. Natural sounds including human speech contain prominent amplitude modulations, and psychophysical studies have shown that good detection of amplitude modulation is essential for a variety of auditory tasks including speech recognition (Krebs et al. 2008; Rosen 1992; Shannon 1992). In the advanced ABI we envisions the primary role of the surface electrodes would be to convey loudness and that of the penetrating microelectrodes to convey (place) pitch, but there is certain to be some overlap of these functions and we propose that the device would function best if both modes of stimulation similarly encode modulation frequency and modulation depth over a wide range of each. A major objective in the present study was to quantify that concordance in our cat model and to determine how the concordance might be affected by the choice of stimulus frequency. Amplitude modulation of a stimulus pulse train can be characterized by its modulation frequency and modulation depth. In this study we investigated the similarities and differences in the manner in which these features of the modulation of the stimulus pulses trains are encoded into neuronal activity in the cat’s ICC in response to electrical stimuli delivered by an electrode implanted on the surface of the dorsal cochlear nucleus and, in the same animal, by microelectrodes implanted into their posteroventral cochlear nucleus.

2. Experimental

2.1 Configuration and Implantation of stimulating and recording electrodes

The procedures for implanting the stimulating and recording electrodes into the cat’s cochlear nucleus were as described previously (McCreery et al. 2013). The purpose-bred cats (age 10 months to 2 years) were from a single source (Liberty Research, LLC). Based on their responses to environmental sounds they appeared to have normal acoustic hearing before and after implantation of the electrodes, but the quality of their acoustic hearing was not formally tested.

In 5 cats an array of 4 silicon substrate microelectrode shanks, each shank with 5 independent activated iridium stimulating sites with a geometric surface area of ~2,000 μm² was implanted into the cats’ right ventral cochlear nucleus. These electrodes were fabricated by Bosch-process deep reactive ion etching (Han et al. 2011). The penetrating shanks extend 2 mm below the array superstructure (Figure 1B) and are configured to place the lower 4 electrode sites of each shank into the cats’ ventral cochlea nucleus (Moore 1987; Osen 1969; Snyder et al. 1997b). In addition to the penetrating shanks, a platinum macrostimulating electrode (200 × 1000 μm²) was affixed to the underside of the array carrier. That large electrode opposed the surface of the dorsal cochlear nucleus and was bracketed by the 4 penetrating shanks (Figure 1A). An array of 16 recording microelectrodes distributed over 4 silicon shanks (Figure 1B) was implanted into the contralateral central nucleus of the cats’ inferior colliculus (ICC) (McCreery et al. 2013; McCreery et al. 2009; McCreery et al. 2010). The occipital cerebrum was retracted to reveal the dorsal surface of the inferior colliculus. Under visual control, the tip of the custom array inserter tool mounted on a stereotaxic manipulator was centered over and in contact with the dorsal free surface of the colliculus. The introducer was retracted, the cerebrum was allowed to retract, and a tunneling tool was passed down the inserter tool to open a path through the cerebrum. The tunneling tool then was retracted and the recording array was inserted through the introducer and into the inferior colliculus at an angle of 45°. This procedure ensures that the microelectrode sites of the implanted array are distributed along the ICC’s tonotopic gradient (Brown et al. 1997).

Figure 1.

A: Implantable cochlear nucleus stimulating array with 4 silicon shanks, each shank with 5 activated iridium microstimulating electrodes spanning 2 mm from the tip of each shank. The distal 4 microelectrodes on each array are electrically active. The geometric surface area of each microelectrode is ~ 2,000 μm². The large rectangular platinum electrode ((0.4 mm²) in the center of the cluster of shanks resides on the surface of the dorsal cochlear nucleus. B: Sketch depicting the cochlear nucleus array on the surface of the cochlear nucleus. The view is rostral-caudal and slightly obliquely along the long axis of the array. The depiction of the internal structure of the feline cochlear nucleus was adapted from Snyder et al (Snyder et al. 1997a). DCN, PVCN, AVCN; Dorsal, Posteroventral, Anterioventral cochlear nucleus C: The array of recording microelectrodes to be implanted into the central nucleus of the cats’ inferior colliculus. Each of its 4 shanks has 4 recording microelectrode sites spanning 2.5 mm from the tip of each shank. Each recording site has a geometric surface area of 2,000 μm².

In a separate acute experiment conducted in a 6^th animal, a single movable 16-site recording electrode was advanced in and out of the inferior colliculus at an angle of 45°. This was done to map the spatial distribution of neuronal activity along the tonotopic gradient of the ICC in response to electrical stimulation applied on and within the contralateral cochlea nucleus.

