Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
INTENSITY ENCODING OF VENTRAL COCHLEAR NUCLEUS NEURONS IN NORMAL AND DEAFENED CATS AND CORRELATES OF LOUDNESS RECRUITMENT by Shanqing Cai A thesis submitted to The Johns Hopkins University in conformity with the requirements for the degree of Master of Science in Engineering Baltimore, Maryland July, 2007 © 2007 Shanqing Cai All Rights Reserved Abstract Loudness recruitment, an abnormally rapid increase of loudness with sound level, is a commonly seen symptom of sensorineural hearing loss (SNHL). Its physiological mechanisms are not well understood. Previous studies failed to find its neural correlates in auditory nerves in acoustically traumatized ears, suggesting that the correlates may primarily reside in the central auditory system. This thesis investigates the rate-level and rate-spectral encoding by neurons in the ventral cochlear nucleus (VCN) of cats with noise-induced hearing loss (NIHL). Permanent NIHL was induced with noise overexposure. Single-unit recordings were performed in traumatized and control cats. Basic physiological properties of VCN neurons in impaired ear, including tuning, response maps, best-frequency (BF) tone post-stimulus-time histograms (PSTH) and phase locking were studied. VCN neurons were classified into primary-like, primary-notch, chopper, onset, locker and unusual categories. Rate-level encoding was studied for simple and complex stimuli including tones, broadband noise (BBN) and the vowel /ε/. The linear-nonlinear weighting model was used to study the rate-spectral encoding in VCN neurons. Acoustic trauma caused elevated thresholds and broadened tuning. The PSTHs were largely unaltered after trauma. In impaired ears, BF-tone rate-level slopes were shallower-than-normal in primary-like and primary-notch units. However, the rate-level slopes for BF-tone, BBN and the vowel were steepened in non-primary-like units. Recruitment-like phenomena, i.e., elevated threshold and abnormally rapid increase in discharge rates with increasing level, were observed in on-BF encoding by non-primary-like neurons, and all-BF encoding by all types of VCN neuron except lockers. Similar phenomena were observed for the rate-level encoding of BBN and the vowel. Abnormally rapid spread of excitation was an important contributing factor to these phenomena. ii These recruitment-like phenomena were quantitatively comparable to the psychophysical properties of loudness recruitment. Other encoding anomalies observed in traumatized ears included increased firing-rate variability, deteriorated rate-place encoding of the vowel spectrum and abnormal weight functions. The results of this thesis indicate that the VCN is the most peripheral structure exhibiting explicit neural correlates of recruitment and supports the post-trauma central plasticity hypothesis. This study provides novel clues to the physiological mechanisms of abnormal intensity and speech perception in people with SNHL. Thesis readers: Eric D. Young, Ph.D., Bradford J. May, Ph.D., and Kechen Zhang, Ph.D. iii Acknowledgements First of all, I would like to thank my thesis advisor Dr. Eric Young. Eric’s encouragement and guidance made possible the completion of this thesis work. He presented a role model of scholarship, which will have a deep impact on me in my academic future. I would also like to thank Dr. Diana Wei-Li Ma for her helpful supervision during the entire course of my thesis work, her instructions and assistance with the experimental procedures, and also for her helpful discussion and support. Thanks are also due to Ben Letham for his helping in several noise overexposure procedures and recording experiments. Technical assistances from Phyllis Taylor, Ron Atkinson and Qian Gao were indispensable to this thesis work. Other people in the Neural Encoding Lab, including Dr. Brad May, Dr. Zachary Smith, Dr. Steven Chase, Dr. Sharba Bandyopadhyay, Dr. Paul Nelson, Dr. Amanda Lauer, Dr. Michael Anderson, Tessa Ropp, Stephanie B. Saylor provide a warm and intellectually nurturing environment, for which I am very grateful for. Dr. Michael Heinz provided me with auditory-nerve data from the previous study for re-processing, along with detailed descriptions, which made possible the cross-comparison between peripheral and central data in this thesis. The constructive comments of the other two readers of this thesis, Dr. Brad May and Dr. Kechen Zhang helped improved the scope and quality of this thesis. This thesis work was supported by a fellowship from the Department of Biomedical Engineering, Johns Hopkins University, and by National Institutes of Health grant DC00109. I would like to give thanks to Wei Chen, whose sharing of pressure and happiness helped me greatly in the process of writing this thesis. Finally, the yăng yù zhī ēn of my parents, Jiliang Cai and Peihua Qiu, is gratefully acknowledged. iv List of figures Figure # Fig. 1.1. pp. 3 Fig. 1.2. Fig. 1.3. 5 7 Fig. 1.4. Fig. 1.5. 13 20 Fig. 1.6. 20 Fig. 1.7. 23 Fig. 1.8. Fig. 1.9 29 33 Fig. 1.10. Fig. 1.11. 36 40 Fig. 1.12. 44 Fig. 2.1. 54 Fig. 2.2. Fig. 2.3. Fig. 2.4. Fig. 2.5. Fig. 2.6. Fig. 2.7. Fig. 2.8. 57 61 66 68 71 74 80 Fig. 2.9. Fig. 3.1. 83 87 Fig. 3.2. 88 Fig. 3.3. Fig. 3.4. 90 91 Fig. 3.5. Fig. 3.6. Fig. 3.7. 96 98 101 Figure description Intensity responses of the basilar membrane and inner hair cells in normal and post-mortem states. Rate-level functions of auditory nerve fibers in normal-hearing ears. Non-linear phase responses of the basilar membrane and a temporal code for intensity. Loudness balance curves in loudness recruitment. Rate-level slopes of auditory nerve fibers before and after acoustic trauma. A schematic of effects of damages to inner and outer hair cells on rate-level slopes of auditory nerve fibers Summed / average rate-level functions of auditory nerve fibers in acoustic trauma. Overexcitability of central auditory nuclei in sensorineural hearing losses. The decision tree for classifying VCN principal neurons based on BF-tone PST histograms Major PST-histogram types of VCN principle neurons The first-spike latency and firing regularity of VCN principle neurons in response to BF-tone bursts. Rate-place encoding of the steady-state vowel /ε/ in auditory-nerve fibers and VCN chopper neurons. Offline processing of compound action potential recordings and estimation of the response thresholds. Examples of amplitude calibration curves of the acoustic drivers The steady-state vowel /ε/ and the spectral manipulation procedure Tuning curves and Q10s generated by two different methods. Construction of rate-level functions and slope analyses Classification of rate-level-function shapes The level adjustment methods for pseudopopulation analyses The random spectral shape stimuli and analyses on the linear weight functions The unexpected acoustic noise in the dynamic driver and its effects A summary of CAP audiograms and recorded VCN units in normal-hearing animals. A summary of CAP audiograms and recorded VCN units in noise-exposed animals. Alignment of frequency axes at the CAP-audiogram edges Individual and average CAP audiograms in the normal and hearing-impaired populations Example PSTHs of primary-like and primary-like-with-notch units Example PSTHs of chopper-type units Example PSTHs of onset-type units v Fig. 3.8. Fig. 3.9. Fig. 3.10. 104 107 109 Fig. 3.11. 112 Fig. 3.12 Fig. 3.13 115 118 Fig. 3.14 Fig. 3.15 Fig. 3.16 Fig. 3.17 Fig. 3.18 122 125 126 129 131 Fig. 3.19 134 Fig. 3.20 136 Fig. 3.21 138 Fig. 3.22 Fig. 3.23 140 143 Fig. 3.24 Fig. 3.25 Fig. 3.26 Fig. 3.27 Fig. 3.28 Fig. 3.29 Fig. 3.30 151 152 156 158 159 161 163 Fig. 3.31 165 Fig. 3.32 Fig. 3.33 Fig. 3.34 169 171 174 Fig. 3.35 Fig. 3.36 175 180 Fig. 3.37 184 Fig. 3.38 187 Example PSTHs of locker-type units Example PSTHs of unusual-type units Quantitative analyses on BF-tone PSTHs at 30 dB re threshold. I. Firing rates. Quantitative analyses on BF-tone PSTHs at 30 dB re threshold. II. Firing regularity and latencies. Minimum first-spike latency versus BF plots for the VCN units Vector strengths of low-BF neurons in response to BF tones at 30 dB re threshold. Relationships between quantitative measures of BF-tone PSTHs and level. Tuning curves of VCN neurons in the normal-hearing ears Tuning curves of VCN neurons in the noise-exposed ears Q10 values of VCN neurons in normal and noise-exposed ears Example response maps of Pri and PriN units in normal and noise-exposed ears Example response maps of chopper units in normal and noise-exposed ears Example response maps of onset and Locker units in normal and noise-exposed ears Example response maps of unusual-type units in normal and noise-exposed ears Spontaneous firing rates of VCN neurons in normal and exposed ears BF-tone rate-level functions of different types of VCN neurons in normal and noise-exposed ears Relationships between chord slopes and fit slopes Rate-level slopes for tones in the PL (Pri/PriN) group Rate-level slopes for tones in the chopper group Rate-level slopes for tones in the onset group Rate-level slopes for tones in the locker group Rate-level slopes for tones in the unusual group Relationships between threshold shift and BF-tone rate-level functions in VCN neurons Analysis of variance on the rate-level slopes for tones with unrestricted ORB ranges in normal and exposed ears. Thresholds for tones Analyses on the thresholds in the tonal pseudopopulations Unnormalized average rate-level functions for the tonal pseudopopulations Normalized average rate-level functions for the tonal pseudopopulations Average rate-level functions for on-BF encoding in the tonal pseudopopulations Average rate-level functions for all-BF encoding in the tonal pseudopopulations Spread of excitation in the tonal pseudopopulations vi Fig. 3.39 190 Fig. 3.40 195 Fig. 3.41 196 Fig. 3.42 Fig. 3.43 Fig. 3.44 Fig. 3.45 198 204 208 211 Fig. 3.46 213 Fig. 3.47 Fig. 3.48 Fig. 3.49 Fig. 3.50 216 219 221 223 The relationships between ORB range and the slopes of rate-matching curves The relations between variances and means of discharge rates for tonal stimuli, I. The relations between variances and means of discharge rates for tonal stimuli, II. Analysis on broadband noise rate-level functions Rate-level encoding of the spectrum-manipulated steady-state vowel /ε/ Rate-place profiles for the spectrum-manipulated steady state vowel /ε/ Example rate-place profiles for the spectrum-manipulated steady-state vowel /ε/ based on single neurons Average normalized rate-level functions for the spectrum-manipulated steady-state vowel /ε/ Example weight functions of VCN neurons in normal-hearing ears Example weight functions of VCN neurons in noise-exposed ears Example weight functions of onset and unusual-type neurons Quantitative analysis on the linear and nonlinear weight functions from the RSS stimuli vii List of tables Table # Tab. 1.1 pp. 26 Tab. 2.1. Tab. 2.2. Tab. 3.1. 50 57 94 Table description Previous observations on central neural overexcitability in animals with SNHL A summary of animals in the NIHL pool. Determining weights for different VCN PSTH types A summary of unit numbers in different PSTH categories and ORE regions in normal and exposed populations. viii Abbreviations Abbreviation Full name / meaning A/D ABR AC AI AM AN ANF ANOVA BBN BC BF BM Ch ChL ChS ChT CI CN CS CV D/A DAS DC DWC EPSC FSL GBC HC HL HRTF HS i.m. IAS IC IHC IIR ISI JND LDWM LNWM Analog to digital Auditory-brainstem response Alternating current Primary auditory cortex Amplitude modulated, or amplitude modulation Auditory nerve Auditory nerve fiber Analysis of variance Broadband noise Bushy cell Best frequency Basilar membrane Chopper Low-firing-rate chopper Sustained chopper Transient chopper Confidence interval Cochlear nucleus Chord slope Coefficient of variation Digital to analog Dorsal acoustic stria Direct current Deaf white cat Excitatory post-synaptic current First-spike latency Globular bushy cell Hair cell Hearing loss Head-related transfer function Hearing status Intramuscularly Intermediate acoustic stria Inferior colliculus Inner hair cell Infinite impulse response Interspike interval Just noticeable difference Level dependent weighting model Linear-nonlinear weighting model ix LSO MSO NC NIHL NRLF OHC On OnC OnI OnL ORB ORE ORF P/S PL Pri PriN PST PSTH RLF RM SBC S.D. S.E. SL SLA SLB SMP SL SNHL SOC SOE SPL SPP SR TB TC TF UT VCN WF WRS Lateral superior olive Medial superior olive Calibration-corrected BBN level Noise-induced hearing loss Normalized rate-level function Outer hair cell Onset Onset-chopper Ideal chopper L-shaped chopper Octaves re best frequency Octaves re (CAP-audiogram) edge Octaves re tone frequency Peak-to-sustained Primary-like + Primary-like-with-notch Primary-like Primary-like-with-notch Peri-stimulus time Peri-stimulus time histogram Rate-level function Response map Spherical bushy cell Standard deviation Standard error Sensation level Level adjustment A (See section 2.6.5) Level adjustment B (See section 2.6.5) Spectral manipulation procedure Sensational level Sensorineural hearing loss Superior olive complex Spread of excitation Sound pressure level Spikes per peak Spontaneous rate Trapezoid body Tuning curve Tone frequency Unit type Ventral cochlear nucleus Weight function Wilcoxon rank-sum (test) x Legend of symbols for VCN PSTH types In many figures of this thesis, different VCN PSTH types are indicated by different symbols. Two sets of symbols are used, one for the coarse PSTH categories, and the other for the fine categories, which are shown below. Which set of symbol is used in a figure should be clear from the context. Coarse categories Fine categories Pri PriN ChS ChT ChL On Locker Unusual Unknown PL Ch On Locker Unusual Unknown xi Table of Contents Acknowledgements ......................................................................................................................... iv List of figures ....................................................................................................................................v List of tables .................................................................................................................................. viii Abbreviations .................................................................................................................................. ix Legend of symbols for VCN PSTH types ....................................................................................... xi Table of Contents ........................................................................................................................... xii I. Introduction....................................................................................................................................1 1.1. Neural Encoding of Sound Intensity ..................................................................................1 1.1.1. Intensity encoding in the auditory periphery ...........................................................1 1.1.2. Intensity encoding in the central auditory system ...................................................8 1.2. Loudness recruitment and its psychophysical properties ................................................. 11 1.2.1. Intensity discrimination in recruitment .................................................................17 1.3. Neuropathophysiology in SNHL and physiological correlates loudness recruitment ......17 1.4. Anatomy and physiology of the ventral cochlear nucleus in the cat ................................31 1.4.1. Bushy cells ............................................................................................................34 1.4.2. Stellate cells...........................................................................................................37 1.4.3. Octopus cells .........................................................................................................41 1.4.4. Regularity and latency of VCN principal neurons ................................................42 1.5. Level and spectral encoding of vowels and broadband sounds by AN and CN neurons in normal and impaired hearing...............................................................................................42 1.6. The current study ..............................................................................................................47 II. Methods ......................................................................................................................................49 2.1. Acoustic trauma in cats ....................................................................................................49 2.2. The compound action potential audiogram ......................................................................51 2.2.1. Edge frequencies of CAP audiograms ...................................................................53 2.3. Surgical preparation .........................................................................................................55 2.4. The acoustic system..........................................................................................................58 2.5. Acoustic stimuli ................................................................................................................58 2.5.1. Simple stimuli .......................................................................................................58 2.5.2. The steady-state vowel /ε/ and the spectral manipulation procedure ....................59 2.6. Data analysis methods ......................................................................................................62 2.6.1. Classification of ventral cochlear nucleus neurons ...............................................62 2.6.2. Quantitative analyses of PSTHs ............................................................................64 2.6.3. Construction of tuning curves and response maps ................................................65 2.6.4. Analyses of rate-level functions ............................................................................67 2.6.5. The pseudopopulation methods .............................................................................72 2.7. The linear / nonlinear weighting model and the random spectral shape stimuli ..............78 2.8. The unexpected acoustic noise in the dynamic speaker and its effects ............................81 III. Results .......................................................................................................................................86 3.1. Acoustic trauma in cats ....................................................................................................86 3.2. Basic unit characterization and classification...................................................................92 3.2.1. Primary-like ...........................................................................................................95 xii 3.2.2. Primary-like-with-notch ........................................................................................97 3.2.3. Chopper .................................................................................................................97 3.2.4. Onset ...................................................................................................................100 3.2.5. Locker .................................................................................................................102 3.2.6. Unusual ............................................................................................................... 105 3.2.7. Quantitative comparisons of BF-tone PST histograms .......................................108 3.3. Latency and phase-locking of VCN neurons.................................................................. 114 3.3.1 First-spike latencies of VCN neurons................................................................... 114 3.3.2. Phase-locking properties of low-BF units ........................................................... 116 3.4 Level dependences of BF-tone PST histograms ..............................................................120 3.5. Tuning curves, frequency-level response maps and spontaneous firing rates ................124 3.5.1. Tuning curves and Q10 measures ........................................................................124 3.5.2. Level-frequency response maps ..........................................................................130 3.5.3. Spontaneous firing rates ......................................................................................139 3.6. Analyses of tonal rate-level functions ............................................................................142 3.6.1. BF-tone rate-level functions ................................................................................142 3.6.2. Slopes of rate-level functions for tones ...............................................................149 3.6.3. The relationship between rate-level slope and BF / threshold shift .....................162 3.6.4. ANOVA analysis on tonal rate-level slopes.........................................................164 3.6.5. Thresholds of tonal rate-level functions ..............................................................167 3.6.6. Average rate-level functions of the pseudopopulation tone .................................172 3.6.7. Spread of excitation .............................................................................................185 3.6.8. The variability of discharge rates in response to tones ........................................193 3.7. Analyses on broadband-noise rate-level functions .........................................................197 3.8. Rate-level and rate-spectral encoding of the vowel /ε/ ..................................................202 3.9. Analysis based on the linear-nonlinear weight model and the random spectral shape stimuli....................................................................................................................................214 IV. Discussion ...............................................................................................................................228 4.1. Summary of findings ......................................................................................................228 4.2. Physiology of VCN neurons in acoustic trauma partially explained by AN abnormalities. ...............................................................................................................................................232 4.3. Central neural alterations following cochlear damage ...................................................236 4.4. The role of VCN in central overexcitability following cochlear damage .......................240 4.5. Neural correlates of loudness recruitment in VCN ........................................................245 4.6. Level discrimination and increased rate variability........................................................249 4.7. Vowel level and spectral encoding .................................................................................250 4.8. Abnormal rate-spectral encoding after acoustic trauma .................................................252 V. Conclusions...............................................................................................................................254 References .....................................................................................................................................257 Biography ......................................................................................................................................275 xiii I. Introduction The motivation for this thesis originates from two unanswered questions. The first question is how neurons in the ventral cochlear nucleus (VCN) respond to sound in the ear with sensorineural hearing loss (SNHL). The second question is what physiological changes in the auditory system give rise to loudness recruitment, the abnormally rapid increase of perceived loudness with sound intensity, a commonly seen symptom of SNHL. These two questions intersect at the problem how VCN neurons encode sound intensity in SNHL, which is the central issue this thesis aims to address. The answers to this question will provide indirect information for solving the long sought problem: how does the auditory system encode intensity and determine loudness? In the Introduction section, previous findings relevant to this thesis and past works that laid the ground for it will be briefly reviewed. These will include neural encoding of sound intensity (Section 1.1), psychophysical properties of loudness recruitment (Section 1.2), peripheral and central neuropathophysiology in SNHL (Section 1.3), anatomy and physiology of the VCN (Section 1.4), and level and spectral encoding of complex stimuli in the VCN (Section 1.5). 1.1. Neural Encoding of Sound Intensity 1.1.1. Intensity encoding in the auditory periphery Sound intensity, or level, is a measure of how much energy is carried by the sound through a unit area in a unit amount of time. It carries useful information, such as proximity of the source, intensity of the ongoing process, and certain prosodic aspects of speech. Loudness is the subjective psychophysical correlate of sound intensity (Moore, 2002). A neural code for sound intensity and a neural correlate of loudness are indistinguishable from each other. Since a 1 thorough understanding of the neural encoding of sound intensity hasn’t been reached, we have to bear in mind that there are a myriad of ways in which the auditory system may encode intensity, as will be reviewed in this section, when we search for a neural correlate of abnormal loudness perception in SNHL. The outer and middle ear, which couples sound waves into the cochlea, is usually assumed to be a filter that behaves linearly in the hearing range (e.g., Zilany and Bruce, 2006). Pressure waves in the cochlear fluids cause vibration of the basilar membrane (BM), which in turn activates the mechanical transduction processes in the inner hair cells (IHCs). In the healthy cochlea, the BM is not a linear mechanical amplifier (Robles and Ruggero, 2001). As shown in Fig. 1.1.A and B, the shape of BM velocity versus level function is dependent on both frequency and sound level. At and near the best frequency (BF), the responses often show a compressive nonlinearity, namely a slope less than 1 dB/dB on a logarithmic scale. This compression is more pronounced at high level than at low level, which causes the function to have a convex shape. However, at very high sound intensity, the slope is closer to being linear, as is the case near threshold. This compression is not seen for frequencies far from the BF. There is evidence suggesting that the compressive nonlinearity in the BM intensity response is a result of the active mechanical amplification by the outer hair cells (OHCs) (Robles and Ruggero, 2001), since this nonlinearity is always associated with low threshold. A strong support of this theory comes from measurement of BM motion in the post mortem cochlea, in which the HC functions are lost. As is shown in Fig. 1.1.C, as the cochlea deviated from its healthy status, the BM velocity versus level function at the best frequency (10 kHz) underwent a loss in compressive nonlinearity, thus appearing as linear as those off-BF functions. This happened in conjunction with an increase in 2 C D Fig. 1.1. Intensity responses of BM and IHCs in healthy and post-mortem states. A, B. BM velocity-level functions for tones at, below (A) and above (B) the best frequency (10 kHz). The dashed lines indicate linearity (1 dB/dB). The measurements were made on a Chinchilla BM (From Robles and Ruggero, 2001). C. Post-mortem BM velocity-level function from a guinea pig. The measurement was based on Mossbauer technique. As in A and B, the dashed line indicates a linear relationship. The BF of the BM site is 10 kHz (From Sellick et al., 1982). D. Input-output functions of DC receptor potentials of IHCs in guinea pigs. The three solid curves show responses to BF tones and the dashed curves show responses to frequencies at tail regions of the tuning curve (From Patuzzi and Sellick, 1983). 3 threshold (Sellick et al, 1982). Intracellular recordings from IHCs, the acoustic transducers, revealed that in response to tone bursts, IHCs show depolarization in membrane potentials. This receptor potential consists of a direct current (DC) and an oscillating (AC) component. Both components are level dependent. The compressive nonlinearity in the BM input-output function is inherited by IHCs, which introduces additional saturating nonlinearity into the system (Patuzzi and Sellick, 1983). As a result of these cascaded nonlinear transfer functions, the IHC DC receptor potential is virtually a linear function of sound pressure level (SPL) on a dB scale (Fig. 1.1.D). That is, the DC receptor potentials of IHCs achieve a logarithmic operation of sound pressure, for frequencies near the best frequency (BF). This relationship underlies the fact that auditory nerve fibers (ANFs) show driven discharge rates that are approximately in linear relationship to SPL (on a dB scale with respect to sound pressure), not to the sound pressure on a linear scale. The exquisitely wide dynamic range of the auditory transducers seen in Fig. 1.1.D (at least 80 dB SPL) is the basis for the extremely wide hearing dynamic range (Smith, 1988). It is also the basis for the theory that sound intensity can be encoded by the neuronal activity inside a narrow BF region (Sachs and Abbas, 1974; Viemeister, 1988), as will be discussed later. The AN afferent fibers are the only channel through which auditory information in the cochlear can be transmitted to the central auditor system in the brain (Pickles, 1988). In contrast to the wide dynamic ranges of the IHCs, the ANFs respond to increase in sound level with increasing discharge rate only for the first 30-40 dB immediately above threshold (Fig. 1.2, Sachs and Abbas, 1974; Liberman, 1978). Above this range, the discharge rates go into partial or complete saturation, in which further increase in level elicits little or no change in mean firing 4 Fig. 1.2. Rate-level functions of auditory nerve fibers in normal-hearing ears (from Sachs and Abbas, 1974). These rate-level functions with about-equal BFs were recorded from a same cat. A correlation of rate threshold and SR can be seen. The rate criterion of threshold was 20 spikes/s. The high-SR, low-threshold fibers show narrow rate-level dynamic ranges and complete saturation. The low-SR, high-threshold fibers show sloping saturation at higher levels and larger dynamic ranges. 5 rate. These narrow dynamic ranges are mainly the consequences of neural refractoriness and adaptation (Westerman and Smith, 1988). The discrepancy between psychophysical and physiological dynamic ranges at the level of AN is called the dynamic range problem (Smith, 1988). This paradox led to the suggestion that encoding medium-to-high intensities require the spread of excitation to off-frequency AN fibers. There are psychophysical data that support the role of spreading excitation. Schroder et al. (1994) showed that in normal-hearing subjects, adding high-frequency masking noise increases the Weber fractions for tone level discrimination at medium-to-high levels. In fact, the classical explanation of the “near miss” to Weber’s law is the effect of spreading excitation (Carlyon and Moore, 1984). However, some other psychophysical observations suggested that spread of excitation is not necessary for large dynamic range of intensity perception and discrimination. Viemester (1988) showed that even under conditions that were assumed to eliminate spread of excitation, intensity discrimination remained relatively good over a large range of SPL. Therefore, one can conclude that spread of excitation is one, but not the only source of information about intensity at medium and high levels. The additional sources of information probably include the low-spontaneous-rate high-threshold AN fibers (Fig. 1.2, Sachs and Abbas, 1974; Liberman, 1978). Those fibers show partial (sloping) saturation even at levels as high as 100 dB SPL. Low- and medium-SR fibers usually have higher thresholds than the high-SR ones. It has been suggested that AN fibers with similar BFs and different thresholds operate at different portions of the BM input-output function to reproduce its wide dynamic range. Apart from the mean discharge rates, the strength of phase-locking at low frequencies 6 A Frequeny (kHz) B C Fig. 1.3. Nonlinear phase response of the basilar membrane and a temporal code for level at low frequencies. A. Relationships between BM phase response and the tone frequency for a BM site with 10-kHz best frequency at a series of sound levels. The phase-frequency relationship at a given level shows a “bow-tie” shape. The dependence of phase on frequency decreases with increasing level (Adapted from Ruggero et al., 1997). B, C. The Carney (1994) model of cross-AN-fiber phase encoding of intensities of low-frequency tones. This model is based on the phenomenon shown in A. The degree of coincidence gets strengthened at higher levels (B), which can be decoded into the mean discharge rate of hypothesized down-stream coincidence detector type neuron (C). C shows the RLFs of the AN fibers and a model neuron as such. 7 (below approximately 5 kHz) is another possible neural code of sound intensity (Johnson, 1980). However, due to saturation, strength of synchrony carries intensity information only at low levels (Colburn et al., 2003). Another difficulty with phase-locking codes or temporal codes in general, is the lack of a straightforward biological decoding mechanism. The phase relations in a group of ANFs with different BFs have been suggested to be a more plausible code for intensity (Carney, 1994, Fig. 1.3). The nonlinear dependence of phases of AN spike trains on intensity and frequency is inherited from that of the IHC receptor potentials. As shown by Fig. 1.3.A, at high sound levels, the phase of the IHC AC receptor potential varies less with frequency than at low sound level (Chatterjee and Zwislocki, 1998). Consequently, there exists a positive relationship between tone level and the degree of temporal overlapping between spikes of neighboring BF channels, which can be decoded into a rate-based code by a hypothesized postsynaptic neuron of coincidence-detector type. It has been suggested that this type of downstream neuron may be located in the ventral cochlear nucleus (VCN) (Carney, 1994), though its existence hasn’t been proven yet. 1.1.2. Intensity encoding in the central auditory system The cochlear nucleus (CN) is the first processing center in the central auditory system. Both major divisions of the CN, VCN (See Section 1.4 for a more detailed review) and dorsal CN (DCN) show diversity in neuronal types and relatively complicated cytoarchitectonics and circuitries (For review, see Rhode and Greenberg, 1991; Young and Davis 2002; and Young and Oertel 2004). Different types of neurons show different intensity responses. One of the most salient distinctions between rate-level responses of CN neurons and AN fibers is the emergence of non-monotonicity in several types of neurons in both VCN and DCN, which originates from 8 neural inhibition and membrane properties. In comparison, DCN type II and IV units show greatest non-monotonicitiy in the CN (Shoftner and Young, 1985). Their rate-level functions for BF tones usually contain very negative slopes and show clear single-peaked shapes. It should be noted that there are not necessarily any ambiguities in the response of those non-monotonic neurons about absolute levels if the changes in the temporal patterns with level are taken into account, though it is not clear yet how these changes can be decoded. It has been suggested that there may exist a place code for sound level (e.g., Suga, 1977), in which maximum firing rate of a neuron signifies its corresponding “best SPL”. In the CN, DCN type IV units are the most possible candidates for the members of such a coding ensemble. Theoretically speaking, such a code, if exists, requires an ensemble of neurons with best SPLs covering the entire audible range of intensities. However, Young and Brownell (1976) showed that the distribution of best SPLs of type IV units is biased toward low levels. But as we will discuss in the following sections, there is strong evidence supporting a place code at higher levels of the central auditory pathway. As is in the CN, there is a mixture of monaural monotonic and non-monotonic tone rate-level functions in the central nucleus of the IC. (e.g., Ramachandran et al., 1999; Rees and Palmer, 1988). Some of the single-peak rate functions show very finely tuned best SPLs, in that the driven discharge rates decrease sharply toward zero when level deviates from the best SPL. The range of the best SPLs has significantly expanded from that of the CN, supporting the general hypothesis that a place code for intensity is gradually forged along the ascending pathway. Unlike in the AN and CN, the average BF-tone rate-level function in the IC shows saturation above 20 dB re threshold (Ehret and Merzenich, 1988), which reflects the large 9 proportion of non-monotonic neurons and a relatively uniform distribution of thresholds and best intensities. However, it is still possible that a subset of the IC neurons encode intensity with their summed rates, since it has been shown that some IC neurons show dynamic ranges (defined as the level difference between the threshold and the saturation of nonmonotonic turning point) can be as large as 80 dB (Rees and Palmer, 1988). In the medial geniculate body of the thalamus, in addition to the fact that there is a mixture of monotonic and non-monotonic rate-level functions (Clarey et al., 1991), a large proportion of the monotonic neurons show broad dynamic ranges of about 80 dB (Rouiller et al., 1983). Similarly broad or even broader dynamic ranges have been observed in the primary auditory cortex (AI) of behaving animals (Pfingst and O’Connor, 1981). Best SPLs of the simple non-monotonic neurons have been show to cover SPL ranges wider than 80 dB in echo-locating bats (Suga, 1977) and cats (Heil et al., 1994). It was also found that in cat AI, quantitative measures of rate-level responses, including thresholds, best SPL, negative slopes and dynamic ranges are not only correlated with each other, but also spatially distributed in an orderly manner. In response to increasing tone level, both local firing rates and spatial distribution of excitation show systematic changes. Interestingly, if one takes the absolute value of the derivative of the individual rate functions, average and integrate them with respect to SPL, the resultant function matches the human loudness growth curve very well (Heil et al., 1994). Thus, on the thalamic and cortical levels, there are data supporting all the major hypotheses of intensity encoding. Which one(s) actually participate(s) in intensity encoding is a question remaining to be answered. One important issue in studying the neural encoding of sound in the central auditory system 10 is the effects of anesthesia. Anesthesia alters the response properties of neurons on many levels of the central auditory pathway. In the VCN, May and Sachs (1992) observed that relative to the awake state, barbiturate anesthesia significantly altered the responses to tones in background noise, but not in quiet. The anesthesia also reduced dynamic ranges for response to vowels in the VCN (May et al., 1998). In the DCN, Young and Brownell (1976) observed that the strongly inhibitory types of response maps (including type IV and V) are commonly seen in decerebrated unanesthetized animals, but rarely in anesthetized ones. Kuwada et al. (1989) reported drastic suppression of responses of the IC neurons caused by barbiturate anesthesia. In the auditory thalamus and cortex, anesthesia causes changes in spontaneous and driven activities (often, but not always suppression) and reduction in the frequency response area (Zurita et al., 1994; Gaese and Ostwald, 2001). Thus one must be careful in choosing a proper anesthetic state and interpreting the experimental data when recording neural activities from central auditory nuclei. 1.2. Loudness recruitment and its psychophysical properties Sensorineural hearing loss, (SNHL), the subtype of hearing loss which originates from damage to the cochlea, auditory nerve or central auditory neurons, is associated with upward shifts in hearing threshold (reviewed by Salvi et al., 1983). Loudness recruitment is an abnormal intensity perception often accompanying SNHL. It is characterized by an abnormally rapid increase of loudness with intensity (Zeng and Turner, 1991; Buus and Florentine, 2001; Moore et al., 2004). In patients with SNHL, the absolute threshold for hearing is elevated in the impaired frequency regions. Any tone with intensity below the threshold is inaudible. A tone with intensity slightly above the threshold sounds softer in the impaired ear than it does in the normal ear. 11 However, when the intensity of the tone is increased above threshold, the difference of loudness in the normal and deaf ears gets smaller. In some cases, at and above certain high absolute level (approximately 90 dB SPL), the loudness in the deaf ear catches up with that in the normal ear, and this is called a “complete” recruitment (Fig. 1.4.A). In other case, the loudness in the deaf ear is always below that in the normal ear, which is termed “incomplete” recruitment or “under-recruitment” (Fig. 1.4.B). Cases in which at certain intensity the loudness in the deaf ear exceeds that in the normal ear are called “over-recruitment”. All three cases are seen in psychophysical measurements in patients with SNHL. Most of the previous psychophysical measurements of loudness recruitment have been based on pure tone stimuli, because of its simplicity and its frequency specificity. Loudness recruitment has been demonstrated by a number of psychophysical paradigms. In one paradigm called absolute magnitude estimation, subjects were required to assign a positive numbers with no imposed upper bounds to their perceived loudness. A similar paradigm called absolute magnitude production requires subjects to adjust the intensity of a tone to match its loudness to a given number (e.g., Hellman and Meiselman, 1990). In both these paradigms, loudness is represented by a number (often the geometric mean of results from individual trials) and loudness-versus-intensity functions can be obtained in different frequency regions. In regions with threshold shifts, the slopes of the functions were observed to be substantially steeper than those in the normal regions. In addition, a positive correlation was observed between the loudness-intensity slopes and the threshold increase measured in dB (Hellman and Meiselman, 1990, 1993). In a paradigm called cross-modality matching (e.g., Hellman, 1994), the subjects were 12 A B Fig. 1.4. Loudness balance curves in loudness recruitment (from Moore, 2004). A. A loudness balance curve obtained by a binaural method. The right ear has a 70 dB HL at 2.5 kHz, while the 2-kHz threshold in the left ear was about normal. The impaired ear shows loudness catching-up (complete recruitment) at about 100 dB SPL. B. A loudness balance curve obtained with the monaural cross-frequency method. The ear has a ~80 dB HL at 6 kHz, while the threshold at 1 kHz in the same ear was about normal. The curve shows under-recruitment. The dashed diagonal curves in the two plots show equality of loudness-intensity relationships in the two ears. The slopes of both loudness balance curves are greater than 1 dB/dB, which is a hallmark of recruitment. 13 instructed to adjust the intensity of a tone to “match” its loudness to the apparent visual length of a line segment simultaneously presented. This method gave length-versus-intensity functions, which are approximately straight lines on a log-length scale. These functions showed greater slopes in the deaf frequency regions than in the normal ones. Another measurement paradigm called loudness matching or loudness balance, can be done for two tones of different frequencies or in the two ears (e.g., Miskolczy-Fodor, 1960; Zeng and Turner, 1991; Moore, 2004). In these two tone loudness balance paradigms, one tone is in the frequency region with normal or near normal threshold, while the other falls into the impaired frequency region. The subjects are instructed to compare the relative loudness of the two tones and judge which one is subjectively louder than the other. The louder one is lowered in intensity or vice versa. And this process is repeated until the subject indicates that he or she hears the two tones as equally loud. The data from these experiments are usually presented in the format of binaural or cross-frequency loudness matching curves, wherein the abscissa is the level (in SPL or SL) in the impaired region and the ordinate is the level in the normal region. As shown in Fig. 1.4., in loudness matching curves, recruitment is manifested as slopes greater than 1 dB/dB, and complete recruitment is manifested as the loudness-matching curve contacting the diagonal. There have been only a few studies on the existence and characteristics of loudness recruitment in animals with SNHL. None of the aforementioned psychophysical paradigms can be easily applied to behavioral measurements of loudness perception in animals. In rhesus monkeys (Stebbins, 1966; Pugh et al., 1979), reaction time has been shown to be a reliable behavioral correlate of loudness. On average, the time delay between the stimulus onset and the 14 motor reaction of the animal (e.g., the release of a previous held operant) is a decreasing function of intensity. Pugh et al. (1979) trained macaque monkeys to perform an operant conditioning paradigm and compared the reaction-time-versus-intensity functions before and after the onset of an acoustic trauma. They observed that the decrease of reaction time with intensity was more rapid after trauma, which suggested the existence of recruitment. In ears with sloping hearing loss, wherein only certain frequency regions (e.g., frequencies above 10 kHz) are affected, recruitment is seen only in the impaired region (Zeng and Turner, 1991). In other words, loudness recruitment is usually associated with threshold elevation. In addition to these measurements using static long-duration tones, dynamic aspects of the steeper loudness-intensity relationship in recruitment has also been shown. Moore et al. (1996) used a scheme of binaural matching of amplitude modulation (AM) depth, which is a straightforward extrapolation of the binaural loudness matching method. The carrier frequencies of the AM tones in the two ears are within the impaired frequency region in the ear with hearing loss. The overall loudness of these two AM tones were matched. The subjects were instructed to adjust the AM depth in one ear so as to match the “coarseness” of the two signals in the two ears. The results indicated that a greater AM depth in the normal-hearing ear was required to match a smaller AM depth in the impaired ear. This phenomenon was relatively invariant with modulation rate. More interestingly, the authors showed that the AM depth matching data fit with predictions based on pure-tone loudness matching curves reasonably well. This result indicates that loudness recruitment is relatively invariant with temporal scale and it is based on a relatively fast acting neural mechanism. 15 Recently there has been some debate on the proper definition of loudness recruitment. The “classical” definition of steeper-than-normal loudness-intensity functions has been challenged. Buus and Florentine (2001) measured the growth of the loudness of a multiple tone complex in the impaired frequency region of five subjects with hearing loss, and observed that the rate of loudness growth with intensity near threshold is similar to normal. Also, based on results from the multiple-tone scheme, they argued that the loudness at threshold is non-zero and greater than normal in the impaired frequency region. Based on a theoretical model, they also predicted that this abnormal increase in loudness at threshold is proportional to the amount of threshold elevation. These evidences led the authors to suggest that an abnormally fast growth of loudness with intensity is not the most essential property of recruitment; instead, recruitment is better defined as an abnormally high loudness at threshold, called “softness imperception”. The reduced loudness difference at a given intensity in the deaf ear and the normal ear (or in the deaf and normal frequency regions) can be at least partially attributed to this upward shift of the loudness-versus-intensity curve associated with threshold elevation. The authors also observed that at moderated sensational levels (SLs), the rate of loudness increase with intensity was faster than normal, which they argued was possibly due to OHC dysfunction and loss of BM compressibility. However, using a pure-tone based binaural loudness matching paradigm, Moore (2004) showed that loudness matching near the threshold didn’t require a substantially higher SL in the normal ear. In fact, equal SL in the two ears near threshold produced similar loudness. This result argued against the notion of softness imperception by Buus and Florentine (2001) and attributed recruitment to a faster-than-normal rate of loudness growth above threshold. 16 1.2.1. Intensity discrimination in recruitment The considerably faster growth of loudness with intensity in recruiting ears, whether due to steeper single-neuron rate-level functions or due to abnormally rapid spread of excitation, led many to suggest that ears with SNHL should be more sensitive to level changes than normal ears are. However, many psychophysical studies observed the just-noticeable differences (JNDs) of level were often found to be normal or near-normal in previous psychophysical studies (e.g., Florentine et al., 1993; Neely & Allen, 1997). Even in a few cases in which the level JNDs were found to be slightly smaller in recruiting ears compared to normal ones at equal SLs, the ratio between the JNDs under two hearing conditions is far less than that would be expected solely based on the ratio between loudness-intensity slopes (e.g., Schroder et al., 1994). It has been suggested that increased variability of neuronal discharge rates (“internal noise”) counteracts the effect of the larger loudness-intensity gains on intensity JNDs after hearing loss. This idea is based on the fact that the discriminability of two stimuli is dependent not only on the difference in means but also on the variability, as described by the d’ measure. Heinz et al. (2005b) tested this hypothesis by fitting power functions to the variance-mean relationships of ANF discharge rates in normal and traumatized ears and found that the powers fits were about equal in different hearing states, which did not support the idea that AN fibers show more rate variability after hearing loss. However, it is still likely that increased internal noise, if indeed exists, primarily originates from central neuronal abnormalities after hearing loss (e.g., Morest et al., 1997; Lee et al., 2003; Wang and Manis, 2005). The current study will provide an opportunity to test the increased-internal-noise hypothesis in the VCN, the first stage of the central auditory system. 