Experiments were conducted at the University of Michigan with approval of the University of Michigan Committee on Use and Care of Animals. Procedures for acoustical and electrical stimulation and for multi-site recording from the inferior colliculus were similar to those described in detail previously (Middlebrooks and Snyder, 2007
) and will be summarized here.
Data presented were from experiments in 5 barbiturate-anesthetized cats. Each experiment involved the following steps: (1) A 32-site silicon-substrate recording probe was implanted in the right ICC, oriented in the coronal plane at 35° from the sagittal plane, approximately parallel to the ICC’s tonotopic axis of increasing CF. (2) Based on the responses to tones presented to the left (contralateral) ear through a calibrated ear bar, the recording probe was adjusted in depth to sample CFs from <1 to >32 kHz, and then the probe was fixed in place. (3) Frequency response areas measured with pure tones were obtained with 1/6-octave resolution in frequency and 5-dB resolution in sound level. The frequency response areas yielded a CF for each recording site that was used in later interpretation of responses to electrical stimulation. (4) The right ear was deafened by removal of the tympanic membrane and ossicles, resulting in a severe conductive hearing loss. The left ear was deafened by injection of neomycin sulfate (10% in water) into the scala tympani. In pilot experiments in other animals, that procedure consistently led to >80-dB elevations of auditory thresholds (i.e., loss of all scalp-recorded auditory brainstem response at levels within the dynamic range of our earphones) within ~10 min. (5) ICC spike activity was recorded in response to single electric pulses and to pulse trains presented through each of three types of stimulating electrode. First, a silicon-substrate intraneural stimulating array was inserted into the auditory nerve through a small hole in the osseous spiral lamina for intraneural stimulation. Second, an 8-channel banded electrode array (Cochlear, Ltd., Lane Cove, NSW, Australia) was inserted in the scala tympani through a small cochleostomy for intrascalar stimulation. Third, the apex of the cochlea was exposed, and a silver apical ball electrode was placed on the osseous spiral lamina of the apical turn. The apical ball electrode was intended to simulate an intrascalar electrode advanced all the way to the extreme apex of the cochlea, which is impossible with present-day cochlear implants. The duration of experiments, from induction of anesthesia to termination, ranged from 18 to 25 hr, typically with only minor reductions in responsiveness observed.
Stimulus presentation and data acquisition used System 3 equipment from Tucker-Davis Technologies (TDT; Alachua, FL) and custom software running in MATLAB (The Mathworks, Natick, MA). Each recording probe (NeuroNexus Technologies, Ann Arbor, MI, USA) had 32 iridium-plated recording sites spaced in 100-μm intervals along a single silicon-substrate shank. The 32 recorded neuronal waveforms were digitized simultaneously, displayed on-line, and stored on computer disk for off-line spike sorting. Electrical artifact from the cochlear stimulus could be detected at the ICC recording sites. Artifact was eliminated by a sample-and-hold function that was programmed into the digital signal processor in the TDT recording system, effectively gating off the amplifier during the brief times of the stimulus pulses (Middlebrooks, 2008
). In a few instances in which adequate artifact rejection could not be attained, and in test cases in which the artifact rejection was disabled, artifact propagated to the ICC with group delay of <2-ms. Such short-delay activity was distinct from spike activity (with group delay >4 ms); recordings were screened to eliminate units in which recordings were contaminated by electrical artifact.
Neural spikes were detected off line using a spike-sorting procedure described previously (Middlebrooks, 2008
). Among the 5 animals, 2 to 4 of the 32 sites on each probe yielded well-isolated single units, and the remaining 28 to 30 sites each yielded unresolved activity from 2 or more neurons. We refer to the spike activity at all recording sites as “unit” activity except in cases in which we specifically refer to “well isolated single units”. Activity was recorded from a total of 160 sites (i.e., 32 in each of 5 cats). The number of units included in specific analyses varied according to the numbers of ICC units that were activated by the various stimuli.
Three types of stimulating electrodes or electrode arrays were used. Intraneural stimulating arrays were NeuroNexus probes similar to the recording probes. Each stimulating array had 16 iridium-plated electrodes centered at 100-μm intervals along a single silicon shank. The shank was inserted into the auditory nerve through a hole in the osseous spiral lamina, approximately perpendicular to the trunk of the auditory nerve as it exited the base of the cochlea. Based on the tonotopic distribution of ICC activation, we selected 2 (in 2 animals) or 3 (in 3 animals) intraneural stimulating electrodes to represent activation of auditory-nerve fibers from basal, middle, and/or apical turns. The intrascalar electrode arrays were similar to the clinical Nucleus 22 cochlear implant, differing from the human version only in that the array was truncated to 8 electrodes. In each animal, the intrascalar array was advanced as far apically as possible, typically lying with the most apical electrode in the basal half of the middle turn. The most apical stimulating electrodes (designated electrodes 7 and 8) exhibited the lowest thresholds, presumably because of their snug fit in the scala tympani. For monopolar stimulation, we used electrode 7 or 8 as the active electrode and a wire in a neck muscle as the return. For bipolar stimulation, electrode 7 was the active electrode and electrode 8 was the return. The apical ball electrode was a ~0.25-mm ball flamed on the end of a silver wire. The ball was positioned on the exposed osseous spiral lamina of the apical turn of the cochlea.