2.2 Stimulation protocol and data acquisition

For the 5 animals with electrodes implanted into and onto their cochlear nucleus and into their contralateral ICC, the procedures for data acquisition and for its subsequent analysis were similar to those described previously. Recording of neuronal activity in the ICC began 3 to 4 weeks after the implant surgery. The cats were lightly anesthetized with Propofol and placed within an acoustic isolation chamber. Data acquisition employed custom software and hardware controlled by a Personal Computer (McCreery et al. 2013; McCreery et al. 2009; McCreery et al. 2010). The software for data acquisition and display of the neuronal recordings during acquisition is written in Microsoft Visual Basic with calls to proprietary subroutines supporting the National Instruments USB6259 Multifunction IO module for USB which controls our custom the 16 channels of controlled- current stimulator and our custom 32 channel recording system. The electrical stimuli applied in or on the cochlear nucleus were trains of charge-balanced, cathodic-first controlled-current pulse. Each phase of the charge-balanced stimulus pulse pair was 100 μs in duration with no interval between the two phases. The stimulus pulse trains were amplitude-modulated (ranging from 0 to 100% modulation depth) using stimulus pulse rates that were greater than 20% of the modulation frequency. When modulated tones were represented as an amplitude-modulated train of electrical stimulus pulses applied in the cochlear of deafened cats, Snyder et al., (Snyder et al. 2000) observed severe distortion of the temporal representation of the modulated signals by neurons in the cats’ inferior colliculus when the modulation frequency exceeded 25% of the stimulus pulse rate. Therefore, we restricted modulation frequency to <= 20% of the stimulus pulse rate. This is consistent with the “sampling ratio” of at least 4-to-1 that is used in most cochlear implants (Wilson and Dorman 2008).

The electrical stimulation applied on the surface of the cochlear nucleus and the microstimulation applied by the microelectrodes in the ventral cochlear nucleus induced neuronal activity that could be recorded by 1 or more of the microelectrodes in the cats’ ICC. In all, data sets were obtained for 52 such ICC neuronal units from 5 cats during 20 recording sessions, each session requiring approximately 6.5 hours of data acquisition. Each of the ICC unit’s activity was recorded in response to 88 combinations of stimulus parameters applied by the macroelectrode on the surface of the dorsal CN and then in response to the same 88 combinations of parameters while stimulating with one of the penetrating microelectrode in the VCN (in all, 176 combination of stimulus parameters and stimulus locations). Stimulus pulse rates were 250, 500, 1000 Hz, modulation depths were 0, 33, 66, 100%. The stimulus was delivered with continuous amplitude modulation of the stimulus pulse trains and then with transient modulation of the trains. Each modulation depth was delivered at 3 or 4 modulation frequencies spanning 10 to 50 Hz when the stimulus rate was 250 Hz, and spanning 10 to 100 Hz when the stimulus rate was 500 or 1000 Hz. Throughout each modulation cycle the mean amplitude of the amplitude-modulated pulses applied through the large platinum electrode on the surface of the cats’ DCN was 600 μA, and was 30 μA for the intranuclear microelectrodes. Those amplitudes were selected because they induced vector strengths of ~ 0.5 at 100% modulation depths of the stimulus pulse trains (Section 2.3). For the transient modulation the train of stimulus pulses was unmodulated for 2 seconds, then amplitude modulated for 0.75 second (or for the closest number of full modulation cycles thereto). The neuronal responses were measured only during the 0.75 second following the 2 seconds of unmodulated stimulation. The responses to the transient modulation are intended to approximate the neuronal responses to the onset of sound (Dugue et al. 2007; Krebs et al. 2008; Zeng and Shannon 1992) and also to approximate the prominent low-frequency fluctuations of human speech. Zheng and Escabi (Zheng and Escabi 2008) showed that the sustained responses of ICC neurons to acoustic stimuli encoded the shape of the stimulus envelope at low modulation frequencies, whist the onset responses more faithfully encoded envelope information at higher modulation rates. Each of the 176 combinations of stimulus parameters was delivered for 100 or 200 repetitions of the stimulus modulation cycle and the corresponding ensemble of neuronal recordings to each combination of parameters is designated a “data frame”.

The neuronal responses induced in the ICC by the surface and by the intranuclear microstimulation were recorded with each of these modes of stimulation delivered separately rather than simultaneously. This decision was based on the difficultly of separating and interpreting the neuronal response induced by simultaneous stimulation with the penetrating and surface electrodes. Also, approximately 6.5 hours was required to acquire the response to the 176 combinations of stimulus parameters and stimulation sites in the ventral CN. Simultaneous stimulation with the surface and penetrating electrodes would have required an additional 3 hours or more and would have required that the neuronal responses to the simultaneous stimulation be acquired while stimulating both types of electrodes with identical modulation depths and modulation frequencies, since including variation in those parameters would have dangerously prolonged the animal’s anesthesia. That type of concordance between the modulation depth and frequency applied to the surface and penetrating electrodes probably would occur only rarely with a clinical device operating in most acoustic environments due to the ambiguities in ranking the pitch of the surface electrodes and thus in assigning the center frequencies of their channels.