1.3. Neuropathophysiology in SNHL and physiological correlates loudness recruitment 17 Acoustic trauma causes damages to the cochlear sensory epithelium and the loss of or damage to the hair cells. The physiological effects of these damages can be seen in ANFs, which show alterations in thresholds, tuning curves and rate-intensity response after trauma. In BM sections where there are substantial HC losses or damages, the BF-matched ANF thresholds are always elevated. There is a correlation between the degree of OHC loss and the amount of threshold increase (Liberman and Kiang, 1978). However, HC loss isn’t always a good predictor of the auditory threshold at the corresponding BF region. In fact, substantial increases in ANF thresholds were often observed in BF regions which corresponded to BM sections with minimal hair cells losses (Liberman and Dodds, 1984b; Liberman and Kiang, 1978). Liberman and Dodds (1984b) showed that damages to the stereocilia atop the inner and outer hair cells are a better correlate of threshold shift and abnormal tuning in ANFs. In noise-exposed ears, the stereocilia of HCs in the damage frequency regions show disorganized and “clumped” shapes, in contrast to the orderly “organ pipe” shape in normal HCs. The stereocilia of the IHCs and first row of OHCs are the most vulnerable to sound overexposure. Damages to stereocilia of inner and outer HCs have different effects on ANF receptive fields. While the anomalies in IHC stereocilia are correlated with hyposensitivity and elevated thresholds at both the tip and tail regions of the tuning curve, OHC stereocilia abnormalities are seen with broadened tuning and occasionally hypersensitive tails (Liberman and Dodds, 1984b) The most frequently cited hypothesis on the physiological origin of loudness recruitment is that ANF firing rates increase faster with intensity in impaired ears, i.e., the steepened-slope hypothesis. Liberman and Kiang (1984) studied the correlation between noise-induced cochlear damage and rate-level functions (RLFs) of ANFs. ANFs show two modes of firing, one called 18 component-1 (C1) response, and the other called component two. At intensities from threshold to 85 - 90 dB SPL, only C1 response can be seen. C2 response dominates at intensities higher than 85 – 90 dB SPL. RLF nonmonotonicity (dips) and phase discontinuity can be seen at the transition points between the two components. ANFs which show elevated threshold at both tip and tail, which suggested IHC damage, showed substantially diminished (C1) responses, while component-2 (C2) responses in this type of units often remained normal. The study by Heinz and Young (2004) based on a similar animal model of NIHL confirmed these observations. In addition, they discovered that in impaired ANFs, tuning curve shape was a fairly good predictor of the rate-level slope changes. Impaired tuning curves could generally be categorized into two groups: one “sharp” group with sharp tuning tips preserved, and a second “broad” group with tuning tips loss. In the first “sharp” category, the slopes were either shallower than normal above threshold, reflecting attenuated C1 response, or very steep but only at high levels (> 80 dB SPL), which presumably came from fibers with only C2 response left. In the second “broad” category, the slopes generally didn’t differ from normal values. When the low-level slopes (slopes of the RLF portion immediately above threshold) were compared across the traumatized and normal-hearing groups, the slopes were shallower in the traumatized group for tones (See Fig. 1.5). These observations are intuitively contradictory to the observation that basilar membrane input-output functions are steepened after cochlear damage, caused by loss of the active mechanical amplifications by the OHCs (Zhang and Zwislocki, 1995; Robles & Ruggero, 2001,). However it is important to remember that level information on the organ of Corti has to be conveyed to the brain by the transducer function of the IHCs. Previous studies have 19 Fig. 1.5. Rate-level slopes of auditory nerve fibers before and after acoustic trauma (From Heinz and Young, 2004). An acoustic trauma was induced in cats by 4-hour exposure to a 2-kHz-centered narrow band noise. Rat-level slopes for BF tones (A) and broadband noise (B) are plotted against fiber BF, for the normal-hearing, mild and moderate/severe HL groups. The distribution functions of slopes for BF tones (C), broadband noise (D), 1-kHz tone (E), and the synthetic vowel /ε/ (F) are shown. For all the stimuli except the vowel, ANF rate-level slopes for tones were shallower than normal after hearing loss (A, C, E). The shallower slopes reflect damages to the IHCs, which lead to reduced transducer functions. Close-to-normal or steeper slopes were observed for BBN and vowel /e/ (B, D, E), which presumably reflected reduced cochlear suppression after cochlear damage. A B C Fig. 1.6. A schematic of effects of damages to inner and outer hair cells on the rate-level slopes of auditory-nerve fibers (From Heinz et al., 2005). A. n: normal.. i: impaired. In acoustically traumatized cochleas, the BM input-output functions are steeper than normal due to the loss of the compressive nonlinearity. B. h: IHC damage. Damages to the IHCs and their stereocilia suppress receptor potentials of the IHCs. C. Solid and dashed curves show RLFs of high and low/medium SR fibers, respectively. The joint effect of OHC and IHC damage are elevated thresholds and diminished rate responses. The vertical lines in A and C indicate the thresholds of the BM compressive nonlinearity. 20 demonstrated that damages to the IHCs diminish ANF responses. For example, Sewell (1984) observed reduction in C1 responses caused by the administration of furosemide, a drug that reduces the transducer function of the inner and outer HCs. Mixed IHC and OHC damages are seen in ears with acoustic trauma (e.g., Liberman and Dodds, 1984). Therefore the effect of acoustic trauma on rate-level encoding in ANFs has to be considered in the context of damages to both types of hair cells, which have opposite effects on ANF rate-level slopes (Fig. 1.6). While pure OHC damages lead to steepened BM input-output functions, IHC damages suppressed the output of the auditory transducers. In cochlear damage with selective and complete OHC destruction (Harrison, 1981; Schmiedt and Zwislocki, 1980) by ototoxic drugs, the authors have reached conclusions that rate-level slopes were steepened in damaged ears. In contrast, other studies in acoustic trauma (Liberman and Kiang, 1984; Salvi et al., 1983; Heinz and Young, 2004) and drug-induced cochlear damage of mixed OHC/IHC damages observed nearly normal or shallower-than-normal slopes. Therefore, if it is assumed that firing rates of ANFs is the primary correlate of loudness, one may conclude that in acoustic trauma, neural substrates of loudness recruitment generally cannot be found in single ANFs. Thus the steepened-slope hypothesis of the physiological mechanism of recruitment is largely questionable on the level and ANF (Heinz and Young, 2005). As such, neural correlates of recruitment have to be searched for either in other aspects of ANF discharges or on higher levels of the auditory pathway. With regard to the first possibility, there are two additional post-traumatic changes in AN physiology, which have been hypothesized to be the causes of recruitment. Both hypotheses depend on the idea that loudness is determined not by listening to the discharge of single fibers, 21 but by looking at the summed rate activities of a group of fibers. This idea is natural corollary of the solutions to the dynamic range problem (Smith, 1988, see Section 1.1). In cochlear damages, it has been suggested that the range span of the thresholds of BF-matched neurons are compressed, leading to steeper-than-normal growth of total firing rate with intensity (Moore et al., 1985; Zeng and Turner, 1991). This hypothesis has been negated by the study of Heinz et al. (2005), which showed that ANF threshold distributions were in fact significantly more dispersed after acoustic trauma. In addition to the compressed-threshold-distribution hypothesis, the broadened ANF tuning curves have been suggested to lead to faster-than-normal spread of neural excitation among different BF channels with increasing tone level (Kiang et al., 1970; Evans, 1975). This hypothesis is related to the theory that the correlate of the level of a tone is the summed firing rates of all ANFs from different BF channels, which has been questioned on the ground of several psychophysical (Viemester, 1988) and physiological studies (Relkin and Doucet, 1997). Heinz et al., (2005) observed that spread of excitation was faster than normal in the moderate/severe HL ANF population. However, when summed (or equivalently, average) rate-level functions for a 2-kHz tone stimulus from AN fibers within a wide or narrow BF windows were generated, these curves didn’t show strong trends of steepening in impaired ears (Fig. 1.7, Top: wide BF range; Bottom: narrow BF range). In addition to these hypothesis based on summed firing rates, the hypothesized cause of recruitment by Carney (1994), which is based on the idea that the degree of synchrony among ANFs of different BFs increases faster with level in damaged cochleas, has been tested based on the same set of data (Heinz et al., 2005b). The non-linear phase-BF relationship was observed to be lost in the impaired ear, which contradicts with the basic assumption of the Carney model. 22 Fig. 1.7. Average rate-level functions of auditory-nerve fibers before and after acoustic trauma (from Heinz et al., 2005). Top. The stimulus is a 2-kHz tone. Fibers with BFs between 0.25 and 9.2 kHz were included. A. Average RLFs of normal-hearing, mild and moderate/severe HL groups plotted on a linear rate scale. B. Rate-matching curves between normal and two HL groups. The dotted curves show results when fibers with BFs greater than 4 kHz were excluded. The black dashed curve is a schematic of a typical loudness-balance curve measured in recruiting ears. C. Average RLFs as in A plotted on a log-rate scale. D. Log-rate-level functions aligned at threshold (on a SL scale). Bottom. The stimulus is also a 2-kHz tone. Only those fibers with BFs falling into a 0.4-octave BF region centered at 2 kHz were included. The format is the same as in Top. 23 There is a copious body of literature on the effects of SNHL on central auditory neurophysiology, many of which contain findings which are very likely to be pertinent to the problem of abnormal intensity perception in SNHL. An increasing body of studies in the central auditory nuclei of deaf animals is lending support to the idea that central auditory responses following SNHL cannot be predicted based solely on peripheral changes (reviewed by Syka, 2002). Loss and re-growth of neurons, axon sprouting, and modification of synapses happen in the central auditory nuclei following cochlea ablation or noise overexposure (e.g., Michler et al., 2002; Illing et al., 2005; Vale & Sanes, 2002; Wang & Manis, 2005). The plasticity has been suggested to be at least partly a homeostatic self-adjustment of which the goal is maintaining a relatively constant level of overall neuronal activity by up-regulating the central gain (e.g., Davis and Bezprozvanny, 2001; Schaette and Kempter, 2006). It has been suggested that recruitment is a consequence of over-compensation by these homeostatic adjustments. A number of previous studies on the evoked response (ER) in central nuclei of hearing-impaired animals lend strong support to this idea. Tab. 1.1 gives a summary of the published physiological studies in the central auditory pathways of animals with SNHL. These studies were based on different types of cochlear damage, including congenital (Bock et al., 1982; Sterbing and Schrott-Fischer, 2002), ototoxic drugs (Qiu et al., 2000), cochlear ablation (Gerken, 1979) and acoustic trauma (Saunders et al., 1972; Popelar et al., 1987; Salvi et al., 1990; Syka et al., 1994; Szczepaniak and Moller, 1996, Wang et al., 1996), and conducted in different animal species. However, they invariably showed enhanced responses and sensitivity on certain stages of the auditory pathway. In the cases where the cochlear is partially damaged by exposure to a narrow-band noise, these steepened amplitude-level function were observed only in the 24 Name of Animal Experimental Central Type the study species method auditory SNHL course and stimuli structure of Time Relevant discoveries (s) Saunders et al., 1972 Mouse ER, click CN, IC Acoustic trauma Permanent Gerken, 1979 Cat CN, SOC, IC, MGB Bilateral cochlear ablation Permanent LonsburyMartin & Martin, 1981 Rhesus monkey Behavioral detection of electrical stimulation Behavioral experiments and neural recording, tones CN, IC Acoustic trauma (< 20 min) Acute Bock et al., 1982 Mouse ER to AN electrical stimulation IC Congenital, dn/dn gene Congenital Popelar et al., 1987 Guinea pig CAP and ER, clicks and tone bursts AN, IC, AC Acoustic trauma Acute & Permanent( 1h-2w) Salvi et al., 1990 Chinchill a ER, tone bursts IC Acoustic trauma Permanent Boettcher & Salvi, 1993 Chinchill a Single-neuron recording, tones on and off BF VCN Acute Syka et al., 1994 Guinea pig ER, clicks and tone bursts AC Acoustic trauma (3-5 min exposure) Acoustic trauma Sczczepan iak & Moller, 1996 Rat ER, clicks IC Acoustic trauma Acute (1.5 h) Wang et al., 1996 Chinchill a Single-neuron recording, tones on and off BF IC Acoustic trauma (~ 30 min exposure) Acute Qiu et al., 2000 Chinchill a CAP, ER, tone bursts AN, IC, AC Drug induced, carboplatin Permanent Salvi et al., 2000 Chin-chil la CAP, ER, tone bursts AN, IC Acoustic trauma Permanent CN, 25 Acute (1h-3d) Larger-than-normal maximum ER amplitude, and steeper ER amplitude-level functions in both CN and IC. “catch-up” at about 90 dB SPL. Decreased behavioral detection thresholds for electrical stimulations in CN, SOC, IC and MGB. 64% CN and IC neurons showed diminished discharge rates; the rest showed higher-than-normal rates at high sound levels. Faster increase of IC ER with increase stimulation intensity than in control strains. Steeper ER amplitude-level functions and larger maximum amplitude in AI 1h after exposure. Similar changes not seen in AN and IC. Steeper ER amplitude/level functions in IC, only in impaired frequency regions. Larger maximal response amplitude. Reduction in inhibitory areas, enhanced rate-level responses in about 25% of the recorded neurons. Steeper AC ER amplitude-level functions and larger maximum responses. Selective hyperexcitability of IC ER. No evidence for overexcitability in CN and SOC. Altered tuning curve shapes, loss of inhibitory responses, and enhanced rate-level responses in about 40-50% of the neurons. Steeper-than-normal ER amplitude-level functions in AC. Shallower-than normal in AN and IC. Steeper-than-normal ER amplitude-level functions in IC, but not in CN and ANF. Sterbing & Schrott-Fi scher, 2002 Norena et al., 2003 Mouse CAP, tone bursts IC Congenital, Kit(W-V) Congenial Steeper single-unit BF-tone RLFas in IC. Cat Single-unit, tone bursts AC Acoustic trauma Acute Altered temporal patterns of neural discharges and increased maximum firing rates. Tab. 1.1. Selected previous studies on central neural overexcitability in animals with SNHL. 26 impaired frequency range (Salvi et al., 1990; Syka et al., 1994). More interestingly, a few studies showed that stronger responses on higher central nuclei can be observed without accompanying response enhancement at lower anatomical levels. For example, of the three structures (AN, IC and AI) studies by Popelar et al. (1987), only AI showed steepened ER amplitude-level function 1 hour after noise exposure. There have been two functional imaging studies in human subjects with SNHL. Morita et al. (2003) observed that magnetoencephalographic signals from the auditory cortices grew with a faster rate with increasing tone level in patients with cochlea damage than in normal-hearing control. Langers and van Dijk (2007) reported faster increase of summed auditory cortical blood-oxygen-level-dependent signal with tone intensity in patients with recruitment. These studies, no matter which methodology was used, all pointed to the existence of strong central neural plasticity (Turrigiano, 1999) following the onset of hearing loss, which makes alterations in neural encoding of sound level different in the central auditory nuclei largely unpredictable from ANF level encoding after acoustic trauma. All these studies show central hypersensitivities and response growth akin to complete or over-recruitment. Although yet to be substantiated, these phenomena are very likely to be the physiological substrates of loudness recruitment. A consensus on the loci of the post-trauma plasticity and overexcitability hasn’t been reached. Some studies pointed to as early as the CN, others indicated little change below the level of AI. Saunders et al. (1972) showed significantly faster increase in ER amplitude with level in both CN and IC of mice with acoustic trauma. Gerken (1979) bilaterally ablated cochleas in cats and found decreased behavioral detection threshold for stimulation in auditory brainstem 27 structures including CN, SOC, IC, and the medial geniculate body. In contrast, Popelar et al. (1987) recorded AN CAP, along with ERs in IC and AI one hour after noise exposure in guinea pigs and observed recruitment-like overexcitability only in AI, but not in AN and IC (Fig. 1.8.A). Szczepaniak and Moller (1996) observed that only the peaks in IC ER which reflected neuronal activity in IC per se increased beyond normal level, while the other shorter-latency peaks which reflected inputs from lower structures including CN and SOC didn’t. In mice treated with carboplatin, Qiu et al. (2000) observed enhanced ERs in AI, but not in IC and AN. Possible explanations for these different observations include (1) different animal species; (2) different types of cochlear damages used, and (3) different durations elapsed between cochlear insult and recording experiment; (4) different electrodes and stimuli used in the ER recording experiments. Single-unit neural recording afford better spatial resolutions and can better locate the neuronal substrates of the central overexcitability, and elucidate the physiological origins of recruitment mechanistically. There have been only a few single-unit neurophysiological studies in subcortical auditory nuclei in hearing-impaired animals. In the DCN, many neurons in acoustically traumatized cats show receptive field shapes unseen in normal animals (Ma & Young, 2006), which suggests weakening of inhibition. Wang et al. (2006) studied the response properties of single units in the chinchilla IC before and after a short-duration (25-30 min) intense noise exposure. They showed that exposure to intense tones with frequencies above the unit BF significantly altered tuning-curve shapes and expanded excitatory regions in the response maps. In general, these changes could be explained by a loss of side-band inhibition. Rate-level responses were found to be substantially enhanced in about 50% of the IC neurons. Sterbing and Schrott-Fischer (2002) observed significantly steepened single-neuron and population average 28 A B Fig. 1.8. Overexcitability of central auditory nuclei in sensorineural hearing losses. A. The CAP and ER amplitude-level functions during the time course of recovery from a temporary noise-induced threshold shift in guinea pigs (From Popelar et al., 1987). In the auditory cortex, faster growth of ER amplitude with level and abnormally larger responses at high levels were seen along with temporary threshold shift, akin to loudness recruitment. This overexcitability gradually disappeared as the animal recovers from the temporary threshold shift. In contrast, overexcitability was not seen in AN CAP and ER in IC, both of which showed diminished response amplitudes and slightly shallower amplitude-level functions. B. Average normalized BF-tone discharge-rate-versus-level function of IC neurons, in three strains of mice with different hearing states. The NMRI and C57BL/6J mice had normal or close-to-normal hearing, whereat the KitW-v mice had a congenital deafness due to the lack of normal OHCs. The average IC RLF of the KitW-v mice was substantially steeper than in the other two strains. 29 rate-level functions in IC of rats that lack OHCs from birth due to genetic defects (Fig. 1.8.B). On the level of cortex, reorganization of tonotopic maps and change in temporal patterns of neuronal discharge and increased maximum responses have been shown on the single-neuron level after acoustic trauma (Rajan et al., 1993; Norena et al., 2003; Norena and Eggermont, 2005). To our knowledge, so far there have been only two single-unit physiological studies conducted in the VCN in animals with SNHL. Both studies were based on acute HL induced by short-duration acoustic over-stimulation. Lonsbury-Martin and Martin (1981) studied the effect of temporary auditory threshold shift induced by short-duration (3 minutes) exposure to intense tones, and found that about two thirds of VCN neurons show diminished rate-intensity responses during the threshold shift, while the other one third showed enhanced responses at high levels. However, since they didn’t classify the recorded VCN neurons into response types, it is difficult, if not impossible, to know the possible anatomical substrates of these two different types of post-traumatic changes. Boettcher and Salvi (1993) studied the effect of acute exposure (5 min in duration) to pure tones at frequencies 0.5 octave above the BF of a recorded neuron on the response properties of this neuron. VCN neurons were classified into different types based on PSTH patterns. It was found that 1) the shapes of PST histograms were not significantly altered by short-duration intense tone exposure, 2) most of the units with inhibitory supra-BF sidebands which were inhibited by the exposure stimuli showed lost or weakened supra-BF inhibition following exposure, 3) most units which had no supra-BF inhibitory regions and thus were excited by the exposure stimuli showed weaker responses to tones above BF after exposure, 4) about 25% percent of the recorded neurons showed enhanced rate-level responses to BF tones 30 after exposure. The mechanism of weakened inhibition was not entirely clear. It may be due to HC damages, depletion of inhibitory neurotransmitters, or other mechanisms. That study didn’t make association between PSTH types and patterns of post-exposure alterations. From these two previous studies it was not clear what effect permanent HLs have on VCN neurophysiology, but they gave some support to the post-acoustic-trauma neural plasticity in VCN. The long-term effects of permanent HL on the single-neuron responses in the VCN have not been studied to date. These effects are likely to differ from the short-term effects reported in the aforementioned two studies, because the recovery from temporary threshold shifts and excitotoxicty and possible neural plastic changes such as axonal sprouting and neuronal degeneration / regeneration will presumably alter neuronal responses to sound dramatically. The current study aims to investigate these effects systematically. 1.4. Anatomy and physiology of the ventral cochlear nucleus in the cat The cochlear nucleus (CN) is the most caudal structure in the central auditory pathway. All AN afferent fibers terminate in the cochlear nucleus. Based on cell morphology, the CN can be divided into several major parts. In the cat, the ventral cochlear nucleus (VCN) is the largest division of the CN in terms of volume and cell count. The VCN can be further divided into anterior and posterior parts based on innervation patterns by the AN. The anteroventral cochlear nucleus (AVCN) receives AN projections through the ascending branch, while the posterior cochlear nucleus (PVCN) receives AN projections through the descending branch (Young and Oertel, 2004). Like the cochlea and other central auditory nuclei, the VCN is tonotopically organized. 31 Bourk et al. (1981) constructed a tonotopic map of the cat VCN based on a block anatomical model. In general, there is a gradient from high BFs to low BFs when one proceeds from the dorsal and medial aspects to the ventral and lateral aspects. Similar BFs are organized into imaginary planar structures called the iso-frequency laminas. The VCN contains several types of principal neurons, which differ in their location, morphology, membrane properties, input and projection patterns, electrophysiology and functions. Each category of principal neurons constitutes a complete representation of the entire tonotopic range. Different types of neurons in the VCN are thought to constitute multiple parallel information processing channels and specialize in processing different aspects of acoustic signals (Young and Oertel, 2004). The tuning curves of VCN neurons resemble those of the AN fibers (Bourk, 1976; Rhode and Smith, 1986). BFs and thresholds can be defined in VCN neurons. PSTHs can be obtained by using multiple presentations of a tone at the BF and at a certain level above the threshold. Different VCN principal neurons show distinct patterns of PSTHs. PSTH has become the standard method of classifying neurons in in vivo electrophysiological studies of the VCN (Bourk, 1976; Young et al., 1988; Blackburn and Sachs, 1989). Blackburn and Sachs (1989) devised a logic decision tree which can be used to reproducibly classify VCN PSTHs into different types (Fig. 1.9). This is based on quantitative measures such as the coefficients of variation (CV) of interspike intervals (ISI) and minimum first-spike latencies. It minimizes the dependence of the classification process on subjective judgments. The classification of neurons in the current study will be basically based on this decision tree. In the following sections, we are going to briefly review the anatomical and physiological 32 Fig. 1.9. The decision tree for classifying VCN principal neurons based on BF-tone PST histograms. Major PSTH types were first divided into onset, primary-like and multimodal (Chopper) categories. Inside each category, further divisions were made according to criteria based on first-spike latencies, regularity of interspike intervals, and firing rates. 33 properties of the major types of VCN principal neurons, and their possible functions in auditory processing. Detailed reviews on these topics can be found in Young & Davis (2002) and Young & Oertel (2004). 1.4.1. Bushy cells The bushy cells (BCs) are among the most numerous type of neurons in the VCN. They are located in the AVCN. The two subtypes of busy cells, spherical and globular bushy cells (SBCs and GBCs) are distinguishable based on cell morphology. SBCs and GBCs share a common morphological feature of short and “bushy” dendritic trees. However, the dendritic trees of the GBCs are more extended and diffuse compared with those of the SBCs. One important feature of the SBCs is the morphology of the synapses formed by afferent fibers onto them. These synapses, which are called endbulbs of Held, are large, paw-like structures which are formed almost exclusively on the somata of the SBCs. Each endbulb covers a large portion of the somatic surface and contains hundreds of synaptic contacts, which makes the endbulbs extremely secure synaptic endings (Ryugo and Sento, 1991). Each input spike from the ANF elicits a post-synaptic spike in the SBC with a probability close to one. The electrophysiological manifestation of these large endbulbs are the prepotentials which invariantly proceed real action potentials (Bourk, 1976) and can serve as a classification criterion for bushy cells (primary-like units, see below). Compared with the SBC endbulbs, the synapses formed by ANFs on GBCs, which are called modified endbulbs, are smaller and not exclusively found on the somata, but also on proximal dendrites (Tolbert and Morest, 1982). The ANF convergence ratios are different in the SBCs and GBCs. On average, 3 AN fibers converge onto one SBC; while the number ranges between 4 and 40 for a GBC (Ostapoff and Morest, 1991; Ryugo & Sento, 1991). In addition to excitatory 34 inputs from the AN, SBCs and GBCs also receive inhibitory inputs, whose origins are not totally clear but probably include the D-stellate and vertical cells in the cochlear nucleus (See section 1.4.2). These inhibitory synapses are mainly formed on the dendrites (Ostapoff and Morest, 1991). The SBC has been associated with primary-like (PL) PSTH shapes. As shown in Fig. 1.10.A, primary-like PSTH are characterized by a relatively high discharge rate near stimulus onset and a gradual decrease (adaptation) of discharge rate (Fig. 1.10.A). This firing pattern closely resemble that of the ANFs, which reflects the one-spike-in, one-spike-out processing mode of the SBCs. The PSTHs of the GBCs differ from the primary-like ones in that there is a precisely timed first driven spike followed by a pause or a notch 1 to 2 ms in duration, which is called primary-like-with-notch (Pri-N) PSTH type (Fig. 1.10.B). This pattern is caused by the convergence of a large number (4-40) of ANF inputs on each GBC. The other aspects of the primary-notch PSTHs are indistinguishable from primary-like. In adult animals, the glutamate receptors found on the bushy cells are mostly of the AMPA type (Bellingham et al, 1998). It has been shown that the kinetics of these auditory AMPA receptors is exceptionally fast and their unitary conductances are large. The time constant the synaptic currents are less than 1 ms, which is the fastest known kinetics anywhere in the central nervous system (CNS) (Gardner et al., 1999). Meanwhile, due to a low-threshold potassium channel, the resting membrane conductances of BCs are exceptionally high, which leads to short time constants (2-4 ms) in these neurons. The short membrane time constants and fast receptor kinetics eliminate temporal integration and cause these bushy cells to specialize in preserving fine temporal structures of impinging AN spike trains. 35 A B C D Fig. 1.10. Major PSTH types of VCN principle neurons. A. A primary-like (Pri) neuron; B. A Pri-N neuron. C. A Chopper neuron. D. An Onset neuron. In all cases, the stimulus is a 50-ms BF tone burst at 30 dB above the threshold. The PST histograms are shown by the main plots in the panels. In each panel, the upper left inset shows the distribution histogram of the first-spike latencies. The upper right inset shows the PST histogram in response to a longer (400-ms) tone burst at the same frequency and level. 36 The SBCs and GBCs differ in their output projections. The SBCs send axons through the trapezoid body (TB, or the ventral acoustic stria, VAS) to the ipsi- and contralateral medial superior olives (MSO), and the ipsilateral lateral superior olives (LSO). The GBCs project to the ips- and contralateral nuclei of the TB via the VAS. The nuclei of the TB in turn send inhibitory inputs to the superior olive. Both cell types of BCs are excitatory and glutamatergic (Cant and Hyson, 1992). The MSO has been shown to be involved in sound localization based on interaural time differences (ITD). The excitatory inputs from SBCs to MSO on both sides are consistent with the coincidence detecting model (Goldberg and Brown, 1969). The extremely secure synaptic connection between afferent fibers and SBCs, along with the low-input resistance and short time constants of those cells, makes the SBCs well suited for conveying the timing of AN fibers (phase-locking) with high fidelity, which is a required by ITD-based sound localization. The LSO, the direct and indirect target region of GBCs, has been shown to be important in sound localization based on interaural level-differences (ILD). 1.4.2. Stellate cells Stellate cells, also called multipolar cells, are another numerous neuronal type in the VCN. These neurons can be found in most parts of the AVCN and PVCN. Stellate cell have polygon-shaped somata and long dendrites which extend in multiple directions away from the soma (Cant and Morest, 1979). The dendritic tree of a stellate cell is largely confined in a plane, the functional importance of which we will discuss later. There are two major subtypes of stellate cells which differ in their cytoarchitecture and projection patterns (Doucet and Ryugo, 2006). One subtype, called T-stellate cells, got their name from their projection through the TB directly to the inferior colliculus (IC) (Schofield and Cant, 1996). These cells are excitatory and 37 glutamatergic (Ferragamo et al., 1998). The other subtype, called D-multipolar cells, are though to be glycinergic inhibitory interneurons. They send inhibitory inputs to various types of neurons in ipsilateral and contralateral CN (Cant and Gaston, 1982). The two subtypes of neurons differ in their sources of ANF inputs. The dendritic planes of T-stellate cells are parallel with the iso-frequency laminas, while those of the D-multipolar cells are perpendicular to and span an array of these laminas. This indicates that, T-multipolar cell receives input from a relatively restricted BF range, whereas D-multipolar cells receives input from a wide range of BFs (Doucet and Ryugo, 2006). These structural distinctions lead to significant differences in the response properties of the two types of neurons to sounds, as will be discussed later. Multipolar cells receive both excitatory and inhibitory inputs. AN fibers form bouton endings on both the somata and dendrites of the multipolar cells. Other synapses are inhibitory and are also found on somata and dendrites. Electrophysiological evidences suggest that as few as five ANFs converge onto a T-multipolar cell (Ferragamo et al., 1998). Anatomical evidences suggest that the number of ANF inputs is larger in D-multipolar cells. Most multipolar cells belong to the T subtype. The ratio of cell counts between T- and D-stellate cells is approximately 15 : 1 (Doucet and Ryugo, 1997). The T- and D-stellate neurons exhibit different PSTH types. T-multipolar cells are associated with chopper (Ch) PSTH types (Rhode and Smith, 1986; Blackburn and Sachs, 1989). As shown in Fig. 1.10.C, chopper PSTHs are characterized by their multimodal shapes, namely several peaks following the stimulus onset. This reflects the regular inter-spike intervals in the T-multipolar neurons. The regular spiking reflects temporal integration and intrinsic membrane properties of the T-stellate cells (e.g., the modeling study by Banks and Sachs, 1991). Previous 38 studies have identified several subtypes of chopper PSTHs, which can be distinguished based on the regularity and time courses of ISI changes (Bourk, 1979; Blackburn and Sachs, 1989). The sustained choppers (ChS) show very regular firing across the entire duration of a 50-ms BF tone burst, where as transient choppers (ChT) show gradual decrease in rate and regularity with increasing time. Low-rate choppers (ChL) fire regularly at a very low rate. As a group, the Chopper neurons generate significantly more regular spike trains than Pri/PriN fibers, whose spiking regularity is close to that of the AN fibers. ChS neurons show more regular firing than ChT neurons do (Fig. 1.11, right). In contrast to the T-stellate cells, most D-multipolar cells show onset-chopper PSTH types. These PSTHs are called onset because spikes are produced only near the onset of the stimulus, the “chopper” part of the name comes from the fact that the interspike intervals between the first few spikes are regular and reproducible (Fig. 1.10.D). The onset-choppers are interesting in that they possess unusually wide dynamic ranges in their rate-level functions for pure tones, which lead some authors to hypothesize that these neurons are intensity encoders (Rhode and Smith, 1986). This wide dynamic range is caused by the increase in the duration of firing with increasing tone intensity. The stellate cells don’t possess short membrane time constants or secure primary endings as the BCs do. Their firing requires temporal integration. In in-vitro studies, a current step elicits sustained and regular firing of action potentials in stellate cells, which is in contrast to the transient onset spiking by bushy cells (Manis and Marx, 1991). Their spike trains of stellate cells don’t preserve the fine temporal structures of the ANF input, and the processing by stellate cells causes loss of information regarding the fine structure of sounds. Based on this, it has been suggested that the primary function of the stellate cells is to process non-temporal acoustic 39 Fig. 1.11. The first-spike latency and firing regularity of VCN principle neurons in response to BF-tone bursts. Left, minimum first-spike latencies as functions of BFs in different types of VCN neurons: Pri and PriN (A), different sub-types of Choppers (B) and Onset neurons (C). In each panel, the dashed and curves shows the minimal latencies in the Chopper group. The solid curve is the dashed curve shifted downward by 1 ms. First-spike latencies are negatively correlated with BF in all three groups. The Chopper neurons consistently show longer latencies than the other two types of neurons. Right, the distribution histograms of CV of ISIs during sustained firing (30-40 ms after tone onset) in different VCN neuronal type, Pri and PriN (A), Ch-S (B), and Ch-T (C). Pri/PriN neurons show the greatest ISI variability. ChS neurons show exquisite regularity in spiking. ChT shows ISI regularities intermediate between ChS and Pri/PriN neurons. 40 information, such as sound spectra. For example, the representation of vowel spectra by firing rates has been shown to be superior in Chopper neurons compared to Pri/PriN neuron (Blackburn and Sachs, 1990; May et al., 1998. See section 1.5). 1.4.3. Octopus cells The octopus cells possess relatively few dendrites extending away from the soma in several similar directions and a long axon. There is a sub-region of the PVCN occupied exclusively by the octopus cells. Each octopus cell receives inputs from a large number of ANFs, through bouton synapses on both the somata and dendrites (Willott and Bross, 1990). Octopus cells send projection to the contralateral superior periolivary nuclei and nuclei of the lateral lemniscus via the intermediate acoustic stria (IAS) (Adams, 1997). The octopus cells are associated with onset responses, which is characterized by one or several short-duration and precisely timed spikes near stimulus onset and little or no sustained firing (Rhode et al., 1983). Similar to the chopper response type, there also seems to be certain degree of inhomogeneity in the onset category. The “ideal” onset (OnI) pattern shows very little sustained activity after the onset spike; where as the OnL pattern contains non-zero sustained firing rate throughout the stimulus duration (Rhode and Smith, 1986). The anatomical causes of these different remain unclear. Similar to bushy cells, octopus cells possess short membrane time constants, due to the interplay between a hyperpolarization-activated cation conductance and a depolarization-activated potassium conductance. Another important functional consequence of the depolarization-activated potassium channel (gKL) in the octopus cells is the sensitivity of these neurons to the rate (i.e., time derivative) of depolarization (Ferragamo & Oertel, 2002). 41 Action potentials will be produced only when the depolarization is fast enough because otherwise the gKL channels will activate and block spiking. Due to this property, octopus cells require synchronized spiking from a large number of ANFs in order to fire, which signifies the onset, or sharp transitions of acoustic signals. 1.4.4. Regularity and latency of VCN principal neurons The different PSTH categories discussed above can be distinguished from each other not only by the shapes of the histograms, but also by regularity of interspike intervals, which is often quantified by the CV of ISIs. However, VCN neurons with low BFs show phase-locking, which can be confused with regular firing seen in chopper neurons. This confounding effect can be ameliorated by presenting tones with randomized initial phases. However, this approach doesn’t always guarantee successful classification of low-BF primary-like and chopper units (Blackburn and Sachs, 1989). Another source of useful information in the classification is the systematic difference between the minimum first-spike latencies between primary-like and chopper neurons. As shown in several previous studies (Bourk, 1976; Young, 1988; Blackburn and Sachs, 1989), chopper neurons with all subtypes showed first-spike latencies approximately 1 ms greater than those of the Pri/PriN and onset units (Fig. 1.11, left). This systematic difference in latency is presumably caused by the differences in the ways in which bushy and T-multipolar neurons integrate ANF inputs and generate spikes. When clues from PSTH shape based on random-phase tones, first-spike latency, and the existence of pre-potentials are jointly considered, reliable classification of low-BF sustained firing VCN neurons is possible (Blackburn and Sachs, 1989). 1.5. Level and spectral encoding of vowels and broadband sounds by AN and CN neurons in normal and impaired hearing 42 To humans, speech is one of the most important and complicated stimuli faced by the auditory system. An important measure of the severity of hearing loss is the degree of degradation in the abilities to recognize and understand speech, especially in noisy background (Moore, 2002). Although formal psychophysical tests of loudness recruitment use pure tone stimuli, a type of sound which is also affected by recruitment and probably more behaviorally relevant to people suffering from SNHL is speech sounds (Huizing, 1948). Vowels are an important component of speech. In the source-filter model (Fant, 1960), glottal pulses generated by the vibrating vocal cord are filtered by the vocal tract, giving rise to the formants and troughs, the defining spectral features of vowels. The formants are spectral peaks and correspond to the resonance frequencies of the vocal tract. Vowels can be identified based on the frequencies and relatively amplitudes of the first three formants. Therefore one important job the auditory system must perform to encode speech is to faithfully convey the information about the spectral shapes of the vowels. The representation of steady-state vowels in ANFs has been extensively studied. Two possible ways of encoding the vowel spectrum have been proposed, one based on mean discharge rates of the AN fibers and the other based on temporal structures of the AN spike trains. In the so called rate-place code (Sachs and Young 1979; Delgutte and Kiang, 1984), fibers with BFs equal to or near the formant frequencies respond by higher driven firing rates than the fibers with BFs falling into the troughs, thereby giving rising to a rate-BF function with a shape similar to the vowel spectrum. The goodness of the rate-place representation can be assessed by the contrast of the rates at formants and troughs. As shown in Fig. 1.12.A, the rate-place code based on absolute rate measures in AN fibers deteriorate at medium to high levels, mainly because of 43 A B Fig. 1.12. Rate-place encoding of the steady-state vowel /ε/ in auditory-nerve fibers and VCN chopper neurons. A. Normalized rate-place profiles of AN fibers from a normal-hearing cat (from Sachs and Young, 1979). Each curve is the moving-window average of driven firing rates of a population of AN fibers from the same animal. The three arrows indicate the frequencies of the first three formants in the vowel spectrum. At low and medium levels, there is a good rate-place representation of the vowel spectrum. At higher levels, the rate contrast at F1, T1 and F2 deteriorates. B. The same stimulus and format as in A. Rate data were obtained from populations of VCN Ch-S and Ch-T type neurons (from Blackburn and Sachs, 1990). Compared with AN fibers (A), better rate contrasts and rate-place representations at wider level ranges are seen in these Chopper neurons. 44 the limited dynamic ranges. At high levels, the fibers with BFs at the formant reach rate saturation and cannot produce higher rates with increasing vowel level, while the rates of fibers with BFs at the troughs continue to grow. Among the different groups of AN spontaneous rates, the low/medium SR fibers (< 20 spikes/s) show better rate-place representations at high levels, because of their higher response thresholds and relatively larger dynamic ranges. In another code called temporal code, the AN fibers preferentially phase-lock onto the spectral component with frequencies nearest to the BF (Young and Sachs, 1979). This code has the advantage over the rate-place code in that it is possible to tell which vowel component is producing the response in a fiber. Moreover, the fidelity of spectral representation of the temporal code is less affected by sound level. In the CN, speech encoding has been studied mainly in only two types of neurons, the Pri/PriN neuron, and Chopper neurons. The rate-place encoding has received more attention than the temporal encoding, mainly because of the technical difficulties associated with sampling enough neurons. As expected from the very secure synaptic connection between AN fibers and bushy cells, primary-like neurons show similar responses to vowels as the AN fibers do (Blackburn & Sachs, 1990; May et al. 1998). However, the chopper neurons show better rate-place encoding in that the rate contrasts are larger and are preserved at high sound levels. The good rate-place code at high levels resembles the behavior of low-/medium-SR AN fibers, but the choppers show thresholds as low as those high SR fibers (Fig. 1.12.B). This led to the suggestion that the T-stellate cells integrate inputs from a group of AN fibers with similar BFs but different SR categories in a way to optimize the rate code at both low and high levels (Lai et al., 1994). 45 The review of vowel encoding has been concentrated on normal hearing animals so far. In the normal cochlea, the BM shows fine frequency tuning and a two-tone suppression nonlinearity, both of which are important for generating the two types of codes discussed above (Deng and Geisler, 1987). In acoustically traumatized ears, both properties of the BM are compromised or lost, leading to significantly worse vowel representation (Palmer and Moorjani, 1993; Miller et al., 1997). Unlike in normal ears, AN fibers with all BFs show phase-locking onto the first formant rather than to formants near their BF. This is caused by loss of fine frequency tuning and a weaker cochlear suppression. Schilling et al. (1998) showed that by using a spectral enhancing modification on the vowel spectrum which emphasized higher-frequency formants relative to F1, the normal temporal coding can be partially restored in impaired ears. To our knowledge, no studies on the encoding of vowels in central auditory nuclei in animals with hearing losses have been published to date. Central neural degeneration and plastic changes may lead to further deterioration of vowel level and spectral encoding, exacerbating the coding abnormalities caused by peripheral pathology. Vowels are a special type of broadband sound. A mathematical model for rate encoding of the spectra of general broadband sounds called the linear-nonlinear weighting model (LNWM) has been devised (Yu and Young, 2000). This model approximates rate of an auditory neuron by weighted sums of energies in different frequency bands and their quadratic interaction. The system identification process for determining the parameters of this model depends on a set of broadband stimuli called random spectral stimuli (RSS). This method has been applied to AN fibers (Young and Calhoun, 2005) and CN neurons (Bandyopadhyay, 2007) in normal-hearing animals. It performs well in modeling and predicting the rate-spectral encoding by neurons in 46 those structures. This method hasn’t been applied to hearing-impaired animal to date. The current study will attempt to use this method to quantitatively study the rate-spectral encoding in the VCN of acoustically traumatized ears, and compare it to its normal counterpart. 