Custom optically isolated 8- or 16-channel current sources were used for electrical stimulation. Output impedances were ~2 MΩ for intraneural electrodes and ~20 kΩ for the intrascalar and apical ball electrodes. Current pulses were biphasic, initially cathodal, 40 μs per phase, with no inter-phase gap. Single pulses were used to derive plots of the tonotopic distribution of ICC activation as a function of current level, known as spatial tuning curves. Measures of transmission of temporal fine structure used pulse trains, 300 ms in duration, repeated every 600 ms. Pulse rates ranged from 39.99 to 602.82 pps in steps of 38.52 to 41.79 pps; hereafter, the rates will be written as 40 to 600 pps in 40-pps steps for convenience of presentation. Current levels of the pulse trains were 4 and 8 dB above the estimated single-pulse detection threshold of units at the most sensitive ICC recording site; in some cases, additional levels at 2-dB increments were tested. Each combination of pulse rate and level was tested 20 times in interleaved order.
Thresholds for detection of single pulses were computed based on trial-by-trial spike counts using a procedure derived from signal detection theory (Green and Swets, 1966
; Macmillan and Creelman, 2005
; Middlebrooks and Snyder, 2007
). Background (i.e., no stimulus) spike counts were measured in the intervals 18 to 3 ms before the pulse, and stimulus-driven spike counts were measured in the intervals 3 to 18 ms after the pulse. For each stimulus level, we formed an empirical receiver operating characteristic (ROC) curve based on the trial-by-trial distributions of background and driven spike counts. The area under the ROC curve gave the probability of correct detection, which was expressed as a z
-score and was multiplied by √2 to obtain the detection index, d’
Analysis of the responses of each unit to pulse trains was conducted at the lowest tested level at which d’ for that unit was ≥1. That relatively low level was favored because many units could not be tested at higher levels, either because of the greater incidence of electrical artifact at higher levels or because the stimulator could not generate sufficiently high currents to stimulate some high-threshold units at high suprathreshold levels. Eleven of the 160 recording sites were judged in off-line analysis to be in the external nucleus of the inferior colliculus (ICX) on the basis of broad frequency tuning and electrically-evoked first-spike latencies longer than 10 ms. In the intraneural condition, units at some of the 149 ICC recording sites were counted 2 or 3 times because they were tested with intraneural electrodes at 2 or 3 depths in the nerve; for that reason, the number of units sampled could exceed the number of physical recording sites. Sample sizes at the minimum suprathreshold level were 170, 100, 90, and 83 ICC units tested with intraneural, monopolar, bipolar, and apical ball configurations, respectively. Whenever possible, the analysis was repeated at an additional level 2 or 4 dB higher than the minimum. Sample sizes at the higher level were 121, 47, 44, and 18 sites, respectively. In most cases, results obtained at the low current level were confirmed at the higher levels, as indicated in the text, although the statistical power often was lower at the higher levels because of the lower sample size or the reduced range of CFs that was sampled.
The strength of phase locking of ICC neurons to electrical pulse trains was represented by the vector strength (Goldberg and Brown, 1969
), which was computed as follows. Each spike was treated as a vector of unit length and of orientation given by its phase relative to the stimulus period. Spikes were summed across 20 trials. The length of the resultant vector divided by the spike count gave the vector strength (possibly ranging from 0 to 1), and the orientation of the resultant gave the mean phase. The statistical significance of the vector strengths were evaluated by the Rayleigh test of circular uniformity (Mardia, 1972
) at the level of p
<0.001. The limiting rate
for each unit was the highest pulse rate at which the vector strength was statistically significant (i.e., at which the vector strength was greater than the Rayleigh criterion). When units phase locked significantly at the highest tested rate, the limiting rate was scored as 600 pps, but we assume that the actual limiting rates of most or all of those units were higher. Mean phase lag tended to increase linearly with increasing stimulus pulse rate across the range of rates that displayed significant phase locking. The slope of the best-fitting phase-versus-frequency line gave the group delay
, for mean phase lag (ϕ) in radians and pulse rate (r
) in s−1
. Units in the ICC responded to the onset of a pulse train with a temporally compact burst of spikes, regardless of the pulse rate, which could have given an erroneous impression of precise phase locking. For that reason, vector strength and mean phase were computed based on spikes falling 50 to 300 ms after the onset of the pulse train. The same post-stimulus time interval was used for measurement of tonic spike rate