2.3 Data Processing and extraction of response metrics

After data acquisition, custom software written in Quickbasic64 was used to extract metrics of the neuronal responses (Vector strength and full-cycle rate) from the recorded neuronal activity. Each stage of data extraction from the recordings and its subsequent analysis was validated using simulated neuronal activity. The recordings were processed offline using custom software (McCreery et al. 2013). The artifacts due to the stimulus pulses were removed from each data frame by template subtraction. The template is the arithmetic average of the 100 or 200 responses constituting the frame. The stimulus artifacts are strongly represented in the template due to their constant latency but other activity including that of neuronal action potentials is strongly suppressed because their time of occurrence tends to vary within each cycle of the modulating stimulus. Thus after subtraction of the template from each member of the data frame the stimulus artifact in each member of the frame is strongly suppressed while most of the events representing neuronal action potentials are preserved (Figure 2). A notable property (and sometimes, liability) of artifact suppression by template subtraction is its strong suppression of action potential induced by the stimulus (in the cochlear nucleus) and recorded (in the ICC) with no intervening synapse. Neuronal activity induced antidromically in descending axons by the stimulus in the CN also is strongly suppressed. In both instances, the nearly constant latency of such APs relative to the stimulus pulses causes them to become part of the template in the same manner as the stimulus artifacts and so during template subtraction they are deleted from the recorded response. Therefore, the detected multiunit activity after subtracting the template represents action potentials induced in the cochlear nucleus and recorded in the ICC with at least 1 intervening synapse. After subtracting the template from the frame, its arithmetic mean was computed as the RMS value of the 100 or 200 data acquisitions at 0 modulation depth of the electrical stimulus and designated as the data frame’s noise. Events with signal-to-noise ratios of 4 or greater were classified as neuronal action potentials (McCreery et al, 2013). Spike sorting of the recorded neuronal action potentials was not used, so all of the neuronal activity recorded in the ICC is classified as multiunit. Figure 2 shows an example of the template subtraction process for traces from one data frame and shows a peristimulus time (PST) histogram illustrating modulation of the neuronal activity induced by the amplitude-modulated stimulus pulse train.

Figure 2.

(A) Responses recorded in a cat’s inferior colliculus during microstimulation in the contralateral cochlea nucleus at 250 Hz. The upper two traces show two of the 200 repetitions of the amplitude modulated stimulus waveform that were averaged to generate the artifact suppression template. The large regularly spaced events are the electrical artifacts of the stimulus applied in the cochlear nucleus. Three much smaller events (*) are presumed to be neuronal action potentials. The lower 3 traces show the recordings after subtraction of the template from each trace. After subtracting the template the stimulus artifacts are suppressed and the neuronal action potentials are clearly seen. (Note the expanded amplitude scale for the lower 3 traces (Modified from McCreery et al, 2013). (B) Post-stimulus time histogram of the neuronal activity evoked by an amplitude modulated pulse train and recorded in the cat’s inferior colliculus

In this report “coding” refers to formulations used by ourselves and by other workers to quantify neuronal responses to acoustic stimuli and to sinusoidally modulated trains of stimulus pulses. The modulated stimulus pulses trains are structured to approximate the outputs of the band-limited channels of the sound processors of cochlear implants and auditory brainstem implants. The topic of which and how specific features of sound are detected and used in normal acoustic hearing to generate various auditory percepts remains an active area of investigation and beyond the scope of this report. The metrics used here to quantify the neuronal responses (mean vector strength (VS) and mean full-cycle rate of the neuronal action potentials) have been used by many researchers to quantify neuronal responses in the central nucleus of the inferior colliculus (Goldberg and Brown 1969; Joris et al. 2004; Krebs et al. 2008; McCreery et al. 2013; Rees and Moller 1983). Vector strength is a measure of the tendency for the electrical stimulation to cause the recorded neuronal action potentials (APs) to become grouped in a restricted portion of the stimulus cycle. The vector components of each recorded AP are x_i = cosθ and y_i = sinθ, where θ is the AP’s time of occurrence within the 0- to-2π window of the modulated stimulus pulse train. Thus $\text{VS}=\frac{1}{n}·⅀({[{(x_i)}^2+{(y_i)}^2]}^{\frac{1}{2}})$ where the summation is over the n recorded neuronal action potentials. Here the summation is over all of the detected action potentials recorded in responses to each of the 176 combinations of stimulus parameters. VS = 1 signifies perfect phase synchrony between phase 0 the stimulus modulation and the n recorded APs whereas VS = 0 indicates a completely random relation (Goldberg and Brown 1969). VS is not affected by the time within the stimulus cycle at which any particular action potential occurs but it is reduced by dispersion of the ensemble of n action potentials across the 2π of the modulation cycle. VS is generally insensitive to the number of neuronal action potentials within each stimulus cycle. Therefore, as a separate metric for full-cycle AP rate we computed the number of detected neuronal action potentials within each data frame for each of the 176 combinations of stimulus parameters.