1.6. The current study In the current study, we studied the encoding of levels and spectra of simple and complex stimuli by VCN neurons in normal-hearing and acoustically traumatized cats. The animal model of NIHL is similar to the one used in a previous study on neural correlates of loudness recruitment in AN fibers in the Neural Encoding Laboratory (Heinz and Young, 2004), so that the effects of acoustic trauma on peripheral and central level encoding can be compared and related. The noise overexposure procedure induced permanent thresholds of 30-50 dB in frequency ranges below 9 kHz. In decerebrated unanesthetized noise-exposed and normal-hearing control animals, single VCN neurons were isolated and classified into major PSTH types. A repertoire of stimuli, including on- and off-BF tone bursts, broadband noise (BBN), the synthetic vowel /ε/ and random spectral shape (RSS) stimuli were presented at series of levels and the responses of VCN neurons were recorded. The effects of acoustic trauma on basic physiological properties of VCN neurons (including thresholds, tuning, response maps and phase locking) were characterized. Attempts to classifying impaired VCN neurons based on methods developed in normal-hearing ears indicated that trauma didn’t dramatically alter VCN PSTH types. The slopes of rate-level functions (RLF) for different types of stimuli were analyzed quantitatively. Differential effects of trauma on slopes of different types of neurons were observed. Generally, PL neurons show shallower rate-level 47 slopes following trauma, which was reminiscent of shallower slopes seen in traumatized AN fibers (Heinz and Young, 2004). In contrast, non-primary neurons show close-to- or steeper-than-normal slopes. To investigate recruitment substrates in VCN, the summed (average) RLF for tones were calculated for different PSTH types and BF channels widths. Recruitment-like phenomena were seen under wide BF channels widths, mainly due to abnormally rapid spread of excitation. For on-BF level encoding, recruitment-like effects were observed in non-PL neurons, but not in PL ones. Other hypotheses regarding the physiological mechanisms of recruitment were also tested. The encoding of the level and spectrum of a synthetic steady-state vowel /ε/ was studied by using the spectral manipulation procedure (SMP, Le Prell et al., 1996). Abnormal rate-place encoding and recruitment-like effects in the impaired ear were found. The abnormal rate-spectral encoding in impaired VCN neurons, as revealed by the linear-nonlinear weighting model (LNWM), will also be described. We investigated and discussed the relationships of post-trauma alterations in rate-level encoding in VCN neurons to the following: (1) psychophysical properties of loudness recruitment, (2) abnormal rate-level encoding by AN fibers, (3) previously observed neural overexcitability in central auditory nuclei in SNHL. The results of this study will add to the current knowledge of central neural pathology in SNHL and the physiological mechanisms of abnormal intensity, spectral and speech perception in people with SNHL. 48 II. Methods 2.1. Acoustic trauma in cats Adult male cats (Catus Felis) weighing between 2.7 and 4.5 kg were used in this study. Animals were divided into two groups, an NIHL group and a normal-hearing control group. Each subject in the NIHL group underwent a noise over-exposure procedure which aimed to induce cochlear damage and permanent hearing loss. Before the sound over-exposure, the healthiness of the outer and middle ear was checked visually with an otoscope. The thresholds for auditory brainstem responses (ABR) was recorded at 2, 4 and 8 kHz monaurally in the right ear. Only cats showing normal ABR thresholds were used in the subsequent procedures. Cats were anesthetized with ketamine and xylazine. The initial dosage of xylazine and ketamine were 0.05 and 0.2 ml / kg, respectively. Follow-ups were given to maintain an areflexive state. The electrocardiogram was monitored during the entire procedure. The exposure stimulus was a 50-Hz wide narrow-band noise centered at 2 kHz. The stimulus was generated a custom program written in the TDT RPvdsEx software (Tucker-Davis Technologies, Alachua, FL), amplified by a Crown D-75A amplifier (Crown International Inc., Elkhart, IL), attenuated by a TDT PA-5 pre-attenuator and fed into a pair of loudspeakers (RCA 5-inch 2-way speaker system, RadioShack, Fort Worth, TX). The head of the cat was placed directly below the speakers with a distance of approximately 14 cm. This position was chosen because of its associated relatively flat HRTF (Rice et al., 1992). A sound meter (Simpson 899 Type 2, Sunshine Instruments, Elgin, IL) was used to monitor sound intensity during the 49 Animal # 06-006 06-009 06-018 06-045 06-047 06-067 06-080 06-081 06-090 07-003 07-034 07-037 Summary Mean exposure level (dB SPL) Left ear Right ear 112.3 111.9 112.4 111.6 111.8 112.2 109.9 111.4 111.8 111.5 110.9 111.6 110.6 111.5 111.1 111.9 111.3 111.7 111.5 111.5 111.5 111.5 111.4 111.5 Mean ± Mean ± S.D. = S.D. = 111.4 ± 111.7 ± 0.7 0.2 Date of noise Recovery overexposure period (days) Acoustic driver used in neural recording 02/10/2006 03/23/2006 04/27/2006 07/05/2006 07/13/2006 09/13/2006 10/20/2006 10/30/2006 11/20/2006 02/05/2007 05/16/2007 05/18/2007 Electrostatic Electrostatic Dynamic Dynamic Dynamic Dynamic Dynamic Dynamic Electrostatic Electrostatic Electrostatic Electrostatic 33 60 34 33 39 33 33 37 111 72 37 39 Median = 37 Tab. 2.1. A summary of animals in the NIHL pool and their neural recording experiments. 50 exposure. Sound intensity was measured directly adjacent to the left and right pinnas, with the probe of the sound meter pointing to the anterior. The measurement was carried out each 15 – 20 minutes. Sound level was maintained at 111-112 dB SPL by adjusting the amount of attenuation and/or the position of the head of the cat. Each noise over-exposure lasted for 4 consecutive hours. The exposures were binaural. Sound intensities measured near the two ears differed less than 2 dB. A summary of sound intensities of the exposure stimuli in animals in the NIHL group can be found in Tab. 2.1. After the animal recovered from anesthesia and regained consciousness and mobility, it was sent back to the animal care facility. The noise over-exposure procedure described above was similar in details to the one used in a previous study on recruitment effects in ANFs in the Neural Encoding Laboratory (Heinz and Young, 2004; Heinz et al., 2005). 2.2. The compound action potential audiogram Noise-exposed animals were given recover periods of at least 4 weeks (Tab. 2.1.), to allow recovery from temporary threshold shifts and stabilization of central neural changes. Before the neural recording in noise-exposed and normal animals, peripheral auditory thresholds were measured with compound action potential (CAP) audiograms. The tympanic bulla was exposed by removing the overlying tissue. A hole approximately 3-mm in diameter is made by a scalpel. After the round window is located visually, a second smaller hole on the bulla was made. A Teflon-coated silver wire electrode (A-M Systems, Carlsborg, WA) was placed through the smaller hole onto the bone near the round window under visual guidance. The wire was secured in place with superglue (Scotch-Weld, 3M, St. Paul, MN). 51 The signal picked up by the wire electrode was amplified, band-pass filtered between 300 and 3000 Hz,A/D converted, and sampled at 12.2 kHz. The stimuli were 10-ms tone pips with 1-ms rise and fall ramps and a repetition rate of 10 per second. Their initial phases were randomized to reduce the amplitude of cochlear microphonic in the average signals at low frequencies (≤ 3 kHz). 17 to 18 frequencies between 0.4 and 20 kHz were covered, which include 2 – 3 frequencies below 1 kHz, 11 frequencies between 1 and 10 kHz and 3 frequencies above 10 kHz. These frequencies were roughly equally distributed on a logarithmic scale. For each frequency, measurements were taken at 3 – 7 sound levels in 5-dB steps with at least one level with a clearly visible CAP response. 50 or 200 repetitions were done for each combination of frequency and level. The average signals were used in future analyses. The CAP data were processed offline to obtain the response thresholds. The offline processing consisted of two steps. The first step was to filter out the cochlear microphonic signals at low frequencies. Because of the pass band of the filter before A/D conversion was set as 0.3 to 3 kHz, stimuli with frequencies at or below 1.5 kHz caused microphonic contamination not only at the fundamental frequency, but also at its second harmonic. Thus, for recordings at frequencies at or below 1.5 kHz, two cascaded infinite-impulse-response (IIR) band-stop filters were applied on the average signal. Each band-pass filter was a 5th-order Butterworth band-stop filter designed with the MATLAB Signal Processing Toolbox. The stop-band of each filter was fC /1.05 < f < f C ⋅1.05 , (2.1) where f C is the stimulus frequency for the first filter and two times the stimulus frequency for the second one. For recordings between 1.5 and 5 kHz, only one band-stop filter centered at the stimulus frequency was applied. For recordings above 5 kHz, no filtering was done on the 52 average signal. Fig. 2.1.A shows one example of the frequency responses of the filters with two stop bands for processing recording at 0.8 kHz. The effect of filtering is shown by the example in Fig. 2.1.B. It can be seen that the filtering effectively reduced the amplitude of the microphonic signal, while preserves the CAP signal, which has a stereotypical shape. The second step in CAP threshold estimating is based on the method for estimating ABR threshold devised by Ngan and May (2001). As shown by the unfilled circles in Fig. 2.1.C, the peak-to-peak signal amplitude inside the time window of 20 to 25 ms after stimulus onset was calculated for each stimulus level, which was taken as the background level. The criterion for judging response threshold was chosen to be two standard deviations of the background levels plus their mean, as shown by the dashed line in Fig. 2.1.C. Another time window of 5 to 10 ms after stimulus onset, which encompasses the delay of CAP peak responses in the frequency range between 0.4 and 20 kHz, was used to obtain an amplitude-versus-level function (solid curve in Fig. 2.1.C). The threshold (the red dot) was the intersection between this function and threshold criterion. In a few cases where there are multiple intersections (due to the noisiness of data), the threshold was taken as the intersection point with the highest level. 2.2.1. Edge frequencies of CAP audiograms In the noise-exposed pool, the CAP threshold at a certain frequency was found to be a good predictor of the single-unit thresholds at the corresponding BF (Fig. 3.2). Therefore, we used an edge frequency method adopted from Ma and Young (2006) to delineate the boundary between BF regions with and without substantial threshold shifts. For the CAP audiogram from a 53 A B Gain (dB) 50 79.4 dB SPL 0 74.4 dB SPL -50 -100 -150 69.4 dB SPL 0 1000 2000 3000 4000 5000 600 64.4 dB SPL -5 59.4 dB SPL -10 -15 0 1000 2000 3000 4000 Frequency (Hz) 5000 600 Voltage (v) Phase (rad) 0 0.1 0 -0.1 Original Filtered 54.4 dB SPL 0 5 10 15 20 Time (ms) C 0.16 0.14 Signal amplitude (V) 0.12 Background amplitude Threshold criterion Response amplitude Estimated threshold 0.1 0.08 0.06 0.04 0.02 0 50 55 60 65 70 Tone pip level (dB SPL) 75 80 Fig. 2.1. Offline processing of compound action potential recordings and estimation of the response thresholds . A. The frequency response (top: amplitude; bottom: phase) of the cochlear microphonic filter at 4 kHz. This filter was constructed by cascading two 5th-order band-stop Butterworth IIR digital filters. B. The effect of the filter shown in A. Data was from noise-exposed animal #07-034. Each trace corresponds to a sound pressure level. Blue curves are the average of 200 repetitions, which was the input to the filter; red curves are output of the filter. C. Calculating CAP response threshold based on comparison between peak-to-peak amplitudes in the response time window (5 - 10 ms after stimulus onset) and in the background time window (20 - 25 ms after onset), after Ngan and May (2001). This graph is based on the same data in A and B. The dashed line indicates the threshold criterion, which is the mean of the background amplitude plus two standard deviations of the background noise amplitudes (the unfilled circles). Level at which the response-amplitude-versus-level curve crosses this criterion was judged to be the threshold for the CAP response. 54 25 hearing-impaired animal, the CAP audiogram was first subtracted by the average normal CAP audiogram (the dashed blue curve in Fig. 3.4) to yield the threshold-shift-versus-frequency curve. The logarithmic center frequency of the rightmost segment in threshold-shift-versus-frequency curve was determined to be the edge frequency (the vertical gray lines in Fig. 3.2). Fig. 3.3.A and B respectively show the alignment of the frequency axes of impaired CAP audiograms and single-unit BFs at zero octaves re edge frequency (ORE), forming a “virtual” frequency axis with the unit of ORE. From these figures, it was evident that this edge-frequency method was an effective way to discriminate between units with normal threshold and shifted ones. Units in the noise-exposed population with BFs below these edge frequencies were regarded as impaired units. For the normal hearing pool, the edge frequency was artificially defined as 10 kHz (vertical gray lines in Fig. 3.1). In subsequent analysis, comparisons between the impaired and the normal populations will be restricted to units with BFs below the edge frequencies (called sub-edge units, the units left to the vertical gray lines in Fig. 3.3.B), unless otherwise stated. 2.3. Surgical preparation During the surgery, the animal was anesthetized with ketamine and xylazine with the same dosages as those used in the noise exposure procedures (Section 2.1). An areflexive state was maintained by follow-up doses during the surgery. Additionally, 2.0 ml dexamethasone was given intramuscularly (i.m.) to reduce tissue edema; and 0.2 ml atropine was given i.m. every 24 hours to reduce mucous secretion. Lactated ringer solution was given intravenously to provide the animal with necessary nutrients and prevent dehydration. The core-body temperature of the animal was monitored with a rectal probe. A feedback controlled heating pad maintains the 55 temperature at approximately 38.2 ۫C. A tracheotomy was performed and a plastic tube was inserted into the trachea to achieve a low-impedance air way and to reduce breath-related noise. The animal was decerebrated by aspirating a complete slice of brain tissue at the level of the thalamus, after which anesthesia was discontinued. The right ear canal was cut open and a hollow metal ear bar was inserted into the opening. Together with another metal bar inserted into the left ear canal and a bone screw secured in the frontal skull, it mechanically stabilized the head of the cat in reference to a metal stereotaxic frame (David Kopf Instruments, Tujunga, CA). A metal plate electrode was inserted into the musculature in the back of the neck and served as the reference. After the completion of CAP recordings (Section 2.2), a part of the bone anterior to the right nuchal ridge was removed and the dorsolateral part of the right cerebellum was aspirated to expose the dorsal surface of the CN. Under visual control, a glass-coated platinum-iridium electrode with platinized tip (impedance range: 1 - 10 MΩ) was inserted near the joining edge of the middle cerebellar peduncle and the CN in the sagittal plane and moved in a dorsal-ventral direction under the control of a hydraulic microdrive (Kopf). The electrodes were often tilted noise-up in the sagittal plane by 1 – 3° in order to achieve better accesses to the low-BF regions of the VCN (Bourk et al., 1981). The search stimuli were 50-ms bursts of tones or broadband noises (BBN). Once a single unit was isolated, its BF and threshold were determined with an automated tuning-curve making program whenever possible. The automated tuning curve program was initially designed for AN fibers and didn’t work well in many VCN neurons with firing patterns substantially different from those of the AN fibers. These include onset unit, and chopper or unusual-type units with 56 Amplitude of calibration (dB SPL) A 100 98 96 94 92 90 10-1 100 Frequency (kHz) 101 Amplitude of calibration (dB SPL) B 120 115 110 105 100 10-1 100 Frequency (kHz) 101 Fig. 2.2. Examples of amplitude calibration curves of the acoustic drivers. Typical amplitude calibrations from the two types of acoustic drivers used in the neural recording experiments. A. The electrostatic driver with equalization, from experiment #07-011. B. The dynamic driver with equalization, from experiment #06-067. In both types of acoustic drivers, the amplitude calibrations fluctuated less than ±5 dB in frequency ranges below 18 kHz. 57 long latencies. Also, evoked responses (ER) picked up by the metal electrode at high SPLs occasionally prevented the utilization of the tuning curve making program. In those cases, the BF and thresholds of the units were determined audiovisually and confirmed later by response map and rate-level function (RLF) recordings. 2.4. The acoustic system Two different types of acoustic drivers were used for CAP and neural recordings in normal and hearing-impaired animals. In the normal-animal experiments and some exposed-animal experiments, an electrostatic speaker (Sokolich, 1977) was used, whereas in the rest of the exposed-animal experiments, another type of dynamic speaker capable of producing sound level up to 120 dB SPL was used. A summary of the types of acoustic drivers used in the impaired-animal experiments can be found in Tab. 2.1, Column 6. The dynamic speaker was later found to produce an artifactual noise that interfered with data in some cases (See Section 2.8). A calibration is done with between 0.04 kHz and 20 kHz (for the deaf-cat experiments) – 40 kHz (for normal-cat experiments) with a sound probe (Bruel and Kjaer, Norcross, GA) placed 2-3 millimeters from the tympanic membrane. An equalizer (GE60, Rane, Mukilteo, WA) was used in several experiments to flatten the amplitude calibration. Fig. 2.2 shows examples of amplitude calibration curves of the two types of acoustic drivers. In both types of speakers, the calibration amplitude fluctuated less than ±5 dB in frequency ranges below 18 kHz. 2.5. Acoustic stimuli 2.5.1. Simple stimuli 58 Acoustic stimuli were presented to the ear ipsilateral to the VCN being recorded from. In each neuron, PST histograms were constructed by presenting 200-300 repetitions of 50-ms BF tone bursts with 5-ms linear rise and fall ramps at a repetition rate of 4 Hz. In about 80 percent of the units, PSTHs were recorded at 20 and 30 dB above threshold. In the rest of the units, PSTHs were recorded only at 30 dB re threshold. In many VCN neurons, offline processing of the threshold revised online estimates of the thresholds, causing many actually levels of PSTHs to deviate slightly from 20 and 30 dB re threshold. Simple stimuli used in the experimental protocol included bursts of tones and BBN. The bursts are 200 ms in duration and have 10-ms linear rise and fall ramps. For constructing RLFs, each level is presented once and the presentation at each level consists of a silent period of 800 ms following the offset of the burst. The sound levels are always arranged in a sequential ascending order in 1-dB steps. RLFs for BBN, BF tone and 1-kHz and 2-kHz tones were constructed. In addition, RLFs for tones with frequencies logarithmically evenly distributed in 0.4-octave steps around the BF of the unit were also constructed. Those were called “LOGTONE” stimuli in the experiments. For example, for a unit with BF equal to 3 kHz, we recorded RLF for tones at 3.96, 2.28, 5.22, 1.72, 6.89 and 1.31 kHz, etc., which were respectively at +0.4, -0.4, +0.8, -0.8, +1.2 and -1.2 octaves and so on, with respect to the BF. Tones were also used in construction of response maps, which were rate-versus-frequency functions at fixed acoustic attenuations. 2.5.2. The steady-state vowel /ε/ and the spectral manipulation procedure The steady-state vowel /ε/ (as in “met”) was initially synthesized at 10 kHz. It had a 59 fundamental frequency (F0) of 100 Hz and duration of 300 ms. Fig. 2.3.A shows one period of its waveform. As shown by Fig. 2.3.B, the first four formants (F1 – F4) of the vowel were at 0.5, 1.7, 2.5 and 3.3 kHz. F2 and F3 were 16 and 27 dB less intense than F1, respectively. The playback rate of the signal was systematically varied in order to align the center frequencies of the formants and troughs with the BFs of the units. Let the original sampling rate be f 0 , the feature (format or trough) center frequency be f F , and the BF of the unit be f B . To align the feature with the unit BF, the playback rate was set to, fP = fB f 0 , (2.2). fF This method was called the spectrum manipulation procedure (SMP), originally proposed and used in Le Prell et al. (1996) and May et al. (1998). Fig. 2.3.C shows a schematic example of the SMP method. By using the SMP method, six pseudopopulation neurons with equivalent BFs equal to the six feature frequencies can be generated from each recorded real-population neuron as long as the BF of the unit falls into a proper range. It has been shown to be a valid method of studying rate encoding of vowel spectra in AN fibers and VCN neurons. The pseudopopulation rate-place profiles generated by SMP closely resemble those measured in real neuronal populations. The duration of each presentation was fixed at 300 ms. Using a playback rate higher than the synthesis rate effectively reduced the duration of the signal. In order to maintain a constant stimulus duration of 0.3 s, the signal was restarted after the offset of the previous cycle repeatedly until the 0.3-s time window was filled. This operation didn’t cause audible discontinuities (clicks). In the current study, units with BFs between 0.3 and 10 kHz were included in this pseudopopulation analysis. 60 Relative amplitude (dB) B A 0.5 0 -20 -40 -60 -80 0.1 -1 0 1 4 -0.5 0.005 Time (s) 0.01 Phase (rad) Wave-file amplitude 1 0 2 0 -2 -4 0.1 1 Frequency (kHz) C Relative amplitude (dB) 0 F1 T1 F2 T2 F3 T3 -10 -20 -30 -40 1 10 Frequency (kHz) Fig. 2.3. The steady-state vowel /ε/ and the spectral manipulation procedure. A. One fundamental period (10 ms) of the vowel waveform. B. Amplitude (top) and phase (bottom) spectra of the vowel. The four gray bars indicate the frequencies of the first four formants (F1 – F4), which were at 0.5, 1.7, 2.5 and 3.0 kHz, respectively. C. Schematic of the spectral manipulation (SMP) procedure. The playback rate of the vowel is systematically varied (Eq. 2.2) to align the BF of the unit (black vertical line, 4 kHz in this example) to the feature frequencies. 61 2.6. Data analysis methods 2.6.1. Classification of ventral cochlear nucleus neurons For VCN neurons in the normal-hearing ear, the BF was defined as the frequency at which the threshold for excitation was the lowest in dB SPL. Because of the alterations of tuning curves shapes after cochlear damage, this definition was inappropriate for a portion of impaired units in which BFs shift downward as the tuning curves broaden. In these cases, the BF was defined as the frequency at the lower-end of the high-frequency high-slope portion of the tuning curve (Liberman and Dodds, 1984b). Classification of VCN units was based on the PSTH patterns of responses to 50-ms BF-tone bursts at 30 dB re threshold (Section 2.5.1). There were a few cases in which reliable single-unit isolation could not be obtained at 30 dB re threshold in units with substantial threshold elevations, either because of the upper bounds of the deliverable sound levels or because of strong neurophonic (evoked) responses at high SPLs. In these cases, PSTHs were obtained up to the highest possible level. A bin width of 0.2 ms was in constructing the PSTHs. A working assumption was made that the basic PSTH patterns were not systematically altered by acoustic trauma, despite threshold shifts and other post-trauma changes. Intracellular recording studies of VCN neurons in animals with cochlear damage have shown that the most basic relationships between cell types and membrane properties were preserved after cochlear damage (Francis and Manis, 2000; Wang and Manis, 2005). As will be shown in Section 3.2, basic PSTH types do seem to be altered following acoustic trauma. Based on this assumption, the decision-tree method proposed in Blackburn and Sachs (1989) was used in classifying neurons in 62 both normal-hearing and noise-exposed ears. The PSTH shapes of a few recorded units showed level dependences (further discussed in Section 3.2). In these cases, classification was based on the shape at or near 30 dB re threshold. As low BFs, it is often hard to discriminate between phase-locking and chopping. In Blackburn and Sachs (1989), two solutions to this problem were proposed, 1) the use of tone bursts with randomized initial phases (asynchronous tones), 2) the use of minimum first-spike latencies in judging between PL and non-PL neurons. They showed that although not true in every case, randomizing the initial phases of the tone can eliminate the multimodal shapes in PL neurons due to phase-locking, whereas the peaks in the chopper neurons will be preserved. Minimum first-spike latency was proposed as a more reliable discriminator because chopper neurons almost always showed longer latencies than PL ones. As will be discussed in greater detail in Section 3.2.5, most of the low-BF neurons we recorded from didn’t show multimodal PSTHs under randomized initial phases, suggesting that they fell into the PL category. However, compared with the low-BF Pri neurons shown in Blackburn and Sachs (1989), they had much longer minimum latencies (Fig. 3.12.D). The reason for this discrepancy was unclear. Different anesthesia states and differences in animal strains were possible causes. Due to this uncertainty, we categorized the low-BF VCN neurons which didn’t show multimodal PSTHs under asynchronous stimuli into a separate group called Lockers. The same categorization was done in some previous studies on VCN single-neuron physiology (e.g., Rhode and Smith, 1986). For neurons that showed multimodal shaped PSTHs, a spike-per-peak (SPP) test was done (Young, 1988). Categorization into the chopper category required an SPP falling into the interval of [0.95, 1.05] for the first peak and an interval of [0.85, 1.1] for the second peak. 63 Units in the onset category were sub-divided into three subtypes, onset-chopper (OnC), ideal onset (OnI), L-shaped onset (OnL). An onset unit must show two or more short-latency and precisely timed onset spikes near the stimulus onset and pass the SPP test for the first two PSTH peaks in order to fall into the OnC subcategory. For non-chopping onset units, the units with little or no sustained firing were classified as OnI; while those exhibiting a positive sustained driven rate were classified as OnL. Due to the way we accessed the VCN with electrodes, it was often inevitable to encounter a few DCN neurons in dorsal portions of the tracks. We judged whether a unit was from the DCN based on the following considerations. (1) Neurons with BF-tone and BBN RLF shapes and response-map patterns resembling those previously described in DCN type-IV or type-II neurons were likely to be recorded in the DCN (Shofner & Young, 1985; Spirou et al., 1999). (2) The pauser-buildup PSTH type is a strong evidence for a DCN neuron. (3) Neurons recorded less deeply than 1000 μm are likely to be located in the DCN. (4) Sudden and clear discontinuities in the BF sequence along the tracks are likely to be the boundaries between DCN and the VCN. Neurons located less deeply than these BF transitions were very likely to be DCN ones. The units which were suspected to be recorded in DCN were not used in further data analysis. 2.6.2. Quantitative analyses of PSTHs To capture the minimum first-spike latency, a cursor was placed by eye at the point where the PST histogram begins to increase. This method has been shown to be effective in extracting minimum latency in spontaneously active units and to be relatively immune to the confounding factor of refractoriness (Young, 1988). For each stimulus repetition, the earliest spike after the 64 cursor was taken as the first-spike latency. Ratios between peak and sustained firing rates, or peak-to-sustained (P/S) ratios, were calculated. For units with BFs above 1 kHz, the peak firing rate was measured as the maximum firing rate inside 0.2-ms bins between 0 and 15 ms after stimulus onset. For units with BFs below or equal to 1 kHz (mostly Lockers), the average firing rate inside the 3-ms time window immediately after the cursor was taken to be the peak firing rate. In both types of units, sustained firing rate was defined as the mean firing rate between 25 and 50 ms after stimulus onset. Mean coefficients of variation (CVs) were calculated for the interspike intervals (ISIs) inside the time window of [30, 40]-ms after stimulus onset. 2.6.3. Construction of tuning curves and response maps Tuning curves were constructed in two different ways. In the first approach, tuning curves were constructed by an automated program. In the second approach, the tuning curves were synthesized by detecting and connecting the threshold points in tonal rate-level functions and response maps (See Section 2.6.4.2 for threshold methods). One reason for using the second method was that in some cases the non-primary-like firing properties (e.g., low firing rates or long latencies, compared to those of the ANFs) and/or strong ER at high sound levels prevented the automatic tuning curve program from working reliably. The validity of the tuning curves and Q-factors obtained from the second approach was justified by comparing the results obtained by the two approaches in the units for which both automatic and synthesized tuning curve data were available (Fig. 2.4). As shown in Panels B and C in Fig. 2.4, due to the different rate threshold criteria used, the synthesized tuning curves (solid curves) are often 5 – 10 dB above the automatically generated ones (dashed curves). Despite the 65 Q10 of synthetic TC A 10 b0=0.13439, b1=0.88905 9 2 =0.74992 r 8 7 p=0.00027072 6 5 4 3 2 2 3 4 5 6 7 8 910 Q10 of automatic TC B. 06-067, Unit 1.07, Pri C. 06-048, Unit 6.05, Unusual-A 90 80 Q10Auto=BF/w10=4.3369 75 Q10Syn=7.4523 Sound level (dB SPL) Sound level (dB SPL) 85 70 w10 65 60 10 dB 55 50 45 40 0 10 80 Q10Auto=BF/w10=3.3103 70 Q10Syn=2.6054 60 50 30 Synthetic TC Automatic TC 20 -2 10 frequency (kHz) w10 10 dB 40 Synthetic TC Automatic TC -1 10 0 10 1 10 frequency (kHz) Fig. 2.4. Tuning curves and Q10’s generated by two different methods. A. Q10s obtained from automatic and synthetic tuning curves, in units for which both types of tuning curves were available. Dashed black line indicates equality. The red solid curve shows linear regression. b0 and b1 are zeroth and first order coefficients of the regression. r2 and p values indicated significant correlation and near-equality between these automatic and synthetic Q10s. B. The two types of tuning curves for the unit in indicated by the red arrow in Panel A. The two Q10’s in this unit are relatively discrepant. As shown in this panel, this discrepancy is mainly a result of the automatic tuning curve failing to capture the threshold at the BF. C. Same format as panel B. Unit indicated by the blue arrow in Panel A. The two types of tuning curves and two types of Q10s agree well with each other except for a difference in threshold. 66 differences in thresholds, the Q10 values obtained with the two methods agree reasonably well with each other (Panel A, slope of the regression line = 0.89, r2 = 0.75, p < 0.001). Therefore, in subsequent analyses, Q10 factors from synthetic tuning curves and automatically generated ones will be pooled. For VCN neurons, the frequency-level response maps (as those shown in Fig. 3.18) were generated by pooling data from tonal RLF and response maps. The two-dimensional frequency-level space is broken up into an array o 0.3-octave-by-4-dB cells, driven rates measured at frequency-level combinations falling into each cell was algebraically averaged to give the driven rate of that cell. 2.6.4. Analyses of rate-level functions 2.6.4.1. Construction of rate-level functions In non-onset VCN neurons, the counting window for calculating discharge rates were [10, 210] ms (re stimulus onset) for the pure tone or BBN stimuli, and [10, 310] ms for the vowel stimuli, unless stated otherwise (Fig. 2.5.A). The counting windows for onset-type neurons were [0, 200] ms for tone and BBN bursts and [0, 300] ms for the vowel stimuli. The rate level functions were smoothed by a running 5-point triangular window (Fig. 2.5.B-D). This processing reduces random rate fluctuations. Driven firing rate was defined as total firing rate minus spontaneous rate. 2.6.4.2. Determination of thresholds Because we typically had only one or two stimulus presentation at each level, simplistic definition of rate thresholds would be susceptible to random variations in firing rates. In an effort 67 100 90 80 dB SPL 70 A 60 50 40 30 0 200 400 600 Peristimulus time (ms) D 200 150 100 50 0 40 1000 C Firing rate (spikes/s) Firing rate (spikes/s) B 800 60 80 150 100 100 50 0 40 Sound level (dB SPL) 60 80 100 Sound level (dB SPL) 06-018, Pic 78 E Firing rate (spikes/s) 150 100 12.34spikes/s/dB 8.42spikes/s/dB 50 Smoothed RLF Linear spline fit Threshold Low-level fit slope 15-dB chord slope 0 30 40 50 60 70 80 90 100 Sound level (dB SPL) Fig. 2.5. Construction of rate-level functions and slope analyses. A. Raster plot of an example RLF (Exposed #06-018, Unit 1.09, 16.7-kHz Pri. Stimulus: 16.7-kHz tone). Spikes in the [10, 210]-ms counting window (magenta) were used to calculate the driven firing rates; spikes in the [500, 1000]-ms counting window (green) in the lowest 20 levels were used to calculate the spontaneous rate. B. The unsmoothed RLF. C. The 5-point triangular kernel for smoothing RLFs. D. The smoothed RLF. E. The smoothed RLF (blue) is fit by a linear spline function (solid black). The threshold (dashed black line) is determined by a Poisson multi-point thresholding (See Section 2.6.4.2). The low-level fit slope (LLFS, dashed red line) is the slope of the segment of the linear spline function which lies immediately above the threshold. The 15-dB chord slope (CS-15) is the slope of a line segment which connects the threshold point and the threshold+15 dB point on the RLF. A three-point averaging is done on each end point. 68 to solve this problem, we defined rate threshold in the following way. For each RLF picture, the mean spontaneous firing rate rs was determined by averaging across levels the time window of [500, 1000] ms after stimulus onset in the lowest 20 levels (The region shown by the green box in Fig. 2.5.A). For EH and Besh stimuli, this time window was [600, 1000] ms after stimulus onset, due to the longer stimulus duration. From rs , a 50% percent central confidence interval (CI) of spontaneous discharge rate was calculated based on the assumption that spontaneous rate follows a Poisson distribution with mean rate rs (exemplified by the green region in Fig. 2.5.E). Threshold was defined as the lowest level of the 5 consecutive levels for which the discharge rates fell above or below the CI (vertical black line in Fig. 2.5.E). Assuming the independence between spontaneous rate across different levels, the probability that five sub-threshold levels all had rates above or below the CI was 0.25 × 2 = 0.002 . This threshold definition was applied to 5 both excitatory and inhibitory responses. 2.6.4.3. Linear spline fitting of rate-level functions The abbreviated search algorithm proposed in Eertel and Fowlkes (1976) was used to fit linear spline functions (piecewise linear functions with connected knots) to the smoothed RLF (Fig. 2.5.E). This algorithm automatically determines the number of knots, or equivalently the number of line segments, and automatically calculates the best-fitting positions (levels and rates) of the knots in a least-square sense. In some cases, the strong random fluctuation in firing rates caused over-segmentation. In these cases, we manually specified the number of segments, without specifying their positions. Under-segmentation rarely occurred. This fitting procedure served as a data reduction method, based on which slope and quantitative analyses of the slopes and shapes of the RLFs could be performed. 69 2.6.4.4. Rate-level slopes The RLFs for simple and complex stimuli consisted of slope portions, i.e., the portions in which the relationships between rate and level are about linear. The levels and level spans of these slope portions varied with frequencies, unit types, spontaneous firing rates and hearing states. Thus for fair comparison between rate-intensity functions, it was not only necessary to quantify the local slopes, but also to quantify the global slopes. Based on this consideration, two types of slopes were calculated for each RLF, whenever possible. A low-level fit slope (LLFS, exemplified by the dashed red line in Fig. 2.5.E) was defined as the slope of the segment in the linear spline function immediately above the threshold. A chord slope (CS) was defined as the slope of a chord that connects the threshold point on the rate-intensity function and the 15 dB re threshold point on the function (dashed magenta line in Fig 2.6.E). In order to counter the effect of random rate fluctuation at single levels, we used the average rate at 3 consecutive levels centered at each end point. The choice of the chord span of 15 dB was based on a tradeoff between the need to capture global trends and the limitation of number of available levels above thresholds (especially in many impaired units). 2.6.4.5. Shape analyses of rate-level functions Based on the linear spline fits, the shapes of the RLFs were classified. RLF were classified into 4 shape categories, (1) Monotonically increasing, (2) Increase followed by saturation, (3) Non-monotonic or decreasing, (4) No response or weakly responsive. RLFs without definable thresholds (Section 2.6.4.2) automatically fell into category 4. The remaining RLFs were classified according to the decision tree shown in Fig. 2.6. The rates and levels were normalized 70 A B Fig. 2.6. The procedures and logical flowchart for classifying the shapes of RLFs. A. An RLF with substantial response (either excitatory or inhibitory) and calculated threshold was transformed into the normalized level and rate axes. The normalized slopes and length of the line segments were calculated. B. Based on the level- and rate-normalized RLF, the shapes are classified into three categories: (1) Monotonically increasing; (2) Increase followed by saturation; (3) Non-monotonic or decreasing, according to the flowchart. 71 into the unit intervals ([0, 1]) according to the maximum and minimum rates and levels in this picture, and the dimensionless normalized slopes and lengths of the constituting segments of the linear spline were calculated. Based on these level- and rate-normalized RLFs, the decision tree shown in B served to automatically categorize the RLF into the categories 1 – 3. 2.6.5. The pseudopopulation methods There were two pseudo-population approaches used in this study. The first one is the tonal pseudo-population approach; and the other was the vowel pseudo-population approach. The vowel pseudopopulation approach (SMP) has been described in Section 2.5.2. The tonal pseudopopulation approach treated tones with frequencies inside certain frequency interval as one “virtual” tone, whose frequency is undefined. The underlying assumption is that the response properties of the units (slopes, saturated rates, etc.) do not vary with tone frequency inside this frequency range (for BF-RLF slopes in ANFs, cf. Fig. 10 in Heinz and Young, 2004). The purposes of this approach are two, 1) to increase the effective amount of data and address the problem of limited data yield in the auditory brainstem single-unit recordings; 2) to systematically study on- and off-frequency level responses. Tones with frequencies below the CAP-audiogram edges were included in the pseudopopulation. So were units with BFs below the edge frequencies. These sub-edge frequency regions were the most impaired frequency region in the noise-exposed pool. In the pseudopopulation, frequencies were no longer measured in Hz. There are two dual ways to express frequencies. In the first scheme, tones with difference frequencies were regarded as one tone, and the unit BFs were expressed in octaves relative to the tone frequency (ORF). This 72 scheme simulated the response of units with different BFs to a fixed-frequency tone. In the dual scheme, units with different BFs were treated as one unit, and the tone frequencies were expressed in octaves re BF (ORB). This approach simulated the response of a neuron to tones at different frequencies. In the data pools, thresholds varied across different BFs and different animals. This would presumably have an effect on the summed or average RLFs generated in the pseudopopulation. To address that problem, two adjusted dB scales based on the CAP audiograms, which are called SLA and SLB corrections, were devised (Fig. 2.7). They are described by the following two equations, LSLA(f)=LSPL(f)-θANIM(f)+θAVG(f), (2.3) LSLB=LSPL(f)-θANIM(f)+θAVG(2kHz), (2.4) In Eqs. 2.3 and 2.4, LSPL(f) was the level in dB SPL; θANIM(f) is the CAP audiogram in dB SPL for the animal; θAVG(f) was the average CAP audiogram of the corresponding population of the animal (normal or exposed). LSLA(f) and LSLB were the two types of adjusted levels. SLA corrected for the variability of thresholds across different animals at the frequency f (Fig. 2.7.A), whereas the SLB adjustment corrected for the variability across different frequencies and different animals (Fig. 2.7.B). A reference frequency of 2 kHz was used because of its central position in the impaired region (Fig. 3.4). The SLA level adjustment will be used when the analysis is based on a real population of neurons. The SLB adjustment will be used in pseudopopulation analyses for which different BFs and frequencies are regarded as one. In the recording experiments, the levels of BBN were initially expressed in dB attenuation. To account for the difference of overall calibration levels across different experiments, BBN 73 Fig. 2.7. The level adjustment methods for pseudopopulation analyses. A. The SLA adjustment. This frequency-specific correction shifts the level by an amount equal to the difference between the CAP thresholds of the individual animal and the population average, which corrects for the variability of hearing thresholds across animals for a given frequency. B. SLB correction. This correction shifts the level by an amount equal to the difference between CAP threshold at frequency f of the animal and the CAP threshold at 2 kHz of the population average. It corrects for the variability of thresholds across animals and over different frequencies. The SLB correction is used in pseudopopulation analyses that ignore difference in BFs or stimulus frequencies, such as the tonal pseudopopulation and the SMP vowel /ε/. 74 levels were expressed in a calibration-adjusted dB scale, which was called dB NC, in population analyses of the broadband noise (e.g., Fig. 3.42). As shown by the following equation, LBBN − NC = LBBN − ATTN f max ⎛ fmax + ⎜ ∫ c( f )df − ∫ c0 ( f )df ⎜f f min ⎝ min ⎞ ⎟ , (2.5) ⎟ ⎠ LBBN − ATTN is BBN level expressed in dB attenuation, with more negative values corresponding to softer stimuli; LBBN − NC is the BBN level in dB NC; c( f ) and c0 ( f ) are the amplitude calibration functions of the specific experiment and a reference calibration (from experiment 06-006); f min and f max are the lowest and highest common frequencies of the two calibration functions, which were typically 0.4 – 20 kHz for the dynamic or 0.4 – 40 kHz for the electrostatic acoustic driver. When doing level corrections as SLA and SLB (Eq. 2.3 and 2.4), we were faced with the problem that CAP thresholds were lacking at very-low frequencies for a few animals. However, single-unit data were available at these frequencies. In these situations, we derived post-hoc CAP thresholds based on the average difference between single-unit thresholds and CAP thresholds at frequencies where both types of thresholds were available. The post-hoc thresholds were set at values such that the threshold differences match the average of those in the existing data. 2.6.5.1. Calculating the average rate-level functions and rate-matching curves Rate-level functions recorded in the experiments terminated at different maximum levels, due to the differences in calibration levels and the necessity to avoid ER contamination at high SPLs. In averaging a group of RLFs, it was necessary to decide the upper bound of levels in the average RLF. A 50-percentile rule was used. The maximum level for which the average rate was calculated was the 50-percentile of the maximum levels of the individual RLFs. The reason why 75 we didn’t use the a most strict 0%-quantile rule was because in a few RLFs with low tone frequencies had low maximum levels (60 – 70 dB) due to rapid increase of ER with level. In the pseudopopulation approach, we also needed to generate the grand average RLF for all VCN principal neuronal types. The numbers of different anatomical types of principal neurons are different, and have been reported in a few anatomical studies in the cat (Osen, 1970; Cant & Morest, 1984; Melcher, 1993). Table 2.2.A gives a summary of the estimates of VCN principal neuron counts from these three studies. Average ratios were calculated based on these cell number estimates. The stellate cell population consist of T and D subtypes. The T-stellate sub-population exhibit chopper PSTH type, while the D type shows a mixture of chopper and onset units. The ratio between the number of T- and D-stellate cells is about 15:1 (Doucet and Ryugo, 2005). Therefore, we assumed that the number of chopper units is equal to fifteen sixth of the number of stellate cells. Both D-stellate cells and octopus neurons have been associated with the onset PSTH type (Young and Oertel, 2004). Therefore, the number of onset units in the VCN was calculated as the sum of the number of octopus cells and one sixteenth of the number of stellate cells. The weights for different subtypes of chopper were based on recorded unit numbers in two previous studies (Blackburn and Sachs, 1989; Bourk, 1979) and the current one (Tab. 2.2.B). To determine the weight for the low-BF Locker units, we pooled the number of units with BFs below 1 kHz and with BFs between 1 and 8.5 kHz in from the three studies and calculate the average ratio (Table 2.2.C). 8.5 kHz was approximately the geometric mean of the CAP-audiogram edge frequencies in the exposed animal group, based on which the tonal pseudopopulation was defined. The weights for unusual type units were determined in a similar way (Tab. 2.2.D). These ratios were integrated and led to the finalized weights for different 76 A. Estimates of numbers of different types of VCN principal neurons SBC GBC Stellate × 15/16 Stellate × 1/16 + Octopus (Pri) (PriN) (Ch) (On) Osen, 1970 36800 6300 35300 × 15/16 = 33090 2210 + 1560 = 3770 Cant and Morest, 1984 31700 7500 39300 × 15/16 = 36840 2460 + 1100 = 3560 Melcher, 1993 28700 6500 35800 × 15/16 = 33560 2240 + 1600 = 3840 Average ratio 1 0.211 1.077 0.116 B. Numbers of different subtypes of chopper units ChS ChT ChL Blackburn and Sachs, 1989 127 127 4 Bourk, 1979 (Table IV-1) 60 79 30 The current study 14 31 3 Average ratio 1 1.510 0.249 C. Numbers of VCN neurons with BF within different BF intervals BF ≤ 1 kHz 1 kHz < BF < 11.3 kHz Blackburn and Sachs, 1989 89 316 Bourk, 1979, (Figure IV-27) 59 424 The current study 30 126 Average ratio 0.220 1 D. Numbers of usual- and unusual-PSTH-type VCN neurons Usual PSTH type Unusual PSTH type Blackburn and Sachs, 1989 544 32 The current study 149 28 Average ratio 1 0.1234 E. Weights for different VCN PSTH types Weight Pri PriN ChS ChT ChL On Locker Unusual 1.000 0.210 0.3904 0.5894 0.0972 0.116 0.5289 0.3618 Tab. 2.2. Determining weights for different VCN PSTH types. A. Numbers of GBC, SBC and stellate cells estimated in the three previous anatomical studies in the cat. The average ratios were used to determine the relative weights for the Pri, PriN and chopper PSTH groups. B. The numbers of three subtypes of chopper (ChS, ChT and ChL) recorded from in the two previous physiological studies in the cat and the current one. C. The numbers of units with BFs below 1 kHz and those with BFs between 1 and 11.3 kHz in the three physiological studies. The average ratio was used to determine the weight for Lockers. 11.3 kHz was approximately the geometric mean of the CAP-audiogram edge frequencies in the noise-exposed group in the current study, based on which the tonal pseudopopulation was defined. D. The numbers of recorded VCN units showing usual (PL, chopper, onset and locker) and unusual PSTH types in the two physiological studies. E. The ratios in Panels A – D were integrated and gave rise to the weights for different PSTH types used in generating average RLFs in the pseudopopulation approach. 77 PSTH types for performing averaging (summarized in Tab 2.2.E). Rate-matching curves, or the simulated loudness balance curves, were generated in the spirit of loudness balance curves in psychophysical studies of loudness recruitment (See Fig. 1.4 for examples). The rate-matching curves were generated for the portions of average RLFs above the thresholds. The thresholds in the average RLFs were determined in the following way. For unnormalized rates, the threshold rate was 2 spikes/s. For normalized rates, the threshold rate was 0.015. The slope of the rate-matching curves was fit by a line in a least square sense over the entire range. 2.7. The linear / nonlinear weighting model and the random spectral shape stimuli According to the linear-nonlinear weighting model (LNWM) (Yu, 2003; Yu & Young, 2001; Barbour and Wang, 2003; Young and Calhoun 2006; Bandyopadhyay, 2007), the mean discharge rate r of an auditory neuron can be approximated by the following equation: m2 m2 m2 r = R0 + ∑ wi S ( f i ) + ∑ ∑ wij S ( f i ) S ( f j ) , (2.6) i = m1 i = m1 j =i The stimulus is a temporally stationary signal. It’s broken up into frequency bands labeled m1 through m2 . S ( f i ) is the amplitude or power inside the i-th frequency band (bin) on a logarithmic (dB) scale. The first term R0 on the right-hand side of Equation (2.6) is the response to a stimulus with 0 dB amplitude (re a reference level) in all frequency bands. The second term characterizes the linear spectral encoding by the neuron, and wi is the linear weight for the i-th frequency bin. The third term reflects quadratic coding nonlinearity. Each element in the matrix [ wij ] reflects the strength of interaction between the energy in the i-th and j-th frequency bins. Higher order nonlinearities are not included in this model because it is 78 impractical to estimate large numbers of weights based on limited amounts of neural data. The linear and nonlinear weights of this model can be viewed as 2nd and 4th order Wiener-Volterra kernels collapsed along the time axis. The reference level can be varied and the parameters of the model can be derived under different reference levels. The random spectral shape (RSS) stimuli are an efficient and standardized way to estimate the parameters, including R0 and the 1st and 2nd order weights in this model. Two sets of RSS stimuli were used in this thesis study (Fig. 2.8.A). The RSS-B stimulus set was made up of 212 stimuli, each of which consisted of eleven 1/8 octave wide bins. The level-versus-bin-number profiles vary from stimulus to stimulus. The standard deviation of levels across bins for a stimulus is approximately 12 dB. 8 tones spaced by 1/64 octaves made up each frequency bin of each individual stimuli. These tones were equal in levels but randomized in phases, which makes the stimuli sound noise-like. The RSS-C stimuli set was similar to the B set, but the width of its frequency bins was 2 times that of the B set, and each bin consisted of 16 tone components. Due to the larger width of the bins, the stimulus set C is capable of capturing weights over wider frequency ranges, but is poorer than the set B in terms of frequency resolution. The RSS-B stimulus set was used for the neurons in the normal-hearing ears and half of the neurons in the exposed ears; whereas the RSS-C set was used for the other half of the neurons in the exposed ears. In a few cases, both sets of stimuli were presented to the same neuron in the impaired ear and comparisons were made between the two sets of model weights. Both sets of RSS stimuli were synthesized at 97.656 kHz and the central bin was centered at 10 kHz. During the recording experiments, the playback rate was chosen so as to align the center frequency with the BF of the neuron, in a way similar to the SMP for the vowel /ε/ (See Section 79 A RSS stim set B Stm. #1 B Stm. #2 Level Stm. #3 Stm. #212 -5 -4 -3 -2 -1 0 Frequency (Oct. re BF) 1 2 RSS stim set C Stm. #1 Stm. #2 Level Stm. #3 Stm. #212 -5 -4 -3 -2 -1 0 Frequency (Oct. re BF) 1 2 Fig. 2.8. The random spectral shape stimuli and analyses on the linear weight functions. A. Comparison of the RSS stimulus sets B and C. Stimulus set B has 0.125-octave wide frequency bins; where set C has 0.25-octave wide bins. The distributions of the levels are similar for these two stimulus sets. B. Quantification of 1st-order weight versus frequency function for the linear/nonlinear weight model. (1) Excitatory bandwidth is the bandwidth between the lowest frequency bin with a significant excitatory weight (> 1 S.D. above zero) and the highest one. (2) BF weight is the linear weight in the frequency bin closest to the BF. (3) Inhibitory ratio is defined as the area of the weight function below zero divided by its total area. 80 2.5.2). The weights and R0 in Eq. (2.7.1) were calculated in a least-square error sense by the method of normal equations. The goodness of fit was quantified by the quality factor (Yu and Young, 2000), 1 Q= , (2.7). n 1+ ∑ (ri − rˆi )2 i =1 n ∑ (rˆ − rˆ ) i =1 2 i in which n is the number of stimuli (212 for both stimulus sets), rˆk and rˆi are actual and model discharge rates in response to the i-th stimulus respectively, and r̂ is the average model firing rates for all stimuli in the set. The quality factor is a real number between 0 and 1. A quality factor equal to 1 indicate perfect modeling performance (zero error in all stimuli), were as lower values indicate worse performances. The standard deviations (S.D.s) of the 1st and 2nd order weight estimates were obtained by bootstrapping the data set for 200 times. Weights with at least 1 S.D. above or below zero were regarded as significant excitatory or inhibitory weights. Three measures were developped to quantify the 1st-order (linear) weight functions (Fig. 2.6.B). (1) Excitatory bandwidth was defined as the sum of the widths of frequency bins width significant excitatory 1st-order weights. (2) BF weight was defined as the 1st-order weight at a frequency bin nearest to the BF of the neuron. (3) Inhibitory ratio was defined as the ratio between the area below zero and the total area in the linear weight function. An entirely inhibitory weight function has an inhibitory ratio of one, while a weight function without inhibitory (negative) weights has a zero inhibitory ratio. 