For each multiunit the difference in the vector strength (VS) induced by two conditions of a single stimulus variable (e.g., by modulation of the stimulus at 10 and then at 100 Hz when delivered at the same modulation depth, stimulus location and stimulus mode) was expressed as the relative difference RD₁₂, where RD₁₂ = (VS₁–VS₂)/MAX[VS₁, VS₂] and VS₁ and VS₂ are the vector strengths induced by stimulating at 10 and 100 Hz, or at 100% and 0% modulation depth, etc. MAX[VS₁, VS₂] is the greater of the two. Because the range of RD₁₂ is limited to ±1, its sample mean and sample variance is less vulnerable to outliers than is the absolute ratio VS₁/VS₂ or the absolute difference VS₁–VS₂. The sample mean and standard deviation of the RDs for all 52 multiunits was computed first, then was recomputed while omitting multiunits whose RD₁₂ (when VS₁ is for 100% and VS₂ is for 0% modulation depth) was greater than 2 standard deviations of the RD₁₂ of the original sample of 52 multiunits. This procedure excluded “outliers” that responded to the modulated stimulus in a manner that was very different from that of most of the multiunits comprising the sample for the two stimulus conditions. Furthermore, the criteria for identifying and excluding outliers is based on RD₁₂ which itself is relatively insensitive to outliers. In nearly all cases the units excluded from the sample were those that exhibited very small VS in response to the stimulation at 100% modulation depth. For different stimulus parameters, these excluded outliers constituted 0 to 12 of the 52 neuronal units. After deleting the outliers from the sample a new MRD₁₂ and its standard deviation was computed for the remaining units. 100*MRD₁₂ is the % change between stimulus conditions 1 and 2. The same procedure for excluding outliers was used to generate the final sample of units for the computations of neuronal firing rate vs. modulation depth (Figure 5 and Table 2)

Figure 5.

Encoding of modulation depth as full-cycle rate of neuronal action potential (APs per second) by 40 to 52 neuronal units in the 5 cats’ ICC in response to electrical stimulation applied on or within the contralateral cochlear nucleus at a rate of 500 Hz. The plots’ ordinate is the change in AP rate relative to the at 0 modulation depth, so a value of 0.25 at 66% modulation depth signifies a 25% increase in full cycle rate relative to that at 0 modulation depth. Left column shows results for continuous amplitude modulation of the stimulus pulse train; Right column shows results for transient modulation. In each frame, the error bars are the standard errors of the mean.

Table 2.

Encoding of amplitude modulation of the stimulus as full-cycle rate of neuronal action potentials. Data are from 40 to 52 neuronal units in the ICC of 5 cats. Columns 2 designate the modulation mode and Column 3 the location of the electrical stimulation that induced the effects on full-cycle rate enumerated in Columns 4 and 5. In Column 4, positive (+) values of MRD signify that the mean full-cycle AP rate increased with increasing modulation frequency. The mean AP rate of each neuronal unit was normalized to 1.0 at 0% modulation depth and the values in Column 5 are the change in AP rate between 0 and 100% modulation depth based on that normalization.

Stimulation rates (Hz)	Modulation mode (Continuous or Transient)	Stimulation location (Surface or Micro)	MRD of AP rate at 100% modulation depth as modulation frequency changed from 10Hz to maximum MF (*)	Mean normalized AP rate at 100% modulation depth, across modulation frequencies from 10Hz to Maximum MF
1000	C	S	−17%±12%	+0.33
		M	−22%±13%	+0.20
	T	S	−27%±20%	+0.10
		M	−4%±2%	+0.38
500	C	S	−20%±12%	+0.50
		M	−3.5%±2%	+0.45
	T	S	−21%±13%	−.060
		M	+12%±7%	+0.60
250	C	S	−14%±10%	+0.74
		M	+12%±5%	+0.33
	T	S	−22%±14%	+0.15
		M	+22%±10%	+0.62

^*Maximum modulation frequency (MF) was 100 Hz when the stimulus rate was 500 or 1000 Hz and was 50 Hz when the stimulus rate was 250 Hz.

3. Results

3.1 Tonotopic span of neuronal activity induced in the ICC by stimulation of the cochlear nucleus with surface electrodes and penetrating microelectrodes

Figure 3 shows contour plots of the time-depth distribution of the vector strength (VS) of the neuronal activity recorded in one cat’s ICC in response to charge-balanced electrical stimulation delivered through a microelectrode in the cat’s contralateral posterior-ventral cochlear nucleus (PVCN), to stimuli applied via a macroelectrode on the cat’s dorsal cochlear nucleus, and then to simultaneous stimulation at both sites. Both sites were stimulated at 50 Hz. These plots were acquired during an acute experiment using a movable recording microelectrode in the cat’s ICC. The plots’ ordinate is along the axis of the recording microelectrode in the ICC and approximately along the tonotopic gradient of the ICC. The correspondences between depth in the ICC (left ordinates) and best acoustic frequencies of the neuronal responses (right ordinate) were determined as described previously (McCreery et al. 2010). Figure 3A and 3B illustrate the large spread of the induced neuronal activity along the tonotopic gradient of the ICC produced by the stimulation on the surface of the dorsal cochlear nucleus (DCN) and the much smaller spread induced by the microstimulation in the PVCN. Figure 3C shows how the response to simultaneous microstimulation in the PVCN and the macrostimulation on the surface of the DCN retained the small focus of the near-maximum response while only slightly reducing the spread of the response due to the surface stimulation (the outer contour line delineates 50% of the maximum response). This illustrates how the intranuclear microstimulation can focus the maxima of the neuronal activity into a small part of the ICC’s tonotopic gradient while retaining the more broadly distributed activity induced by the surface stimulation.

Figure 3.

Contour plots of the rate of action potentials elicited from a multiunit in a cat’s ICC in response to: (A) stimulation by a microelectrode in the contralateral posteroventral cochlear nucleus, (B) stimulation via a macroelectrode on the surface of the dorsal cochlear nucleus, and (C) when the microelectrode and the surface electrode were pulsed simultaneously. The 100 μs charge-balanced stimulus pulses were applied at time 0 on the abscissa at a rate of 50 pulses per second, and repeated 60 times at each depth increment of 100 microns in the ICC. The amplitude of surface stimulation was 600 μA and that of the microstimulation was 30 μA. The inner and outer contour line delineate respectively, 75% and 50% of the unit’s maximum discharge rate.

3.2 Temporal encoding of modulation depth by intranuclear microstimulation and by surface stimulation

Figure 4 shows plots of the vector strengths (VS) of the multiunit neuronal responses recorded in the 5 cats’ ICCs across a range of modulation depths for stimulus pulse rates of 500 Hz. Table 1 summarizes the data for stimulus rates of 250, 500 and 1000 Hz. For each modulation mode, modulation rate and stimulus rate the outliers were deleted from the original sample of 52 multiunits as described in Section 2.3, leaving 48 to 52 multiunits in the sample. Non-zero values of VS are always positive, and in Figure 4 the finite mean values of VS at 0 modulation depth (typically ~0.04 to 0.05) represents the arithmetic mean of the variance across the sample from which the vector strengths were calculated. Presumably this reflects the variability intrinsic to the finite number of action potentials detected in each modulation cycle at 0 modulation depth and the finite number of modulation cycles used to acquire the data. The standard deviations and standard errors at 0 modulation depth are much smaller than those at non-0 depth, indicating that nearly all of the variances at non-0 modulation depths reflect real variance across the sample of neuronal multiunits in their response to the modulated stimulus. Part of that variance may be due to the tendency for individual neuronal unit in the ICC to exhibit various amounts of trade-off in their responses to spectral vs. temporal modulation (Rodriguez et al. 2010). The study by Rodrigues et al also revealed the tendency for the neurons whose responses favor temporal vs. spectral modulation to be somewhat spatially segregated in the ICC. That segregation would allow their preferential responses (spectral vs. temporal) to be reflected in their multiunit activity and thus in the plots of vector strength shown in Figure 4 which reflect only their response to temporal modulation.

Figure 4.

Encoding of modulation depth as the vector strength (VS) by 48 to 52 multiunits recorded in 5 cats’ ICC, in response to charge-balanced pulsatile electrical stimulation applied at a rate of 500 Hz on (surface stimulation) or within (microstimulation) the contralateral cochlear nucleus. Left column shows responses to continuous amplitude modulation of the stimulus pulse train. Right column shows responses to transient amplitude modulation of the stimulus. Mean amplitude of the modulated intranuclear microstimulation was 30 μA and that of the surface stimulation was 600 μA. For each modulation depth (0,33,66, 100%), the larger error bars are standard deviations of the vector strength and the smaller bars are standard errors of the mean. At 0 modulation depth the error bars show only the standard deviation and the bar representing the very small standard error are hidden behind the symbols. The symbols for the surface and microstimulation are offset slightly from the abscissa to reveal the error bars representing the standard errors and standard deviations.

Table 1.

Encoding of stimulus modulation as vector strength for stimulus rates of 250, 500 and 1000 Hz. Data are from 48 to 52 neuronal units in the ICC of the 5 cats. Column 2 specifies the modulation mode of the stimulus (Continuous or Transient) and Column 3 specifies the location of the electrical stimulation that induced the effects enumerated in Columns 4 and 5. Column 4 shows the change in the Mean Relative Difference (MRD) of the multiunit’s vector strength as the modulation frequency of the electrical stimulus changed from 10 Hz to its maximum of 50 or 100 Hz (depending on the stimulus pulse rate). Positive (+) values in Column 4 signify that the modulation depth increased with increasing modulation frequency. Column 5 shows the mean of the vector strengths at 100% modulation depth across the same range of modulation frequency of the electrical stimulation (10Hz to 50Hz or 100Hz).

Stimulation rate (Hz)	Modulation mode (Continuous or Transient)	Stimulation location (Surface or Micro)	MRD of VS at 100% modulation depth as modulation frequency changed from 10Hz to maximum MF (*)	Mean VS at 100% modulation depth, across modulation frequencies of 10Hz to maximum (*)
1000	C	S	−17% ±7%	0.41
		M	−17% ±4%	0.29
	T	S	+23% ±17%	0.33
		M	−11% ±1%	0.29
500	C	S	−9% ±1%	0.41
		M	+7% ±10%	0.31
	T	S	+15% ±3%	0.34
		M	+13% ±10%	0.36
250	C	S	+6% ±4%	0.34
		M	+34% ±13%	0.33
	T	S	+20% ±9%	0.27
		M	+20% ±10%	0.32

^(*)Modulation frequencies were 10, 25, 50 and 100 Hz when stimulus rate was 500 or 1000 Hz, and 10,25 and 50 Hz when the rate was 250 Hz.