2.8. The unexpected acoustic noise in the dynamic speaker and its effects An unexpected acoustic noise was found in the dynamic speaker, which was used in five 81 recording experiments in noise-exposed cats. The numbers of these cats are 06-045, 06-047, 06-067, 06-080 and 06-081. This noise can be described as a hissing sound. It was measured by A/D converting the output of the Bruel and Kjaer sound probe. Five 1-second long measurements were made at 100 kHz for five times for each of the two conditions: 1) the sound delivery system turned off; 2) the sound delivery system turned on, but presenting no sound signals. The blue curve in Fig. 2.9.C shows the power spectrum of the probe amplifier when the acoustic system was turned off, which reflects the intrinsic noise of the sound probe / amplifier system and background noise of the recording chamber. The red curve shows the power spectrum of the signal picked up by the probe when the sound delivery system was on and playing no sound. The noise spectrum spans the frequency range from 0.3 to 4 kHz, and was not entirely flat, with several spectral peaks near 1.2 kHz. After correcting for the frequency response of the sound probe – amplifier system, we calculated the total level of the signal to be 28.1 dB SPL and its mean spectrum level to be -7.7 dB re 20uPa / Hz . The origin of the acoustic noise was determined to be the Crown amplifier. We were not aware of the noise during the five experiments. It drew our attention during an experiment, during which we decided to test the consistency of the two types of speakers by using the dynamic speaker on a normal-hearing cat. We observed that the CAP thresholds were about 10 – 15 dB above usual normal values. After we repeated the CAP measurements with the electrostatic driver, the thresholds were found to be normal. In that and all subsequent recording experiments, the electrostatic speaker was used. To study the effect of the speaker on the CAP and single-unit data, we first compared the background noise level of the analogue CAP recordings. Fig. 2.9.A shows the distributions of 82 Impaired Electrostatic x Impaired DCD x 100 Unit threshold (dB SPL) 0.5 Normal-hearing Electrostatic B 0.4 0.3 0.2 0.1 0 80 60 40 20 0 06-023 06-024 06-036 06-035 06-048 06-054 06-079 06-085 07-011 06-006 06-009 06-018 06-090 07-003 06-045 06-047 06-067 06-080 06-081 CAP background amplitude (V R.M.S.) A -20 0.1 C Threshold spectrum level (dB SPL) Power spectrum density(V2/Hz) fL = 0.3 kHz; fH = 4 kHz Total level = 28.41 dB SPL Mean power spectrum level = -7.27 dB re 20 uPa / sqrt(Hz) 10-10 10-12 10-14 10 Noise floor (Crown amp. off) Crown amp. on 0.1 1 Frequency (kHz) 1 Unit BF (kHz) 10 D 10-8 -16 Normal-hearing Exposed uncontaminated Exposed contaminated 40 20 0 -20 -40 Speaker noise spectrum 1 BF (kHz) 10 10 Fig. 2.9. The unexpected acoustic noise in the dynamic driver and its effects. A. Root-mean-square amplitudes of CAP background (20 - 25 ms after stimulus onset) in different experiments. Two experiments with impaired cats (06-067 and 06-081) using the deaf-cat speaker had unusually large background amplitudes, indicating contamination by the acoustic noise (spectrum shown in C and D). B. Comparing the single-unit thresholds gotten from normal (Blue), impaired-contaminated (Purple, 06-067 and 06-081), and impaired-uncontaminated (Red, the rest of the impaired pool) experiments. C. Measurements of background acoustic signals under Crown amplifier off and on conditions. The spectra are each average of five 1-second long recordings by the sound probe. D. Comparison of single-unit noise thresholds with the amplitude spectrum of the speaker noise. Three units from animals 06-067 and 06-081 whose BBN thresholds are with in 10 dB from the BBN spectrum (red arrows) were discarded. 83 the root-mean-square (RMS) amplitudes of the CAP recordings (the last 50 ms of each stimulus cycle, during which no stimulus was played). Among the five experiments with the deaf-cat driver, two (06-067 and 06-081, called CAP-contaminated animals hereafter) experiments yielded CAP recordings with noisy background, while the other three generated CAP background noise levels within the normal range. When comparing the single-unit thresholds from the two contaminated experiments from the rest of the data in the impaired pool, as shown by the purple data points in Fig. 2.9.B, it was evident that these two exposed animals had the least threshold increases among all the noise-exposed animals, which explained their microphonic response to the speaker noise. All subsequent data figures will indicate the data from the two noise-contaminated experiments by a purple color. In Fig. 2.9.D, the spectrum of the acoustic noise was plotted in the same axes with the single-unit thresholds for BBN. The purple data points indicate data from the two noise-contaminated experiments. Three units (black arrows) in the frequency range from 0.3 to 4 kHz from these two experiments had BBN thresholds within 10 dB of the noise spectrum. We excluded those three units from our data pool, and they were not used in any further analysis. The rest had BBN thresholds at least 20 dB above the noise spectrum, indicating that the neural activities of these units were not significantly affected by the acoustic noise. BBN rate-level functions were not taken for three other units with BF between 0.3 and 4 kHz from the noise-contaminated animals (blue arrows in Fig.2.10.B). These units were not discarded, based on the reasoning that another unit with similar BF and similar or lower BF-tone threshold as those three units (blue arrow in Fig.2.10.D) had a BBN threshold more than 20 dB above the noise spectrum. 84 The CAP threshold masking effect observed in the aforementioned normal-cat experiment with the dynamic speaker presumably also occurred for the two noise-contaminated experiments, because our CAP threshold calculation algorithm uses background signal amplitude to determine CAP amplitude threshold criteria. In order to correct for the upward biases in the CAP thresholds caused by this effect, we respectively shifted the two CAP audiograms downward uniformly for all frequencies from those two experiment by amounts such that the average differences between the single-unit thresholds and the (shifted) CAP thresholds of these two animals equals the average differences from the uncontaminated noise-exposed experiments. In the panels in Fig. 3.2 corresponding to animals 06-067 and 06-081, the dashed curves are the original CAP audiograms from these two animals; and the solid curves show the ones after this correction. 85 III. Results 3.1. Acoustic trauma in cats Acoustic trauma was induced in 10 previously normal-hearing cats by exposure to a 2-kHz centered narrowband noise at 111-112 dB SPL for 4 hours while the animals were anesthetized by xylazine and ketamine (see Methods Section 2.1 for details). Table 2.1 summarizes the sound levels of the exposures and the recovery periods between deafening and the acute recording experiments. 9 unexposed normal-hearing cats were used as controls. The hearing status of these control animals were confirmed to be normal based on the CAP audiograms measured before VCN neural recordings, as shown by solid blue curves in Fig. 3.4. Fig. 3.4 also summarizes the CAP audiograms of the noise-exposed animals (solid red curves). As shown in Fig. 3.2., the CAP audiograms of the two CAP-contaminated exposed animals (06-067 and 06-081) were shifted downward from the original values (dashed curves) to equate the average difference between single-unit thresholds and CAP thresholds between these two animals and the other exposed animals (See Section 2.8 for details). The average audiograms of the normal-hearing and noise-exposed groups (blue and red dashed curves in Fig. 3.4) were calculated by moving window averaging along the frequency axis. On average, the largest threshold shifts occur between 1 and 4 kHz. In the exposed group, there was substantial variability in the severity of hearing losses across different animals. About half of the exposed animals had threshold shifts between 25 – 35 dB relative to the average normal audiogram (dashed blue curve); whereas others had 35 – 55 dB threshold increases. This observation parallels the considerably variable single-ANF threshold 86 Threshold (dB SPL) 06-023 100 50 50 0 1 10 BF (kHz) 06-035 50 0 1 10 06-048 100 10 06-079 100 10 06-085 10 10 07-011 50 0 1 1 100 50 0 06-054 0 1 100 50 10 50 0 1 1 100 50 0 06-036 100 50 0 100 06-024 100 0 1 10 1 10 6 4 2 0 0 Unknown Pri/PriN Chopper Onset Locker Unusual 5 Fig. 3.1. A summary of CAP audiograms and recorded VCN units in normal-hearing animals. Each panel corresponds to a normal-hearing control animal, whose number is shown at the top. The CAP audiograms (Section 2.2) are shown by the black curves. The blue dots show BF and thresholds of recorded VCN neurons. Different symbols represent different PSTH types (See the legend).The vertical line in each panel indicates the edge frequency (see Section 2.2.1). 87 Threshold (dB SPL) 06-006 100 50 50 0 1 10 BF (kHz) 06-045 50 0 1 10 06-047 100 10 06-080 100 10 06-081 10 07-003 100 10 07-034 10 10 07-037 50 0 1 1 100 50 0 06-090 0 1 100 50 10 50 0 1 1 100 50 0 06-067 0 1 100 50 10 50 0 1 1 100 50 0 06-018 100 50 0 100 06-009 100 0 1 10 1 10 6 4 2 0 0 Unknown Pri/PriN Chopper Onset Locker Unusual5 Fig. 3.2. A summary of CAP audiograms and recorded VCN units in noise-exposed animals. The format is the same as in Fig. 3.1. The vertical gray lines show the CAP-audiogram edge frequencies. In each impaired audiogram, the edge frequency was defined as the logarithmic center frequency of the line segment with the most negative slope (See Section 2.2.1 for details). The CAP audiograms from the two experiments (06-067 and 06-081) with noise-contaminated CAP recordings were shifted downward to match the average differences between CAP thresholds and single-unit thresholds (See Section 2.6) for details. The dashed curves show the original threshold estimates. 88 shifts reported in a previous study in our laboratory using a similar animal model of NIHL (Heinz and Young, 2004). The variability of hearing loss is also reflected in the bandwidths of frequency regions with substantial threshold shifts. For example, in animals 06-018 and 06-080, significant threshold shifts can be seen from below 0.5 kHz to above 20 kHz; while in animals 06-047 and 07-003, threshold increases were confined to the frequency region between 0.5 and 10 kHz. There was a consistent correlation between the severity of threshold shift between 2 – 4 kHz and the frequency span of threshold elevations. Larger threshold shifts at 2 – 4 kHz are always associated with higher upper boundaries and lower lower boundaries of the threshold-shifted frequency regions. Despite the inconsistency of the amount and frequency range of threshold shifts in the noise-exposed group, the audiograms of the impaired animals shared a common shape, containing a steep down-sloping portion at the higher end of the most impaired region. Due to the large variance of the upper bound of the impaired frequency regions, it was impossible to choose a single frequency (e.g., 10 kHz) to delineate the boundary between impaired and relatively normal BF regions for the entire population. To address this problem, we took advantage of the common shapes of the exposed CAP audiograms, and used an edge-frequency method adapted from Ma and Young (2006) (see Section 2.2.1 for details). The edge frequencies of the exposed animals had a population geometric median of 11.3 kHz. Fig. 3.3.A shows the exposed CAP audiograms aligned at the edge frequency. Panel B shows the thresholds of the recorded VCN neurons versus their BFs in octaves re edge (ORE). It can be seen that the edge frequencies effectively divided neurons with substantial threshold shifts from those with little or 89 B Unit threshold shift (dB) 80 60 40 20 -4 -2 0 Frequency (oct. re edge) Pri PriN ChS ChT ChL On Locker Unusual Unknown Unknown PL 60 60 40 40 20 20 0 0 -20 -6 -20 -4 -2 2 Unit threshold shift (dB) -6 Unit threshold shift (dB) CAP threshold (dB SPL) A 0 2 -6 -4 Ch 60 40 40 20 20 0 0 -20 2 0 2 -20 -4 -2 0 2 -6 60 60 40 40 20 20 0 0 -20 -20 -4 -2 0 2 BF (oct. re edge) -4 -2 Locker Unusual -6 0 On 60 -6 -2 -6 -4 -2 0 2 BF (oct. re edge) Fig. 3.3. Alignment of frequency axes at the CAP-audiogram edges. Edge-aligned CAP audiograms of the noise exposed cats (A), and single-unit threshold shift versus unit BF in octaves re CAP-audiogram edge frequencies (ORE) (B). The threshold shift of a neuron was defined as the difference between its BF threshold and the threshold on the average normal-hearing CAP audiogram at the corresponding frequency. The vertical gray lines show the zero OREs. Each panel in B corresponds to a unit response type. Color code for hearing conditions: Blue: normal-hearing; Red: noise-exposed, CAP uncontaminated; Magenta: noise-exposed, CAP contaminated. The CAP audiograms reliably delineated threshold-shifted and near-normal-threshold BF regions. All units with negative OREs were units with substantial threshold shifts. 90 80 70 70 60 60 CAP Threshold (dB SPL) CAP Threshold (dB SPL) 80 50 40 30 20 10 0 50 40 30 20 10 n(Normal)=9 n(Exposed)=12 1 10 Frequency (kHz) 0 Exposure stimulus Normal average n(Normal)=9 Exposed average n(Exposed)=12 06-023 06-024 06-036 06-035 06-048 06-054 06-079 06-085 07-011 06-006 06-009 06-018 06-045 06-047 06-067 06-080 06-081 06-090 07-003 1 10 07-034 Frequency (kHz) 07-037 Fig. 3.4. Individual and average CAP audiograms in the normal and hearing-impaired populations. Solid blue and red curves show the CAP audiograms of individual normal-hearing and noise-exposed animals. Blue and red dashed curves are the average audiograms (population arithmetic mean on a dB scale) for the normal-hearing and noise-exposed populations, respectively. The 2-kHz centered narrow-band noise for inducing acoustic trauma is indicated by the gray vertical bar. The width of this bar is drawn to scale. Purple color indicates two experiments in which CAP recordings were contaminated by the acoustic noise in the dynamic driver (Section 2.8). 91 no threshold shifts. In Heinz and Young (2004), the animals in the noise-exposed group were divided into a mild HL sub-group and moderate/severe HL one according to the single-fiber threshold shifts. Based on the CAP threshold shifts, the exposed animal group in the current study could be subdivided in a similar way. However, it was impractical to perform this fine division due to the limited amount of neuronal data. Therefore, we addressed the problem of variability in threshold shifts in a different way, namely, to adjust the sound levels in ways such that the variability of single-neuron thresholds across different animals in each of the two populations was accounted for. Details of the SLA and SLB level adjustments are described in Method Section 2.6.5, and will be used in a number of the following analyses. 3.2. Basic unit characterization and classification Single VCN neurons were recorded from in normal-hearing and acoustically traumatized animals. As shown in Figs. 3.1 and 3.2, for both populations, CAP thresholds faithfully reflected the single-unit thresholds at corresponding BFs, which justified the use of the CAP audiograms and their averages for level adjustment in pseudopopulation analyses (Section 2.6.5). Units were classified into five major categories: primary-like (Pri), primary-notch (PriN), Choppers (Ch), onset (On), unusual (Un) and Lockers based on analyses of their BF-tone PSTHs at 30 dB re threshold. Because Pri and PriN PSTHs both had anatomical correlates of bushy cells and share many properties, they were treated as one group denoted PL in some following analyses. The Unknown category contained units for which unit holding times didn’t allow completion of the PSTH collection. Fig. 3.3.B shows threshold-versus-BF plots for the six types 92 of neuron. Their BFs are expressed in octaves re edge frequencies (ORE). Regardless of PSTH type, the sub-edge units were all units with substantial threshold shifts (> 20 dB). There is a slight trend for the threshold shifts to be larger for OREs between -4 and -2 than for OREs between -2 and 0. The supra-edge neurons shows variability in their threshold shifts. While some of the supra-edge units (especially those with OREs less than 0.4) had threshold shifts falling into the ranges of the sub-edge units, others have thresholds indistinguishable from those of the unexposed neurons. Overall, the edge frequencies provided a relatively conservative way to derive an upper bound for the impaired frequency region. It will be used in most of the following neuronal population analyses unless otherwise stated. As shown in Fig. 3.2.B, the recorded Locker neurons in both populations all had best frequencies less than 0.8 kHz and therefore their OREs are often very negative. The Lockers are the unit group that showed the largest variability in threshold shifts. The Lockers from the two CAP-contaminated cats (purple data points), which had mild HL, had thresholds almost falling into the normal range. Lockers from other more impaired animals had greater threshold shifts. This is one reason why the low-BF phase-locking neurons were treated as a separate group in the current study. Generally speaking, the decision tree for classifying VCN neurons described in Blackburn and Sachs (1989) applied to the units from both unexposed and exposed ears in the current study, which suggests that major PSTH types were preserved after acoustic trauma. However, this is not equivalent to saying that the PSTH categories of VCN neurons were completely unaltered by NIHL. Tab. 3.1 summarizes the numbers of neurons recorded from in the two populations by PSTH types and positive/negative OREs. One salient distinction between the PSTH type 93 Normal-hearing BF ≤ fEdge BF > fEdge BF ≤ fEdge BF > fEdge Pri 4 1 21 7 PriN 7 5 13 6 ChS 4 1 7 2 ChT 7 5 8 10 ChL 1 0 1 1 OnI 2 0 0 1 OnL 5 2 1 3 OnC 2 0 2 3 5 0 11 0 Unusual-A 6 1 6 3 Unusual-B 3 1 6 1 18 10 23 9 64 26 99 46 PL Ch On Locker Unusual Noise-exposed Unknown Total 90 145 Tab. 3.1. A summary of unit numbers in different PSTH categories and ORE regions in normal and exposed populations. Another 15 recorded units were judged to be of DCN origin and were excluded from further analyses. 94 distributions in the normal and impaired ear was the significantly greater prevalence of the PL units and a scarcity of onset neurons in the impaired ears (Normal-hearing population: 11 PL and 9 On; Noise exposed population: 34 PL and 3 On. p = 0.002, two-tailed Fisher’s exact test). If it can be assumed that unit type sampling was commensurate in the two populations, this jibed with a conjecture that acoustic trauma induced the transition of onset PSTHs into other types. One possibility is that they turned into PSTH types that are indistinguishable from PriN neurons. An effort was made to sample only the impaired neurons in impaired ears. However, a large number of supra-edge neurons were nonetheless recorded from because of the way we access the VCN with electrodes and the tonotopic gradient of the structure. The chopper, onset, and unusual groups were heterogeneous categories and further divided into subtypes, which will be discussed shortly. 3.2.1. Primary-like The Pri units are one of the most commonly encountered unit types in VCN. Panels A and B in Fig. 3.5 show two representative PSTHs of the Pri category, respectively from a normal-hearing and a noise-exposed one. Both units had BFs lower than the CAP audiogram edge frequencies. The unit shown in Panel B had a substantial threshold elevation. Both PST histograms were taken at 30 dB above thresholds. The defining features of a Pri PSTH included relatively a high onset peak followed by a smooth decay (adaptation) of firing rate; relatively short minimum first-spike latency (FSL); and a dispersed distribution of FSLs across stimulus repetitions (upper left insets), reflected in their large standard deviation. These features were seen regardless of hearing states. The upper right inset in each panel shows the time course of changes in mean ISI and its standard deviation. Firing regularities, as quantified by CV of the ISIs during 95 B. Pri (Exposed) 07-011, Unit 1.12 BF = 4.26 kHz, Th = 23.0 dB SPL 06-018, Unit 2.01 BF = 10.20 kHz; Th = 58.0 dB SPL 0 10 F.R. (spike/s) 600 0 20 0 20 40 FSLMin/STDFSL=3.73/3.495ms P/S=3.670 CVAVG =0.751 200 0 0 20 40 Time (ms) ISI (ms) 5 400 0.05 Fraction 0.1 10 0 10 500 400 F.R. (spike/s) 0.2 ISI (ms) Fraction A. Pri (Normal) 0 20 0 20 40 FSLMin/STDFSL=3.79/2.595ms P/S=4.266 CVAVG=0.918 300 200 100 0 60 0 20 40 Time (ms) 60 C. PriN (Normal) D. PriN (Exposed) 06-048, Unit 2.03 BF = 6.21 kHz, Th = 7.1 dB SPL 06-067, Unit 1.04 BF = 6.64 kHz, Th = 43.0 dB SPL 10 0 10 1500 0 20 0 20 40 FSLMin/STDFSL=3.41/0.226ms P/S=10.839 CVAVG=0.680 1000 500 0 0 20 40 Time (ms) ISI (ms) Fraction 5 0.1 0 1500 F.R. (spike/s) Fraction ISI (ms) 0.05 F.R. (spike/s) 20 0 20 0 20 40 FSLMin/STDFSL=2.80/0.886ms P/S=5.246 CVAVG=0.652 1000 500 0 60 10 5 0 20 40 Time (ms) 60 Fig. 3.5. Examples PSTHs of Primary-like (Pri) and Primary-like-with notch (PriN) units, in response to 50-ms BF tones bursts at 30 dB above threshold, from normal-hearing (left column) and noise-exposed (right column) ears. All units shown in this figure had negative OREs. In each panel, the main plot on the bottom shows PSTH in 0.2-ms time bins. The inset on the top left corner shows the histogram of the first-spike latencies (FSLs) for all the stimulus repetitions. The inset on the top right shows mean interspike interval (ISI) versus time (blue), along with standard deviation of the ISIs (magenta). FSLMin and STDFSL are the minimum and standard deviation of the first-spike latencies, respectively. P/S is the ratio between the peak onset firing rate during and the sustained firing rate during the second half of the stimulus duration. CVAVG is the average coefficient of variances of the ISIs inside the [30, 40]-ms time window following stimulus onset. See Methods Section 2.6.2 for details on how these quantities were derived. The BF and thresholds of the units are shown in the captions to the panels. No significant differences in the PSTH patterns were seen in the PL group between normal and impaired hearing conditions. 96 the sustained portion of firing, are usually greater than 0.6, sometimes approaching 1 (e.g. in Fig. 3.5.B). In normal hearing ears, the Pri PSTH type has been shown to be correlated with the anatomical cell type of spherical bushy cells (SBC). These properties of Pri PSTHs closely resemble those of AN afferent fibers, which reflects the functions of the endbulbs of Held, the extremely secure AN endings onto SBCs. 3.2.2. Primary-like-with-notch PriN units have morphological correlates of globular bushy cells (GBCs) (Rhode et al., 1983). The PSTHs of an exemplar unexposed PriN unit and an exposed one are shown in Panels C and D of Fig. 3.5. Their shapes differ from Pri ones in the existence of a brief notch immediately after the onset peak. The notch often but not always brought the firing rate down near zero. The onset peaks are usually very sharp, because of the precise timing of the first driven spikes, as shown by the sharp FSL histograms and small SDs of FSLs. Pri and PriN neurons share the property of relatively short FSLs, but are discernable by the preciseness of first-spike timing. The precise timing of the first spikes in PriN units caused the absolute refractory periods to line up and gave rise to the notch. The resumption of firing after the notch is often abrupt, giving rising to a second peak, which can be seen in both examples in Fig. 3.5. PriN types and their stereotypical features were observed in both hearing states. Similar to Pri neurons, the PriN units also had low discharge regularities. 3.2.3. Chopper Fig. 3.6 shows representative PST histograms of the chopper neurons recorded in normal-hearing and noise-exposed ears. There are several major features that distinguished 97 06-047, Unit 1.03 BF = 4.49 kHz; Th = 47.4 dB SPL 10 F.R. (spike/s) 1500 0 20 0 20 40 FSLMin/STDFSL=4.33/0.234ms 0 P/S=4.767 CVAVG=0.163 1000 10 1000 F.R. (spike/s) 0 2 0.05 ISI (ms) 0.05 Fraction B. ChS (Exposed) 06-079, Unit 1.04 BF = 9.19 kHz, Th = 15.9 dB SPL ISI (ms) Fraction A. ChS (Normal) 500 800 5 0 20 0 20 40 FSLMin/STDFSL=4.32/0.397ms P/S=2.665 CVAVG=0.291 600 400 200 0 40 Time (ms) 0 60 0 20 40 Time (ms) 60 06-067, Unit 1.10 BF = 5.67 kHz, Th = 40.5 dB SPL 0.05 10 0 20 0 20 40 FSLMin/STDFSL=4.04/0.252ms P/S=6.728 CVAVG=0.565 1000 500 0 0 20 40 Time (ms) 0.05 0 2000 F.R. (spike/s) 0 5 ISI (ms) D. ChT (Exposed) 06-079, Unit 2.01 BF = 7.97 kHz, Th = 2.7 dB SPL Fraction C. ChT (Normal) 1500 F.R. (spike/s) 20 ISI (ms) Fraction 0 1500 0 20 0 20 40 FSLMin/STDFSL=3.53/0.309ms P/S=3.430 CVAVG=0.424 1000 500 0 60 10 5 0 20 40 Time (ms) 60 ISI (ms) Fraction C. ChL (Exposed) 06-081, Unit 1.07 BF = 0.45 kHz, Th = 52.1 dB SPL 0.05 0 10 F.R. (spike/s) 1000 800 5 0 20 0 20 40 FSLMin/STDFSL=12.14/0.656ms P/S=5.250 CVAVG=0.397 600 400 200 0 0 20 40 Time (ms) 60 Fig. 3.6. Examples of Chopper-type units. Three sub-types of Chopper PSTHs from sub-edge BF regions in normal and exposed ears are shown. The formats are the same as in Fig. 3.5. 98 chopper units from PL ones. (1) As can be seen by comparing minimum FSLs in Fig. 3.6 to those in Fig. 3.5, chopper units had response latencies significantly longer than the PL neurons at similar BFs (See Section 3.3.1). (2) The timing of the first driven spikes of chopper units are significantly more precise than Pri units, but about as precise as those in PriN units. (3) Interspike intervals were by far more regular in choppers than in PL units, as can be seen in the low CVs of ISIs, which usually fall below 0.6. The first 3 or more driven spikes were not only well timed, but also highly reproducible across stimulus repetitions. Each unit in the chopper category passed the spikes-per-peak (SPP) test by showing SPPs between 0.95 and 1.05 for the first peak and between 0.85 and 1.10 for the second peak. In impaired ears, the two intervals were relaxed to [0.9, 1.1] and [0.85, 1.1]. There were several subtypes within the chopper category, which could be distinguished from each other based on the regularities of firing and the time courses of firing rate changes. In both hearing states, the chopper branch of the decision tree in Blackburn and Sachs (1989) was readily applicable to the vast majority of the chopper units. They were classifiable into sustained and transiently subtypes. Sustained choppers (ChS) have mean ISIs that stay constant or increase slowly in a linear manner throughout the course of the 50-ms stimulus (Fig. 3.6.A and B). In ChS units, the regularity of firing was largely time-independent. The gradual disappearance of discernable peaks in their PST histograms merely reflects the accumulated randomness with increasing spike counts. In contrast, transient choppers (ChT, Fig. 3.6.C and D) show gradual sagging in discharge rate and regularity with increasing time. In some ChT units as the one shown in Panel D, the mean ISIs exhibited a steep up-ramp 5 – 10 ms following response onset. Unsurprisingly, ChT units showed higher ISI CVs during sustained portion of firing compared 99 with ChS units. There was no evidence these properties regarding latency and discharge regularity differ between the normal-hearing and exposed populations. The low-firing-rate chopper (ChL) is an occasionally encountered chopper subtype. The ChL units showed multimodal PSTH patterns similar to the other subtypes and passed the SPP tests as other subtypes of choppers did. However, they usually exhibited long latencies and chopped at significantly longer time intervals (> 5 ms). We recorded from one ChL unit from an impaired ear (Fig. 3.6.E). The unit had a low BF (0.45 kHz). The PST histogram was acquired with randomized initial phases. Therefore the multimodal shape of the PSTH reflects intrinsic chopping of the neuron, instead of phase-locking (See section 3.2.5 for further discussion). 3.2.4. Onset The defining feature of onset units were precisely timed one or few onset spikes followed by little or no sustained firing. There were a few marginal cases in which onset PSTH shapes resemble PL ones. The following criteria were used to demarcate the boundary. (1) Onset units had sustained rates below 100 spikes/s. (2) Unlike in PL neurons, onset units had abrupt, instead of exponential-like decay of firing rates. (3) As shown in the PL branch in the decision tree in Fig. 1.9, units with large onset timing irregularities and no visible notches irregularity fell into the Pri category. Fig. 3.7 exemplify the features of onset PSTHs. Unsurprisingly, onset units always had very large, sometimes infinite P/S ratios. Due to the lack of sustained firing, ISI CVs were undefined for many of the onset units (Fig. 3.7.A). In the normal-hearing ears, the onset units with sustained firing showed large the ISI CVs (e.g., Fig. 3.7.B & C). Interestingly, this was untrue in all three onset neurons recorded in the sub-edge regions in the impaired ears, which all possessed ISI CVs falling below 0.6 (e.g. Fig. 3.7.D & E, also see Fig. 3.11.A). 100 B. OnL (Normal) 07-011, Unit 1.03 BF = 6.29 kHz, Th = 22.8 dB SPL 06-048, Unit 5.02 BF = 2.46 kHz; Th = 10.5 dB SPL 10 0 20 40 Time (ms) 10 40 20 0 20 0 20 40 FSLMin/STDFSL=3.30/0.931ms P/S=19.147 CVAVG=0.974 1000 500 0 60 0 20 40 Time (ms) 60 07-037, Unit 4.06 BF = 6.43 kHz, Th = 50.4 dB SPL 0.05 0 10 1500 20 0 20 40 Time (ms) 10 3000 P/S=20.373 CVAVG=0.828 500 0 0.05 0 0 20 0 20 40 FSLMin/STDFSL=4.11/0.311ms 1000 ISI (ms) D. OnC (Exposed) 06-024, Unit 3.01 BF = 7.85 kHz, Th = -13.5 dB SPL Fraction C. OnC (Normal) F.R. (spike/s) Fraction 0 500 ISI (ms) F.R. (spike/s) 1000 0.1 1500 P/S=Inf CVAVG=NaN 1500 0 F.R. (spike/s) 0 20 0 20 40 FSLMin/STDFSL=3.46/0.162ms F.R. (spike/s) 0 2000 2 0.2 ISI (ms) 0.02 Fraction 4 0.04 ISI (ms) Fraction A. OnI (Normal) 10 0 20 0 20 40 FSLMin/STDFSL=3.29/0.093ms P/S=31.677 CVAVG=0.536 2000 1000 0 60 0 20 40 Time (ms) 60 D. OnL (Exposed) 07-034, Unit 2.14 ISI (ms) Fraction BF = 3.5229 kHz, Th = 64.2 dB SPL 0.1 F.R. (spike/s) 2000 0 10 1500 10 5 0 20 0 20 40 FSLMin/STDFSL=3.23/0.991ms P/S=11.328 CVAVG=0.317 1000 500 0 0 20 40 Time (ms) 60 Fig. 3.7. Example PSTHs of Onset-type units. Subtypes of the chopper PSTH, including OnI, OnL and OnC are shown. The layout is the same as in Fig. 3.5. 101 Similar to the Chopper group, the onset category could also be subdivided. Ideal onset (OnI) units exhibited one extremely sharp timed onset peaks (sometimes followed by one less reproducible second spike), and total silence of discharge during the rest of the stimulus duration (e.g., Fig. 3.7.A). L-shaped onset units (OnL) differ from OnI units in their non-zero sustained firing rates (e.g., Fig. 3.7.B & E). Some OnL units had PSTH shapes resembling the PriN ones. However, notches in OnL PSTHs are usually wider and the following resumption of firing is gradual, making the absence of the second peak. Onset-chopper units (OnC, Fig. 3.2.C and D) had 2-3 precisely timed and uniformly spaced spikes shortly after tone onset. All the OnC units passed the SPP tests (Section 2.6.1), which confirmed the reliability of the onset chopping patterns. It has been suggested that the morphological correlates of Onset neurons are diverse, including radiate stellate cells in the AVCN and octopus cells in the PVCN (Young and Oertel, 2004; See Sections 1.4.2 and 1.4.3). Interestingly, in exposed animals, most onset units were sampled only in supra-edge BF region; only three onset units were found in the sub-edge regions (Tab. 3.1). Because the sample of onset units in this study was small, it was not clear whether this merely reflected sampling unevenness or suggested a decrease in prevalence of onset firing patterns in the VCN in permanent SNHL. This observation raises the possibility that radiate stellate cells and/or octopus cells with substantial threshold shifts lose their onset firings patterns and transform into sustained PSTH types after acoustic trauma, contributing to greater and abnormally fast growing total spike counts in the VCN, which could be a neural correlate of loudness recruitment. However, this is merely a wild conjecture awaiting further investigation. 3.2.5. Locker 102 Most recorded VCN neurons with BFs lower than 1 kHz showed phase-locking. Fig. 3.7 shows that for a phase-locking low-BF neuron, tone repetitions with fixed initial phases generated a multi-peak PSTH shape (Panel A), which was unseen if the random initial phases were used (Panel B), indicating that the pattern is a manifestation of phase-locking, instead of intrinsic chopping dynamics of the neuron. Phase-locking also gave rise to low ISI CVs, which should also be distinguished from intrinsic regular firing in Choppers. As shown in Panels C and D, under both fixed and randomized initial phases, almost all ISIs were multiples of the period of the tone (1 / f). In Blackburn and Sachs (1989), low-BF units were categorized into PL and Choppers based on two clues other than PSTH shapes, (1) low-BF chopper neurons tended to have longer latencies than low-BF Pri units (Fig. 1.11, Left); (2) unlike low-BF Pri units, low-BF Choppers show multimodal PSTH shapes even under asynchronous (random phase) tone stimuli. Another clue that could facilitate the categorization of low-BF VCN neurons is the presence of pre-potentials (Bourk, 1976). However, as will be discussed in detail in the next section, low-BF units recorded from in the current study all had latencies above the lower limit of chopper latencies reported in Blackburn and Sachs (1989), suggesting that all these units should be categorized as choppers. However, contradictorily, when asynchronous tone stimuli were used, most of these low-BF neurons didn’t show multimodal PSTHs. Possible causes of this inconsistency were differences in animals strain and/or differences in states of anesthesia (in Blackburn and Sachs (1989), a ketamine-anesthetized non-decerebrated preparation were used). Since it was impractical to derive a new latency boundary between the low-BF Pri and Chopper neurons based on our rather small unit samples, we treated these units as a separate category called Locker. Another justification for separately analyzing low-BF VCN neurons was that the 103 10 F.R. (spike/s) 3000 P/S=11.014 CVAVG=0.431 0 20 40 Time (ms) 10 800 1000 0 0.02 0 0 20 0 20 40 FSLMin/STDFSL=6.28/0.356ms 2000 4 0.04 F.R. (spike/s) 0 4000 5 ISI (ms) 0.1 B. Randomized initial phase 0.464-kHz tone Fraction ISI (ms) Fraction A. Fixed initial phase 0.463-kHz tone 600 400 200 0 60 0 20 40 Time (ms) 60 D 800 1000 800 600 Number of ISIs Number of ISIs 0 20 0 20 40 FSLMin/STDFSL=5.61/0.536ms P/S=2.193 CVAVG=0.403 C 600 400 200 0 2 0 2 4 6 ISI (ms) 8 10 400 200 0 12 0 2 4 6 ISI (ms) 8 Fig. 3.8. Example PSTHs of Locker-type units. An example of a low-BF phase-locking unit (Locker), for which PSTH was constructed for a BF tone presented 30 dB above threshold, in fixed initial (A) and uniformly randomized (B) initial phases. This unit comes from a normal-hearing animal (06-079, Unit 2.03). Its BF is 0.463 kHz and its threshold is 15.3 dB SPL. The two panels (C & D) at the bottom show the distribution histograms of the interspike intervals under fixed and random initial phases. 104 10 low-BF units showed less threshold shifts than units with higher BFs, because they were farther away from the CAP-audiogram edge frequencies (Fig. 3.3.B). 3.2.6. Unusual Units that fell into the unusual category mostly had deep spatial locations (greater than 2000 μm from the dorsal surface of the CN). Many unusual-units were located in close vicinities of units showing regular VCN PSTH types. For example, the unusual unit shown in Panel C of Fig. 3.9 was located within 200 um from a Pri and a PriN unit, and its BF agreed with the BF trend of the electrode track. This evidence strongly suggested that the units labeled unusual in the current study were not sampled from DCN or the granular cell domain. Unlike the typical PSTH type categories, the unusual category is rather heterogeneous, consisting of units with wildly different PSTH contours. However, there are several common properties which can be said of the unusual neurons, (1) long-latency and imprecisely timed first driven spikes; (2) lower mean firing rates (as compared with regular VCN PSTH types). These properties are exemplified by the four Unusual PSTHs shown in Fig. 3.9. However, the second property didn’t hold in a few unusual units recorded from in impaired ears. For example, the impaired unit shown in Panel D had a sustained rate greater than most of the regular-type VCN units, which can be seen in the previous five figures. This unit was located in close vicinity (< 200 um distances) to a Pri and a chopper unit. Also, its precisely timed first spike and relatively low ISI CV suggested that it might be recorded from a stellate cell. However, no obvious chopping pattern could be observed and the SPP test couldn’t be passed. This unit may be an example of abnormal PSTH type in traumatized ears. As will be shown later (Sections 3.5.2 and 3.6.1), this unit also had an abnormal level-frequency response map and a steep RLF. 105 Regularity of spiking can be high in certain unusual neurons (e.g., the normal unusual unit shown in Fig. 3.9.A). But units of this type didn’t show well-timed first spikes and typical chopping patterns; neither could they pass the SPP test. In addition to these “barely chopper” units, the unusual category also included units which showed Pri-looking PSTHs but had long latencies (06-009, T5U1), according to the PL branch in the decision tree shown in Fig. 1.9. The unusual category was subdivided into two subtypes based on the relative strengths of their onset and sustained firing. Unusual-A type neurons had P/S ratios greater than 3 (e.g., Fig. 3.9.A and C), and can be described as onset-like or “peaky”; Unusual-B type PSTHs had P/S ratios less than 3 (e.g., Fig. 3.9.B and D), can be described as sustained-firing or “boxy”. Both Unusual subtypes were seen in both hearing states. The Unusual-A unit shown in Fig. 3.9.C which had a high P/S ratio was representative of a number of PSTHs recorded from in the impaired ears. These PSTHs had several relatively precisely-timed spikes within 10 ms after stimulus onset, followed by a sudden and step-like drop of firing rate to a much lower value. All PSTHs of this type were recorded from in the sub-edge BF regions in impaired ears. Similar PSTHs were not encounters in normal ears. These units were qualitatively similar to OnC units but quantitatively distinct because of their imprecise onset-spike timing and inability to pass the SPP tests. Three other units of this type had short latencies consistent with the PL category and were classified as Pri units. The one shown in Fig. 3.9.C had a long latency and was therefore put into the unusual category according to the decision tree in Fig. 1.9. 106 A. Unusual-A (Normal) B. Unusual-B (Normal) 06-048, Unit 7.03 BF = 4.28 kHz, Th = 45.9 dB SPL 06-036, Unit 2.01 BF = 6.0 kHz, Th = 10.2 dB SPL 0 10 F.R. (spike/s) 600 10 0 20 0 20 40 FSLMin/STDFSL=6.51/1.209ms P/S=7.636 CVAVG=0.292 400 200 0 0 20 ISI (ms) Fraction 0.05 40 Time (ms) 0 250 F.R. (spike/s) ISI (ms) Fraction 0.05 100 50 0 20 F.R. (spike/s) 10 600 P/S=7.692 CVAVG=0.532 0 20 40 Time (ms) 0.1 ISI (ms) Fraction 0 20 0 20 40 FSLMin/STDFSL=5.02/3.161ms 200 0 80 100 07-003, Unit 2.07 BF = 3.98 kHz, Th = 68.6 dB SPL 10 400 40 60 Time (ms) D. Unusual-B (Exposed) 0.05 0 1000 F.R. (spike/s) ISI (ms) Fraction 0 800 0 20 40 FSLMin/STDFSL=10.46/4.394ms P/S=1.138 CVAVG=0.610 06-067, Unit 1.05 BF = 4.67 kHz, Th = 52.2 dB SPL 0.1 0 150 C. Unusual-A (Exposed) 0.2 20 200 0 60 10 10 20 800 600 0 0 20 40 FSLMin/STDFSL=4.04/1.096ms P/S=2.081 CVAVG=0.562 400 200 0 60 10 5 0 20 40 Time (ms) 60 Fig. 3.9. Example PSTHs of unusual-type units. The unusual category was subdivided into Unusual-A and Unusual-B sub-categories based on P/S ratios. Unusual-A units had high P/S ratios (> 3, panels A and C); Unusual-B units had low P/S ratios (< 3, panels B and D). Latencies were generally long and variable in unusual-type units. Most unusual-type units had much lower sustained firing rates compared to regular PSTH types (A, B & C). However, this didn’t hold true for some impaired unusual units (D). 107 3.2.7. Quantitative comparisons of BF-tone PST histograms Fig. 3.10 shows the peak and sustained firing rates in neurons of different PSTH types in the two hearing states. These rates were calculated from BF-tone PSTHs recorded between 26 and 34 dB re threshold. This deviation from 30 dB re threshold was because the thresholds calculated offline (Section 2.6.4.2) often slightly differed from those estimated online audiovisually or by the automatically generated tuning curves. For both normal-hearing and exposed populations, only those units with BFs below the CAP-audiograms were included. Notice that not all units we recorded from are included in this figure, because in some units it was impossible to record responses at about 30 dB above threshold, due to strong ER or limited deliverable sound level. The error bars in Fig. 3.10 indicate ± 1 S.E. The p-values shown on the upper-right corner of each figure are the p-values of a two-way ANOVA, in which the two factors are hearing status (unexposed versus exposed) and unit types, respectively. Inside each unit category, a two-tailed Wilcoxon rank-sum (WRS) test was done. Asterisks indicate significant differences across hearing status in the response groups at the significance level of 0.05. Peak firing rates of the major VCN unit types (Pri, PriN, Ch and On) were all high, compared with the unusual-type neurons. The low peak rates in the Lockers were due to their phase-locking and the use of asynchronous initial phases (See Fig. 3.8). As expected, the unit types with more precise FSLs (PriN, Ch, and On) showed significantly higher peak rates than the types with less reliable FSL timing (Pri and Unusual). As indicated by p(HS) of the two-way ANOVA, the peak rates at 30 dB re threshold generally didn’t differ significantly between the normal and exposed populations. PSTH-type-wise, the PriN units in impaired ears showed significantly lower peak firing rates than the normal-hearing PriN units (p<0.05, two-tailed WRS 108 A1 3 * 2500 8 2000 1500 7 1000 9 14 12 3 12 500 8 0 Pri PriN Ch 2500 2000 1500 1000 500 0 1 10 5 7 6 On Unsl.Locker 2500 2000 1500 1000 500 0 Unit BF (kHz) A3 1000 800 600 400 200 0.1 1 Unit BF (kHz) 10 A4 1 Peak firing rates (spikes/s) p(HS) = 0.317 3000 p(UT) = 1.110e-016 Peak firing rates (spikes/s) Peak firing rates (spikes/s) 3500 A2 Peak firing rates (spikes/s) Peak firing rates (spikes/s) A 1000 800 600 400 200 0 Unit BF (kHz) 1 Unit BF (kHz) 10 p(HS) = 0.210 500 *14 p(UT) = 2.176e-009 7 6 3 12 PriN Ch 8 On Unsl.Locker C log10 (P/S ratio) 3 * 1 12 7 12 0.5 0 9 Pri PriN Ch 500 200 400 Sustained rate (sp/s) 2000 1500 1000 500 0 0 600 C3 8 14 1000 8 3 3 2500 1500 0 0 2.5 p(HS) = 0.710 p(UT) = 0.000e+000 2 1.5 C2 2000 8 Pri C1 3 100 0 7 Peak rates (sp/s) 200 5 9 12 Peak rates (sp/s) 300 6 5 7 On Unsl.Locker 1000 1000 800 800 600 400 200 0 0 100 200 300 Sustained rate (sp/s) 109 200 400 Sustained rate (sp/s) 600 C4 Peak rates (sp/s) 400 Peak rates (sp/s) Sustained firing rates (spikes/s) B 400 600 400 200 0 0 200 400 Sustained rate (sp/s) 600 Fig. 3.10. Quantitative analyses on BF-tone PSTH at 30 dB re thresholds. I. Firing rates. Units in normal and impaired ears with BFs below the CAP-audiogram edge frequencies were included in this figure. In the left column, data are plotted in the format of mean ± 1 S.E. Asterisks indicate significant intra-PSTH-category differences across the between the two hearing states (Two-tailed Wilcoxon rank-sum test, α = 0.05). A. Peak firing rates, measured as the maximum firing rates during the first 15 ms after stimulus onset in 0.2-ms time bins. In Panels A1 – A4, the peak firing rates are plotted against unit BFs for the PL, chopper, locker and unusual units, respectively. B. Sustained firing rates, measured as the mean firing rates between 25 and 50 ms after stimulus onset. C. P/S ratios, viz., the ratio between peak and sustained rates. Panels C1 – C4 show peak-versus-sustained rates for PL, chopper, locker and unusual units. A two-way ANOVA (factors: hearing status and unit type) was performed on each qualitative measure. The p-values for the hearing states and unit types are shown as p(HS) and p(UT) in each plot in the left panel. 110 test. This difference might have resulted from a deterioration in first-spike timing in the PrN category after HL, as can be seen in Fig. 3.11.B. The unusual category was another exception. On average, impaired unusual units showed higher peak rates than their normal counterparts, although this difference didn’t reach statistical significance. The three panels A1 – 4 on the right-hand side showed the relationships between peak firing rates and unit BF, in the PL, Ch, On and Locker categories. In the Chopper category, there was some evidence that higher BFs were associated with higher peak rates. Also, the ChL subtypes showed much lower rates than ChS and ChT subtypes. In other PSTH types, there was no significant indication that peak rates were dependent on BF. The higher peak rates seen in the impaired unusual category didn’t seem to be caused by a discrepancy of BF distributions between the two hearing states. As shown in Fig. 3.9.B, sustained rates were invariably less than peak rates. This trend is most salient in the Onset group. Sustained rates didn’t differ significantly between normal and impaired ears in the PL category. The Locker neurons showed a non-significant trend to have lower sustained rates in impaired ears. In contrast, chopper and unusual neurons tended to show higher sustained rates following acoustic trauma. These differences reached statistical significance (p = 0.013, two-sided WRS test). Onset and unusual types were the other two categories which exhibited enhancement of sustained firing rates after HL, although these changes didn’t reach statistical significance, probably due to small sample sizes. As shown in Panel C, P/S ratios were generally very similar across hearing conditions in PL units. Due to the increased sustained in choppers in the exposed population, the chopper neurons show significantly lower P/S ratios in impaired ears (p = 0.0042). Panels C2 shows peak-versus-sustained rate plots for the Ch units, which clearly indicated the trend for discharge 111 A B Average CV 1 0.8 12 3 0.6 S.D. of first-spike latency (ms) 2 p(HS) = 0.377 p(UT) = 1.998e-008 5 7 12 9 7 14 0.4 3 6 5 7 0.2 0 Pri PriN Ch On Unsl.Locker 10 p(HS) = 0.545 p(UT) = 3.793e-003 6 12 8 1 10 8 12 3 7 9 14 0 5 7 3 10 -1 10 Pri PriN Ch On Unsl.Locker Fig. 3.11. Quantitative analyses on BF-tone PSTHs at 30 dB re threshold. II. Firing regularity and latencies. Comparison of average CVs of ISIs and SDs of FSLs for BF tones 30 dB above threshold in different response types and the two hearing status. The layouts are the same as in Fig. 3.10. A. Average CV of ISI during 30 to 40 ms after stimulus onset. B. Standard deviations of the first-spike latencies. Individual data points are shown. The bars show mean ± 1 S.E. 112 to shift in favor of sustained responses in the impaired choppers. For the chopper category, this was seen for all chopper subtypes (ChS, ChT and ChL). As expected, the ANOVA indicated highly significant effects by response types in all the three rate measures discussed above. But it indicated no significant effects by hearing status. p(HS) was the closest to being significant for sustained firing, mainly due to the strengthening of sustained firing in the Ch and Unusual categories. Fig. 3.11 shows the comparison of average CVs of ISIs (between 30 and 40 ms after stimulus onset) and standard deviations of FSLs for different unit categories and the two hearing states. The normal onset units exhibited the greatest discharge irregularity, followed by Pri and PriN neurons. It was not surprising that chopper neurons showed much lower CVs of ISIs than the three PSTH types. Interestingly, the unusual neurons also exhibited low CVs of ISIs regardless of hearing states. In the locker category, the low ISI CVs reflected phase-locking and entrainment to the tones. Significant differences in ISI CVs between the two hearing states were not seen for any unit types. However, one can see a slight trend of choppers in exposed ears to show greater discharge regularity than those in the unexposed ears. As can be seen in Fig. 3.10.C2, this was unlikely to be an effect of different compositions of chopper subtypes. An even greater degree of decrease of CV (increase of regularity) after HL can be observed in the onset category, although statistical significantly wasn’t reached due to small sample size. Generally, as indicated by the p(HS) value of the two-way ANOVA, the acoustic trauma didn’t significantly alter the discharge regularity in various types of VCN neurons. As shown by Panel B in Fig. 3.11, timing precision of first driven spikes varied across unit types. The Pri units showed the greatest FSL variability. The PriN and chopper categories had 113 comparatively low SDs of FSLs (notice the logarithmic ordinate in the panel). The situation within the normal onset category was more complicated. Of the eight onset neurons in normal-hearing population shown here, six showed SDs of FSLs in the lower range of the other categories. However, the other two showed large values of FSL SDs. These two neurons had unreliable first spikes, not showing up for every stimulus presentation, which caused their large FSL SDs. In both hearing states, the unusual type was the category showing greatest variation in the FSLs. Except for the onset category in which impaired data were unavailable, all PSTH categories showed overlapping distributions of SDs of FSLs in normal and exposed conditions, which indicated that acoustic trauma didn’t affect the first-spike timing precision in any VCN unit types. This was confirmed by the p(HS) value of the two-way ANOVA. 