In Table I, Columns 2 and 3 together specify each of the 4 stimulation modes (surface or microstimulation, each delivered with continuous modulation and then with transient modulation and by the surface electrode and by a penetrating microelectrode). Column 5 summarizes the mean strength of this temporal encoding across the full range of modulation frequencies. In Column 5 the mean VS for the 4 modes were similar at stimulation rates of 250, 500, and 1000 Hz (mean of 0.33), but slightly greater (mean of 0.35) at 500 Hz. Column 4 of Table I summarizes how the strength of this encoding was affected by changes in modulation frequency. At a stimulus frequency of 250 Hz the MRDs of the vector strengths across the range of modulation frequencies (Column 4) was a large as 34% and as large as 23% at 1000Hz, but did not exceed 15% when the stimulus rate was 500 Hz. With reference to a future clinical device, we would prefer the values in Column 5 to be large and the values in Column 4 to be small and to be similar across the 8 stimulation modes specified in Columns 2 and 3. In a clinical device in which the intranuclear microstimulation is intended to primarily encode place pitch and the surface stimulation to encode loudness, large differences in VS between the surface stimulation and the microstimulation due to changes in modulation frequency could render it difficult to maintain a consistent balance between pitch and loudness in listening environments in which modulation frequency and depth must be expected to vary frequently and rapidly.

3.3 Encoding of modulation depth as full-cycle rate of neuronal action potentials

In the central nucleus of the Inferior colliculus temporal modulation of sound is represented as a temporal code but these is evidence for the beginning of a transition to a rate code (Frisina 2001; Frisina et al. 1990; Joris et al. 2004; Krebs et al. 2008). In this study the unmodulated full cycle AP rate of each multiunit was highly variable so the full cycle AP rate of each multiunit was normalized (to 1.0) on the its rate at 0 modulation depth. Figure 5 shows plots of the normalized full-cycle rate of the multiunits recorded in the cats’ ICC (“full-cycle AP rate”) as a function of the depth of the modulation of the stimulus pulses trains applied in or on the contralateral cochlear nucleus at 500 Hz. The data are from 40 to 52 multiunits. Units for which the MRD of the normalized AP rate at 100% modulation depth of the continuously modulated stimulation differed by more than 2 standard deviations from the MRD of the sample mean were excluded from the original sample of 52 units. The excluded multiunits were those whose AP rate changed very little as the modulation depth changed from 0 to 100%.

Table 2 summarizes the findings for stimulus rates of 250, 500 and 1000 Hz. Column 5 summarizes how, across the range of modulation frequencies, the normalized full-cycle AP rate changed as the depth of the modulation of the stimulus pulses changed from 0% to 100%. Column 4 summarizes how the strength of that encoding of modulation depth was affected by changes in the modulation frequency. Large positive values in Column 5 signify strong encoding of modulation depth as full cycle rate and small values in Column 4 signify relative stability in this encoding as modulation frequency changes. We postulate that both of these conditions would be desirable in a future clinical device. As shown in Column 5, increasing modulation depth of the amplitude-modulated stimulus pulse trains delivered to the ventral cochlear nucleus via the microelectrodes produced an increase in the full-cycle AP rate in the contralateral ICC. In Figure 5 the full-cycle AP rate increased monotonically with increasing modulation depth of the continuously modulated stimulation and for the transiently modulated intranuclear microstimulation. However, when the stimulation was applied on the surface of the dorsal cochlear nucleus, the normalized AP rate changed little with increasing depth of the transient modulation of the stimulus pulse train. It is possible that the units’ full cycle AP rate at the onset of the transient stimulation always increased to near it maximum upon the onset of the modulation after being unmodulated for 2 seconds. This implies that the depth of transient modulation would be poorly represented in the ICC as a rate code when the stimulus is delivered via macroelectrodes on the surface of the dorsal cochlear nucleus. However Figure 5 and Column 5 of Table 2 show that modulation depth of the stimulus delivered by intranuclear microstimulation would be relatively strongly encoded in the ICC as full-cycle AP rate across the full range of modulation frequencies. This rate coding of modulation depth by the microstimulation was especially strong at a stimulus rate of 500 Hz (Column 5) while at 500 Hz the MRD of the full-cycle AP rate induced by the surface stimulation tended to decrease moderately with increasing modulation frequency (Column 4). At 500 Hz the greatest change across modulation frequency (of ~ 20%) was for the continuous and transient modulation of the surface stimulation. In contrast, the mean MRD of the intranuclear microstimulation changed by 12% or less across the range of modulation frequencies (Column 4). Excluding the transient surface stimulation, the means of the normalized full-cycle AP rates at 100% modulation depth (Column 5) were similar at stimulus rates of 250 and 500 Hz but were slightly smaller at 1000 Hz.

4. Discussion

For several years we have been implanting multisite silicon substrate microstimulating arrays of the type depicted in Figure 1 into the ventral cochlear nucleus of adult cats (Han et al. 2011; McCreery et al. 2010; McCreery et al. 2007). Physiologic data presented in this report, and histologic evaluations of the implant sites indicate that these multisite silicon substrate probes can be implanted into the ventral cochlear nucleus with minimal tissue injury and are able to activated neurons in a restricted portion of the tonotopic gradient of the ventral cochlear nucleus (VCN). The multi-site silicon substrate devices we now are developing will continue to function in vivo for more than 1 year (Han et al. 2012) and soak test now underway indicate lifetime in excess of 3 years (manuscript in preparation). In the clinical Auditory Brainstem Implant (ABI) that we envision, the primary role of the surface stimulation would be to convey a wide range of loudness while the primary role of the microstimulation in the VCN would be to convey place pitch. The frequency bands (processor channels) assigned to the surface electrodes would be pitch-ranked to the extent possible, but their sound processor channels would be relatively wide and likely would encompass one or more of the narrower pass-bands of the pitch-ranked channels serving the penetrating electrodes. Pitch ranking of the penetrating microelectrodes by users should be relatively unambiguous due to the tonal quality of their percepts (Abe and Rhoton 2006; Otto et al. 2008).

The primary issue addressed in this report is how the dynamic temporal structure of most listening environments might affect the ability of the surface stimulation and intranuclear microstimulation to function harmoniously. We investigated how the choice of stimulus parameters, and particularly the choice of stimulus pulse rate might affect the balance between loudness (conveyed by the surface electrodes) and place pitch (conveyed by the intranuclear microelectrodes) as modulation frequency and modulation depth vary, and when there are frequent transitions between onset and more sustained amplitude modulation of the sound. Our metric for temporal encoding was the vector strength (VS) of the multiunit neuronal response in the ICC. As shown in Table 1, a stimulus frequency of 500 Hz yielded the least variation in VS (Column 4) and the greatest mean value in VS (Column 5) across the range of modulation frequencies, and also produced the least difference in VS across the 4 modulation modes (stimulation on the surface of dorsal cochlear nucleus or microstimulation within the ventral cochlear nucleus, each with continuous and then with transient amplitude modulation of the stimulus pulse trains). At 500 Hz, the differences in VS across and within those stimulus modes were <= 24% across the full range of modulation depths and modulation frequencies. Cochlear implants and ABIs typically employ a sampling rate of at least 4 to 1 relative to the highest modulation frequency, so a stimulus rate of 500 Hz should allow encoding of acoustic modulation up to ~ 125 Hz. However, in an earlier study (McCreery et al. 2000) we showed reversibly depression of neurons in the feline ventral cochlear nucleus when the continuous stimulation exceeds 100 Hz and the stimulation was prolonged, so in the proposed clinical device it would be necessary to restrict the amplitude of the microstimulation in noisy environments. Should this depression of neuronal excitability prove to be troublesome in a clinical device, the stimulus rate could be reduced, but at the expense of reducing the range of modulation frequencies that could be encoded. As shown in Table 1, the inter-and intra-modal variances in the vector strength across modulation frequency were moderately greater at 250 Hz than at 500 Hz (column 4), and the mean vector strengths for each mode were slightly smaller (Column 5). However, the lower stimulation rate of 250 Hz might be preferable in a noisy environment in which the stimulus amplitude often would be high.

The neuronal responses in the cats’ ICC induced by the surface and by the intranuclear microstimulation were recorded with each of those modes of stimulation delivered separately rather than simultaneously (Section 2.2). This limitation notwithstanding our findings do support the concept of a clinical ABI in which the two modes of stimulation could be integrated to convey pitch and loudness as described above. In particular, we found that when stimulation in the cat’s cochlear nucleus via microelectrodes in the VCN or by surface stimulation of the DCN with a macroelectrode, the neuronal responses in the ICC representing the encoding of modulation depth were similar across a range of modulation frequencies and modulation depths.

At the level of the inferior colliculus, temporal modulation of sound is represented as a temporal code, but there is evidence for the beginning of a transition to a rate code (Frisina 2001; Frisina et al. 1990; Joris et al. 2004; Krebs et al. 2008). In the present study the one exception to the good concordance of the neuronal responses to the surface stimulation and to the intranuclear microstimulation was for the rate coding of the depth of transient modulation (Figure 5 and Column 5 of Table 2). Modulation depth of the transient stimulation was well encoded as AP rate when the stimulus was delivered by the microstimulation but was very poorly encoded when the stimulus was delivered by the surface electrodes. However, the weak encoding of modulation depth as full-cycle neuronal firing rate may be of little consequence to the functioning of the proposed clinical device in which the role of the surface stimulation would be to convey most of the range of loudness. The role of neuronal firing rate in the coding of loudness is unclear even at the level of the auditory nerve (Ehert and Merzenich 1988; Moore 1993; Relkin and Doucete 1997; Rohl and Uppenkamp 2012; Zeng and Shannon 1994). In and above the midbrain the role of neuronal firing rate in encoding loudness appears to be subordinate to the role of the volume of neuronal tissue activated by the sound (Rohl and Uppenkamp 2012).