3.3. Latency and phase-locking of VCN neurons 3.3.1 First-spike latencies of VCN neurons The minimum FSLs for BF-tones at approximately 30 dB re thresholds in different VCN response types and hearing states are plotted versus BFs in Fig. 3.12. The lower bound of chopper latencies from Blackburn and Sachs (1989) are shown by the dashed black curves for comparison. The solid black curves were the chopper latency bounds shifted downward by 1 ms, which were approximately the lower limit of PL and onset latencies reported in Blackburn and Sachs (1989). On average, minimum latencies were negatively correlated with BF, and were systematically longer in Chopper and Unusual units than in PL and Onset ones. These 114 A. PL B. Chopper 10 Minimum first-spike latency (ms) Minimum first-spike latency (ms) 10 8 6 4 2 0 1 8 6 4 2 0 10 1 Unit BF (kHz) C. Onset D. Locker 10 Minimum first-spike latency (ms) Minimum first-spike latency (ms) 10 8 6 4 2 0 10 Unit BF (kHz) 1 8 6 4 2 0 10 Unit BF (kHz) 1 10 Unit BF (kHz) E. Unusual Minimum first-spike latency (ms) 10 8 6 4 2 0 1 10 Unit BF (kHz) Fig. 3.12. Minimum first-spike latency versus BF plots for the VCN units. Minimum FSLs were extracted from BF-tone PSTHs recorded at between 25 and 35 dB re threshold. The dashed black curve: lower bound of chopper latencies from Black burn and Sachs (1989). Solid curve: the dashed curve shifted downward by 1 ms. Hearing status is indicated by colors: Blue: normal-hearing; Red: noise-exposed (uncontaminated CAP recording); Magenta: noise-exposed (contaminated CAP recording, from animals #06-067 and #06-081. As shown by the legend, different PSTH types are indicated by different symbols. 115 observations were consistent with previous findings. In the PL group, response latencies tend to be more dispersed in a BF range between 10 and 20 kHz in the impaired ear in normal ears. The few Pri/PriN neurons with unusually long latencies were all neurons with considerable threshold shifts. In contrast, no impaired chopper neurons showed apparently abnormal response latencies. There was evidence that the latency bounds from Blackburn and Sachs (1989) didn’t fit the current data at low BFs. If it was assumed that the latency-BF relationships observed in Blackburn and Sachs (1989) were applicable to the strain of cats used in this study, all the low-BF phase-locking neurons shown in Panel D should fall into the chopper category. However, as discussed before, the majority (12/15) of these neurons showed no clear chopping patterns when PST histograms were recorded with asynchronous tone stimuli. Possible causes of this discrepancy included differences in animal strains and/or anesthesia states. This discrepancy led us to treat the low-BF (< 1 kHz) as a separate category. Another reason for this categorization was that threshold shifts were on average smaller in low-BF units (See Fig. 3.3, Panel B), compared with neurons in higher BF ranges in the impaired ears. All the unusual neurons showed significantly longer latencies, when compared with BF-matched neurons with typical response types. In general, no systematic effects by acoustic trauma can be seen in any of the PSTH types. 3.3.2. Phase-locking properties of low-BF units In response to pure tones with frequencies below certain limits, AN fibers and AVCN neurons in normal-hearing ears produce spikes at preferred phases of the tone. This phenomenon is called phase-locking (Johnson, 1988; Blackburn and Sachs, 1989). The strength of phase-locking can be quantified by a measure called vector strength, which can be calculated 116 according to the following equation (Johnson, 1988), VS = 1 N N N i =1 i =1 ∑ sin 2 (2π fti ) + ∑ cos2 (2π fti ) , in which N is the total number of spikes and ti is the time of the i-th spike in the spike train with respect to the onset of the tonal stimulus. The vector strength (VS) is such a measure that perfect phase locking, i.e., all spikes produced at a fixed phase, corresponds to a value of 1, and no phase locking, i.e., spikes produced equally likely at all phases between 0 and 2 pi radians corresponds to a value of 0. Blackburn and Sachs (1989) showed that different types of VCN units show distinct phase-locking properties. In response to BF tones, Pri and PriN units with BFs below approximately 4.5 kHz show VS significantly different from zero; while the BF range for phase-locking is much narrower (up to about 2 kHz) in the chopper category. Low-BF onset neurons show similar VS-BF relations as the PL units do. They showed that the VS-BF relations can be well fit by quadratic functions. These best fitting function for PL/Onset and chopper categories from Blackburn and Sachs (1989) are shown by respectively by the solid and dashed curves in Fig. 3.13.A and B for comparison. Fig. 3.13.A and B show the vector strengths in response to BF tones at 30 dB re threshold for different types of VCN units in unexposed and exposed animals. The vector strengths were calculated from rate-level recordings. Each VS was calculated by pooling data from 5 consecutive levels surrounding 30 dB re threshold. We required the mean discharge rate of the 5 levels to be greater than 50 spikes/s. This was for meeting the Rayleigh criterion to rule out the possibility that the periodicity arose out of chance. The reasons why we didn’t use the PSTH recordings were, 1) most of the PSTHs were recorded for random initial phases, however, the initial phase information was not stored due to technical problems, which made the 117 B 1 1 0.8 0.8 Vector strength Vector strength A 0.6 0.4 0.2 0 0.1 0.6 0.4 0.2 PL fit (B&S 1989) Ch fit (B&S 1989) 1 Unit BF (kHz) 10 0 0.1 PL fit (B&S 1989) Ch fit (B&S 1989) 1 Unit BF (kHz) C 2 Relative vector strength 1 0 -1 -2 -3 -4 -5 -6 -7 1 10 Relative minimum FSL (ms) Fig. 3.13. Vector strengths of low-BF neurons in response to BF tones at 30 dB re threshold. A. The relationship between vector strength and unit BF of VCN units in unexposed normal-hearing ears. The fit lines for the Pri/PriN and chopper populations from Blackburn and Sachs (B&S) (1989) are overlaid for comparison. B. The same as A, but for VCN units in exposed ears. C. The relationship between relative minimum first-spike latency (FSL) and relative vector strength in response to a BF-tone at 30 dB re threshold. The relative vector strength was defined such that a VS falling onto the solid (PL) curve in Panel A corresponds to a value of 1 and that falling onto the dashed (chopper) curve in Panel A corresponds to a value of zero. The mapping is linear. The relative minimum FSL was defined as the difference between the minimum FSL and the lower limit of FSL for PL in B&S (1989) (the solid curve in Fig. 1.11). 118 10 determination of VS not straightforward; 2) in a few units, PST histograms at 30 dB re threshold were unavailable. The comparison with fitting results from Blackburn and Sachs (1989) shown in Fig. 3.13.A validates this operation. It can be seen that the VS data of the PL, onset and chopper units agreed well with the previous results. For the very-low-BF units, which fell into the Locker category in the current study, some showed vector strengths consistent with the PL category; others showed vector strengths more suggestive of chopper type. However, similar agreement with previous results was unseen for units from the exposed animals (Fig. 3.13.B). The impaired Locker neurons appeared to have lower-than-normal vector strengths, and deviated from the fit curves from Blackburn and Sachs. Unfortunately, there was a lack of data inside the BF interval between 0.8 and 2 kHz. The reason for this was our inability to get recordings at 30 dB re threshold in the very-high-threshold impaired units with BF falling into this range. In order to study the relationships between unit type and VS in normal and impaired hearing states, we plotted the relationship between the relative FSL and the relative VS in Fig. 3.13.C. The relative FSL was defined as the minimum FSL of the unit in response to a BF tone at 30 dB re threshold subtracted by the lower limit of the minimum FSLs of PL neurons from Blackburn and Sachs (1989) (The solid curves in Fig. 3.12). The relative VS VS rel was defined by the following equation, VS rel = VS − VSCh ( BF ) , VS PL ( BF ) − VSCh ( BF ) in which VS PL ( BF ) and VSCh ( BF ) are respectively the latencies defined by the solid (PL) and dashed (Ch) curves in Fig. 3.13.A. The relative VS equals 1 for a unit with a VS falling onto the PL fit curve and equals 0 for a unit with a VS falling onto the Ch fit curve. Because Chopper 119 units show systematically longer latencies than PL units (Fig. 3.12), we expected to see a negative correlation between relative FSL and relative VS, which was indeed the case as can be seen in Fig. 3.13.C. However, at corresponding relative FSLs, about half of the impaired units show relative vector strengths substantially lower than normal values. Therefore, although the amount of data is small, these observations suggest the existence of a deterioration of phase-locking abilities of VCN neurons following acoustic trauma. This observation was surprising at the first glance because previous studies showed that phase-locking abilities of AN fibers in acoustically traumatized ears were not significantly affected (Miller et al., 1997, Heinz et al., 2004; Furman et al., 2006). In fact, the relative invulnerability of phase locking to acoustic trauma is a basic assumption of the model of loudness recruitment based on phase relationships between AN fibers by Carney (1994). We suggest that possible mechanisms of the compromised phase locking are disrupted synaptic transmission between AN afferents and VCN neurons, which is expected based on previously reported morphological and synaptic degeneration in VCN neurons in SNHL (e.g., Lee et al., 2003; Redd et al., 2002). At the meantime, it seems reasonable to predict that phase-locking to complex stimuli, such as vowels, would show similar degeneration, thus affecting the representation of these stimuli in the central auditory system. This is potentially an important contributing factor to compromised speech perceptions in deafened ears if a synchrony code in VCN unit is important for speech processing at higher stages of the pathway (Blackburn and Sachs, 1990). 3.4 Level dependences of BF-tone PST histograms Previous literature showed that a VCN unit could show different PSTH types at different 120 levels (Rhode and Smith, 1986). Similar level dependences of PSTH contour patterns were observed in the current study. For example, some PriN units showed notches after onset peaks only at 30 dB re threshold, but not at 20 dB; a few onset units showed more sustained firing patterns at lower than 30 dB re threshold. Other than these observations, all the units with usual PSTH types displayed consistent PSTH categorization at 20 and 30 dB re threshold. In a number of recorded VCN neurons, BF-tone PST histograms were obtained at about 20 and 30 dB re threshold, making possible the study of level dependences of the quantitative aspects of PST histograms. A simple hypothesis is that correlates of loudness recruitment in VCN, if they exist, should be manifested in the steeper rates of change with level in certain measures of the PSTHs. Possible measures include peak and sustained firing rates, first-spike latency, discharge regularities and precision of the first-spike timing. In Fig. 3.14, the relative changes in qualitative measures of the BF-tone PST histograms with levels, such as, x(l1 ) − x(l2 ) x( l ) l1 − l2 , (3.1) were calculated for different VCN unit types and hearing states. In the above equation, x can any of the quantitative measures of PSTH including peak rates, sustained rates, P/S ratios, mean ISI CVs, minimum FSL and SD of FSLs, which are shown by panels A – F in Fig. 3.14. l1 and l2 were the levels at which the PST histograms were taken. Data shown in Fig. 3.14 were based on two or more levels respectively at approximately 20 and 30 dB. l was the mean of l1 and l2 . As in Figs. 3.10 and 3.11, only units with sub-edge BFs were included in the analyses here. Unsurprisingly, with a few exceptions from onset units, peak and sustained rates were positively correlated with tone level (Fig. 3.14.A and B). In PriN categories, increases of peak and sustained rates with level appeared to be slightly more rapid in the impaired ear. To see the 121 0.1 0.05 B. Sustained firing rate Rel. sust. rate change / level (dB-1) Rel. peak rate change / level (dB-1) A. Peak firing rate p(HS) = 0.617 p(UT) = 3.995e-003 3 4 7 10 9 4 7 3 12 0 4 Pri PriN Ch On Unsl.Locker 0.15 0.05 Rel. mean CV change / level (dB-1) Rel. PS Ratio change / level (dB-1) 4 0.15 0.1 0 7 9 3 4 10 3 7 12 -0.05 -0.1 4 Pri PriN Ch On Unsl.Locker 7 -0.03 Rel. FSL SD change / level (dB-1) Rel. mininum FSL change / level (dB-1) 3 9 7 10 7 12 4 -0.05 3 Pri PriN Ch On Unsl.Locker 4 0.02 9 0 7 12 -0.02 3 7 10 4 4 -0.04 p(HS) = 0.198 p(UT) = 3.067e-001 -0.06 Pri PriN Ch On Unsl.Locker F. SD of FSL 0 -0.02 10 7 0 E. Minimum FSL -0.01 4 9 D. CV of ISI p(HS) = 0.729 p(UT) = 2.954e-003 0.05 3 0.1 C. P/S ratio 0.2 p(HS) = 0.948 p(UT) = 1.602e-003 4 12 4 4 -0.04 p(HS) = 0.155 p(UT) = 2.471e-001 3 -0.05 Pri PriN Ch On Unsl.Locker 0.1 0 -0.1 7 12 7 9 4 4 3 4 10 -0.2 -0.3 -0.4 -0.5 p(HS) = 0.164 p(UT) = 3.628e-001 3 Pri PriN Ch On Unsl.Locker Fig. 3.14. Relationships between quantitative measures of BF-tone PSTHs and level. Slopes of relative changes of several PSTH quantitative measures (A: peak rate; B: sustained rate; C: P/S ratio; D: mean ISI CV; E: minimum FSL; F: SD of FSL) with tone level were calculated from BF-tone PSTHs pictures taken at approximately 20 and 30 dB above thresholds. As in Figs. 3.10 and 3.11, only units with BFs below the CAP edge frequencies were included in this analysis. Error bars show ±1 S.E. A two-way ANOVA (factors: hearing status and unit types) was performed on the slope data. The p-values for the hearing status and unit type factors are shown as p(HS) and p(UT) in the plot. 122 cause of this effect, compared the threshold-aligned RLFs shown in Fig. 3.23, Row 2 Column 5, which shows that normal PriN units reach saturation at 20 dB re threshold, which didn’t happen in the exposed units. The fact that relatively change in peak and sustained rates with tone level were less rapid in impaired Choppers than in normal ones may was superficially inconsistent with the tendency of Chopper units to show greater sustained rates in impaired ears at 30 dB re threshold (See Fig. 3.10.B). This superficial discrepancy was caused by inability of the level choice (20 and 30 dB re threshold) to capture the range of significant rate changes, because most chopper neurons often dynamic ranges less than 20 dB (See Fig. 3.23, Row 3 Column 5), beyond which rate saturations occurred. There were no systematic effects by hearing condition on the rate of change of P/S ratios with stimulus level. On average, CVs of ISIs were decreasing functions of level in the PriN and chopper units (Fig. 3.14, Panel D). In other words, regularity of discharge generally tended to increase with sound level in VCN neurons. However, it should be noted that in the Pri category and normal population, there were units with ISI CVs changing in both directions with increasing level. The trend for ISI CVs to decrease with level was more consistent and pronounced in impaired PriN and chopper neurons. Similar phenomenon can be seen in the unusual-type units. Similar trends were observable in the onset and unusual categories, although the scarcity of data prevented any conclusions. These observations suggested that in many types of VCN neurons, discharge regularities became more level-dependent following acoustic trauma, which was probably caused by a greater degree of convergence of ANF input on individual VCN neurons (Banks and Sachs, 1991). However, it is unclear whether this can be a correlate of recruitment. As shown in Fig. 3.14, Panels E and F, minimum FSLs of all VCN neurons were decreasing 123 functions of stimulus level, and first-spike timings were more precise higher levels (Fig. 3.14.F). PL and chopper unit in the exposed ears appeared to have similar rates of FSL decreasing with increasing level between 20 and 30 dB re threshold, in comparison with their normal counterparts. Interestingly, the exposed PriN and Onset neurons showed faster gain in FSL precision with increasing level. This suggested that the latency in PriN neurons may show correlates of recruitment in impaired ears. This effect is potentially a potential correlate of loudness recruitment, because minimum FSL has been shown to be monotonically decreasing functions of tone level in VCN neurons (confirmed by our data) and may be a code for intensity (Rhode and Smith, 1986). However, the ANOVA performed on the FSL data (Panel E) indicated no significant effects by acoustic trauma on the dependence of minimum FSL on tone level. 3.5. Tuning curves, frequency-level response maps and spontaneous firing rates 3.5.1. Tuning curves and Q10 measures Boundaries of the excitatory response areas, i.e., the tuning curves, were constructed for most of the recorded VCN neurons. As detailed in Section 2.6.3, two different methods were used in constructing tuning curves due to technical reasons. These two methods yielded tuning curves with slightly different thresholds (< 10 dB difference), but consistent Q10 measures. Exemplar tuning curves of different types of VCN neurons from the normal and impaired ears are shown in Fig. 3.15 and 3.16, respectively. In the normal ears, the tuning curves of the majority of PL, Chopper and Onset neurons closely resembled ANF tuning curves, with sharply tuned tips at BFs and occasionally seen low-frequency tails at high levels (Kiang et al., 1965). It needs to be mentioned that the tuning curves shown in the two figures only captured excitatory 124 B. Chopper 100 100 80 80 60 40 20 0 06-048 U 7.10 PriN 06-048 07-011 U 7.7 06-048 07-011 U 1.5 U 1.1 PriN U 5.1 PriN Pri Pri 1 Threshold (dB SPL) Threshold (dB SPL) A. PL 60 40 20 0 10 06-048 06-048 U 6.5 U 1.2 ChL ChT 1 C. Onset D. Unusual 100 80 80 Threshold (dB SPL) 100 60 40 20 06-048 U 5.2 OnL 60 06-048 U 7.3 Unusual:A 40 20 06-048 U 6.3 OnL 1 06-048 U 6.4 Unusual:A 0 10 06-054 U 2.2 Unusual:B06-048 U 6.8 Unusual:B 1 Frequency (kHz) 10 Frequency (kHz) E. Locker 100 80 Threshold (dB SPL) Threshold (dB SPL) 10 Frequency (kHz) Frequency (kHz) 0 07-011 U 1.4 ChT 06-079 U 1.4 ChS 60 40 06-048 U 1.1 Locker 20 06-048 06-079 U 6.9 U 2.7 Locker Locker 0 0.1 1 Frequency (kHz) Fig. 3.15. Tuning curves of VCN neurons in the normal-hearing ears. The tuning curves shown in this figure capture only the boundaries of excitatory regions. 125 A. PL B. Chopper 80 60 40 06-067 Unit 2.1 Pri 06-067 Unit 1.7 06-047 Pri Unit 1.4 PriN 20 06-067 Unit 1.11 Pri 100 80 Threshold (dB SPL) Threshold (dB SPL) 06-081 Unit 1.6 PriN 06-081 Unit 1.8 PriN 100 06-047 Unit 1.3 ChS 60 40 06-081 Unit 1.7 ChL 06-067 Unit 1.9 ChT 20 0 0 1 1 10 D. Onset C. Locker 100 80 80 07-034 Unit 2.14 OnL 40 Threshold (dB SPL) 100 60 10 Frequency (kHz) Frequency (kHz) 07-037 Unit 2.1 OnC 20 0 Unit 9.5 06-018 Locker 60 06-081 Unit 3.3 Locker 40 06-018 Unit 11.1 Locker 20 0 1 10 0.1 Frequency (kHz) 1 Frequency (kHz) E. Unusual 100 Threshold (dB SPL) Threshold (dB SPL) 06-067 Unit 1.8 ChT 80 06-047 Unit 2.2 Unusual:A 60 40 06-081 Unit 4.1 Unusual:B 06-067 Unit 1.5 Unusual:A 06-067 Unit 1.2 Unusual:A 20 0 1 10 Frequency (kHz) Fig. 3.16. Tuning curves of VCN neurons in the noise-exposed ears. The layout is the same as in Fig. 3.15. Note that the tuning curves shown in this figure capture only the boundaries of excitatory regions. Acoustic trauma had two major effects on the tuning curves of VCN neurons: (1) threshold elevations, (2) loss of sharply tuned tips. No onset neuron is shown in this figure because of the lack of sampled sub-edge onset units. 126 responses, thus they may not reflect the existence of inhibitory side-bands in many types of VCN neurons (e.g., PriN and chopper units), as will be discussed in the next section. When plotted on a logarithmic frequency scale, neurons with higher BFs tended to show sharper tuning curves. In a typical normal tuning curve, the edge above BF was steeper than the edge below BF. A few Unusual-type units in the normal ear showed unusually high thresholds and broad tuning (Panel D, Fig. 3.15). Due to their low BFs, the Lockers displayed the broadest tuning of all the response types. Unlike high-BF neurons, their tuning curves were often symmetric with respect to the BF, or even showed sub-BF edges steeper than supra-BF ones. Threshold shifts were manifested in the tuning curves of neurons in the impaired ears. This was seen across all response types. Compared with other categories, the Locker group showed smaller threshold increases. Consistent with the CAP audiograms, threshold shifts were more severe at middle frequency ranges than at higher frequencies (> 10 kHz). Losses of sharply tuned tips were observed in many, but not all impaired neurons. This suggests that cochlear damages in these animals were of mixed IHC and OHC abnormality (Liberman and Dodds, 1984b). The bowl-shape tuning curves are the extreme cases of loss of tips (e.g., 06-067, Unit 2.01 Pri in Panel A, Fig. 3.16). The same type of abnormal tuning curves have been observed in previous ANF studies, which were suggested to correspond to total loss of OHC functions (Liberman and Dodds, 1984b). A few threshold-shifted Unusual neurons exhibited extremely wide tuning (e.g., 06-067, Unit 1.5 in Panel D, Fig. 3.16). Lockers lacked sharp tips in normal ears to begin with, which caused broadening of tuning hard to see in many exposed Locker units (06-081, Unit 3.3 and 06-018, Unit 11.1 in Panel E, Fig 3.15) In Fig. 3.17, Panel A, Q10 values of all recorded VCN neurons were plotted against their 127 BFs. For comparison, the upper and lower limits (5% quantiles) of Q10s of normal cat AN fibers (Miller et al., 1997) were shown by the dashed curves. In the BF region below 10 kHz, except for a few unusual-type neurons, most normal VCN units had Q10 values falling into the ANF range. The moving-window average of normal Q10s over BF is shown by the blue curve in Panel A. In the impaired population, almost all neurons with BF between 0.5 and 8 kHz showed Q10s below the normal range. A few lockers in the impaired ears had Q10s falling into the normal range. These Locker units had small but non-negligible threshold shifts (See Fig. 3.3). The fact that many high-BF (> 10 kHz) units in the exposed ears had lower-than-normal Q10s was not surprising, since single-unit and CAP thresholds suggested that some exposed animals had hearing losses at high frequencies as well (e.g., see animals 06-006 and 06-009 and 06-018 in Fig. 3.2). Panel B in Fig. 3.17 shows the relationship between thresholds and Q10 values of the VCN neurons. This panel was produced in a similar way as Fig. 5 in Heinz and Young (2004), which is reproduced in Panel C here for comparison. Here, threshold shifts were defined as the difference between single-unit threshold and the threshold on the average normal CAP audiogram (Fig. 3.4, dashed blue curve) at the corresponding BF. The Q10s were normalized by the average BF-matched Q10 in the unexposed population (the blue curve in Panel A). The horizontal dash-dotted line was an artificially drawn border between sharp and broad tuning (0.55, the same as the one in Fig. 5 in Heinz and Young, 2004). As shown in Panel B, in the normal-hearing population, all VCN neurons had sharp frequency tuning according to this criterion, with exceptions of a chopper and a few unusual-type units. The population of exposed units with substantial threshold shifts exhibited a mixture of sharp and broadened tuning. For greater 128 A ANF limits from Miller et al. (1997) Normal average Q10 10 1 0.1 1 Unit original BF (kHz) PL Ch On Locker Unusual Unknown 10 B C. Q10 re normal (Fig. 5, Heinz and Young, 2004) 1 0.1 -40 -20 0 20 40 Threshold shift (dB) 60 80 Fig. 3.17. Q10 values of VCN neurons in normal and noise-exposed ears. A. A summary of tuning curve Q10s of units in the normal and noise-exposed populations. The gray bar indicates the frequency region of the overexposure stimulus. The upper and lower limits of Q10s of normal AN fibers from Miller et al. (1997) are overlaid for comparison. The blue curve shows the moving average of Q10s versus BF, which was used in the normalization in Panel B. B. Relative Q10s (normalized by average Q10s at corresponding BF in the normal-hearing population) versus threshold shifts (threshold relative to the average normal CAP audiogram, see Fig. 3.4). The horizontal dashed line indicates the normal average. The horizontal dash-dotted line is an artificially drawn boundary drawn between sharp and broad tuning. C. Fig. 5 (Right) from Heinz and Young (2004) reproduced for comparison. Notice that threshold elevation was defined in a different way. 129 threshold shifts, more severe broadening of tuning can be seen (the lower right quadrant). Overall, for exposed units with threshold shifts greater than 0 dB, there were about equal numbers of sharply and broadly tuned units. By comparing Panel B to Panel C, it can be seen that the maximum degree of threshold shifts and distribution of relative tuning sharpness of AN fibers in the moderate/severe HL population in Heinz and Young (2004) were comparable to those of the exposed VCN neurons in the current study. If it is assumed that the relationships between threshold shifts and tuning losses are similar in the AN fibers and VCN neurons, one may reach the conclusion that the population of impaired neurons in the current study are similar to the moderate/severe HL pool in Heinz & Young in terms of the degree of broadening of tuning and elevation in thresholds, which is evidence for that the degree of IHC and OHC damages were comparable in the animals deafened in the current study and those moderate/severe HL animals in the previous one. This will be the basis for comparisons between our VCN data and ANF data from the previous study in the following sections. 3.5.2. Level-frequency response maps For many of the recorded neurons, we were able to construct the level-frequency response maps by integrating data from multiple tonal RLF and RM picture. Figs. 3.18 – 3.21 show driven firing rates (total – spontaneous rates, color code) as a two-variable function of tone frequency (x-axis) and level (y-axis) in different VCN unit types and the two hearing states. Driven rates below zero indicate inhibitory responses. Most of the response maps of normal Pri units and some PriN units in normal-hearing ears showed excitatory-only (type I or type I/III) responses (Fig. 3.18.A & B). In these cases, the tuning curves agreed well with the boundaries of the response areas. At a given level, maximum 130 B. Normal PriN A. Normal Pri 06-048, T7U10, PriN, 1.46 kHz @ 9.4 dB SPL, SR = 23.5 94 400 Sound intensity (dB SPL) Sound intensity (dB SPL) 06-048, T5U1, Pri, 3.08 kHz @ 9.1 dB SPL, SR = 2.7 89 200 150 65 100 50 41 0 17 -50 300 66 200 100 38 0 10 -100 -100 0.770246 -200 1.76956 4.06538 9.3398 Tone frequency (kHz) 0.364127 0.679485 1.26796 2.3661 Tone frequency (kHz) C. Normal PriN D. Normal PriN 69 06-048, T7U8, PriN, 11.70 kHz @ 8.8 dB SPL, SR = 103.1 93 300 Sound intensity (dB SPL) Sound intensity (dB SPL) 07-011, T1U5, PriN, 4.33 kHz @ 17.0 dB SPL, SR = 41.5 93 150 100 50 45 0 21 -50 1 200 65 100 37 0 9 1.86607 3.4822 6.49802 Tone frequency (kHz) -100 1 E. Exposed Pri F. Exposed PriN 6-047, T1U4, PriN, 6.21 kHz @ 40.1 dB SPL, SR = 25.3 105 200 Sound intensity (dB SPL) 06-047, T2U4, Pri, 1.47 kHz @ 85.2 dB SPL, SR = 7.1 106 100 Sound intensity (dB SPL) 2.2974 5.27803 12.1257 Tone frequency (kHz) 86 50 66 0 46 -50 150 85 100 65 50 0 45 -50 -100 25 0.484919 0.735 1.11405 1.68859 Tone frequency (kHz) 1 2.2974 5.27803 12.1257 Tone frequency (kHz) Fig. 3.18. Example response maps of Pri and PriN units in normal and noise-exposed ears. Level-frequency response maps of driven firing rates of PL neurons in unexposed (A-B) and exposed (E-F) ears. Driven firing rate = total firing rate – spontaneous rate. Warm colors indicate excitation and cool colors inhibition. The black curves show the boarders of the responses (excitatory only) areas, which are the tuning curves shown in Figs. 3.15 and 3.16. The white vertical lines indicate BFs and thresholds. The BFs, thresholds and spontaneous rates (unit: spikes/s) were shown above the response maps. 131 rates occurred at BF (Fig. 3.18.A), or sometimes below the BF (Fig. 3.18.B). The responses to tones above the BFs were often relatively weaker. There was some evidence that maximum driven rates tended to be higher in PriN units compared with in Pri units, probably reflecting the greater convergence of AN fibers in GBCs than in SBCs (Ostapoff and Morest, 1991; Ryugo & Sento, 1991). Inhibitory responses (negative driven rates, type III RMs) were seen in a subpopulation of spontaneously active PriN units, in both normal and exposed ears (Panels C, D, and E). Inhibition in PriN units can be manifested in two ways: (1) inhibitory sidebands, in which the neurons response by firing rates under the spontaneous rates (Panel D); (2) non-monotonic rate-level functions at and below BF (Panel C). In the latter case, the unit fired relatively rigorously to BFs slightly above the threshold, but responses sagged and/or even went below SR at higher levels, but could show a rebound at very high levels (> 90 dB SPL). Panel D shows that in units with strong inhibitory sidebands, the tuning curves failed to capture the entire response areas. These different inhibition patterns were probably correlated with different arrangement of excitatory and inhibitory BFs in GBCs. Pri and PriN neurons in impaired ears exhibited threshold shift and broadened tuning (Fig. 3.18.E & F). Like Pri units in impaired ears, most expose Pri units showed relatively little inhibitory responses. The Pri unit shown in Panel E had a very large threshold shift (about 80 dB). No sharply tuned tips could be seen in the response area and the tuning curve shows a bowl shape. The driven rates at all tone frequencies were relatively low in this response map, which is suggestive of severe OHC and IHC damage. This type of response map is representative a substantial subpopulation of Pri units recorded in exposed animals. Threshold shift was less severe in the PriN unit shown in F. And a tuned tip can still be seen, though it’s less sharp 132 compared with the tips of normal PriN neurons. Inhibition response can be seen at high levels above the BF in this unit. As shown in Fig. 3.19, inhibitory regions were commonly seen in both normal chopper units and some deaf Chopper units (in both Ch-S and Ch-T subtypes). A large subset of the chopper group possessed SRs below 1 spikes/s. Inhibitory responses were not readily observable in these neurons (Panel B). However, in normal chopper neurons with SR above 15 spikes/s, sideband inhibitions could be clearly seen (Panels A and C). These inhibitory responses were often seen both below and above BFs in normal chopper units. However, inhibitory sidebands below BF were rarely, if ever, observed in chopper units with substantial threshold shifts. Unlike in PL units, the maximum rates in chopper response maps were almost always seen at or near BF. As shown by the example in Panel C, there was no substantial evidence that neural inhibition became diminished in the VCN chopper units following acoustic trauma. Another interesting observation in the chopper units from impaired ears is that a considerable subset (but not all) of the choppers with substantial threshold shifts (> 40 dB) showed unusually high driven firing rates and steep rate increases with tone levels. Driven rates and rate-level slopes as high were rarely seen in unexposed chopper neurons (e.g., A and B in Fig. 3.19). Panel D-F in Fig. 3.19 shows three representative cases, in which maximum driven rate exceeded 400 spikes/s and approached 600 spikes/s, as compared to the typical maximum driven rates of 150 – 300 in normal choppers. Also, interestingly, theses extremely high rates were often confined to the frequencies at and near BF. In the examples shown in D and E, there was a sharp drop in rate-level slope from the BF to frequencies slightly below and above. Inhibition in this type of rate-enhanced choppers couldn’t be readily seen, probably because it was lost or masked 133 A. Normal ChT B. Normal ChS 06-024, T6U1, ChS, 10.15 kHz @ 6.0 dB SPL, SR = 0.1 96 Sound intensity (dB SPL) Sound intensity (dB SPL) 6-079, T2U1, ChT, 7.97 kHz @ 2.7 dB SPL, SR = 33.4 92 150 64 100 50 36 0 8 -50 200 150 68 100 50 40 0 12 -50 -100 1 1.86607 3.4822 6.49802 12.1257 Tone frequency (kHz) 1 C. Exposed ChS D. Exposed ChT 200 100 46 0 18 -100 1 06-067, T1U10, ChT, 5.41 kHz @ 45.2 dB SPL, SR = 38.1 600 100 Sound intensity (dB SPL) Sound intensity (dB SPL) 06-047, T1U3, ChS, 4.49 kHz @ 47.4 dB SPL, SR = 48.7 102 300 74 400 80 200 60 0 40 -200 1.86607 3.4822 6.49802 12.1257 Tone frequency (kHz) 1 E. Exposed ChT 200 100 0 48 -100 06-018, T2U3, ChS, 12.90 kHz @ 40.2 dB SPL, SR = 1.5 400 119 Sound intensity (dB SPL) Sound intensity (dB SPL) 300 68 1.86607 3.4822 6.49802 Tone frequency (kHz) F. Exposed ChS 06-067, T1U9, ChT, 5.67 kHz @ 40.5 dB SPL, SR = 21.4 108 400 88 1.86607 3.4822 6.49802 12.1257 Tone frequency (kHz) 300 91 200 100 63 0 35 -100 -200 28 1 1.86607 3.4822 6.49802 Tone frequency (kHz) 0.5 1.7411 6.06287 21.1121 Tone frequency (kHz) -200 Fig. 3.19. Example response maps of chopper units in normal and noise-exposed ears. The layout is the same as in Fig. 3.19. The white vertical lines indicate BFs and thresholds. Inhibition is more prevalent in Choppers compared with in PL neurons (Fig. 3.19). A subpopulation of exposed choppers (exemplified by D, E & F) showed abnormally large driven firing rates despite substantial threshold shifts. 134 by strong excitatory inputs. This type of enhanced rate-level response was seen in both sustained and transient chopper subtypes and in all chopper SR categories. In Fig. 3.20, Panels A and B show the response maps recorded from onset units. The unexposed onset units recorded in this study generally showed sharp tuning comparable to the PL and chopper units. The onset neurons showed strong trends to have close-to-zero spontaneous firing rates as the choppers did. However, inhibitory responses were observed in all of the few spontaneously actively (SR > 20 spikes/s) onset neurons, in exposed and unexposed ears (e.g., Fig. 3.20.A). The unit shown in Panel B was one of the three onset units with sub-edge BFs recorded from in the current study. It had a considerable threshold shift and an interestingly wide response area and broad tuning curve. At levels less than 20 dB re threshold, the firing rates were relatively low (< 50 spikes/s), reflecting the neuron’s onset firing pattern. However, the firing rates were substantially higher at higher levels and at lower frequencies. At around 1 kHz, the rates could reach over 200 spikes/s. The very high firing rates at low BFs (< 1 kHz) reflects the ability of the onset neurons to phase-lock. This type of discharge rate gradient from low to high tone frequencies was representative of response maps of many onset neurons recorded from, in both normal and exposed ears. The response maps of Locker units (Fig. 3.20.C & D) confirmed the symmetric and broad tuning curve shapes shown in the previous section. As shown, unlike neurons with higher BF, the maximum firing rates in Lockers sometimes were elicited by tones above BF. Low-BF neurons tended to show low spontaneous rates, for which reason inhibitory responses were not seen in most of the recorded Locker units. However, as shown by the example in Panel C, some Lockers 135 A. Normal onset B. Exposed onset 07-037, T2U1, OnC, [email protected] SPL, SR = 0.0 103 200 Sound intensity (dB SPL) Sound intensity (dB SPL) 06-048, T6U3, OnL, 6.93 kHz @ 4.7 dB SPL, SR = 30.4 90 80 60 66 40 42 20 0 18 -20 150 79 100 50 55 0 31 -50 -40 -6 1 -100 2.2974 5.27803 12.1257 27.8576 Tone frequency (kHz) 1 06-018, T6U1, Locker, 0.38 kHz @ 44.0 dB SPL, SR = 5.8 107 400 Sound intensity (dB SPL) D. Exposed locker 06-079, T2U3, Locker, 0.46 kHz @ 15.3 dB SPL, SR = 1.0 95 500 Sound intensity (dB SPL) C. Normal locker 2.2974 5.27803 12.1257 Tone frequency (kHz) 400 67 300 200 39 100 0 11 -100 -200 300 83 200 100 59 0 35 -100 -200 0.201402 0.375829 0.701322 1.30871 Tone frequency (kHz) 0.095 0.218253 0.501413 1.15194 Tone frequency (kHz) Fig. 3.20. Example response maps of onset and Locker units in normal and noise-exposed ears. Panels A and B show the response maps of two example onset units, one from a normal-hearing animal and the other from a noise-exposed one. Panel C and D show the response maps of Locker units, from a normal-hearing ear and an exposed one. 136 did show non-monotonic tonal rate level functions, which were probably attributable to C1/C2 transitions. The unusual category showed the strongest heterogeneity of response map patterns among all PSTH groups. Some very non-primary-like patterns, such as the multiple-tip response area shown in Fig. 3.21.A, were seen in both Unusual-A and Unusual-B types in normal animals. Compared with usual PSTH types, unusual-type neurons in the normal ear mostly showed low driven firing rates (maximum < 150 spikes/s, Panels A & B). Inhibitions were also prevalent in unusual units (e.g., Panels B, C and E). Many Unusual neurons showed response maps resembling those of the non-unusual neurons. These neurons showed PSTHs that matched the criteria of regular PSTH categories except for their inability to satisfy certain quantitative conditions. For example, the neurons shown in Panels B, C and D showed RMs indistinguishable from PriN or Chopper units. In fact, the unit shown in D showed a PST histogram consistent with the Chopper classification in terms of its high discharge regularity and latency, however, it was categorized as an Unusual because of its imprecisely timed first-spike timing and its failure to pass the SPP test. This suggested that these neurons probably had anatomical correlates of usual cell types, such as bushy and stellate cells. Thus it should be bear in mind that the Unusual group contained a portion of units which were marginally non-unusual. The response map of the impaired neuron shown in Panel E (07-003, T2U7) was unseen in any other CN neurons in either normal or noise-traumatized ears. A central excitatory area, with BF at approximately 3.98 kHz, was flanked by to wide inhibitory areas. The excitatory area overlapped with the most impaired BF region in that animal; the maximum driven rate was greater than 450 sp/s; and the rate-level slope for the 3.98-kHz tone was extremely steep (> 20 137 A. Normal Unusual-A B. Normal Unusual-B Sound intensity (dB SPL) Sound intensity (dB SPL) 06-048, T7U4, Unusual:A, [email protected] SPL, SR = 2.4 06-036, T2U1, Unusual:B, [email protected] SPL, SR = 10. 103 93 100 100 73 50 53 0 33 13 0.652976 1.8469 5.22381 14.7752 Tone frequency (kHz) 75 50 47 0 19 -50 1 C. Exposed Unusual-A 2.2974 5.27803 12.1257 Tone frequency (kHz) -50 D. Exposed Unusual-B 86 Sound intensity (dB SPL) Sound intensity (dB SPL) 06-047, T1U5, Unusual:A, [email protected] SPL, SR = 29. 06-018, T5U3, Unusual:B, [email protected] SPL, SR = 0. 80 110 119 150 60 40 20 62 0 -20 38 0.805596 1.85077 4.25196 9.76844 Tone frequency (kHz) -40 100 95 50 71 0 47 -50 1 2.2974 5.27803 12.1257 Tone frequency (kHz) E. Exposed Unusual-C Sound intensity (dB SPL) 003, T2U7, Unusual:B, [email protected] SPL, SR = 58.4 94 400 300 78 200 100 62 0 46 -100 -200 1 1.86607 3.4822 6.49802 12.1257 Tone frequency (kHz) Fig. 3.21. Example response maps of unusual-type units in normal and noise-exposed ears. Panels A and B show response maps of Unusual-A and Unusual-B type units from the normal-hearing ears. Panels C-E show response maps of unusual-type units in impaired ears. The unit shown in panel E had an exotic RM pattern and a high maximum driven rate. 138 spikes/s/dB). Inhibition was seen on both sides of the excitatory area and both their thresholds extended below the threshold for the 3.98-kHz tone. No similar response maps had been described in previous physiological studies in the VCN or DCN. We therefore set the best frequency at 3.98-kHz, which was at the high-frequency edge of the impaired central excitation area. 3.5.3. Spontaneous firing rates Fig. 3.22 shows the spontaneous firing rates of different VCN neuronal types in the two hearing states. Row A shows the data from Pri/PriN and low-BF phase lockers; Row B shows the data from Chopper and Onset neurons; the data from Unusual and Unknown types are shown in Row C. In both hearing states, SRs of the each unit category show trends to increase with increasing BFs, which is consistent with previous results (Bourk, 1976). The p-values shown in the figures are from comparisons of the SRs of sub-edge units in the two hearing states in each PSTH category. As those p-values for Wilcoxon rank-sum tests show, no significant differences in the SR distributions between the normal and exposed ears was seen for PL/Locker and Unusual/Unknown categories. However, the in the PL/Locker and unknown+unusual categories, the mean SRs were slightly greater in the impaired population. Moreover, the effect that SRs tended to increase in sub-edge BF regions following trauma was significant in Chopper/Onset and Unusual/Unknown categories. If all VCN unit types were considered together, statistical comparison indicated a significantly elevated spontaneous firing rate in the sub-edge BF regions in the impaired population (Normal mean: 14.4 spikes/s; exposed mean: 29.1 spikes/s; p = 0.003, two-tailed Wilcoxon rank-sum test). As can be seen in the SR-versus-BF plots, this difference was mainly due to a relatively small proportion of low-SR (< 10 spikes/s) units in the BF regions 139 150 100 50 0 0.1 1 Unit BF (kHz) Normal mean = 24.3 50 Sub-edge: p=0.436 Percentage of neurons A Spontaneous rate (spikes/s) 200 25 0 Exposed mean = 35.6 50 25 0 0 100 Spont. rate (sp/s) 200 10 Sub-edge: p=0.090 150 Percentage of neurons B Spontaneous rate (spikes/s) 200 100 50 0 0.1 1 Unit BF (kHz) Normal mean = 9.5 50 25 0 Exposed mean = 20.5 50 25 0 0 150 100 50 1 Unit BF (kHz) Normal mean = 11.6 50 Percentage of neurons Spontaneous rate (spikes/s) Sub-edge: p=0.071 0 0.1 200 10 200 C 100 Spont. rate (sp/s) 25 0 Exposed mean = 24.5 50 25 0 0 100 Spont. rate (sp/s) 200 10 Fig. 3.22. Spontaneous firing rates of VCN neurons in normal and exposed ears. Left column: SR-versus-BF plots. Right column: the distribution histograms of SRs of units with BFs below the CAP-audiogram edge frequencies. A. Pri/PriN and Locker neurons. B. Chopper and onset neurons; C. Unknown and unusual neurons. The p-values for comparison of the SRs of sub-edge units by Wilcoxon rank-sum tests in different PSTH categories are shown in the left column. The difference in SRs reached statistical significance in the Chopper/Onset category (B). 140 with substantial threshold shifts. It is not entirely clear what the cause of this absence of low-SR neurons. Although this post-trauma increase in SR may be a genuine phenomenon, the possible explanation that it was due to difficulties with sampling low-SR units with high thresholds cannot be entirely dismissed. The acoustic noise in the dynamic driver was unlikely to be the cause for the following reasons. (1) In those units recorded in the noise-contaminated experiments (purple), there were units with SRs less than 10 spikes/s and BFs falling into the frequency range of the noise. (2) As discussed before, the units from those contaminated experiments with BBN thresholds less than 15 dB above the noise spectrum level had been discarded. All the remaining neurons used in subsequent analysis had BBN thresholds at least 20 dB above the noise spectrum (See Fig. 2.9). (3) A few impaired neurons with very high SRs (> 150 spikes/s) had BFs much higher than the frequency range of the noise. (4) There were high-SR impaired units from impaired animals for which recordings were done with the electrostatic driver. The observation that SR of VCN neurons tended to increase after hearing loss appeared to agree with the previous finding by Sumner et al. (2005) that VCN units showed significant increased SRs 1 – 14 days after unilateral conductive hearing loss in guinea pigs. In the same study it was also observed that the abnormally high SR showed gradual decrease with post-HL time. Therefore the comparatively longer recovery period in the current study was a possible explanation for the lack of significance in several PSTH types in our SR data. 141 3.6. Analyses of tonal rate-level functions 3.6.1. BF-tone rate-level functions Most psychophysical studies on recruitment were based on pure tone stimuli (e.g., Moore et al., 1985; Zeng and Turner, 1991; Moore 2004). There has been some physiological and psychophysical evidence suggesting that for encoding the intensity of a pure tone, the most important neural substrates are the discharge rates of the auditory neurons with BFs at or near the frequency of the tone (Sachs and Abbas, 1974; Viemeister, 1988). Therefore, it is important to analyze the RLFs for BF tones of VCN neurons in normal and noise-exposed animals, and look for post-trauma alterations that are potentially neural correlates of recruitment. For each recorded VCN neuron, a BF-tone RLF was recorded from 10 -20 dB below the response threshold to the highest deliverable level of the speaker (90 – 105 dB SPL for the electrostatic speaker; 105 – 115 dB SPL for the dynamic speaker), or the highest level at which single-unit triggering wasn’t affected by ER. Fig. 3.23 summarizes the BF-tone RLFs, their shape categories, rate-level slopes and population averages. In the chopper and onset categories, different subtypes (e.g., ChS and ChT) were pooled, because inspection revealed no systematic differences between the RLFs of these subtypes. Rates shown in the figure were driven rates. Only units with BFs at or below the CAP-audiogram edge frequencies (Section 2.2.1) were shown. As can be seen in the figure, all sub-edge units in the impaired ear, except for a few Locker units, had substantial threshold shifts. The first column shows the population plots of the RLFs by PSTH types. The RLFs of the Pri neurons in the normal ears had shapes resembling those of normal ANFs, in that monotonic and rapid rate increases over the first 20 – 30 dBs above threshold were followed by abrupt 142 0 400 200 0 0 50 100 Tone level (dB SPL) 400 200 0 0 50 100 Tone level (dB SPL) 50 0 Nml. Exp. p = 0.330 50 0 Nml. Exp. p = 0.311 50 0 Nml. Exp. 25 Exp. p = 0.292 20 15 10 5 Nml. 8 Exp. p = 1.000 6 4 2 0 15 Nml. Exp. p = 0.799 10 5 Nml. 20 Exp. p = 0.939 10 0 143 Nml. Exp. 200 100 0 0 20 40 60 80 Level (dB SPL) 400 200 0 0 20 4060 80 Level (dB SPL) 150 100 50 0 -20 0 20406080 Level (dB SPL) 200 100 0 0 20 40 60 80 Level (dB SPL) 100 50 0 0 20 40 60 80 Level (dB SPL) Avg. rate (spikes/s) Avg. rate (spikes/s) 0 20 40 60 80 Level (dB SPL) Avg. rate (spikes/s) Nml. 0 Avg. rate (spikes/s) 5 Avg. rate (spikes/s) 100 -20 0 20406080 Tone level (dB SPL) p = 0.159 10 50 Avg. rate (spikes/s) 100 Nml. Exp. p = 0.113 100 Avg. rate (spikes/s) 200 0 CS-15 (spikes/s/dB) 100 0 50 100 Tone level (dB SPL) 50 Exp. 150 Avg. rate (spikes/s) 0 p = 0.007 Nml. Avg. rate (spikes/s) 200 Nml. Exp. 0 Avg. rate (spikes/s) 400 0 15 5 Avg. rate (spikes/s) 600 50 CS-15 (spikes/s/dB) 100 0 50 100 Tone level (dB SPL) p = 0.630 CS-15 (spikes/s/dB) 0 Nml. Exp. p = 0.277 10 CS-15 (spikes/s/dB) 200 0 CS-15 (spikes/s/dB) Driven rate (spikes/s) Driven rate (spikes/s) 400 50 15 Avg. rate (spikes/s) Unsl 100 0 50 100 Tone level (dB SPL) p = 0.378 CS-15 (spikes/s/dB) D Pct. nonmonotonic Lckr Pct. nonmonotonic E 100 Pct. nonmonotonic On 0 Pct. nonmonotonic D 100 Pct. nonmonotonic Ch 200 100 Pct. nonmonotonic C Driven rate (spikes/s) PriN Driven rate (spikes/s) B Driven rate (spikes/s) Pri Driven rate (spikes/s) A 300 150 100 50 0 -20 0 20 40 Level (dB re Threshold) 200 100 0 -20 0 20 40 Level (dB re Threshold) 400 200 0 -20 0 20 40 Level (dB re Threshold) 150 100 50 0 -20 0 20 40 Level (dB re Threshold) 200 100 0 -20 0 20 40 Level (dB re Threshold) 100 50 0 -20 0 20 40 Level (dB re Threshold) Fig. 3.23. BF-tone rate-level functions of different types of VCN neurons in normal and noise-exposed ears. Shape and slope analyses of the BF-tone RLFs for units with BF below the CAP-audiogram edges, for the normal-hearing and noise-exposed animals. Units with different PSTH types are summarizes in different rows. Color code: Blue: normal-hearing; purple: noise-exposed with contaminated CAP recordings (animals #06-067 and #06-081); red: noise-exposed with uncontaminated CAP recordings. Column 1: Population plots of the BF-tone RLFs. Column 2: The percentage of non-monotonic RLFs in the normal and exposed groups. The criterion and algorithm for classifying RLF shape has been discussed in Section 2.6.2 and illustrate in Fig. 2.6. p-values are from Fischer’s exact test. Column 3: 15-dB chord-slopes of the rate-level functions shown in the first column. In each box plot, the lines show upper quartile, median and lower quartile; the whiskers show 1.5 interquartile ranges above and below the. Outliers are shown by red crosses. p-values are from two-tailed Wilcoxon rank-sum tests. Column 4: Average of the rate-level functions shown in Column 1, with sound level measured in dB PSL. Equal weights were assigned to the individual RLFs. Column 5: Average firing rate versus level re threshold. In other words, the individual RLFs were first aligned at their thresholds and then an averaging was done. Vertical gray lines indicate threshold level. The circles and error bars show mean ± 1 S.E. of rates at 30 dB re threshold. 144 decreases in slopes, i.e., saturation (Kiang et al., 1965; Sachs and Abbas, 1974). The maximum (saturated) driven rate was usually between 100 – 300 spikes/s. Pri neurons in impaired ears showed similar rate increases at levels close to the thresholds. However, probably due to the elevated thresholds (40 – 70 dB), complete saturation couldn’t be demonstrated in most of the impaired neurons. Compared with Pri units, the PriN units showed more complicated and diversified BF-tone RLF shapes. Non-monotonic RLFs were more commonly seen in the PriN units than in the Pri ones. While about two-thirds of the PriN RLFs had the same increase-and-saturate shapes as the ANF and Pri neurons, the rest showed strongly non-monotonic responses, with a local rate maximum at 20 – 30 dB above threshold, above which driven rate sloped down but rebounded at very high levels (> 70 dB SPL). In the BF-tone RLF of many Pri and PriN units, rate dips at about 80 – 90 dB SPL could be seen, which were probably caused by C1/C2 transitions. The rate-level slopes near threshold were analyzed by the 15-dB chord slopes (CS-15, Section 2.6.4.4). As can be seen from Column 1, the choice of the 15-dB span avoided the confounding factor of non-monotonicity in the RLF shapes. As shown by the panel in the third column, the medians of the chord slopes were lower in the impaired ear than in the normal ear for both Pri and PriN categories, and the ranges of rate-level slopes of impaired PL neurons were concentrated at the lower half of the normal range. However, due to the small sample sizes, the difference in slopes didn’t reach statistical significance in either category (Two-tailed Wilcoxon rank-sum tests, significance level: 0.05). However, when the Pri and PriN groups were considered as one group, the trend for exposed neurons to show shallower chord slopes was significant (p = 0.029. See Fig. 3.25). 145 The average RLFs on a dB-SPL scale and a dB-re-threshold scale are shown in the 4th and 5th columns. When averaged on a dB-SPL scale (4th column), the average RLFs of Pri/PriN neurons in impaired hearing were significantly shallower than their normal counterparts, which partially reflected the large variability of threshold shifts of neurons in impaired ears. This threshold variability originated from pooling data from different BF ranges and different animals. To account for this variability and single out the effects of rate-level slopes, we aligned the individual BF-tone RLFs at their thresholds before averaging (Column 5). In both Pri and PriN categories, the slope of the average RLFs on a dB-re-threshold scale showed similar steeps of rate growth with level in the first 10 dB from threshold for the two hearing states. However, at levels greater than 10 dB re threshold, the rates and rate-level slopes of impaired average RLFs fell below the normal ones. This was consistent with the result of chord-slope comparison shown in Column 3, and consistent with the fact that maximum driven rates were conspicuously smaller in the impaired populations, especially in the PriN group, which can be seen in Column 1. These results jibed with the previous observation that on average, AN fibers exhibited shallower BF-tone RLFs following acoustic trauma (Heinz and Young, 2004). In normal hearing, as in the individual RLFs, the average BF-tone RLFs for Pri and PriN types showed rate saturation beyond 20 dB above threshold. In contrast, saturations weren’t obvious in the impaired average RLFs. As shown in Row 3, Column 1 of Fig. 3.23, the typical shape of a normal Chopper neuron was a steep rate increase over a relatively small dynamic range (about 20 dB), followed by a nearly flat saturated portion, which was reflected in the average RLFs. A good few of chopper RLFs showed saturated portions with slightly negative slopes, leading to the high prevalence of 146 non-monotonicity (Column 2, see Section 2.6.2 and Fig. 2.6). Apart from this typical shape, a few other Chopper neurons showed complicated RLF shapes, such as a shape with multiple turning points. These reflect the more complicated neural integration and interplay between excitation and inhibition in choppers than in Pri units. These normal shapes were less frequently seen in choppers in impaired ears, whose RLF shapes showed higher prevalence of monotonicity (Column 2, significant). One interesting observation was that exposed choppers tended to show close-to-normal or even steeper rate-level slopes compared with normal ones (Column 3). Also, the maximum driven rates were above the normal range in a few impaired choppers. This was in contrast to the situations with the PL neurons. Together with alterations in RLF shapes, these enhanced rate-level slopes and maximum driven rates caused the average RLFs to be steeper in the impaired population than in the normal one, whether calculated on an aligned or unaligned scale. These observations are qualitatively consistent with loudness recruitment, in that threshold elevation is accompanied by suprathreshold overexcitability. Quantitative analysis on this phenomenon will be done in later sections. As shown in Fig. 3.23, Row 4, onset units in normal hearing showed a mixture of monotonic and non-monotonic rate-level responses to BF-tones. The RLFs of many (but not all) onset neurons differed from those of the other typical VCN neuronal types in their extremely wide dynamic ranges (> 80 dB). However, what’s untold by the RLFs was the change in their temporal firing patterns from onset to more sustained ones as the tone levels increased. Similar previous observations led Rhode and Smith (1986) to suggest that certain subtypes of onset neurons may play important roles in level encoding. When averaged on an SPL scale, the average RLF of normal onset unit showed monotonic rate increase over a level range greater 147 than 80 dB (Column 4), which was in contrast to the saturated shapes in PL and chopper groups. The saturated shape in threshold-aligned averaging was superficially self-contradictory (Column 5). In fact, it reflected a balance between monotonic and non-monotonic rate-level responses in onset units. We were able to sample only three onset neuron in the sub-edge BF regions in the impaired ears. Thus comparison across hearing states was feeble. However, all the three onset units recorded from exhibited monotonically increasing shapes in their BF-tone RLFs, in contrast to the mixture of monotonic and non-monotonic RLFs in the normal population. On average, driven rates at 20 dB re threshold and higher were greater in the impaired population than in the normal-hearing one. Locker neurons in the normal-hearing ears showed a mixture of saturating and unimodal non-monotonic shapes; whereas most Locker units in the impaired ears possessed monotonically increasing or saturating RLFs. The rate-level slopes of Locker units were more dispersed after acoustic trauma. As can be seen in the third column, the median and maximum value of Locker rate-level slopes were greater in the impaired population, however, impaired ears also contained a few unusually shallow Locker RLFs. One likely explanation for this dispersed distribution of rate-level slopes in the impaired population was that we didn’t distinguish between PL and chopper sub-types in low-BF neurons. As suggested by the data shown in the previous three rows, the steeper-than-normal slopes might originate from phase-locking chopper units, where as those shallower-than-normal ones might be recorded in phase-locking PL units. When averaged on a SPL scale, the impaired function was slightly shallower than the normal one. However, the steepness of average RLFs was essentially equal between the two hearing states when the averaging was done on a dB-re-threshold scale. 148 The unusual-type units formed a heterogeneous group in terms of RLF shapes, which ranged from increase-and-saturate, monotonically increasing, to barely responsive. It was noteworthy that a number of unusual neurons also had large (> 60 dB) dynamic ranges. As in the onset group, unusual neurons showed the large dynamic range in their average RLFs (Row 6, columns 4 and 5). Interestingly, on average, driven rate increased faster with level in the impaired in the impaired population than in the normal one. This led to a rate catching-up and overtaking at above 90 dB SPL. As in the case of choppers, this was very analogous to loudness increase in recruiting ears. Important contribution to this phenomenon came from a few exposed unusual-type neurons which showed very large driven rates and steep rate-level slopes. For example, the outlier shown in Column 3 corresponded to the unusual unit whose response map has been shown in Fig. 3.21.E. 3.6.2. Slopes of rate-level functions for tones Compared with ANF recording, single-unit recording in the auditory brainstem is faced with the problem of limited data yield. In order address with this problem, we used a pseudopopulation approach in the analysis of tonal RLFs. Detailed rationale and methodologies of this approach can be found in Methods Section 2.6.5. In this approach, we treated all tones with frequencies below the CAP-audiogram edge frequency as one “virtual” tone. In the impaired ear, this virtual tone mimics a tone falling into the frequency regions with substantial threshold elevations. In the tonal pseudopopulation, BFs of the neurons were no longer expressed in kHz, but in octaves relative to the tone frequency (ORF). In a dual sense, tone frequencies can be expressed in octaves re BF (ORB), if we choose to use real frequency units on BFs. One potential caveat of this approach was that the octave difference between unit BF and tone 149 frequency has a systematic effect on the steepness of rate-level functions. For a given AN fiber, the lower the tone frequency is relative to the unit BF, the steeper the rate-level slope tends to be (See Fig. 6. in Heinz et al., 2005). The same is likely to be true in the VCN neurons. Therefore, it was important to ensure that when comparing rate-level slopes in normal and impaired ears, the ORB distributions are in agreement between the two samples. To illustrate the relationship between the two kinds of slope measured, Fig. 3.24 shows the comparison between low-level fit slopes (LLFSs) and 15-dB chord slopes (CS-15) measured from the same RLFs. The gray line in each panel indicates equality between the two slope measures. It can be seen that these two slope measures are generally in linear relation to each other, and analyses based on these two measures were expected to lead to commensurate conclusions under most conditions. However, the fit slopes showed a systematic trend to be greater than or equal to the chord slopes. This reflects the fact that local slope of RLFs tend to be steeper than their global slopes near BF (e.g, the RLF shown in Fig. 2.5). Primary-like and primary-like-with-notch Fig. 3.25 shows the statistical comparisons of the rate-level slopes for BF and off-BF tones of the Pri and PriN neurons in normal and impaired ears. In Panel A, the 15-dB chord slopes (CS-15) of individual RLFs (i.e., neurons in the pseudopopulation) were plotted against tone frequency in ORB. The yellow vertical band showed the 0.4-octave on-BF region. No significant differences in slopes between the Pri (triangles) and PriN (inverted triangles) units were seen and therefore they were pooled. The blue and red curves in Panel A are the running averages (arithmetic means) of CS-15 versus ORB. Panel E is in the same format as Panel A, but it shows the low-level fit slopes (LLFS). The chord slopes and fit slopes were not based on exactly the 150 A. PL B. Chopper 30 20 LLFS (spikes/s) LLFS (spikes/s) 40 10 0 0 5 10 15 CS-15 (spikes/s) 30 20 10 0 20 C. Onset 10 15 20 25 CS-15 (spikes/s) D. Locker 10 20 LLFS (spikes/s) LLFS (spikes/s) 5 5 0 -5 15 10 5 0 0 2 4 6 8 CS-15 (spikes/s) 5 10 15 CS-15 (spikes/s) E. Unusual LLFS (spikes/s) 30 20 10 0 0 5 10 15 CS-15 (spikes/s) 20 Fig. 3.24. Relationships between chord slopes and fit slopes. Comparison of 15-dB chord-slopes (CS-15) and low-level fitting slopes (LLFS) for BF-tone RLFs of different types of VCN units (shown in different panels). The gray line in each panel indicates equality between the two types of slope measures. 151 Pri/PriN 20 0.5 15 Fraction of RLFs CS-15 (spikes/s/dB) 11 33 10 5 0 -5 Median=6.49 Median=5.70 Median=7.14 Median=5.27 0.4 0.3 0.2 0.1 -4 0 -2 0 2 Tone freq. (oct re BF) D 0 5 10 15 CS-15 (spikes/s/dB) E 60 0.5 Fraction of RLFs 40 30 20 10 p = 0.055 0 -2 -4 Nml. 100 Exp. p = 0.022 50 0 Nml. Exp. All-freq Nml. p = 0.496 Exp. On-BF Nml. p = 0.115 Exp. Median=9.46 Median=8.49 Median=9.66 Median=6.53 0.4 0.3 2 p = 0.166 0 -2 -4 Nml. Exp. 0.2 0.1 0 -10 2 F 11 33 50 LLFS (spikes/s/dB) All-freq Nml. p = 0.248 Exp. On-BF Nml. p = 0.022 Exp. Tone f. (oct. re BF) C Pct. nonmonotonic B Tone f. (oct. re BF) A -4 -2 0 2 Tone freq. (oct re BF) 0 0 20 40 LLFS (spikes/s/dB) Fig. 3.25. Rate-level slopes for tones in the PL (Pri/PriN) group. Comparison of 15-dB chord slopes (CS-15), low-level fit slopes (LLFS) and RLF shapes for tonal RLFs in the PL (Pri and Pri-N) group in the two hearing states. Data with unit BFs and tone frequencies below the CAP-audiogram edges were included. No restriction was set on tone frequencies in octaves re BF (ORB). A. CS-15 plotted against tone ORB. Unit type symbol and hearing-condition color code conventions are as in previous figures. The curves are moving-window averages over frequencies. The yellow band shows the on-BF frequency region (less than 0.2 octaves away from the BF). The blue and red numbers at the top left corner of the plot are the number of real-population neurons in the normal-hearing and exposed populations. B. CS-15 distribution histograms for the two hearing conditions. Medians for BF-only and all-frequency (all-ORB) cases under the two hearing conditions are shown. The p-values are from two-tailed Wilcoxon rank-sum (WRS) tests. C, Top. Comparison of ORB distributions of the two populations. p-values are from WRS tests. For Pri/PriN units, impaired chord-slope data with ORBs below 2 were excluded to achieve congruent distributions of ORBs in the two populations. C, Bottom. Fractions of non-monotonic RLFs for the two populations. The p-values is from a two-tailed Fisher’s exact tests. E. Same as A, but for the low-level fit slopes (LLFS). F. Same as B, but for the LLFS. G. Same as C, top, but for LLFS. 152 same sample of pseudopopulation neurons, because it was not always possible to record at 15 dB above threshold, especially at large absolute values of ORBs and in impaired units with large threshold shifts. A trend of slopes to decrease with increasing ORB was seen in the unexposed population. The same slope-ORB gradient wasn’t as readily observable in the impaired population. Because of the very high thresholds, no 15-dB chord slopes could be calculated for positive ORBs in the impaired data pool. This was remedied by the fit slopes, whose calculation didn’t depend as heavily on level ranges. The fit slopes at positive ORBs were steeper in the impaired ears. Therefore, it can be concluded that after hearing loss the dependence of rate-level slopes on the distance between BF and tone frequency gets weaker in the PL category, which probably reflects similar post-trauma changes in AN fibers. This was in agreement with a previous observation that the rate-level slopes of AN fibers in ears with ototoxic damage were similar across different frequencies, in contrast to the RLFs of normal fibers (Harrison, 1981). In Panels B and E, the distribution functions of the chord and fit slopes for unrestricted ORBs are shown, respectively. The ORB distribution for chord-slope data showed a difference between the two populations, in which the ORBs in the impaired population were biased toward more negative values. Therefore, the data in the impaired CS-15 population with ORBs below -2.5 were discarded for the analysis in Panel B. No data were discarded for fit slope comparison. Panels C and G showed that the range of ORBs for comparison of chord and fit slopes agreed across the two populations. Visual inspection of the data indicated that the ranges of slopes of the two hearing states were largely overlapping. This is confirmed by the values of the median slopes for all-frequency cases, which were essentially equal in the two populations. 153 As mentioned earlier, the medians of the chord slopes (Panel B) for on-BF tones (ORB = 0) were significantly shallower in the impaired ear (p = 0.029, WRS test). This trend of on-BF RLFs showing shallower slopes was confirmed by the fit-slope analysis (Panel E), although this difference didn’t reach significance. Based on these observations, conclusion can be reached that in VCN Pri/PriN units, acoustic trauma significantly reduced the slopes for tones at or near the BF, had less effect on the slopes for tones below BF, and increased the slopes for tones above BF. These alterations were most saliently manifested in the moving-window average of LLFS (red curve in Panel D), which shows an “M” shape, with its dip situated at and slightly below the on-BF region. With unrestricted ORBs, there were no significant differences between Pri/PriN tonal rate-level slopes before and after acoustic trauma. Panel D shows the percentage of non-monotonic RLFs in the normal and impaired Pri/PriN units for the entire ORB ranges. Compared with on-BF RLFs (Row 1 & 2, Column 2 in Fig. 3.23), the fractions of non-monotonic RLFs were less for the entire ORB range. Acoustic trauma significantly decreased the percentage of non-monotonic rate-level functions (p = 0.027, Fisher’s exact test). It can be observed in Panels A and D that the slopes of the two experiments showing noise-contaminated CAP recordings (06-067 and 06-081, shown in purple, see Section 2.8 for details) were systematically steeper than those from uncontaminated exposed animals (red). This difference was unlikely to be a consequence of the speaker noise, for several reasons: (1) as mentioned earlier only units with BBN thresholds at least 20 dB above the level of the speaker noise were included; the rest have been discarded; (2) this difference was seen for units with BFs 154 falling out of the noise frequency band; (3) the expected effect of noise, was to reduced the slopes through adaptation and rate suppression, not to cause steeper slopes as seen here. Therefore, it is likely that the differences seen here reflect the effects of threshold shift on rate-level slopes, because as mentioned earlier, the units from the noise-contaminated experiments generally had less threshold shifts (Fig. 3.3). One possible explanation is that, in more severe hearing losses, the damage to the IHCs were more severe, and/or the morphological and synaptic changes in the primary endings in VCN were more dramatic, leading to weaker responses and shallower rate-level slopes in the CAP-uncontaminated hearing-impaired population. Chopper The tonal rate-level slopes of chopper neurons were shown in Fig. 3.26, with the same format as in Fig. 3.25. Comparison of the slope distributions of the ChS and ChT subtypes didn’t indicate significant differences in either hearing condition (p > 0.11, WRS test). The two subtypes were then treated as one group in this analysis. In the normal population, the negative correlation between slopes and ORBs observed in ANF and VCN Pri/PriN units were seen in the Chopper category as well (in both chord and fit slopes). However, the pattern of post-trauma slope change was different from that observed in the Pri/PriN category. For ORBs at and near zero, the mean fit and chord slopes were similar to or slightly greater in the noise-exposed population. However, the slopes at sufficiently negative ORBs (< -1.5) decreased after HL. These changes at and below BF were observed in both types of slope analysis. After impairment, slopes at positive ORBs also showed a tendency to decrease. In fact, in the exposed ears, all the Chopper rate-level slopes at positive ORBs were below zero. 155 Chopper 1 11 15 0.8 20 Fraction of RLFs CS-15 (spikes/s/dB) 25 15 10 5 0.6 0.4 0 -3 -2 -1 0 1 Tone freq. (oct re BF) D 0 10 20 CS-15 (spikes/s/dB) E 60 0.5 Fraction of RLFs 40 30 20 10 p = 0.094 -2 -4 100 Nml. Exp. p = 0.000 50 0 Nml. Exp. All-freq Nml. p = 0.866 Exp. On-BF Nml. p = 0.157 Exp. Median=14.43 Median=12.77 Median=14.23 Median=16.36 0.4 0.3 p = 0.114 0 -2 -4 Nml. Exp. 0.2 0.1 0 -10 0 F 11 15 50 LLFS (spikes/s/dB) Median=10.71 Median=9.22 Median=12.17 Median=12.70 0.2 0 -5 All-freq Nml. p = 0.467 Exp. On-BF Nml. p = 0.292 Exp. Tone f. (oct. re BF) 30 C Pct. nonmonotonic B Tone f. (oct. re BF) A -3 -2 -1 0 1 Tone freq. (oct re BF) 0 0 10 20 30 40 LLFS (spikes/s/dB) Fig. 3.26. Rate-level slopes for tones in the chopper group. Data with unit BFs and tone frequencies below the CAP-audiogram edges were included. ChS, ChT and ChL units were included. The format is the same in Fig. 3.25. 156 The decreases in slopes at negative and positive ORBs were probably based on the same mechanism, i.e., the strengthening of sideband inhibition after HL. When all ORBs were pooled, no statistically significant difference between the chopper slopes in normal and impaired ears were seen, although the decrease in slopes at frequencies far away from BF caused the impaired median to be slightly lower. For on-BF RLFs, the medians of chord slopes were about equal in the two populations, whereas the fits slopes were greater in the noise-exposed pool, although this didn’t reach significance. As in the Pri/PriN category, one significant and consistent effect of acoustic trauma was to reduce the number of non-monotonic RLFs in chopper units. Onset As in the PL and chopper categories, the normal onset units exhibited an average trend to possess larger rate-level slopes for tones at more negative ORBs (Fig. 3.27.A and D). The trend was preserved in the impaired onset units. However, the moving-window averages were greater in the impaired onset units than the unexposed ones. For unrestricted ORB ranges, cross-hearing-status differences in both fit and chord slopes reached statistical significance (Panels B and E). Moreover, there was a significant decrease in the prevalence of non-monotonic RLFs after trauma (Panel C, bottom). These observed changes pointed to an up-regulation of rate-level responses in the onset category, especially for frequencies below BF. However, as indicated in the Panel A and D, the small number of real-neurons used in this analysis (only 3 in the exposed pool) prevented any solid conclusion from being reached. Lockers As shown in Fig. 3.28, the Locker slopes were more dispersed after trauma, mainly because 157 Onset 1 9 3 15 10 5 0 -5 Median=1.62 Median=5.41 Median=2.40 Median=1.40 0.6 0.4 0.2 -3 0 -2 -1 0 Tone freq. (oct re BF) 0 5 10 CS-15 (spikes/s/dB) E 93 15 10 5 -1 -2 -3 Nml. Exp. p = 0.000 50 0 Nml. Exp. All-freq Nml. p = 0.011 Exp. On-BF Nml. p = 0.600 Exp. Median=2.04 Median=6.96 Median=1.98 Median=1.39 0.4 0.3 0 p = 0.918 -1 -2 -3 Nml. Exp. 0.2 0.1 0 -5 Fraction of RLFs 0.5 20 p = 0.395 F 30 25 0 100 Tone f. (oct. re BF) D LLFS (spikes/s/dB) All-freq Nml. p = 0.008 Exp. On-BF Nml. p = 1.000 Exp. 0.8 Fraction of RLFs CS-15 (spikes/s/dB) 20 C Tone f. (oct. re BF) B Pct. nonmonotonic A -3 -2 -1 0 Tone freq. (oct re BF) 0 0 5 10 LLFS (spikes/s/dB) 15 Fig. 3.27. Rate-level slopes for tones in the onset group. The format is the same as in Fig. 3.25. The rate-level slopes of exposed Locker neurons showed a bimodal distribution, with one peak showing steeper-than normal slope and the other showing shallower-than-normal slopes. This phenomenon was observable by both chord-slope and fit-slope methods. 158 Locker 1 5 11 0.8 Fraction of RLFs CS-15 (spikes/s/dB) 20 15 10 5 -1 0.6 0.4 0 0 1 2 Tone freq. (oct re BF) D 5 10 15 CS-15 (spikes/s/dB) E 25 5 11 0.5 Fraction of RLFs 20 15 10 5 p = 0.106 0 -1 Nml. 100 Exp. p = 0.007 50 0 Nml. Exp. All-freq Nml. p = 0.590 Exp. On-BF Nml. p = 0.959 Exp. Median=9.48 Median=9.50 Median=9.90 Median=10.02 0.4 0.3 3 2 1 0 -1 p = 0.210 Nml. Exp. 0.2 0.1 0 -5 1 F 30 LLFS (spikes/s/dB) Median=6.66 Median=5.80 Median=7.70 Median=8.71 0.2 0 -5 All-freq Nml. p = 0.294 Exp. On-BF Nml. p = 0.799 Exp. Tone f. (oct. re BF) 25 C Pct. nonmonotonic B Tone f. (oct. re BF) A -1 0 1 2 3 Tone freq. (oct re BF) 0 5 10 15 20 LLFS (spikes/s/dB) Fig. 3.28. Rate-level slopes for tones in the locker group. The format is the same as in Fig. 3.25. The rate-level slopes of exposed Locker neurons showed a bimodal distribution, with one peak showing steeper-than normal slope and the other showing shallower-than-normal slopes. This phenomenon was observable by both chord-slope and fit-slope methods. 159 there were a few exposed Lockers with very shallow rate-level slopes (< 5 dB spike/s/dB). Unlike all the other slope distribution histograms shown in Figs. 3.25 – 3.27, the distribution of the slopes of exposed Locker units was bimodal. In one mode, the slopes were steeper than normal; and in the other mode, the slopes were shallower than normal. When considered as a whole, Locker units didn’t show significantly altered rate-level slopes after acoustic trauma. A change in the Locker units post-trauma is that slopes at positive ORBs were larger than normal, similar to the changes in Pri/PriN neurons. Also as in the Pri/PriN category, the ORB dependence of rate-level slopes in normal-hearing was lost after trauma. Furthermore, as in other categories, percentage of non-monotonic RLFs were significantly reduced in the exposed population. Unusual In the unexposed population, the two subtypes of Unusual units, Unusual-A (high P/S ratio) and Unusual-B (low P/S ratio) showed significant differences in their rate-level slopes, with the Unusual-B slopes larger than the Unusual-A ones (p < 0.0001, WRS test, not shown). However, no significant differences were observed for the Unusual units in exposed ears (p = 0.57, WRS test). And since the proportions of A and B subtypes in the two hearing states were not substantially different, we treated them as one group in Fig. 3.29. In the normal-hearing ear, the unusual-type neurons showed shallower slopes than Pri/PriN and Ch type neurons, which reflected their lower driven firing rates. The dependence of average slope on ORB wasn’t clearly observable. Interestingly, the unusual-type neurons are the unit category in which HL had the strongest effect on the distributions of rate-level slopes. Except at positive ORBs, the mean slopes in the impaired ear were substantially above those in the normal ear at all ORB ranges (Fig. 3.27, Panels A and D). This effect was more dramatic at very more 160 Unusual 1 9 12 0.8 Fraction of RLFs CS-15 (spikes/s/dB) 20 15 10 5 0.6 0.4 0 -2 0 2 Tone freq. (oct re BF) D 0 5 10 15 CS-15 (spikes/s/dB) E 50 9 12 Fraction of RLFs 30 20 10 0 -10 20 1 p = 0.063 0 -1 -2 Nml. 100 Exp. p = 0.223 50 0 Nml. Exp. F All-freq p = 0.194 0.5 On-BF p = 0.649 40 LLFS (spikes/s/dB) Median=1.78 Median=2.21 Median=3.63 Median=3.49 0.2 0 -5 All-freq Nml. p = 0.460 Exp. On-BF Nml. p = 0.939 Exp. Pct. nonmonotonic 25 C Tone f. (oct. re BF) B Nml. Exp. Nml. Exp. Median=2.62 Median=2.88 Median=3.85 Median=6.11 0.4 0.3 Tone f. (oct. re BF) A 2 p = 0.032 0 -2 -4 Nml. Exp. 0.2 0.1 -2 0 2 Tone freq. (oct re BF) 0 0 10 20 30 LLFS (spikes/s/dB) Fig. 3.29. Rate-level slopes for tones in the unusual group. The format is the same as in Fig. 3.25. The on-BF and all-frequency rate-level slopes of exposed unusual-types were steeper than their normal counterparts. 161 negative ORBs. The exclusively negative slopes at positive ORBs were similar to those seen in the impaired Chopper population. The medians were higher in the impaired population for both on-BF and all-ORB cases; and there was a good agreement between the chord- and fit-slope analyses. For all-ORB comparison, the differences reached significance (p = 0.001 for fit slopes). However, the effect of deafening on monotonicity wasn’t significant, mainly because the Unusual type had a relatively low percentage of non-monotonic RLFs to start with (Fig. 3.23). 3.6.3. The relationship between rate-level slope and BF / threshold shift To investigate the effect of threshold shift on slopes, we defined threshold shift of a unit as the difference between its BF threshold and the average CAP threshold at the same BF in the unexposed population (the dashed blue curve in Fig. 3.4). Panels A, B, C and D in Fig. 3.30 summarize the relationship between BF rate-level slopes and threshold shifts for Pri/PriN, Chopper, Locker and Unusual units, respectively. Unlike in the previous figures, sub-edge and supra-edge BFs were all included in this figure. Due to large data dispersions, the correlation coefficients were generally close to zero. Nevertheless, a few trends could be seen. A positive correlation between rate-level slope and threshold shift can be seen in the normal PL and Locker populations (although the correlation coefficient was lower due to a dispersion of data). This observation jibed with the previous finding that low-/medium-SR AN fibers tend to have higher thresholds, greater driven firing rates and therefore steeper rate-level slopes near threshold than high-SR fibers (e.g., Sachs and Abbas, 1974. See Fig. 1.2 for an example). In contrast, the impaired PL and Locker populations showed a negative correlation between rate-level slope and threshold increase. The larger the threshold shift was, the shallower the slopes tended to be. A possible explanation for this is that greater 162 A. PL 15 30 y = 0.067 x + 8.840 r2 = 0.012 y = -0.048 x + 6.782 r2 = 0.067 CS-15 (spikes/s/dB) CS-15 (spikes/s/dB) 20 B. Chopper 10 5 0 -5 25 20 15 10 5 0 -20 0 20 40 60 Unit threshold shift (dB) y = -0.108 x + 10.679 r2 = 0.045 y = 0.035 x + 12.162 r2 = 0.012 -20 0 20 40 Unit threshold shift (dB) C. Lockers D. Unusual 25 y = 0.250 x + 13.332 r2 = 0.926 y = -0.083 x + 9.086 r2 = 0.050 CS-15 (spikes/s/dB) CS-15 (spikes/s/dB) 25 20 15 10 5 0 60 -20 0 20 Unit threshold shift (dB) 20 15 10 5 0 -5 40 y = -0.069 x + 4.052 r2 = 0.226 y = 0.089 x + 1.977 r2 = 0.136 -20 0 20 40 60 Unit threshold shift (dB) E. Onset CS-15 (spikes/s/dB) 10 5 0 -5 y = -0.206 x + 0.983 r2 = 0.674 y = 0.069 x + 0.946 r2 = 0.166 -10 -20 0 20 40 Unit threshold shift (dB) Fig. 3.30. Relationships between threshold shift and BF-tone rate-level functions in VCN neurons Units with BFs below, at and above CAP-audiogram edge frequencies are all included in this analysis. The four panels show data in the four PSTH groups. Threshold shift was defined as the difference between single-unit BF threshold and the threshold of the normal averaged CAP audiogram at the corresponding frequency (the dashed blue curve in Fig. 3.4). 163 threshold shifts are correlated with more severe IHC damages, whose primary effect is to suppress the rate responses of the AN fibers and subsequently the responses of the bushy cells. In contrast, little correlation between slope and threshold shift could be seen in the impaired chopper units. The regression line had a close-to-zero but positive slope. A positive correlation between slopes and threshold shifts can also be seen in the impaired unusual and onset populations. In fact, the chopper and unusual units with the steepest BF rate-level slopes were those units with severe threshold shifts. The post-trauma alteration in the onset category was particularly interesting because in the normal-hearing onset population, a strong negative relationship between rate-level slope and threshold was seen (r2 = 0.675). Considering the negative correlation between threshold increases and average rate-level slopes in PL neurons and possibly also in AN fibers, the slightly positive correlations observed in non-primary-like unit types are suggestive of working of certain neural plastic mechanisms, which function to compensate for suppressed rate responses following HL and had stronger effects for more significant threshold shifts. 3.6.4. ANOVA analysis on tonal rate-level slopes The previous several sections were concerned with slope comparisons on a unit-type-by-unit-type basis. Fig. 3.31 shows the box plots of the whole sub-edge RLF data body by unit type and hearing status. Intra-category comparisons of slopes between the two hearing states were done (WRS test, significant level of 0.05). As shown, in none of the unit types, hearing state had a significant effect on the slope median. When a two way ANOVA (See figure caption for details) was performed, no significant effect by noise exposure on the low-level fit slopes was seen (df = 1, F = 1.33, p = 0.25). Significant effects by hearing condition were not 164 A CS-15 (spikes/s) 20 15 10 5 Nm l. P Ex L p. P Nm L l. C Ex h p. C Nm h l. O n Ex p. O Nm n l. U n Ex p. U Nm n l. L Ex ckr p. Lc kr 0 B 60 LLFS (spikes/s) 50 40 30 20 10 PL l. Nm -10 Ex p. P Nm L l. C Ex h p. C Nm h l. O Ex n p. O Nm n l. U Ex n p. U Nm n l. L Ex ckr p. Lc kr 0 Fig. 3.31. Analysis of variance on the rate-level slopes for tones with unrestricted ORB ranges in normal and exposed ears. Summary and comparisons of 15-dB chord-slopes (A) and low-level fit-slopes (B) for tonal RLFs. Only RLF data with BF and tone frequencies below the edge frequencies were included. ORB truncations for the Pri/PriN and Choppers as those performed in Figs. 3.25 and 3.26 are done here. Asterisk: significance (alpha < 0.05) for a two-tailed Wilcoxon rank-sum test. See the caption to Fig. 3.23 for the description on the format of the box plots. For each type of slope measures, a two-way ANOVA (factors: 1). hearing conditions and 2). PSTH types) was done. This analysis indicated no significant effects by hearing condition on the chord slopes (df = 1, F = 0.32, p = 0.5738), and no significant effects on fit slopes either (df = 1, F = 1.33, p = 0.25). This analysis also indicated very significant effects of unit type (df = 4, F = 36.03, p ≈ 0 for CS-15; df = 4, F = 25.76, p ≈ 0 for LLFS). 165 seen for chord slopes either (df = 1, F = 0.32, p = 0.5738). As expected, for both fit and chord slopes, the effect of unit response types were highly significant. The same analysis was performed on BF-tone rate-level slopes, and the results also indicated no significant effects by hearing status (df = 1, F = 0.24, p = 0.6246 for CS-15; df = 1, F = 0.12, p = 0.7311 for LLFS, not shown). This reflects the differential effects of trauma on the slope distributions in different PSTH types. As a summary of the several proceeding section, we found that the distributions of VCN tonal rate-level function slopes did differ between normal-hearing and traumatized ears, but only under certain conditions. The effect of acoustic trauma on the steepness of rate-level functions of VCN neurons was not uniform and cannot be generalized in a simplistic way. In PL, chopper and locker response types, slope comparison based on unrestricted ORB range indicates no significant differences. However in the onset- and unusual-type units, the distribution of rate-level slopes for all-ORB range was elevated in the exposed population. One must carefully differentiate between not only unit types, but also between ORB ranges to see the effects of acoustic trauma. The most salient effects of noise exposure were to reduce the PL rate-level slopes at and near BF and to increase the at-/near-BF slopes of chopper neurons. Curiously, chopper neurons exhibited slightly decreases in their far-ORB slope following trauma. The Unusual type showed a significant increase in rate-level slopes over a wide range of ORBs after HL. When all PSTH types and all ORB regions were considered as a whole, as indicated by the ANOVA discussed in Section 3.6.4, steepness of the RLFs at low and medium levels above threshold was not significantly altered after NIHL. 166 3.6.5. Thresholds of tonal rate-level functions When summing or averaging a population of RLFs, the factors that determine the slope or steepness of the resultant summed or averaged RLF include not only the shapes and slopes of the individual RLFs, but also the spread of their thresholds. The more concentrated the thresholds are, the steeper the average RLF tends to be; and vice versa. In fact, according to one hypothesis of the origin of loudness recruitment, the distribution of BF-tone thresholds for a group of ANFs with similar BFs and different SRs gets more compact following hearing loss, and leads to steeper increase of average rate with sound level (Moore et al., 1985; Zeng and Turner, 1991). The same hypothesis is applicable to the VCN neurons and can be tested based on the data in the current study. Another major hypothesis is also based on the effect of threshold distribution on the steepness of summed rate activity, but is concerned with the distribution over a range of BFs for a fixed frequency. This hypothesis holds that due to the loss of fine frequency tuning after cochlear damage, for a tone of certain given frequency, the thresholds of the units with BFs at or near the tone frequency are substantially elevated, while the thresholds of the units with BFs away from (mainly above) the tone frequency are largely unchanged, or even gets slightly decreased. These two trends cause the distribution of thresholds for a tone frequency to be compressed over the whole tonotopy, which leads to faster increase of number of excited BF channels and steeper summed rate-level responses (Kiang et al., 1970; Evans, 1975). In fact, the term “recruitment” itself comes from the idea that increase in recruited psychophysical auditory filters by a given increase in level are greater-than-normal in impaired hearing, which is closely related to the second hypothesis cited above. 167 Therefore, it is important to analyze the slope distributions, for both the on-BF case and for all-BF population coding. The pseudopopulation approach provides a convenient way of performing these analyses. Fig. 3.32.A shows the pseudopopulation threshold distributions in the two hearing states for different unit types. The abscissa of each panel is the virtual tone frequency expressed in ORB; the ordinate is the threshold of the pseudopopulation neurons measured in uncorrected dB SPL. The threshold-ORB relations in the normal-hearing data of all types show fine tips with very low thresholds near zero ORB. In most unit types except the Lockers and Unusual units, there were rapid increases of minimum threshold with ORB increasing above zero. The minimum-threshold-versus-ORB ramp at negative ORBs were less steep. In the Locker and Unusual neurons, the sub-BF and supra-BF minimum-threshold-versus-ORB ramps were about equal in steepness, which reflected the broad and near symmetric tuning curves of these neurons. These observations confirmed that the pseudopopulation approach preserved the basic threshold-frequency configurations of the real neuronal population in each unit category. The minimum-threshold-versus-ORB slopes in the impaired data set were less steep, especially at negative ORBs. Overall, the span of thresholds over the entire ORB range was compressed after hearing loss. This was caused by (1) increased thresholds at zero and positive ORBs; (2) essentially unchanged or even lowered thresholds at negative ORBs (e.g., Panel A, Pri, ORB at around -3 oct.). Observations similar to the second change have been made in previous studies in AN fibers, which was referred to as “tail hypersensitivity” (Liberman and Kiang, 1978). The compression wasn’t observed in the Locker category, consistent with the fact that the Q10 decrease of exposed Locker neurons wasn’t as dramatic as high-BF neurons (Fig. 3.17, Panel B). 168 Threshold (dB SPL) Threshold (dB SPL) A 100 PL Ch On 80 60 40 20 0 -4 -2 100 Locker 0 2 4 Unusual 0 2 4 -4 -2 0 2 4 -4 Tone freq. (Octaves re BF) Unknown 80 60 40 20 0 -4 -2 -2 0 2 4 B Threshold (dB SPL) 100 80 60 40 20 0 0.1 1 BF (kHz) 10 Fig. 3.32. Thresholds for tones. A. A summary of thresholds in the tonal pseudopopulation as a function of tone frequency relative to unit BF (ORB). Thresholds in this figure include both excitatory and inhibitory thresholds, which causes the low thresholds at frequencies below BFs seen in the Chopper, Onset and Unusual categories. B. A population summary plot of real-neuron thresholds for the 2-kHz tone (dashed vertical line) as a function unit BF. 169 In Fig. 3.32.B, the real-population threshold distributions for the fixed frequency (2-kHz) tone are shown. These data are from all the neurons in which response to a 2-kHz tone were recorded; neurons with no rate responses within recorded level range were not shown in this figure. Notice the mirror-image relationship between the abscissa of this panel and the abscissas in Fig. 3.32.A. Following trauma, thresholds were elevated for BFs near and below 2 kHz, and essentially unaltered or even lower for BFs well above BF. In fact, tail sensitivity can be seen around BF = 15 kHz, and the neurons with the lowest threshold for a 2-kHz tone had BFs greater than 2 kHz in the impaired ear. These paralleled observation in Panel A, again supporting the validity of the RLF pseudopopulation approach. Fig. 3.33.A shows the threshold-versus-ORB plot for the tonal pseudopopulations. All response types except Lockers are included in this plot. The solid curves show mean ± 1 S.D. of the thresholds; the dashed curves show minimum threshold inside the 0.4-octave ORB channels. The asterisks at the top indicates significant differences in median of thresholds in the ORB channel below (two-tailed WRS test, p < 0.05); whereas the pentagram indicates significant differences in threshold distribution (two-tailed F-test, p < 0.05). It can be seen that significant threshold shifts occurred only for the ORB range between -2 and 1.0. For the rest of the ORB ranges, threshold changes were not significant. Except for one ORB bin centered at 0.8, significant differences in the standard deviations of the threshold distributions were not seen. The minimum threshold curves (dashed) in Panel A can be viewed as the tuning curves of the tonal pseudopopulations. Panel B shows the number of channels with minimum thresholds below certain tone SPL as a function of the SPL, which can be viewed as a spread-of-excitation curve in a crude sense. An approximately 50-dB threshold increase was apparent. However, the steepness 170 B 15 # of responsive ORB bins A 100 10 5 0 60 0 50 100 Tone Level (dB SLB) C 40 20 100 Nml. mean Nml. min Exp. mean Exp. min 0 -4 -2 0 Threshold at BF (dB SLB) Threshold (dB SLB) 80 Normal Exposed 2 Tone freq. (Octaves re BF) SD(Normal)=9.42dB SD(Exposed)=11.23dB p(F-test)=0.9136 80 60 40 20 0 Normal Exposed Fig. 3.33. Analyses on the thresholds in the tonal pseudopopulation. A. The pseudopopulation tuning curve: threshold versus ORB. The solid and dashed curves are respectively moving-window average and minimum thresholds in the 0.4-octave ORB bins. Error bars indicated ± 1 S.D. Asterisks indicate significance of threshold differences by Wilcoxon rank-sum tests (α = 0.05). Threshold included both excitatory and inhibitory ones. Pentagrams indicate significant differences in the standard deviations by F-test (α = 0.05). B. The number of responsive 0.4-octave ORB bins as a function of tone levels. A bin in which the minimum threshold is below the tone level was counted as responsive. C. Testing the compressed-threshold-distribution hypothesis. Thresholds for tones at BFs are expressed on a dB scale corrected for inter-frequency and inter-animal variability of audiogram thresholds by Eq. (2.4). Although the standard deviation is greater in the exposed population, the F-test indicates no significant differences between the two hearing states. 171 of the impaired curve significantly exceeded that of the normal curve. In the normal hearing population, it took about 75 dB of SPL increase for the number of excited channels to from 0 to 13; whereas the same amount of excitation spread took only about 35 dB. To test the compressed-BF-threshold-distribution hypothesis, we first performed SLB level adjustment on the thresholds of the RLFs inside the 0.4-octave central on-BF ORB bin to eliminate the effect of variability of threshold shifts across different BFs and different animals (See Section 2.6.5), and then compared the standard deviations of the two threshold distributions. As shown in Fig. 3.33.C. the S.D. of SLB-adjusted thresholds were slightly greater in the impaired population, but the difference wasn’t significant (F-test, p > 0.8). Therefore, there was no evidence that BF-threshold distributions became more compact after HL. Together with the result shown in Panel A, this result indicated that our data didn’t support the compressed-threshold-distribution hypothesis in the VCN neurons. 3.6.6. Average rate-level functions of the pseudopopulation tone As reviewed in Introduction Sections 1.1 and 1.3, it is often assumed that the summed firing rate is the code for stimulus intensity, or the correlate of loudness. However, it is debated whether the loudness of a narrow-band stimulus, e.g., a tone, is determined by only the firing rates of ANFs with BFs at or close to the tone frequency (on-BF encoding), or by the firing rate of all ANFs with BFs over the entire tonotopic range (all-BF encoding) (Smith, 1988). The same uncertainty exists in the VCN, the processing stage immediately after the ANF. Another question concerned with level encoding in VCN is which type(s) of neurons participate(s) in level encoding or loudness calculation; and if more than one unit type jointly encode sound level, how information from the different neuronal types is integrated. For these reasons, we have to distinguish between ORB 172 ranges and unit types when analyzing summed (or average) RLFs in VCN. In Fig. 3.34, Panel C shows the average RLFs for combinations of different response types and different ORB ranges. In the main area (bottom right), each row corresponds to a VCN response type. Each column corresponds to an ORB range. The entire virtual tonotopy in the pseudopopulation (i.e., the ORB axis) was divided into 5 intervals as shown, which contained approximately equal numbers of RLFs. The second column from the right contains average RLFs for on-BF and near-BF encoding, in which ORBs were between ± 0.4. In each row, the subtypes were given different weights according to estimated neuron counts (See Section 2.6.5.1 and Tab. 2.2 for details. These include Pri and PriN neurons in the first row and the ChS/ChT/ChL neurons in the second row. The SLB level adjustment was done to correct for the different variability of thresholds across different BF ranges and the two hearing states. When averaging multiple RLFs with different maximum levels, a 50%-quantile rule was used to determine the maximum level of the average RLFs (Section 2.6.5.1). The leftmost column shows the all-ORB average RLFs for the unit types, which were obtained from weighted averaging across the panels in the corresponding row according to the octave widths of the ORB intervals. The topmost row contains the all-unit-type average RLFs for different ORB ranges, obtained from weighted average of different unit types in each column. The weights for different PSTH types were discussed in Section 2.6.5.1 and shown in Tab. 2.2. The grand average RLFs shown in the upper left panel were obtained from averaging all panels in the main area according the unit type weights. For clarity, its expanded version and corresponding rate-matching curve are shown in Panels A and B. A linear regression was done on the rate-matching curve, whose slope (in dB / dB) is also shown. Rates in this figure are not normalized. This would potentially lead to biases toward unit 173 A B Normal Exposed 80 dB SLB in Nml. ear Driven rate (spikes/s) 100 60 40 20 0 -20 0 80 60 2.22 40 20 20 40 60 80 20 Tone level (dB SLB) 40 60 80 dB SLB in Exp. ear C 200 [-3.6,-2) oct. [-2,-1.2) oct. [-1.2,-0.4) oct. [-0.4,0.4) oct. [0.4,3.6) oct. 7 /17 7 /18 11/31 4 /9 4 /11 5 /11 12/16 5 /7 8 /2 6 /2 9 /3 4 /1 6 /7 6 /8 9 /11 7 /5 1 /0 2 /6 5 /11 3 /9 100 Driven firing rate (spikes/s) 0 100 PL 50 0 150 100 50 200 6 /22 100 Ch 0 40 On 20 0 -20 40 Unusual 20 0 -20 150 300 0 300 6 /12 200 100 0 100 3 /2 50 0 100 4 /8 50 0 300 0 /0 200 100 0 0 -20 0 20 40 60 80 0 40 100 50 Locker 80 0 40 80 0 40 80 0 40 80 0 40 80 Tone level (dB SLB) Fig. 3.34. Unnormalized average rate-level functions for the tonal pseudopopulation. A. Grand average unnormalized RLFs. The circles indicate thresholds according to a rate criterion of 2 spike/s, which determined the lower limit of the levels over which the rate-matching curve shown in B was fit over. B. The rate-matching curve. The rate-matching function was fit by a linear regression and its slope in dB/dB is shown. C. Decomposition by PSTH types and ORB channels. In the main area, each column corresponds to an ORB range. The second column from the right shows on-BF responses. Each row corresponds to a PSTH type. For the purpose of visualization, the subtypes (e.g., Pri and PriN in the PL category) are combined. The top row shows the averages across PSTH types for different ORB ranges. The bottom group in the left column shows the averages across ORB ranges in different PSTH types. The grand average RLF for all ORB ranges and all PSTH types is shown on the top left panel. The numbers colored in blue and red at the upper left corner in each panel in the main area are the number of pseudopopulation neurons (RLFs) for each combination of unit type and ORB range. Each real-population unit is used in one cell no more than once. 174 B Normal Exposed 0.6 dB SLB in Nml. ear Normlized driven rate A 0.4 0.2 0 -20 0 80 60 2.35 40 20 20 40 60 80 20 Tone level (dB SLB) 40 60 80 dB SLB in Exp. ear C 1.5 1 [-3.6,-2) oct. [-2,-1.2) oct. [-1.2,-0.4) oct. [-0.4,0.4) oct. [0.4,3.6) oct. 6 /22 7 /17 7 /18 11/31 4 /9 6 /12 4 /11 5 /11 12/16 5 /7 3 /2 8 /2 6 /2 9 /3 4 /1 4 /8 6 /7 6 /8 9 /11 7 /5 0 /0 1 /0 2 /6 5 /11 3 /9 0.5 Normalized driven rate 0 0.6 PL 0.4 0.2 0 0.6 Ch 0.4 0.2 0 0.6 On 0.4 0.2 0 0.6 Unusual 0.4 0.2 0 0.6 Locker 0.4 0.2 0 -20 0 20 40 60 80 1.5 1 0.5 0 1.5 1 0.5 0 1.5 1 0.5 0 1.5 1 0.5 0 1.5 1 0.5 0 0 40 80 0 40 80 0 40 80 0 40 80 0 40 80 Tone level (dB SLB) Fig. 3.35. Normalized average rate-level functions for the tonal pseudopopulation. The layout is basically the same as in Fig. 3.34. A. Grand normalized average RLF for the normal and noise-exposed populations. The rate criterion for calculating the thresholds (circles) is 0.02. B. The rate-matching curve. C. Decomposition by PSTH types and ORB channels. The format is the same as in Fig. 3.34.C. 175 types with higher driven rates when averaging over different unit types. To address this problem, we show the average normalized average RLFs in Fig. 3.35. The layout of this figure is the same as Fig. 3.34, but the driven firing rates are normalized by the average driven rate at 40 dB re threshold in response to BF tones in the normal population (See the 5th column in Fig. 3.23). The reason for the choice of a level of 40 dB re threshold was because in most common VCN neuronal types, driven firing rates for BF tone reached saturation at this level. 