We should not expect that our findings from cats with normal acoustic hearing would be quantitatively applicable to persons with deafness of etiologies that render them poor candidates for cochlear implants. However, the correspondences between our physiological findings from cats implanted with surface and with penetrating cochlear nucleus electrodes and the percepts reported by human patients who had received auditory brainstem implants that included surface electrodes and penetrating microelectrodes (Otto et al. 2008) does support our cat model as a vehicle for addressing some of the key pre-clinical issues. Specifically, the human patients with the ABIs that included penetrating microelectrodes in their VCN described the percepts elicited by the macroelectrodes on the surface of their dorsal cochlear nucleus as loud but not tonal while the percepts elicited by the intranuclear microstimulation were described as tonal but as much less loud than the percepts from the surface electrodes. In our cats, and as shown in Figure 2, the surface electrodes induced neuronal activity in the central nucleus of the inferior colliculus (ICC) that was distributed widely over the tonotopic gradient of the ICC, and thus consistent with the volume model of loudness perception, while the microstimulation in the cats’ ventral cochlear nucleus induced neuronal activity that was restricted to a small region of the ICC’s tonotopic gradient (McCreery et al. 2009; McCreery et al. 2010; McCreery et al. 2007), consistent with the tonal quality of the percepts elicited by the microstimulation in the patients’ ventral cochlear nucleus. On that basis we expect that the findings from out feline model would be generally applicable to human patients.

A cautionary note regarding the feasibility of the device we propose is our assumption that the patients’ cochlear nucleus would retain essentially normal function. In our animal model we did not attempt to duplicate the clinical conditions that are indications for an ABI. By far the most common clinical indication for an ABI is destruction of the auditory nerve by a tumor or during surgical resection of the tumor of Type II Neurofibromatosis which often develop on both auditory nerves and which continue to grow and become life-threatening if not removed. However, it has been shown that many patients who received ABIs to restore hearing after traumatic injury to the auditory nerves demonstrate significantly better auditory performance with their ABI than do patients who have received an ABI after resection of auditory nerve tumors. Some of these non-tumor ABI users have achieved “open set” speech perception without the assistance of lip reading, a level of function never achieved by the tumor patients. (Colletti et al. 2009a; Colletti and Shannon 2005). This suggest that an auditory nerve tumor, and/or its surgical resection may damage the cochlear nucleus in a manner that limits the function of an ABI. Nevertheless, in the clinical trial noted earlier in which 10 patients received arrays of penetrating and surface electrodes after resection of auditory nerve tumors, the patients who received auditory sensations from more than 2 of their penetrating electrodes performed significantly better on the NU-Chip word identification test than did patients who had received only the array of surface electrodes (Otto et al. 2008). This implies that even after tumor resection the auditory system often retains the capacity to utilize the integrated device we propose. However, it also implies that the array of penetrating electrodes should include a large number of independent microelectrodes sites on multiple shanks to ensure that a sufficient number of microelectrode sites reside in the appropriate part of the ventral cochlear nucleus. This raises the question of how best to allocate the stimulator’s channels to the surface and penetrating electrodes. Speech processors and implanted stimulators now used in cochlear implants and ABIs have up to 22 channels and each of the stimulator’s hermetically sealed conductor feed-throughs are “hard wired” to individual stimulating electrodes. In one series of 61 ABI users conducted at the House Ear Institute and Clinic (Los Angeles, CA), the user’s performance increased moderately when they could to some degree pitch rank at least 2 of their 24 surface electrodes and their performance improved further when they could pitch rank up to 8 of their surface electrodes. (Otto et al. 2002). On that basis, allocation of 8 or 10 channels to surface electrodes and 12-14 channels to penetrating electrodes seemingly would not compromise the benefits conveyed by the surface array. The advantage of placing a large number of microelectrodes into the ventral cochlear nucleus while minimizing the number of penetrating shanks could be addressed by the multisite silicon substrate arrays now being developed in our program and in other programs. A possible solution to the problem of accommodating a large number of microelectrodes with a 22-channel sound processor would be to include a small ancillary implant that is external to the implanted stimulator module. In the ancillary implant 5 or 6 simultaneously active or inactive channel lines from the 22-channel stimulator would direct the stimulus on a 6^th or 7^th channel to any one of 32 to 64 microelectrode sites, leaving 15 or 16 channels for the surface electrodes. That approach certainly would require revisions to the software of the external sound processor and possibly to the software of the implanted stimulator, but should require no hardware modifications to either.

Bullet points.

• Auditory brainstem implants (ABI) may restore useful hearing to persons with deafness and who are not candidates for a cochlear implant.

• ABIs now in clinical use deliver electrical stimulation by an array of electrodes on the surface of the cochlear nucleus

• The quality of hearing restored by an ABI may be improved by integrating the stimulation from the surface electrodes with intranuclear stimulation delivered by multiple penetrating microelectrodes

• Results from an animal model indicate that such a functional integration between surface electrodes and penetrating microelectrodes is possible in a dynamic listening environment