3.6.6.1. Average RLFs of different response types and ORB ranges In the normal-hearing population, for all the four unit types (PL, Ch, unusual and locker), rate saturation was seen in the near-BF ORB channel (-0.4 – 0.4 oct.). The dynamic ranges of these normal average RLFs for PL and Ch units were 30 – 40 dB, which are greater than the dynamic ranges of individual BF-tone RLFs, because ORB channels in this column included some slightly off-BF RLFs. The average responses in the ORB channel (-1.2 – -0.4 oct.) immediately below the central one were almost as strong as in the central ORB channel, although the thresholds were 20 – 40 dB higher and less saturation effects were seen. For ORB ranges more negative or more positive to these two ORB intervals, average response rates fell monotonically with increasing distance between the tone frequency and BF. Generally, all average RLFs were excitatory. Many of these trends were unseen in the impaired ears. First of all, rate saturation was rarely seen after trauma, which held true for all the response categories. Secondly, the differences in maximum rate responses in different ORB channels were not as large as in normal hearing. For example, in the Pri/PriN category, the maximum rates in the [-3.6, -2] and [-2, -1.2] ORB channels were about equal to those in the [-1.2, -0.4] and the central channels. This post-trauma change was 176 even more dramatic in the unusual group. The strongest responses of the unusual neurons deviated from the central ORBs toward very negative ORBs post trauma. Another property of the average RLFs lost in the exposed ear was the difference of thresholds between different ORB channels. The difference between the thresholds in the central ORB channel and that of the [-3.6, -2]-octave channels was approximately 50 dB in the normal PL group, but was only 20 dB in the impaired PL group. This change reflects broadened frequency tuning, and was observed in all response types shown in Fig. 3.34. When comparing post-trauma changes in the average RLFs of different PSTH types, dramatic difference behaviors were seen. Let us first concentrate on the central ORB channel, which corresponds to on-BF encoding of tone level. The Pri/PriN units showed diminished rate responses after acoustic trauma. Both the maximum driven rate and the average rate-level slope were about half of the normal value (Row 1, Column 4 in the main area of Panel C in Fig. 3.34). In contrast, the Chopper category showed very little changes in the maximum rate following trauma in the central BF channel. The impaired average RLF resembled a horizontally translated version of its normal counterpart (Row 2, Column 4), although in certain level ranges, the impaired slope were steeper than the normal slope. For the unusual-type units, acoustic trauma led to steeper average rate-level slope in the central ORB bin (Row 2, Column 5). These observations jibed with analysis on the slopes of individual RLFs, which led to conclusions that for chopper and unusual neurons, slopes distributions were close to or above normal in the impaired population. The average RLFs of the locker neurons before and after exposure were similar in slopes near threshold, but the maximum rate in the exposed ear were substantially lower (Row 4, Column 5). For very negative ORB ranges, the average Chopper RLFs can be seen to be slightly 177 shallower in the impaired ear (Row 2, Columns 1 and 2), which was consistent with the observation shown in Fig. 3.26 that chopper rate-level slopes at ORBs more negative than - 2 were reduced after acoustic trauma. When the response of different unit types were weighted-averaged (Fig. 3.34.C, top row, Column 4), the average rate-level slopes in the on-BF ORB bin were very similar in the two hearing states. The two curves could almost be described as horizontally translated versions of each other. However, average RLFs in the impaired ear in negative ORB ranges showed slightly steeper-than-normal slopes, these changes were mostly attributable to the effects seen in the non-primary-like groups. There were also interesting post-trauma changes at positive ORBs. In chopper and unusual categories, average rates in the [0.4, 3.6]-octave ORB bin were slightly negative, suggesting enhanced inhibition at frequencies above BF following NIHL. This observation was consistent with the observation that in the two unit types, the rate-level slopes for positive ORBs changed toward negative values in exposed ears (Figs. 3.26 and 3.29). 3.6.6.2. The grand average RLFs and rate-matching curves Panel A in Fig. 3.34 showed the grand average RLFs for the pseudopopulation tone under the two hearing conditions. The normal grand average curve showed a monotonically increasing shape, which supports the idea that all-BF encoding of tone intensity is a solution to the dynamic range problem (Smith, 1988). The extent of threshold shift in the impaired ear was approximately 40 dB. Increase of average driven rate with level was more rapid in the impaired population. The corresponding rate-matching curve shown in Panel B has a regression slope of 2.22 dB/dB, much greater than 1 dB/dB. In other words, at levels above the threshold, a given amount of tone level 178 increase incurred 2.22 times as much rate increase in the impaired ear as in the normal ear. This observation was qualitatively and quantitatively analogous to loudness recruitment. For normalized driven rates shown in Fig. 3.35, the comparison between the average curves under the two hearing conditions yielded similar recruitment-like phenomena with essentially the same rate-matching slope. In both the unnormalized and normalized situations, the maximum driven rates at 80-90 dB were similar between the two populations, indicating a rate catching-up and suggesting full recruitment. 3.6.6.3. Average rate-level functions for on-BF encoding Fig. 3.36 shows the average RLFs and corresponding rate-matching curves for on-BF level encoding in different types of VCN units and the AN fibers. The average RLFs shown in this figure were based an ORB range of [-0.2, 0.2]. The choice 0.4 octave window width was based on psychophysical results, which showed that the critical bands in cats were about 0.4 octave wide (Pickles, 1979). Slopes fit to these rate-matching curves were qualitative measures of the strengths of recruitment-like phenomena. Due to significantly lower rate-level slope and maximum driven rate in the impaired PL population, the corresponding rate-matching slope was 0.30, significantly below unity, suggesting anti-recruitment effects in the PL category. Because of the secure synaptic connections between ANFs and spherical and globular bushy cells in the VCN, one might expect that similar post-traumatic changes seen in rate-level responses should be seen in the ANFs and VCN PL units, given that the cochlear damages were similar in the animals used in the current study and the previous one (Heinz and Young, 2004 Fig. 3.17). The ANF data was initially processed in real-population ways in the original publications. 179 A. PL (on-BF) 500 150 100 50 60 40 20 0.32 Driven rate (spikes/s) 80 0 100 20 Tone level (dB SLB) 40 60 dB SLB in Nml. ear Driven rate (spikes/s) 150 100 50 3.49 20 0 0 20 40 60 80 0 Tone level (dB SLB) 80 60 40 20 0.62 20 40 60 80 dB SLB in Exp. ear 80 20 Tone level (dB SLB) 40 60 80 80 60 40 20 80 0 dB SLB in Exp. ear E. VCN all-type (on-BF) Normal Exposed 100 0 0 20 40 60 40 120 200 0 100 60 D. Unusual (on-BF) Normal Exposed 250 200 dB SLB in Exp. ear C. Locker (on-BF) 300 300 0 -20 80 80 dB SLB in Nml. ear 50 Driven rate (spikes/s) 0 Normal Exposed 400 dB SLB in Nml. ear Normal Exposed dB SLB in Nml. ear Driven rate (spikes/s) 200 B. Chopper (on-BF) 20 40 60 70 60 50 3.61 40 30 20 80 20 Tone level (dB SLB) 40 60 80 dB SLB in Exp. ear F. ANF (on-BF) 0.4 0.2 0 40 1.05 20 0 -20 0 20 40 60 80 Tone level (dB SLB) 0 20 40 60 150 100 50 0 80 dB SLB in Exp. ear 80 dB SPL in Nml. ear 0.6 60 Driven rate (sp/s) 1 0.8 ANF Normal ANF Impaired 80 dB SLB in Nml. ear Normlized driven rate 200 Normal Exposed 1.2 60 40 20 0.52 0.56 0 0 20 40 60 80 Tone Level (dB SPL) 0 20 40 60 80 dB SPL in Exp. ear Fig. 3.36. Average rate-level functions for on-BF encoding (0.2 ≤ ORB < 0.2) in the tonal pseudopopulation. The circles indicate thresholds according to a rate criterion of 2 spikes/s, which determined the minimum levels in the rate-matching curves. A. Pri/PriN units. B. Chopper units. C. Locker units. D. Unusual units. The Unusual-A and Unusual-B sub-types were treated as one group. E. VCN on-BF average RLFs. F. ANF on-BF average RLFs. See text for details. 180 In order to facilitate comparison, we re-processed the ANF data in a pseudo-population way. Because no CAP audiograms were available from those ANF experiment, we used the tone frequency and unit BF ranges of [0.5, 6]-kHz to generate the ANF tonal pseudopopulation. In the rate-matching plot in Panel F we overlaid the rate-matching curve from the original real-population processing, along with its slope, which was based on the response to a 2-kHz tone by all fibers inside an 0.4-octave BF bin centered on 2 kHz,. The agreement of the positions, shapes and slopes between the original real-population result and the re-processed pseudopopulation result was satisfactory. Apart from the noisiness of the normal PL on-BF average RLF shown in Panel A, the PL units and ANFs showed very similar RLF shapes, maximum rates, and dynamic ranges in normal hearing, which was expected based on the nearly one-spike-in-one-spike-out firing of these neurons. Comparison indicated that the post-trauma changes in VCN PL units were similar to that in the ANFs, to the degree that the average on-BF rate-level slopes in both populations of neurons became shallower. However, there was evidence that the decrease in the average slope in PL units was more dramatic. If it is assumed that the peripheral damages were similar in the exposed animals in the current study and those in the cited previous study, which is supported by the result shown in Fig. 3.17, this additional decrease in rate-level slope probably reflected morphological and synaptic degeneration in bushy cells following NIHL (e.g., Lee et al., 2003; Redd et al., 2002). The on-BF rate-matching slope in the Chopper unit type was 3.25, which was much greater than both the PL rate-matching slope and unity. In the unusual category, both the maximum rate and rate-level slope was substantially higher in the impaired population, which was reflected in 181 the very steep slope (3.95) of the corresponding rate-matching curve, and a rate catch-up at about 70 dB SPL. Although the level at which this catching-up happened was lower than the levels of loudness catch observed in psychophysical experiments, which were usually between 90 and 100 dB SPL (Zeng and Turner, 1991; Moore, 2004), the shape of the unusual-type rate-matching curve was very suggestive of recruitment in the impaired ear. The spectacular distinction between the change in average on-BF rate-level encoding after HL in ANFs and different types of VCN neurons shown above suggested that different synaptic and cellular changes occurred after the onset of the trauma. The rate responses in the PL neurons were suppressed by cochlear damage, as in ANFs, and probably also by central neuronal changes. In contrast, the post-trauma changes enhanced responsiveness in the non-primary-like types in the VCN and led to steeper individual and summed RLFs. This compensation was strong enough to override the significantly diminished spike inputs from ANFs. Panel E in Fig. 3.36 shows the weighted average RLFs for all VCN response types, including the onset units. The rates were not normalized. The same set of weights as that used in Fig. 3.34 was used here. In the normal-hearing ear, the on-BF average RLFs of the overall VCN neuronal population and the ANF population were similar in shapes, which were both sigmoid shapes in the first 30 dB above threshold followed by saturation. However, in impaired ears, VCN neurons and ANFs showed apparently different slopes in their average RLFs. As mentioned earlier, the slope of the ANF on-BF average RLF was diminished following acoustic trauma. In comparison, the VCN overall on-BF average RLF showed almost no changes in its slope after HL, which led to a rate-matching slope close to but slightly greater than unity. This slope reflected a balance between the diminished rate-responses in PL category and the enhanced responses in other neuronal 182 classes. Based on the results shown in Fig 3.35, one may reach the conclusion that for on-BF encoding, no explicit correlates of loudness recruitment can be found in the overall VCN rate response. However, if it is true that only certain response types in VCN participate in determination of loudness or generating recruitment, then it’s possible that certain unit types (e.g., choppers and unusual neurons) possess substrates of recruitment in their responses. Although it is not entirely clear whether neural substrates of loudness recruitment exist in the VCN, one thing which is clear is that changes in rate-level responses in the VCN after acoustic trauma cannot be explained solely by peripheral changes. 3.6.6.4. Average rate-level functions for all-BF encoding The average RLFs for all-BF encoding are shown in Fig. 3.37, unlike the on-BF case, each of the three VCN neuronal categories: PL, chopper, and unusual, showed rate-matching slopes greater than 1 dB/dB in their all-BF RLFs. This reflects the effect of faster-than-normal spread of excitation across ORB bins in the impaired population (further discussion in Section 3.6.8), for which the loss of fine frequency tuning was probably the major cause. The Locker neurons showed little broadening of tuning in their response areas following acoustic trauma (Figs. 3.17), and consequently, the all-ORB-range average rate activity in the impaired Lockers didn’t increase faster with level after trauma, which was reflected in the below-unity rate-matching slope. Of all the VCN response types, the PL-type units showed the strongest recruitment-like alterations, with their all-BF rate-matching slope as steep as 2.98. This was in sheer contrast to the strong anti-recruitment effect seen in the on-BF encoding of PL neurons (Fig. 3.36.A). Similar to the PL neurons, the impaired ANF population also showed a rate-matching slope above unity, although 183 A. PL (all-BF) B. Chopper (all-BF) 200 20 0 60 2.98 40 20 20 40 60 80 20 200 150 100 50 40 60 80 60 1.76 40 20 -20 0 20 40 60 80 20 Tone level (dB SLB) 40 60 80 dB SLB in Exp. ear 50 80 60 40 0.50 20 0 0 50 80 D. Unusual (all-BF) dB SLB in Nml. ear Driven rate (spikes/s) Normal Exposed 250 100 dB SLB in Exp. ear C. Locker (all-BF) 300 150 0 Driven rate (spikes/s) 0 Tone level (dB SLB) Normal Exposed dB SLB in Nml. ear 40 80 20 40 60 80 20 Tone level (dB SLB) 40 60 40 30 20 10 0 -10 80 0 20 40 60 80 70 60 50 2.21 40 30 80 40 Tone level (dB SLB) dB SLB in Exp. ear E. VCN all-type (all-BF) 90 Normal Exposed dB SLB in Nml. ear 60 Driven rate (spikes/s) Normal Exposed dB SLB in Nml. ear Driven rate (spikes/s) 80 60 80 dB SLB in Exp. ear F. ANF (all-BF) 0 -20 0 20 40 60 80 Tone level (dB SLB) 60 2.35 40 20 ANF Normal ANF Impaired 80 60 40 20 0 20 40 60 60 1.14 40 20 1.46 0 -20 80 0 20 40 60 80 Tone Level (dB SPL) dB SLB in Exp. ear 80 dB SPL in Nml. ear 0.2 80 Driven rate (sp/s) 0.4 dB SLB in Nml. ear Normlized driven rate 100 Normal Exposed 0.6 0 20 40 60 80 dB SPL in Exp. ear Fig. 3.37. Average rate-level functions for all-BF encoding (unrestricted ORB ranges) in the tonal pseudopopulation. The formats are the same as in Fig. 3.34. A. Pri/PriN units. B. Chopper units. C. Locker units. D. Unusual units. The Unusual-A and Unusual-B sub-types were treated as one group. E. VCN all-BF average RLFs, the same plot as in Fig. 3.33A. F. ANF all-BF average RLFs. The blue curve in the rate-matching axes shows the original rate-matching result from Heinz et al., (2005). The thresholds and slopes at low sensational levels are similar in this curve and the black solid curve, which is the result from a pseudopopulational re-processing. The slight deviation seen at high levels was probably caused by the different ranges of BFs used in the two different analyses. In the original processing, the included BF range was [0.25, 9.2] kHz. The one used in the pseudopopulational re-processing as [0.5, 6]. But these two curves agree reasonably well with each other and lead to identical conclusions. See text for details. 184 the slope of ANF on-BF rate matching was only about 0.55 (Fig. 3.36.F). This indicates that spread of excitation also became faster in the ANFs following trauma, which played a role in steepening the all-BF average RLF. However, the fact that ANF all-BF rate-matching slope was less than that of the VCN PL one probably reflected the fact that the post-trauma increase in the spread of excitation was greater in the VCN than in the AN fibers, which will be discussed in Section 3.6.8. Panel E shows the all-type, all-ORB grand average RLFs of the VCN populations. The slope of the grand rate-matching curve was approximately 2.35, significantly greater than the VCN on-BF and the ANF all-BF rate-matching slopes. The difference between VCN and ANF was most likely caused by the following responses: (1) the presence of a subset of VCN neurons, which showed steepened individual and on-BF average RLFs after acoustic trauma, (2) the slightly more rapid spread of excitation in the impaired VCN neuronal population, compared with the impaired ANFs (See Section 3.6.8). Therefore, if one assumes that tone levels are encoded by the summed firing rates of all neurons along the tonotopic axis (in certain neuronal type(s)), then conclusion can be reached that most VCN response types (except onset and locker neurons) exhibited recruitment-like RLF alterations after acoustic trauma. The all-BF summed RLFs in the impaired ANFs was also greater than one, but not as steep as VCN neurons following HL, again giving hints to the potentially important roles of post-trauma central neuronal changes in generating loudness recruitment. 3.6.7. Spread of excitation To analyze the effect of level increase on the excitation pattern over the entire tonotopy, equal-level contours of average normalized rates were generated and shown in Fig. 3.38.A. In the 185 normal hearing ear, maximal driven rates were seen at the central BF channel for low to medium levels; where as at high levels (> 70 dB), the rate peaks shifted to neurons with BFs slightly above tone level. The spread of excitation to neurons with BFs above the tone frequency was a gradual process. Near the threshold, a level increase of 40 dB recruits a 1.2 octave BF range above the tone frequency. In the impaired ear, spread of excitation to high-BF neurons was significantly more rapid. As can be seen in the bottom plot of Fig. 3.38, a 30 dB level increase from the threshold (~ 40 dB) recruits essentially all the BF channels above the tone frequency. However, as exposed equal-intensity contours shows, the activation of high-BF neurons was less vigorous than normal after trauma. This reflects the changes in response areas and enhancement of off-BF inhibition in certain types of units (Figs. 3.26 and 3.29). Panel B of Fig. 3.38 shows the spread-of-excitation (SOE) curves, which are the numbers of excited 0.4-octave ORB channels as functions of tone level. To rule out the possible effects by different unit type constituents in different ORB channels, the derivation of these spread-of-excitation curves was based on normalized rates. A normalized driven rate of 0.05 was used for judging excitation, corresponding to a driven rate of about 8 spikes/s in the normal PL group. The threshold elevation for the first excited ORB channel (the central one) was approximately 40 dB. However, at 70 dB, the number of excited ORB channels equaled in the two hearing states. Overall, it took about 60 dB for the number of excited channels to increase from 0 to 10 in the normal-hearing ear, whereas the same amount of excitation spreading took only about 20 dB in the impaired ear. This observation was consistent with the finding in Fig. 3.33.B that difference in minimum thresholds in different ORBs were smaller in the impaired population, which resulted from broadened tuning curves of the impaired units. This led to the conclusion that 186 Normal Exposed 1 80 0.8 B 70 0.6 20 0.4 15 40 0.2 10 30 50 0 1 -1 080 1 2 3 70 0.8 0.6 90 0.2 10 5 0 60 0.4 Normal VCN Exposed VCN Normal ANF Exposed ANF From Heinz et al, 2005 60 # of excited channels Avg normalized driven rate Avg normalized driven rate A 0 20 40 60 80 Tone level (dB SLB) 50 40 0 -1 0 1 2 BF (octaves re tone freq.) 3 Fig. 3.38. Spread of excitation in the tonal pseudopopulation. A. Equi-level rate contours for the tonal pseudopopualtion. The rates shown in these contours are average normalized driven firing rates of all VCN response types, except onset and ChL. The numbers show levels in dB SLB. B. Comparison of spread of excitation (SOE) in ANFs and VCN neurons (solid curves). The dashed curves show the SOE curves obtained from re-processing of the ANF data (the moderate/severe HL population). The threshold for excitation was 8 spikes/s in the ANF data. Note that the abscissa for AN data is in the unit of dB SPL. The gray dash-dotted curve is a re-production of the real-population SOE curve for AN fibers in the moderate/severe HL population from Fig. 7 in Heinz et al. (2005). The unit for its abscissa is dB SPL. 187 SOE was approximately 3 times as fast in the exposed VCN as in the normal one. Notice that the area below each equal-intensity contour shown in Panel A is proportional to the total driven rate. Therefore the abnormally rapid SOE would contribute significantly to the abnormally steep rate-level slopes in many VCN neuronal types and the overall VCN neuronal type for all-BF encoding of level. In Fig. 3.38.B, the SOE curves of the re-processed ANF data is also shown for comparison. Only the moderate-/severe-HL animals were included. The normal-hearing ANF and VCN SOE curves matched each other reasonably well in thresholds and steepness. The slopes of the exposed curves were also close. In the exposed ear, VCN neurons and AN fibers also exhibited SOE curves that were very similar in threshold and steepness. This agreement probably followed from the commensurate degrees of threshold elevations and broadening of tuning curves illustrated in Fig. 3.17.B and C. This observation suggested that the difference in the steepness of the all-BF average RLFs in VCN an AN after acoustic trauma as shown in Fig. 3.37.E & F was not primarily caused by different extents of tuning deterioration in the two studies. Instead, the post-trauma up-regulation of rate-level responses in the non-primary-like neurons probably played the most important roles. The AN real-population SOE curve for a 2-kHz tone in the moderate/severe HL grouping (From Fig. 7 in Heinz et al., 2005) was re-produced and shown in Panel B for comparison in Fig. B for comparison. A discrepancy was seen between this SOE curve and the pseudo-population one. The real-population SOE curve was steeper than its pseudo-population counterpart for the first 20 dB above threshold, but was significantly shallower at higher sensation levels. The cause for this discrepancy was unclear. Possible explanations include different data sets and different rate 188 criteria used in the current and the previous studies (The pseudo-population analysis included tone frequencies other than 2 kHz between 0.5 and 6 kHz). This was not a challenge to the validity of the pseudopopulation approach because as shown in Fig. 3.37.F, the AN all-BF average RLFs by the pseudopopulation method wasn’t steeper than the one generate by the real-population method at high sensation levels. A natural question that follows the observation that all-BF encoding incurred steeper rate-matching slopes than on-BF encoding did (Figs. 3.36 & 3.37) is what happens if intermediate BF window widths were used, and whether there existed any systematic relationships between the strength of VCN recruitment-like phenomena and ORB window width. Fig. 3.39.B shows the rate-matching curve slopes as function of ORB window widths for different VCN unit types. The abscissa, ORB window width, was the difference between the maximum and minimum ORBs of the ORB window centered on zero used for generating the rate-matching curves. For each ORB window width, the entire ORB range was sub-divided into 0.4-octave wide bins. For example, for a 1.2-octave ORB window with, three consecutive ORB channels, [-0.6, 0.2], [-0.2, 0.2], and [0.2, 0.6] were used. Normalized firing rates were used in generating VCN overall rate-matching curves. ANF data from the previous study (Heinz and Young, 2004, moderate/severe HL group, BF and tone frequency between 0.5 and 6 kHz) were reprocessed in a similar way and the resultant rate-matching slopes were also shown in the figure for comparison (the dashed magenta curve in Panel B). The cyan band in Fig. 3.39.B shows the range (from minimum to maximum) of psychophysical loudness-balance curves observed in human subjects with unilateral SNHL in the impaired frequency region. These data were extracted from Zeng & Turner (1991), Moore et al. (1985) and Moore (2004). Only loudness balances obtained from comparable threshold shifts 189 80 0.4 oct. 1.2 oct. 2 oct. 2.8 oct. 40 3.6 oct. 4.4 oct. 5.2 oct. 6 oct. 6.8 oct. 2.20 2.26 2.07 1.83 1.41 1.17 1.05 1.11 1.05 0 0 40 80 0 40 80 0 40 80 0 40 80 0 40 80 0 40 80 0 40 80 0 40 80 0 40 Tone level in exposed ear(dB SLB) B 4 Rate-matching slope (dB / dB) Level in normal ear A 3.5 3 2.5 2 VCN all types VCN PL VCN Ch VCN Unusual VCN Locker ANF 1.5 1 0.5 0 0 2 4 6 8 10 BF window width (oct. re tone f.) Fig. 3.39. The relationships between ORB range and the slopes of rate-matching curves. The effect of tone frequency range (in octaves re BF, ORB) on simulated loudness matching curves and the strength of the recruitment-like phenomena. Tonal pseudopopulation data (with BFs and tone frequencies below the CAP-audiogram edge frequencies) were included in the analysis. A. Rate-matching curves under different ORB window widths (shown on the upper left corner of each panel). The linear regression slopes were shown. The diagonal line indicates equality of loudness at a given level in normal and exposed ears. Tone levels were SLB-adjusted. B. Rate-matching slopes versus ORB window width. Data for Pri/PriN, Chopper, Unusual and Locker sub-populations are also shown. The cyan region band shows the range (from minimum to maximum) of loudness-matching-curve slopes in human subjects with hearing losses comparable to those of the exposed cats. The same analysis was performed on ANFs in the normal-hearing and moderate/severe HL loss pool from Heinz and Young (2004), and the results are plotted (the dashed curve) for comparison. ANF rate-level data with tone frequencies and BFs between 0.5 and 6 kHz were used. 190 80 (between 35 and 55 dB loss) were used. To visualize the change of the shape and slope of rate-matching curves with increasing ORB window width, Panel A shows the family of rate-matching curves for VCN all unit types under window widths from 0.4 to 6.8 octaves, with 0.4 octave steps. The narrowest window widths (i.e, 0.4 and 1.2 octaves) correspond to on-BF encoding, while the widest window widths (i.e. 6 and 6.8 octaves) correspond to all-BF encoding. It can be seen the level in the normal ear needed to match the average rate produced by a 80 dB tone in the impaired ear increased monotonically with widening ORB windows. Concurrently, the rate-matching slope was a monotonically increasing function of ORB window width. For 0.4-octave ORB, the slope barely reached 1.10, whereas the slope for a 6.8-octave ORB window was as steep as 2.20. When the different VCN PSTH types were analyzed individually, distinct slope-ORB-window relationships were observed. The Pri/PriN category showed a increasing slope-ORB-window function. At window widths less than 2 octaves, the rate-matching curves showed anti-recruitment. However, for window widths wider than 2 octaves, the rate-matching slopes fell into the psychophysical recruitment region. A rapid and monotonic rise of rate-matching slope with increasing ORB window width was seen. The ANFs and VCN PL neurons showed similar rate-matching slopes at narrow ORB windows. The ANF data also showed a positive relationship between the rate-matching slope and the ORB window width, although this dependence was far less strong than that seen in the PL group. At even the largest window width, the ANFs showed rate-matching slopes falling below the psychophysical range of loudness balance slopes. These strongly positive correlations between rate-matching slopes and ORB windows in the 191 PL and VCN overall categories were mainly effects of faster-than-normal spread of excitation. This point is illustrated by the curve for Locker neurons, the VCN response category that showed relatively little threshold shift and far less broadening of frequency tuning after acoustic trauma. For narrow window widths, the Locker rate-matching slopes were steeper than that of the PL category. However, rate-matching curves of the Locker neurons showed little dependence on ORB window width. The slope-versus-ORB-window-width functions of the choppers and unusual-type neurons differed in both values and shapes from that of the PL neurons. These two types of neurons showed slopes falling above the psychophysical recruitment range even at the narrowest window width. However, slight increases of ORB window width beyond 0.4 octaves brought the slopes of both types down. Rebounds were seen for further increases of ORB window widths. The decrease of rate-matching slopes at narrow ORB window widths reflected the reduced average rate-level slopes in off-BF frequency bins (For chopper, see Fig. 3.26; for unusual, see Fig. 3.29) in impaired ears. In summary of this section, there were systematic relationships between the slopes of the rate-matching curves and their resemblance to psychophysical loudness recruitment, and the ORB window widths. These relationships were dictated by two factors (1) differences in rate-level slopes between the two hearing states; (2) increase in the rapidness of spread of excitation. Where the second factor dominates, the rate-matching slopes exhibited monotonic increases with ORB window widths. This was the case for VCN PL neurons, ANFs (from the previous study) and VCN overall response types. In other types, including Chopper and Unusual neuron, the first factor caused the shapes of the slope-ORB-window curves to be non-monotonic. But both these two 192 types showed rate-matching analysis results consistent with recruitment at both very narrow and very wide ORB windows. When considered as a whole, the VCN overall neuronal type showed no recruitment-like changes at narrow ORB window widths, and showed recruitment-like phenomena at wide ORB window widths (> 4 octaves) (The thick curve in Fig. 3.38.B). 3.6.8. The variability of discharge rates in response to tones As has been discussed in Introduction Section 1.2.1, the increased internal noise, i.e., larger variability of neural responses, has often been hypothesized as an explanation for the close-to-normal intensity discrimination limens in the face of steeper loudness-intensity relationships in human patients with SNHL (e.g., Florentine et al., 1993; Neely & Allen, 1997). To our knowledge, the existence of the physiological correlates of this increased internal noise hasn’t been found to date. In the previous AN study, it was shown that after acoustic trauma, AN fibers show rate variances indistinguishable from those in the unexposed ears (Heinz et al., 2005b). Nonetheless, it is still possible that increases in response variability result mainly from alterations in central auditory nuclei following trauma. Based on a similar analytic method used in Heinz et al. (2005b), we studied the effect of acoustic trauma on the variance-mean relationships in different types of VCN neurons. Iterations were made through each level in each unsmoothed tonal RLF. For a level showing regression slope below a certain slope threshold in its 5-dB neighborhood, the standard deviation of the rates in this neighborhood was taken. The slope threshold θ S was determined on a unit-type basis according to the following equation, θS = 0.0125 × DFRNorm , Eq. (3.2) 40dB in which DFRNorm was the average driven rate of threshold-aligned BF-tone RLFs at 40 dB re 193 threshold in the sub-edge frequency region in normal-hearing ears (See Fig. 3.23, Column 5). When applied to normal AN fibers, which show DFRNorm of about 160 spikes/s (See Fig. 3.35.F), Eq. (3.2) yielded a slope threshold of 2 spikes/s/dB, which equaled the slope criterion used in Heinz et al. (2005b). In this analysis, only firing rates recorded at levels above the response thresholds were included. Each panel in Figs. 3.40 – 3.41 corresponds to a combination of VCN neuronal type and BF/TF region. The dashed black curve shows the prediction from a Poisson-process model without refractoriness, viz. 1 σ (rate) = ⎡⎣ rate ⎤⎦ 2 , (3.3). The dashed blue and red curves show running averages of S.D. versus its mean. Power functions of the form, p σ (rate) = ⎡⎣ rate ⎤⎦ , (3.4) was fit to the normal and impaired data sets in a least-square sense. The solid blue and rate curves show the fits. For each population, a 20000-time bootstrapping (Efron and Tibshirani, 1993) was performed to generate a bootstrap sample of the best-fitting powers. The distributions of these samples were shown on the right plots in Figs. 3.40 and 3.41. Based on the assumption that the distributions of the normal-hearing and impaired populations are approximately normal and independent, the one-tailed p-value for testing under the Neyman-Pearson paradigm the hypothesis that the power p was greater in the exposed population was calculated and shown. In normal-hearing ears, the variability of PL (Fig. 3.40.A) and chopper (Fig. 3.40.B) neurons are less than both that predicted by a Poisson model and that of an average ANF. The powers for these two types of neurons were approximately 0.32 and 0.26, as compared to 0.37 in ANFs. It can 194 A 18 Standard deviation of spike count 16 14 12 10 Prob. frequency 0.08 Poisson model ANF fit Normal avg. Exposed avg. Normal fit Exposed fit 8 p = 0.0000 0.06 0.04 0.02 0 6 0.3 0.32 0.34 Best-fitting power 4 2 0 0 20 40 60 80 Mean spike count (in 200 ms) B 14 Prob. frequency 0.08 Standard deviation of spike count 12 10 8 0.06 0.04 0.02 0 6 4 p = 0.0000 0.26 0.28 0.3 0.32 Best-fitting power 2 0 0 20 40 60 80 100 Mean spike count (in 200 ms) 120 Fig. 3.40. The relations between variances and means of discharge rates for tonal stimuli, I. A. Pri/PriN units, BFs and tone frequencies below the CAP-audiogram edge frequencies. B. ChS/ChT units, the same frequency region as A. Data were taken from 5-dB sections of tonal RLFs with average slopes < 0.0125 * (normalizing rate) / dB (See text for details). The no-refraction Poisson model (dashed black curve) and previous AN results (gray dashed curve, from Heinz et al., 2003) are shown for comparison. The running averages and best power-function fits (least square) are shown by the dashed and solid curves (Blue: unexposed, Red: exposed). 20000-time bootstraps were done on the data sets and the distributions of the best-fitting powers are shown in the plot on the right. The p-value on the upper left corner is for one-sided testing of the hypothesis that the best-fitting power is greater in the exposed population. 195 A 0.08 Prob. frequency Standard deviation of spike count 25 20 15 0.06 0.04 0.02 0 10 p = 0.0000 0.15 0.2 0.25 Best-fitting power 5 0 0 20 40 60 80 100 Mean spike count (in 200 ms) 120 B 0.08 Prob. frequency Standard deviation of spike count 15 10 p = 0.7208 0.06 0.04 0.02 0 0.26 5 0 0 20 40 60 Mean spike count (in 200 ms) 0.28 0.3 Best-fitting power 80 Fig. 3.41. The relations between variances and means of discharge rates for tonal stimuli, II. The same format as Fig. 3.39. A. All unit types, BFs and tone frequencies below the CAP-audiogram edges. B. All unit types, BFs and tone frequencies above the CAP-audiogram edges. 196 be seen that the chopper units showed less variance in their rates than the Primary-like neurons did. However, the impaired (sub-edge) PL and Chopper units showed significantly higher powers (0.34 for impaired PL units and 0.30 for impaired chopper units. This effect on rate variability by acoustic trauma was significant (p ≈ 0), as shown by the almost non-overlapping bootstrap samples. The moving-window averages showed consistent results. Fig. 3.41.A shows the comparison of variance-mean power relationships for all VCN neuronal types in the sub-edge frequency regions. The best-fitting power value of normal VCN neurons in this frequency range was very low, mainly because of the locker units, which entrained onto low-frequency tones and showed reproducible rates across repetitions. However, the best-fitting power for impaired VCN neurons was significantly larger than the normal one, showing that trauma increased response variability in the overall VCN neuronal types. This trend was also manifested by the moving-window average curves. In contrast, the power fits in the supra-edge frequency regions were overlapping in the two hearing states (Fig. 3.41.B), which suggested that the increase in rate uncertainty was restricted to the BF areas with substantial threshold elevations. 3.7. Analyses on broadband-noise rate-level functions In addition to tones, broadband noise was also used in the current study. We paid special attention to responses to tones because they are the most frequently used stimuli in previous psychophysical studies on recruitment. Since many behaviorally relevant sounds (such as speech) are broadband in nature, it is also important to study the changes in the level responses to broadband stimuli after acoustic trauma. Panel A in Fig. 3.42 show the BBN thresholds of single units plotted against BF in octaves 197 B Average driven rate (spikes/s) 0 Threshold (dB NC) -20 -40 -60 -80 -6 -4 -2 0 BF (Octaves re edge) 2 Lv. in nml. ear (dB NC SLB) A 150 100 50 0 -100 -50 0 0 -20 -40 1.68 -60 -60 BBN level (dB NC SLB) -40 -20 0 Lv. in exp. ear (dB NC SLB) 150 100 50 0 300 200 100 0 -100 -50 0 Level (dB NC SLB) p = 0.231 0 Nml. 6 4 2 0 50 Nml. Exp. 12 10 8 6 4 2 Exp. 198 0 100 0 150 2.452 100 50 0 60 2.355 40 20 0 150 100 50 0 0.477 Avg. rate (sp/s) 0 -10 0 5 BBN CS-15 (sp/s) 150 100 50 0 10 0 0 10 20 BBN CS-15 (sp/s) 10 -50 0 50 Level (dB re Thresh) 60 40 20 0 -60-40-20 0 20 Level (dB re Thresh) 200 -100 -50 0 Level (dB NC SLB) 150 100 50 0 r2=0.009 p=0.782 5 0 0 10 BBN CS-15 (sp/s) r2=0.045 p=0.469 20 10 0 -5 0 5 BBN CS-15 (sp/s) 20 -40-20 0 20 40 Level (dB re Thresh) r2=0.383 p=0.003 20 30 -100-80-60-40-20 Level (dB NC SLB) 200 BF CS-15 (sp/s) -60-40-20 0 20 40 Level (dB re Thresh) -100 -50 0 Level (dB NC SLB) 80 BF CS-15 (sp/s) 100 200 r2=0.200 p=0.025 10 30 Avg. rate (sp/s) 2.515 -60-40-20 0 2040 Level (dB re Thresh) -100 -50 0 Level (dB NC SLB) Exp. p = 0.373 Nml. 200 Exp. p = 0.620 Nml. Avg. rate (sp/s) CS-15 (sp/s/dB) 5 Nml. Exp. p = 0.769 p = 0.145 0 -100 -50 0 Level (dB NC SLB) Exp. 10 Nml. Exp. 50 0 0 50 BF CS-15 (sp/s) 50 0 5 100 BF CS-15 (sp/s) -100 -50 0 Level (dB NC SLB) 10 15 0 150 20 BF CS-15 (sp/s) 0 p = 0.382 p = 0.205 Nml. 50 Exp. 15 Nml. Exp. 100 Avg. rate (sp/s) 100 20 150 200 Avg. rate (sp/s) 200 50 0 Nml. 0.463 Avg. rate (sp/s) -100 -50 0 Level (dB NC SLB) 2 Avg. rate (sp/s) 0 4 200 Avg. rate (sp/s) 200 100 6 Nml. Exp. p = 0.143 p = 0.322 Avg. rate (sp/s) 300 8 Avg. rate (sp/s) Driven rate (spike/s) 0 CS-15 (sp/s/dB) 100 -100 -100 -50 0 Level (dB NC SLB) CS-15 (sp/s/dB) 0 50 CS-15 (sp/s/dB) -50 -100 -50 0 Level (dB NC SLB) 0 p = 0.705 CS-15 (sp/s/dB) Pct. nonmonotonic Pct. nonmonotonic Pct. nonmonotonic Pct. nonmonotonic L c k r 100 Pct. nonmonotonic U n sl 100 100 Driven rate (spike/s) O n 100 200 300 Driven rate (spike/s) C h 100 300 400 Driven rate (spike/s) P L Driven rate (spike/s) C 15 r2=0.446 p=0.018 10 5 0 0 5 10 BBN CS-15 (sp/s) Fig. 3.42. Analysis on broadband noise rate-level functions. A. Threshold-versus-BF plot. BFs were plotted in octaves re CAP-audiogram edges (ORE). B. The grand average RLF for BBN of all VCN PSTH types. In generating thesis average RLFs, weighted average (same as in Fig. 3.33, see Tab. 2.2) was performed on different PSTH types. The right panels shows the corresponding rate-matching curve and its slope in dB / dB. C. Analysis on BBN responses by PSTH types. Only the unit with BFs below the CAP-audiogram edge frequencies were included. Onset neurons were not shown due to the lack of onset data in the sub-edge region. The format is similar to that of Fig. 3.23. Column 1: population summary of BBN RLFs. Column 2: Percentage of non-monotonic RLFs. Column 3: 15-dB chord slopes (CS-15). Column 4: average BBN RLFs. The noise level was expressed in a dB scale corrected for different speaker calibrations with SLB adjustment. The blue and red circles indicate thresholds according to a 3 spikes/s rate criterion. Rate-matching curves were produced for each pair of average BBN RLFs and its slope (in dB / dB) is shown in the panel. Column 5: averages of threshold-aligned BBN RLFs. Column 6: correlations between CS-15 for BBN and CS-15 for BF tones. Linear regression was performed on the data. The correlation coefficients and p-values are shown on the upper left corner of each plot. 199 re edge (ORE) in different VCN unit types and the two different hearing states. Noise levels (NC) in this figure are adjusted for differences in amplitude calibrations of the acoustic drivers in different experiments (Section 2.6.5, Equation 2.5). It can be seen that in the impaired population, all of the units with sub-edge BFs showed substantial (25 – 50 dB) BBN threshold elevation, whereas the supra-edge BF region showed a mixture of normal and elevated thresholds. Subsequent analysis of BBN RLFs will be concentrated on data in the sub-edge regions, in both populations. The layout of Fig. 3.42.C is similar to that of Fig. 3.23. SLB correction was performed on the corrected BBN levels (yielding a dB scale called NC SLB) to account for the variability of thresholds across different BFs and different animals in each population. Each row corresponded to a specific unit category. Compared with BF-tone RLFs, RLFs for BBN showed less prevalence of non-monotonic shapes. However, the unusual-type neurons showed a higher percentage of non-monotonic RLFs after hearing loss. As can be seen in Row 5 Column 1, BBN RLFs of many impaired unusual-type neurons were purely inhibitory or inhibitory near threshold and excitatory at higher levels. For most of the individual neurons, the firing-rate dynamic ranges for BBN were greater than the BF-tone dynamic ranges. The PL, Locker and Unusual unit category showed no significant changes in BBN rate-level slopes after acoustic trauma. The choppers and onset units were the VCN neuronal category which exhibited a steepened BBN RLFs in the exposed ear. As for BF-tones, the average RLFs for BBN were less steep in the impaired ear in the PL group (Row 1, Column 4). However, when the RLFs were processed on a dB-re-threshold scale 200 (Row 1, Column 5), the steepness of the average RLFs were similar between the two populations. This suggested that for given BF-tone thresholds, the distribution of BBN thresholds were more dispersed in the exposed ear. The Locker neurons (Row 3) showed close-to-normal BBN average rate-level slope the impaired ear, which was consistent with their close-to-normal BF-tone average rate-level slope after hearing loss (Fig. 3.23). The average BBN RLFs in the chopper and onset categories exhibited recruitment-like behavior: a significantly steeper rate-level slope was seen in the impaired ear; the average driven rate in the impaired ear caught up with that in the normal ear at certain high level; the average threshold-aligned RLF was also apparently significantly steeper in the exposed population. Similar trends of changes were seen in the impaired unusual-type neurons. However, one unique and surprising property of the Unusual neuron was their negative driven rates at levels near the thresholds. Despite these inhibitory responses, the strength of excitation at higher levels could still produce rate catching-ups at high levels. As can be seen in Column 6, the chord slopes of BBNs RLFs were generally shallower than the BF-tone rate-level slopes in most VCN response types. This was most likely due to the effect of two-tone rate suppression and the neural inhibitions in off-BF frequency bands. Linear regressions between the BF-tone and BBN rate-level slopes were performed. The correlations between these two types of slopes were positive in all the three usual VCN response types, although in two of the cases, correlation didn’t reach significance because of dispersion of the data. The correlation in the unusual group was negative. This was mainly caused by an unusual (07-003, Unit 2.07. See Fig. 3.21.E) which exhibited a very steep BF-tone RLF and a negative BBN 15-dB chord slope. If this unit was taken out of the calculation, the correlation would be 201 also positive, as in other VCN neuronal categories. To study the overall effect of acoustic trauma on the rate-level encoding of BBN level in the VCN neurons, we weighted averaged the average BBN RLFs of different PSTH types according to the weights shown in Tab. 2.2 to produce the grand average BBN RLF (Panel B). The growth of driven rate above threshold was slightly more rapid in the impaired ear, as indicate by the rate-matching slope of 1.68, which fell into the psychophysical range as shown in 3.39.B. Less rate saturation was seen after trauma; and the maximum driven rate was greater in the impaired ear. These findings indicate that in VCN after acoustic trauma, recruitment-like phenomena existed for broad-band stimuli as well. However, it is noteworthy that the rate-matching slope for BBN was slightly less than the rate-matching slope generated from the grand average tonal RLFs (Figs. 3.35.A and B). This was presumably due to the fact that spread-of-excitation played no role in level encoding of BBN due to the flat spectrum of the stimulus. Based on this observation, a prediction can be made that if level of a tone is encoded in a all-BF way, loudness recruitment in an impaired ear should be stronger for a tone than for a BBN; whereas if a tone is encoded in an on-BF way, then loudness recruitment for BBN should be as strong or even stronger than the recruitment of a tone. Following this line of reasoning, psychophysical experiments in recruiting ears can be designed to answer the question of in which way is the intensity of a tone encoded. 3.8. Rate-level and rate-spectral encoding of the vowel /ε/ Although most published studies on loudness recruitment used pure-tone stimuli, the stimuli that are most relevant to the everyday life of patients suffering from SNHL and recruitment are speech sounds. Many SNHL patients complain of their reduced intensity dynamic range for 202 speech perception: the talker has to speak at a louder-than-normal level to make the voice audible and recognizable, but he or she also has to be careful not to raise the level too much to make the speech sounds painfully loud (Huizing, 1948). In this section, we will show the response of VCN neurons to the vowel /ε/ in normal and acoustically traumatized hearing, and the abnormality in the encoding of vowel spectrum and vowel level in the impaired ear. The vowel /ε/ (as in “met”), whose amplitude spectrum is shown in Fig. 3.43.B, was digitally synthesized. The spectral manipulation procedure (SMP) was used to construct a pseudopopulation of neurons with effective BFs located at the feature frequencies (Le Prell et al., 1997. See section 2.5.2 for details). In noise-exposed animals, only units with BFs below the CAP-audiogram edge frequencies and with substantial threshold shifts were used in the SMP; in the normal-hearing animals, the SMP method was performed only on the units with BFs below 10 kHz. Panel A in Fig. 3.43 shows a summary of the units from which the SMP vowel data were obtained. The relatively small amount of data didn’t allow an analysis by neuronal type. Therefore, we used normalized driven firing rates, and treated the data from different unit types indifferently. For each neuronal category, the normalizing rate was the average driven rate in response to broad-band noises at 30 dB above threshold in the normal-hearing population (threshold-aligned average, See Fig. 3.42, Column 5). Vowel levels were adjusted by the SLB method to correct for the variability of single-unit thresholds across different animals and different BF regions. Row C in Fig. 3.43 shows the individual normalized RLFs (NRLFs) for the six feature alignments. The numbers shown at the upper left corner of each panels are the number of 203 Unit threshold (dB SLB) A 100 80 60 40 20 0 Relative level (dB) 0.1 Normalized driven rate B E F 10 0 -20 -40 0.1 F1 1 Frequency (kHz) T1 3 9 /28 T2 F2 9 /25 F3 9 /21 9 /24 T3 8 /18 8 /17 2 1 0 0 100 50 100 0 50 100 0 50 100 0 50 100 0 50 100 0 50 100 Vowel level (dB SLB) p=0.440 p=0.014 p=0.034 p=0.005 p=0.072 p=0.231 50 0 0.1 Nml. Exp. 0.058 0.054 Nml. Exp. Nml. Exp. Nml. Exp. Nml. Exp. Nml. Exp. 0.034 0.036 0.052 0.047 0.037 0.043 0.040 0.045 0.025 0.043 p=0.861 p=0.745 p=0.769 p=0.820 p=0.370 0.05 Avg. normalized driven rate D -1 Normalized CS-15 (dBPct. ) nonmonotonic C 1 Unit BF 0 p=0.778 0 50 1 100 0 2.27 50 100 0 50 100 0 50 Threshold (vowel level dB SLB) 1.15 100 0 2.04 1.04 50 100 0 1.82 50 100 50 100 2.36 0.5 0 0 50 100 0 50 100 0 50 100 0 Vowel level (dB SLB) 204 50 100 0 50 100 0 Fig. 3.43. Rate-level encoding of the spectrum-manipulated steady-state vowel /ε/. Levels were SLB-adjusted. Driven firing rates were normalized by unit-type-specific normalizing rates obtained from average driven rates for BBN (See Fig. 3.42. See text for details). Only data from real neurons with BF below the CAP-audiogram edge frequencies were included in these analysis. A. A summary of real-neuron BF and thresholds. B. Relative spectrum of the (unshifted) steady-state vowel /ε/. The solid and dashed vertical lines indicate formant and trough frequencies, respectively. C. Summary plot of individual SMP rate-level functions. D. Percentage of nonmonotonic RLFs and their comparison. The p-values are from two-tailed Fisher’s exact test. E. Normalized CS-15 versus thresholds in vowel levels. ANOVA indicates no significant effect by hearing condition (df = 1, F = 0.12, p = 0.7268). F. Average normalized rate-level functions for the six feature alignments. The circles indicate thresholds according to a normalized-rate criterion of 0.015. The slopes of the rate-matching curves (not shown) are shown in the upper left corners. 205 individual RLFs from the two hearing conditions. The normalized 15-dB chord slopes are plotted against vowel-level thresholds Row E. These slopes were the differences between normalized driven rates at 0 and 15 dB re threshold divided by 15 dB, and had units of (dB-1). As expected, all impaired units had substantially elevated thresholds. As the mean slopes for the two hearing states shown on the upper left corners of the panels show, rate-level slopes were generally similar between the two hearing states. T-tests indicated no significance of rate-level slope changes after hearing loss in any of the feature alignments. However, comparison inside individual PSTH categories did reveal some interesting post-trauma changes. While PL and Locker neurons tended to show shallower rate-level slopes for all feature alignments in the impaired ear, the chopper and onset units exhibited the converse. As can be noticed Column 1 in Row E, three impaired Chopper neurons showed very steep slopes (> 0.1 dB-1). Similar effects can be seen in the onset units for other feature alignments. The mixture of steeper and shallower post-trauma rate-level slopes in different PSTH types led the ANOVA to indicate no significant effects of acoustic trauma on the slope distributions (df = 1; F = 0.12; p = 0.7268). Furthermore, rate-level slopes didn’t tell the whole story about RLF changes after hearing loss. The percentage of non-monotonic RLFs decreased after HL (Row D). Row F in Fig. 3.43 shows the average NRLFs for the six alignments, which summarizes how the different factors, including RLF slope, shapes and threshold distributions were integrated to determined the summed rate activities in the VCN neurons as a code for vowel level encoding. In the normal-hearing population, responses to the first three formants (F1-3) all showed rate saturation at medium-to-high vowel levels. In contrast, in the impaired population, average NRLFs for all the formants and features exhibited monotonically increasing shapes. 206 Rate-matching curves were generated for the pair of average NRLFs for each feature alignment, and the resultant rate-matching slopes were shown on the upper left corner of each panel in Row F. Rate-matching slopes above 1 were seen for all feature alignments. Furthermore, there were rate catching ups at 70 – 80 dB SLB vowel level, the level at which the normal average NRLFs entered saturation. To better illustrate the relationships between rate-responses to different feature alignments, the average NRLFs shown in Row F were presented in a color-coded form in Fig. 3.44.A. If it is assumed that as in all-BF encoding of tone intensities, the level of the steady-state vowel is encoded by the total firing rate of the entire population of VCN neurons, then the factors that determine the loudness-level relationship are not only the steepness of rate-level growths for the individual vowel features, but also the dispersion of the response thresholds for different feature alignments. The average normalized driven rates were shown by color code and plotted versus feature alignment (x axis) and the vowel level (y axis). As can be seen in the normal plot (left), the threshold profiles approximately reflected the spectral shape of the vowel. Formant 1 was the strongest feature and it as associated with the lowest response threshold. The difference between peak amplitudes of F1 and F2 was about 15 dB, and an approximately 15-dB threshold difference was seen between F1 and F2. The same was true for other features. However, in the impaired population (the right graph in A), the threshold-feature relationship was lost. The difference between thresholds for the three formants differed less than 10 dB, and the thresholds for troughs were about equal to those for formants. The major cause of this abnormal threshold-feature profile was the reduction of frequency selectivity after and the loss of fine tips in the tuning curves after hearing loss, the same factor that caused abnormally rapid spread of 207 80 80 70 70 Vowel level (dB SLB) Vowel level (dB SLB) A 60 50 40 30 20 10 60 50 40 30 20 10 F1 T1 F2 T2 F3 T3 Avg. normalized driven rate 0 0.5 F1 T1 F2 T2 F3 T3 Avg. normalized driven rate 1 0 0.5 1 Avg. normalized driven rate B 70 dB 75 dB 60 dB 70 dB 50 dB 65 dB 40 dB 30 dB 60 dB 20 dB 10 dB 0.2 0 dB SLB F1 55 dB 0.2 50 dB SLB F2 F3 F1 1 Frequency (kHz) F2 F3 1 Frequency (kHz) Fig. 3.44. Rate-place profiles for the spectrum-manipulated steady state vowel /ε/. Data were obtained in the same ways as the previous figure. A. Average normalized firing rates (color codes) for the six feature alignments at corresponding vowel levels. The abscissas are feature alignments; the ordinate are vowel level. The dashed red curves in each column indicate the threshold level by a rate criterion of 0.075 for the feature alignments. B. SMP rate-place profiles at different vowel levels for the normal-hearing (Blue) and noise-exposed pools (Red). The error bars show ±1 S.D. These are horizontal cross-sections over plots in Panel A. Each curve corresponds to the average over a 6-dB range centered at a specific level as shown by the axis on the left. Note that the zero rate positions for different curves are different. 208 excitation with increase tone level in the tonal RLF analyses. In other words, for a given vowel spectrum, the differences between thresholds for BF-encoding of different formants and troughs were smaller after acoustic trauma, which as will be shown shortly, contributed to a more rapid increase of total driven rate in the impaired ear, i.e., a recruitment-like phenomenon. In addition to the abnormal thresholds in the impaired pool, the supra-threshold rate-feature profiles lost their faithful representation of the spectral feature of the vowel. Panel B in Fig. 3.44 shows the average NRLF data plotted in the format of rate-place (rate-feature-alignment) profiles at fixed vowel levels. Each curve shows the relationship between the pseudopopulation averaged normalized driven rate and BF (or feature alignment) at a specific vowel level, as indicated by the corresponding axis on the left. In practice, generation of each curve involved average over a 6-dB level range centered on the nominal level in order to suppress noise in the data. Notice the zero rates for different vowel levels were located at different positions along the ordinate. Faithful rate representation requires agreement of orders of feature level and responding driven rates at the corresponding BF-feature alignment, which was the case in the normal hearing population over a wide range of levels (from 10 – 70 dB SLB). However, as shown in the left panel of B, the rate-feature profile in the impaired ear differed greatly from the normal one. The responses to F2 were often indistinguishable from the responses to T1, which makes the F1-T1-F2 contrast unseen in most levels above threshold. In this abnormal representation of vowel spectrum, information regarding the frequency of F2 is largely lost, which would seriously compromise the performance of vowel perception. Despite the abnormal drop in responses to F2, response to F1 was abnormally strong, especially at high vowel intensities. The most salient change that happened in the impaired ear was the abnormally rapid increase of average rate for 209 F1. Notice that the level spacing in the normal (left) plot is 10 dB, whereas the spacing in the exposed (right) one is 5 dB. Nevertheless, the F1 average rate at about 25 dB above threshold in the exposed ear was as strong as the F1 average rate at 70 dB above threshold. Therefore, one can conclude that after acoustic trauma, population rate responses to the vowel stimulus became disproportionate for different spectral feature. Growths in responses to major and strong spectral features, such as F1 in the vowel, were significantly faster than normal. The response to minor spectral features became relatively weaker after HL, and their increases with level became less rapid, compared to the growth at F1. Interestingly, these observations were consistent with results from a behavioral study currently ongoing in the Neural Encoding Laboratory on cats deafened with a similar exposure paradigm as the one used in this thesis, which suggested that the deafened cats showed loudness recruitment at F1 (B.J. May, personal communication). These spectral encoding abnormalities were not only seen for population average, but also for individual neurons in the SMP vowel data pool, as shown by the exemplar data in Fig. 3.45. Previous studies have shown that ANFs and VCN PL neurons afforded satisfactory rate encoding of vowel spectra only at low-to-medium levels, over a relatively small dynamic range (Sachs and Young; 1979, Blackburn and Sachs, 1990; May et al., 1998). Rate saturation at medium-to-high levels destroys the rate-feature relationships. In comparison, the VCN chopper population has been shown to be a level-tolerant encoder of vowel spectra, maintaining faithful rate-place representations even at very high levels (Blackburn and Sachs, 1990; May et al., 1998). Panels A shows the rate profiles of a PriN unit over a 70-dB level range. Satisfactory rate spectral encoding was seen only for levels below 50 dB SLB; at higher levels, the F1-T1-F2 contrast 210 A. 06-048, Unit 7.10, 1.46-kHz PriN SR = 23.5 spikes/s B. 06-079, Unit 1.04, 9.19-kHz ChS SR = 0.0 spikes/s 50 sp/s 50 sp/s 80 dB 70 dB 80 dB 60 dB 70 dB 50 dB 60 dB 40 dB 50 dB 30 dB 40 dB 20 dB 30 dB 10 dB 20 dB 0 dB SLBBF F1 10 dB SLBBF F1 F2 F3 1 Frequency (kHz) F2 F3 1 Frequency (kHz) C. 06-047, Unit 1.04, 6.21-kHz PriN SR = 25.3 spikes/s D. 06-047, Unit 1.03, 4.49-kHz ChS SR = 48.7 spikes/s 50 sp/s 80 dB 50 sp/s 75 dB 85 dB 80 dB 70 dB 75 dB 65 dB 70 dB 60 dB 65 dB 60 dB 55 dB 55 dB 50 dB SLBBF F1 50 dB SLBBF F1 F2 F3 1 Frequency (kHz) F2 F3 1 Frequency (kHz) Fig. 3.45. Example rate-place profiles for the spectrum-manipulated steady-state vowel /ε/ based on single neurons. Unnormalized driven rates are shown. Vowel levels are SLB corrected. A. A PriN unit from a normal hearing ear. At moderate levels, rate-BF profiles faithfully reflected spectral shape of the vowel. At higher levels, rate saturation and degradation of rate-BF representation is seen. B. A ChS unit from a normal-hearing ear. Rate saturation and deterioration of rate spectral representation was nonexist in this neuron. C. A PriN unit from an impaired ear. Rate-BF contrasts are low even near threshold. D. A ChS unit from an impaired ear. As in C, rate-BF profile encodes the vowel spectrum poorly. Saturation is seen at levels > 80 dB. 211 became weaker, and highest driven rates drifted toward higher formants. By using an SMP method similar to the one used in the current study, May et al. (1998) showed that low-/medium-SR PL units afforded better rate-place encoding of the vowel spectrum of /ε/ than high-SR PL units in normal-hearing cats. However, similar analysis was not possible based on our data due to the very small number of Pri neurons recorded from in the unexposed animals. The unexposed Chopper unit shown in Panel B exhibited satisfactory rate spectral representation at levels from 30 to 80 dB SPL, inside which he F1-T1-F2 contrast stay essentially unchanged. In fact, even at levels as high as 70 – 80 dB SPL, this profile showed little tendency to deteriorate. These observations were consistent with published findings and confirmed the validity of the SMP. Some other types of recorded VCN neurons, such as a few Unusual-type neurons in the unexposed ears also show dynamic ranges greater than those of the VCN PL neurons (not shown). Panels C and D in Fig. 3.45 show the rate-spectral encoding by an impaired PriN unit and an impaired Chopper unit. In neither unit was there a clear rate contrast between the first two formants. Instead, responses were dominated by firing near F1. As mentioned early, response to F1 increased abnormally fast with vowel level in both units (notice that the level spacing in Panels C and D were only half the spacing in Panels A and B). To generate the overall summed rate-level functions for the pseudopopulation of vowel-encoding neurons, we weighted different feature alignments by the frequency bandwidths (measured in octaves) of the features. For example, F2 of the vowel /ε/ was centered at 1.7 kHz; and the center frequencies of T1 and T2 were 1.2 and 2.2 kHz, respectively. Thus the lower and upper boundaries of the formant were taken as the geometric mean between 1.2 and 1.7 kHz, and 212 A Lv in normal ear (dB SLB) Avg. normalized rate 1 Normal Exposed 0.8 0.6 0.4 0.2 0 -20 0 20 40 60 80 Vowel level (dB SLB) 80 60 40 2.04 20 20 40 60 80 Lv in impaired ear (dB SLB) B Avg. driven rate 1 Lv in normal ear (dB SLB) Normal Exposed 1.2 0.8 0.6 0.4 0.2 0 0 50 100 80 60 40 2.27 20 20 Vowel level (dB SLB) 40 60 80 Lv in impaired ear (dB SLB) Fig. 3.46. Average normalized rate-level functions for the spectrum-manipulated steady-state vowel /ε/. A. The grand average normalized RLFs for all six features in the SMP (F1-3, T1-3).The weights for the features were based on their frequency bandwidths (See text for details). B. The average normalized RLFs for the first formant (F1) only. The corresponding rate-matching curve is shown on the right. 213 between 1.7 and 2.2 kHz, respectively. From these boundary frequencies, the bandwidth of F2 could be calculated. Fig. 3.46.A shows the grand RLFs for /ε/ under the two hearing conditions. The normal curve showed a saturation at levels > 50 dB. No rate saturation was seen for the impaired curve. A rate matching between these two curves yielded a slope of 2.04, which was greater than unity and consistent with faster loudness growth in the impaired ear, and also fell into the psychophysical range of slopes (for tones) in recruiting ears (Fig. 3.39.B). Fig. 3.46.B shows the average NRLFs for F1 and the corresponding rate-matching curve. As mentioned before, the growth of response to F1 was abnormally rapid in impaired ears and devoid of any saturation. A rate-catching up was seen at about 80 dB SLB. The rate-matching curve had a slope of 2.27, falling well inside the psychophysical recruitment range (for tones). The rate-matching curves for F1-only and all-feature cases were different, with the F1-only slope being steeper. The reason for this difference was the weakened responses to higher-frequency features in the impaired ear. The weighted average of the six rate-matching slopes for the six feature alignments (Fig. 3.43.F) was 1.78, which was below 2.04, the rate-matching slope of the overall population response. This discrepancy was mainly caused by the fact that thresholds for different feature alignments were more compactly distributed in exposed ears (Fig. 3.44.A). 3.9. Analysis based on the linear-nonlinear weight model and the random spectral shape stimuli Previous studies have applied the linear-nonlinear weighting model (LNWM) (See Method section 2.7) to study the encoding of spectral shapes by discharge rates in AN fibers (Young and 214 Calhoun, 2005), CN neurons (e.g., Yu and Young 2000, Bandyopadhyay, 2007) and auditory cortical neurons (Barbour & Wang, 2003). The RSS stimuli are the standard way of obtaining the parameters (weights) of this model. In normal-hearing ears, this model has been shown to work very well in modeling the rate-spectral encoding of AN fibers and principal neurons in AVCN. The predicted response rates to novel stimuli (e.g., HRTF filtered noise) based on the derived weight functions agreed well with actual firing rates (Young and Calhoun, 2005, Yu and Young 2000). However, in DCN principal and cortical neurons, the modeling and prediction performances were shown to be relatively worse (Yu and Young, 2000; Bandyopadhyay, 2007; Barbour & Wang, 2003). Further nonlinearity beyond the quadratic terms, such as the level-dependent weighting model (LDWM, Bandyopadhyay, 2007), need to be added to allow better model performances. In the current study, responses to RSS stimuli were recorded in different types of VCN neurons in both normal-hearing and noise-exposed ears. In the LNWM, each set of weights applies to only one particular overall stimulus level, because this model is essentially a small-signal analysis method. Whenever possible, RSS stimuli were presented at multiple levels above the threshold to allow analysis of the dependence of the rate-spectral encoding on overall level. To capture the often wider weight functions in impaired neurons, the RSS-C stimulus set, which samples the response over a wider frequency range, was used in experiments on impaired animals. Fig. 3.47.A and B show the representative linear and quadratic weight functions from a PL and a Chopper neuron from the unexposed population. The basic characterizations including PST histograms for 30-dB-re-threshold BF tones and RLFs for BF tones and BBN were shown. The 215 1000 0 0 0.05 Time (s) 0.1 24.5dB 400 200 1 sp/s/dB -0.6 -15.5dB -0.2 0 Oct re BF Oct. re BF Oct. re BF -5.5dB -50 Sound level (dB attn.) 0.5 24.5dB 0 0.05 14.5dB 0 0 -0.02 -0.5 -0.05 0.5 4.5dB 0 -5.5dB 0.05 0 -0.5 -0.04 0.1 0 -0.05 0.5 -15.5dB 0 -0.5 0.4 0.04 0.02 0 14.5dB 4.5dB BF BBN 0 -100 Oct. re BF 06-048 (normal), T7U10, RSS_B PriN, 1.4565 kHz, 9.44 dB SPL Firing rate (sp/s) Firing rate (sp/s) A 2000 -0.1 0.05 0 -0.05 -0.5 0 0.5 Oct. re BF 06-079 (normal), T1U4, RSS_B ChS, 9.185 kHz, 15.9 dB SPL 1000 0 0 0.05 Time (s) Firing rate (sp/s) Firing rate (sp/s) B 2000 0.1 400 200 BF BBN 0 -100 -50 Sound level (dB attn.) 26.4dB Oct. re BF 0.5 26.4dB -0.6 -0.2 0 Oct re BF 0.5 6.4dB Oct. re BF 6.4dB -0.5 16.4dB 0.2 0.1 0.1 0 0 -0.1 -0.1 -0.2 16.4dB 1 sp/s/dB 0 0 0 0 -0.5 0.4 0.1 -0.1 -0.5 0 0.5 Oct. re BF Fig. 3.47. Example weight functions of VCN neurons in normal-hearing ears. A PriN (A) and a chopper (B) units are shown here. PST histograms for BF tones at 30 dB above thresholds, and RLFs for BF-tones and broad-band noise are shown. Error bars in the linear weight function are ± 1 bootstrap S.D. Red / cyan regions indicated significant excitatory / inhibitory weights. Levels are in dB SPL / component. The X’s in 2nd-order weight plots indicate significant weights. 216 1st-order (linear) weight function of a PL neuron (PriN in this case) showed a well-defined central excitation region, which had a peak shape. Over different levels, the bandwidth and magnitude of this excitatory region often changed. At low-levels near the threshold for the RSS stimuli, the weights were generally small and distributed in a narrow interval surrounding the BF. As level increases, the excitatory weights increased in magnitude and the excitatory region gradually expanded to higher and lower frequencies. However, at very high levels, the excitatory weights decreased and inhibitory areas often appeared, presumably reflecting effects of rate saturation. At low and medium stimulus levels, the maximum weights are always found at or close to the BF. With further increase in level, the peak level gradually shifts to lower frequencies. These findings were similar to the ANF weight functions shown in Young and Calhoun (2005), which also showed downward shifting of the peaks in weight functions at high levels. However, this shift typically happened only at 80 dB re threshold or higher in the AN fibers. Therefore although this shift of weight-function peak in the neuron shown Fig. 3.47.A may be qualitatively similar to ANF features, it is presumably due to other mechanisms, such as inhibitory inputs. The central excitatory peaks were also observed in the Chopper neuron. However, as exemplified by these to sample neurons, Chopper neurons generally exhibited wider and deeper inhibitory regions than PL neurons do. These inhibitory regions in chopper neurons can be wider than the excitatory center and quite level-independent. These results confirmed the findings in Yu (2000). This was also consistent with observations in level-frequency responses maps, which also showed relatively more pronounced inhibition in the chopper group (See Section 3.5.2). The 2nd-order weight functions for PL and Chopper neurons were similar. On the diagonal 217 line, which quantified the quadratic interaction within the same frequency bins, there were often significantly excitatory weights near BF. Most significant off-diagonal weights, which captured the interaction between different frequency bands, were inhibitory. Again, these were consistent with previously shown weighting functions for VCN principle neurons (Yu, 2003; Yu and Young, 2000). The LNWM weight function obtained by RSS-C stimuli from a PL and a Chopper neuron in impaired ears are shown in Fig. 3.48.A and B. The linear weight functions were substantially wider in these impaired neurons than in normal ones with the same response types. In a few cases, the maximum possible frequency range over which the weights could be computed, as determined by the number of stimuli, couldn’t even cover the entire excitatory region (e.g., the WF at 25.5 dB in A). These wider functions are unlikely to be a result of using a different set of stimuli, for the following reasons. (1) Extending the computation range substantially over a neuron’s true responsive frequency region usually doesn’t lead to wider weighting functions, as exemplified by the linear weighting functions at the lowest three levels shown in Fig. 3.46.A. (2) It has been shown previously that the frequency resolution of the RSS stimuli doesn’t have a significant or systematic effect on the resultant weighting functions (Barbour and Wang, 2003; Young and Calhoun, 2005). (3) In a few neurons in the impaired ears, the RSS-B stimulus set was used, which also yielded wider-than-normal weight functions. These widened linear weighting functions were not surprising because they are expected from broader frequency tuning of VCN neurons after hearing loss (Section 3.5.1). Another abnormal feature of the linear weights of impaired neurons is the relatively weaker or total absence of inhibitory weights in the chopper category. No systematic differences in the quadratic weight profiles were seen between 218 1000 06-047 (2-k Deaf), T1U4, RSS_C PriN, 6.21 kHz, 40.1 dB SPL Firing rate (sp/s) Firing rate (sp/s) A 500 0 0 0.05 Time (s) 0.1 300 200 100 BF BBN 0 -80 -60 -40 -20 Sound level (dB attn.) 30.5dB Oct. re BF 30.5dB 20.5dB -1.8 -1.4 -1 Oct. re BF 1 sp/s/dB 0.05 25.5dB 0.04 0.02 0 0 0 -1 -0.02 -2 -2 -1 0 Oct. re BF 20.5dB 25.5dB 0 0 -0.05 -2 -1 0 Oct. re BF -0.04 0.02 0 -1 -0.02 -2 -2 -1 0 Oct. re BF -0.6 -0.2 3.3e-0160.4 Oct re BF 1500 06-047 (2-k Deaf), T1U3, RSS_C ChS, 4.485 kHz, 47.4 dB SPL 1000 500 0 0 0.05 Time (s) Firing rate (sp/s) Firing rate (sp/s) B 0.1 400 200 BF BBN 0 -80 -60 -40 -20 Sound level (dB attn.) Oct. re BF 38.6dB 38.6dB -1 -2 -2 -1 0 Oct. re BF 0.1 28.6dB 0.1 0.05 0.05 0 0 -0.05 -0.05 -0.1 -0.1 -2 -1 0 Oct. re BF 28.6dB 1 sp/s/dB -1.8 -1.4 0 0 -1 -0.6 -0.2 0 Oct re BF 0.4 Fig. 3.48. Example weight functions of VCN neurons in noise-exposed ears. A PriN (A) and Chopper (B) units are shown here. In both cases, the RSS-C stimulus set was used. The foramts are the same as in Fig. 3.47. 219 the normal and exposed units. The typical on-diagonal-excitatory, off-diagonal-inhibitory pattern was seen in exposed units as well. Responses to the RSS stimuli were obtained from a few onset- and unusual-type neurons in the normal hearing ears. Fig. 3.49.A shows representative weight functions of an OnC type neurons. As the PST histogram shows, this unit responded to BF-tone bursts at 30 dB re threshold with short-latency and very precisely timed onset spikes and a very low but discernable sustained firing. Its rate-level function showed strong nonmonotonicity, with a peak firing rate at about 15 dB re threshold. At low reference levels, its linear weight functions showed narrow excitatory regions. At higher levels, inhibitory areas appear at frequencies below BF, whose bandwidth increases with level. At the meantime, the excitatory widths appeared to be “pushed” away from BF toward higher frequencies, but they never entirely disappeared. Also, unlike the quadratic weights of PL and Chopper neurons, most diagonal quadratic weights at and near BF are negative. These patterns of linear weight functions were very different from those of the PL and Chopper neurons, and consistent with the weight functions of VCN onset neurons described in Yu (2000). The weight functions of unusual-type neurons in the normal hearing population were very heterogeneous. Some units showed WFs indistinguishable from those of the PL units, except for relatively lower weight magnitudes. Others showed low-magnitude weights which barely reached significance. Still others exhibited exotic weight-function patterns unseen in other neuronal types (including ANF, VCN and DCN neurons) described before. For example, the Unusual-B unit in Fig. 3.49.B showed inhibitory weights in its linear weight functions only at frequencies below BF. Accompanying these inhibitory regions are the significant excitatory 220 1000 0 0 0.05 Time (s) 0.1 26.4dB 16.4dB 6.4dB -3.6dB 300 200 100 -100 -50 Sound level (dB attn.) 0.02 0.5 26.4dB 0 -0.2-1.6e-016 Oct re BF 0.4 Oct. re BF -0.6 -23.6dB 0 16.4dB 0 0.5 6.4dB 0 -0.02 0.02 -3.6dB 0 -0.5 0 0.02 -0.02 0.02 -23.6dB 0 -0.5 -0.02 0.04 0 -0.02 0.5 -13.6dB 0.02 0 -0.5 -13.6dB 1 sp/s/dB BF BBN 0 -150 Oct. re BF 06-024 (normal), T3U1, RSS_B OnC, 7.85 kHz, -13.5 dB SPL Oct. re BF 2000 Firing rate (sp/s) Firing rate (sp/s) A -0.02 -0.04 0.02 0 -0.02 -0.5 0 0.5 Oct. re BF 06-048 (normal), T2U2, RSS_B Unusual:B, 5.5 kHz, 32.2 dB SPL 100 50 0 0 0.05 Time (s) 0.1 29.6dB 20 10 0 -80 Oct. re BF 150 Firing rate (sp/s) Firing rate (sp/s) B BF BBN -60 -40 -20 Sound level (dB attn.) -3 x 10 0.5 24.6dB 5 0 -3 19.6dB x 10 2 0 0 0 -0.5 -2 -5 24.6dB -3 0.1 sp/s/dB 19.6dB -0.6 -0.2 0 Oct re BF Oct. re BF 0.5 29.6dB 0 x 10 5 0 -0.5 -5 0.4 -0.5 0 0.5 Fig. 3.49. Example weight functions of onset- and unusual-type neurons. The weight functions of an On-C (A) and an Unusual-B (B) units are shown here. Both units were recorded from normal-hearing ears. In both cases, the RSS-B stimulus set was used. The formats are the same as in Fig. 3.47. 221 weights at and above BF. This was distinct from PL and Chopper neurons, which showed inhibitory regions at both sides of the BF or only at supra-BF regions. Unfortunately, no RSS responses were recorded from unusual-type neurons in the impaired ear. Fig. 3.50.A is a summary plot of linear weighting functions obtained from PL, Chopper and Locker neurons from normal and exposed ears. Because weight magnitudes and significant bandwidths were level dependent, only the weight functions from narrow ranges of levels are shown (See text in the panel). In the impaired population, only sub-edge units, i.e., the neurons with substantially threshold shifts, are shown. In addition to level dependence, BF also had a systematic effect on the linear weight functions. As shown in Panel B, the BF weights showed a positive correlation with BF in both normal and impaired ears. These weights were from the same level ranges as in Panel A. In these low/medium level ranges, the BF weights were almost always maximum weights in the linear WF. As the relative positions of the two regression lines show, impaired units showed lower BF weights than normal units at corresponding BFs. It is noteworthy that a few chopper units, which is the unit shown in Fig. 3.48.B, were the only exceptions. These units had BF weights comparable to the weights of normal choppers at nearby BF regions. When weight magnitudes of different types of neurons were compared, Chopper neurons show larger weights than other types of neurons, as expected from the relatively larger driven firing rates in this type of neurons. As the blue dashed line in Panel C shows, the excitation bandwidth of linear weight functions also showed strong dependence on BF in the normal ear. This was consistent with the observations in ANFs (Calhoun and Young, 2005). The higher the BF was, the narrow the 222 4 BF weight (spikes/s) Linear weights (sp/s/dB) PL / Ch / Locker BF: [0.3, 20) kHz Normal: -4 - 15 dB / comp. Exposed: 19 - 40 dB / comp. 2 0 B. Linear weights at BF 6 W=1.65*x+1.16 W=1.14*x+0.85 4 Excitatory bandwidth (octaves) A. Linear weight functions 6 2 0 -2 0.1 10 BF (kHz) Quality factor 0.6 0.4 0.2 Normal avg. Exposed avg. E. Qual. factor: full vs linear 1 0.8 0.6 0.4 0 -20 0 20 40 60 RSS stim. level (dB / comp.) G. R0-level functions 200 100 0 -20 0 20 40 60 RSS stim. level (dB / comp.) 0 0.1 10 6 4 2 0 -20 0 20 40 60 RSS stim. level (dB / comp.) F. Normalized RMS of quad. wts. 0.2 0.15 Normal avg. Exposed avg. 0.1 0.05 0 -20 0 20 40 60 RSS stim. level (dB / comp.) H. R0-level slopes p=0.08 6 4 2 0 0.1 10 BF (kHz) Excitatory bandwidth (octaves) I. BF weight vs. level Linear weight at BF (sp/s/dB) 0 0 0.2 0.4 0.6 0.8 1 Quality factor of linear model 8 R0-level slope (sp/s/dB) R0 (spikes/s) 300 kNormal=0.25 kExposed=0.63 0.2 Norm. RMS of quad. wts (dB-1) D. Qual. factors vs. level 0.8 0.5 1 Quality factor of full model 1 -1 0 Octaves re BF 1 BF (kHz) J. Excitatory width vs. level 2 1.5 K. Inhibition ratio vs. level 0.8 Inhibition ratio -2 -2 C. Excitatory bandwidths 2 B=-0.50*x+0.81 B=-0.14*x+0.74 1.5 1 0.5 0 -20 0 20 40 60 RSS stim. level (dB / comp.) 223 0.6 0.4 0.2 0 -20 0 20 40 60 RSS stim. level (dB / comp.) Fig. 3.50. Quantitative analysis on the linear and nonlinear weight functions from the RSS stimuli. Pri/PriN, Chopper and Locker neurons were included in this figure. In the impaired data pool, only units with BFs below CAP-audiogram edge frequencies were included. A. 1-st order (linear) weight functions in certain level and BF range. B. The relations between BF weights (for definition, see Section 2.7) and BF. Linear regressions for the two populations are shown. Symbols: W: weight; x: log10(BF), BF expressed in kHz. C. The relations between excitatory bandwidths and BF. As in B, the regression lines were shown. Symbols: B = bandwidth; x: log10(BF), BF expressed in kHz. D. Quality factor for the full model plotted against RSS reference level (in dB SPL per component). The data points connected by lines are from the same neurons. The running averages for the two hearing states are shown. E. The relationships between quality factor of the linear (1st order weights only) model and the full (1st + 2nd order weights) model. The gray diagonal line indicates equality. The regression line and their slopes are shown. F. Normalized root-mean-square values of quadratic (2nd-order weights) plotted against RSS reference level. The running averages are shown. G. R0 (calculated from the model) versus RSS reference level. Their slopes were extracted and plotted in H. H. Relationships between R0-versus-level slopes in the two hearing states and their comparison. The p-value was from a Wilcoxon rank-sum test. I. BF weights versus RSS reference level. J. Excitatory bandwidths versus RSS reference level. K. Inhibition ratios versus RSS reference level. 224 excitatory region tended to be. However, this relationship didn’t hold in the impaired population, mainly because a few impaired neurons which show abnormally broad linear weight functions. The quality factors (Q) as defined by Eq. 2.7 was used to quantify the goodness of fit by the LNWM. Fig. 3.50.D shows the dependence of the quality factors for the full model (with 1st and 2nd order weights) as functions of reference levels. Most normal and impaired neurons show arch-shaped Q-versus-level functions. The running averages of both populations confirmed this feature. The average fitting performance of the model was the best at medium levels, and sagged at higher levels. This was probably due to the nonlinear turning points near the threshold at low levels and saturation at high levels. In other words, responses were most linear at medium levels, which led to the best model performances. As shown in Panel F, the root-mean-square (RMS) value of quadratic weights normalized by the RMS value of linear weights, when plotted against stimulus level, showed the opposite shapes as those Q-versus-level functions, on both single-unit and population level. This indicates that the strength of nonlinear terms in the model was somehow negatively correlated with goodness of fit. This model worked better at capturing linear responses than at modeling nonlinearities. This was probably due to the fact that when substantial nonlinearities exist, merely using the second order terms, as the LNWM did, captured only a relatively small fraction of the nonlinearity. The quality factors for the full LNWM model and the linear-weights-only model are compared in Fig. 3.50.E. Incorporation of quadratic terms improved model goodness of fit in all cases. There was a positive relationship between Q factors for the two versions of model. This relationship was stronger in the impaired population. In the unexposed ears, there were a few units in the with linear-only Q factors below 0.1, but with full-model Q factors approaching 225 perfect. On average, the Q factors were low in the impaired population than in the normal one. One possible cause of this different was the larger rate variability in impaired neurons (Section 3.6.9). As shown in Fig. 3.50.G, the zeroth-order weights (R0), i.e., the average response to the all-zero-dB RSS stimulus were monotonically increasing functions of reference levels in all recorded VCN neurons, except in one normal chopper neuron, in which the saturation was reached at high levels. For units in which more than one level were available, line fits were made to quantify their rate-level slopes for the RSS stimuli. The slopes were plotted against BF in Panel H. No systematic relationship between rate-level slopes and BFs were seen in either hearing condition. No significant effects of hearing loss on the slope median were seen (WRS test, p = 0.08). The dependence of the linear weight functions on reference level is shown in the bottom row in Fig. 3.50. There were three trends of the linear WFs in the normal hearing population. (1) BF weights-versus-level functions showed arch shapes. Maximum BF weights (and all the other excitatory weights, not shown) usually occurred at medium levels. (2) Excitatory bandwidths tended to be small at low levels near threshold. The bandwidths were larger at medium levels and may or may not decrease at higher levels. (3) Inhibition ratios were often increasing functions of level. Some unit-type dependence of linear weight functions can be seen in these three panels. Chopper neurons tend to exhibit larger linear weights, narrower excitatory bandwidths and greater inhibition compared with PL and Locker neurons. The dependence of BF weight on level was less obvious in the impaired population, in which the weights were often low at all levels. As mentioned earlier, the three impaired chopper 226 neurons were the only exceptions to this trend. Excitatory bandwidths were on average much larger in the impaired neurons, but the level dependence could not be clearly seen, because of larger noise in the impaired data. The trend for inhibition to strengthen at higher levels can also be seen in many impaired neurons. However, if the three outlying chopper neurons were disregard, the inhibition ratio of the units from exposed ears occupied a much lower range than those from unexposed ears did, reflecting a loss of inhibitory linear weights after HL, for which a possible explanation was the reduction in two-tone suppression in the cochlea following acoustic trauma (Franck, 1994; Miller et al., 1997). 227 IV. Discussion 4.1. Summary of findings This thesis presents a systematic investigation of basic in vivo physiological properties of VCN neurons in acoustically traumatized cats, and the effect of NIHL on encoding of the levels and spectra of simple and complex stimuli in the VCN. Differential alterations were observed in different VCN neuronal types. This study also systematically investigated the relationship between post-traumatic level encoding abnormalities in VCN and the psychophysical symptom of loudness recruitment. Possible neural correlates of loudness recruitment were found in certain aspects of the rate-level response in the VCN. Acoustic trauma was induced in cats by overexposure to a 2-kHz centered narrowband noise. The patterns of hearing losses were measured by compound-action-potential audiograms. These matched previous results based on similar types of noise exposure. Single neurons were recorded from in the VCN of unanesthetized decerebrated brains, in both exposed and normal-hearing control animals. The single-unit threshold-BF relationships matched CAP audiograms in individual animals. The maximum threshold shifts and frequency spans of HL showed considerable variability across animals. In data pooling for single-unit analyses, this variability was corrected for by applying level adjustments, which eliminated threshold variance across animals and BF regions. VCN neurons in both hearing conditions were classified into primary-like, primary-like-with-notch, choppers and onset (with different subtypes), low-BF phase lockers and unusual-type neurons. This classification was based on BF-tone PST histograms, and basically 228 followed the methodology previously described by Blackburn and Sachs (1989). Pri, PriN, phase-locker and various subtypes of Chopper units were present in both normal and exposed animals. Except for a few differences in rate, latency, regularity, and phase-locking measures and their level dependences, no substantial differences were seen for basic characterizations of the BF-tone PSTHs. Analyses based on the vector strength measure suggested a weakening of phase-locking to BF tones of VCN units in exposed ears. Very few onset neurons were recorded from in the impaired BF regions in the exposed animals. A few unusual-type neurons with properties unseen in unexposed units were found in impaired ears. VCN neurons in the impaired BF regions showed elevated thresholds, as well as broadening and shape changes in their tuning curves. The extent of threshold shift and loss of tuning matched those of the ANFs of the moderate/severe hearing-loss group of animals deafened with a similar method (Heinz and Young, 2004). Overall, there was no strong evidence that neural inhibition in VCN was lost or weakened after trauma. In addition to changes in tuning curves, the level-frequency response maps of the VCN neurons also showed alterations after hearing loss. The threshold, shape and slope patterns of BF and off-BF tonal rate-level functions were analyzed. Different effects of acoustic trauma on RLFs were found in different types of VCN units. The slopes of BF-tone RLFs of PL and phase lockers became shallower after trauma, whereas chopper and unusual-type units show near normal or steepened BF rate-level slopes in impaired ears. Some systematic changes were also seen for the off-BF RLFs. Off-BF inhibition appeared to be slightly stronger than normal in Chopper and Unusual units in impaired ears. In all unit types, the percentages of non-monotonic rate-level functions significantly decreased after hearing loss. 229 A pseudopopulation approach was used in the analyses on rate-level encoding of tone frequencies. Both on-BF and off-BF encoding were studied. Rate-matching curves between the normal and impaired populations were produced to quantify the strength of recruitment-like phenomenon. Although no evidence was found that the threshold distribution was more compact in a narrow BF region showing threshold shift, the variance of thresholds for a single tone over the entire tonotopy was significantly reduced, owing to broadened tuning. This led to significantly faster-than-normal expansion of activated BF regions with increasing tone levels. For on-BF encoding, PL and Locker units showed shallower average RLFs after hearing loss, while chopper and unusual-type neurons showed steepened average RLF. For all-BF encoding, due to faster-than-normal spread of excitation, all unit types, except Locker neurons, showed steepened average RLFs in impaired ears. The grand VCN tone RLF, which summarized the coding by all BF channels and all unit types (except onset), showed significantly steepened slopes in impaired hearing, consistent with recruitment. More than one type of possible neural correlates of loudness recruitment in VCN was discovered. According to our observation, the two strongest correlates are, 1) an abnormally rapid spread of excitation which leads to substantially faster increase of summed rates of VCN neurons across the entire tonotopic axis with tone level; 2) the abnormally steep BF-tone rate-level functions of individual neurons in the non-primary-like PSTH categories including chopper, onset and unusual ones. Rate-level encoding was also studied for broadband noise. As in the tone RLF case, abnormally steep RLFs were found in the chopper and unusual-type categories. Rate-level and rate-spectral encoding of the steady-state synthetic vowel /ε/ were also 230 abnormal in exposed ears. The spectral manipulation procedure was used to construct a vowel-encoding neuronal pseudopopulation. Rate-place encoding of the vowel spectrum in the impaired ear showed abnormal formant-trough organization and was dominated by responses to F1. Responses to minor spectral features were weaker after hearing loss. The grand vowel RLF was slightly steeper in the impaired ear, consistent with recruitment. The linear-nonlinear weighting model and random spectral shape stimuli were used to study the rate encoding of spectral shapes at different levels in the VCN after acoustic trauma. The patterns of linear weight functions of impaired VCN neurons systematically differed from those of the normal ones in their significantly broadened excitatory region, reduced weight magnitudes and smaller and weaker inhibitory areas. As for the question of whether inhibition was weakened after acoustic trauma in the VCN neurons, there was no strong evidence in favor of this hypothesis. Inhibitory response areas persisted in non-primary-like neurons such as chopper and unusual units after HL. Moreover, there was some evidence that off-BF inhibition was slightly strengthened in chopper and unusual categories after HL. The increased prevalence of monotonic rate-level functions wasn’t strong evidence for the decreased inhibition because the narrower level ranges over which rate responses were recorded in the impaired ears could not be ruled out as a possible cause for this alteration. Also, the reduced width and magnitude of inhibitory areas in the weight functions in the LNWM may be primarily caused by a weakening of two-tone suppression in the damaged cochlea. Variance-mean analyses revealed larger rate variability in response to tones in VCN neurons after acoustic trauma, supporting the increased internal noise hypothesis. 231 4.2. Physiology of VCN neurons in acoustic trauma partially explained by AN abnormalities. The classical physiological alterations in AN fibers following acoustic trauma, including threshold elevation and broadening of tuning (Liberman and Kiang, 1978; Miller et al., 1997; Heinz and Young, 2004) were also observed in VCN neurons in this study. This was not surprising, because previous studies show that tuning curves of most VCN principal neurons are indistinguishable from those of the ANFs in normal-hearing ears (Bourk, 1976; Rhode and Smith, 1986), which was also confirmed by our results. The primary effect of acoustic trauma on the BF-tone RLFs of ANFs is to reduce the component-1 (C1) response (Liberman and Kiang, 1984; Heinz and Young, 2004). In normal-hearing ears, the synaptic transmissions between AN afferent fibers and VCN bushy cells are very secure because of the endbulb endings (Ryugo and Sento, 1991). Based of this, if one assumes that no dramatic changes happen in the synaptic or electrical properties of bushy cells after acoustic trauma, it should be expected that Pri/PriN neurons in the VCN also show shallower rate-level slopes after trauma. This was indeed the case according to our observation. One of the most striking nonlinearity of ANF RLFs is the rapid phase-transition and rate non-monotonicity at levels between 85 – 100 dB SPL (Liberman and Kiang, 1984; Wong et al., 1998). There is evidences that AN responses below and above this transition are of two different modes, called component-1 and component-2 (C1-C2), which are out-of-phase and cancelling each other during the C1/C2 transition. The RLFs of a few recorded PL and Locker VCN neurons show signs of C1/C2 transition as manifested in the non-monotonicity (notches) observed at levels above 80 dB SPL. However, these rate-level notches are harder to see in other types of VCN neurons. It is possible that the rate-notches at the C1/C2 transitions are eliminated 232 in the VCN neurons through convergence of input from multiple afferent fibers. It is unlikely that the recruitment-like phenomena observed in certain conditions in the current study are due to the invulnerable C2 responses. In our recording experiments, it was rarely possible to maintain reliable single-unit isolation at levels above 80 - 90 dB SPL, in both normal and impaired hearing, especially in BF regions between 0.2 and 6 kHz. The steep rate-matching functions all reflect abnormally rapid rate increases between 40 – 80 dB (Figs. 3.35 & 3.36), which was lower than the thresholds of C2 response. The observation that choppers and other non-primary-like neurons didn’t show shallower rate-level slopes after trauma was contradictory to abnormalities seen in AN fibers. It gave hints to the existence of up-regulation of neural excitability or changes in the circuitry after hearing loss. There are several possible ways this increased excitability can be realized. The first is changes in electrical properties of the stellate and other types of cells. The study by Francis and Manis (2000) in cochlea-ablated rats indicate seems to negate this possibility because they found that electrical properties of VCN neurons, including voltage thresholds, time constants and input resistances are largely unaltered by cochlear damage. The second possibility is increase in the synaptic weights after trauma. There haven’t been enough studies on the synaptic properties of stellate cells after hearing loss. So this possibility still awaits testing. A stellate cell receives relatively few ANF inputs in normal ears (Ferragamo et al., 1998). The third possibility is that the number of AN fibers a stellate (or other non-primary-like) cell receives input from increases as a consequence of axon sprouting and synaptogensis (e.g, Kim et al., 2004; Benson et al., 1997). The slightly lower ISI CV shown in Fig. 3.11.A and the stronger-than-normal dependence of ISI CV on level in the impaired chopper units as shown in Fig. 3.14.D is in principal 233 consistent with this hypothesis, as it has been shown by a modeling study that the high regularity of discharge in choppers is mainly a consequence of integrating converging spike-train inputs from a few AN fibers, and the regularity of firing is positively correlated with the degree of convergence (Banks and Sachs, 1991). Heinz and Young (2005) showed that the three major hypotheses regarding the physiological origin of loudness recruitment have not been supported by experimental ANF data. These three hypotheses are: (1) AN fibers show steepened RLFs, mainly due to loss of the compressive nonlinearity of the BM (e.g., Harrison, 1981); (2) the distribution of thresholds of a population of ANFs within a narrow BF window gets more compact (Moore et al., 1985; Zeng and Turner, 1991); (3) the spread of excitation to off-frequency fibers become faster, due to loss of tuning (Kiang et al., 1970; Evans, 1975). Heinz and Young (2005) showed that AN fibers exhibited shallower rate-level slopes for tones, reflecting IHC damages, therefore the first hypothesis was negated. The threshold distributions were observed to be significantly larger after trauma, probably because of the unevenness of the effect of noise exposure on different hair cells. So the second hypothesis wasn’t supported either. With regard to the faster-than-normal spread of excitation hypothesis, they did observed a slightly more rapid recruitment of off-frequency channels in the moderate/severe HL group, although this effect wasn’t strong enough to cause significantly steeper-than-normal summed rate-level function for all-BF encoding, which has been reproduced in the pseudopopulation re-processing done in the current study. In VCN neurons, for BF-tones, shallower rate-level slopes are seen in primary-like neurons after trauma, while near normal or steeper ones are seen in non-primary-like ones. However, the latter should not be viewed as supporting evidence for the first hypothesis because it’s primarily 234 caused by post-traumatic changes in the central neurons, not by changes in BM input-output function and the transduction functions of the IHCs. The second hypothesis on threshold distribution was specifically designed for AN physiology, and doesn’t have much meaning to the VCN neurons, because convergences of inputs can destroy the patterns of threshold distributions. But when we disregard these caveats and compare the standard deviations of thresholds in normal and impaired ears, we still could not find any evidence of more compact threshold distributions in VCN. Our data supports the faster-than-normal spread of excitation hypothesis (Fig. 3.37). The recruitment of off-frequency BF channels was about 3 times as fast in the impaired ear as in the normal one. Moreover, the increase in the rapidness of SOE appeared to be faster in the VCN neurons than in the ANF (based on re-processed data of Heinz and Young, 2004, Fig. 3.37.B). The current study and the previous one were based on very similar animal models of NIHL, except for that the average intensity of the exposure noise was 2 – 3 dB greater in the current study. Therefore it’s unlikely that this difference was caused by different patterns of hair cell damages. We attempted to construct BM input-output functions from comparing rate-level slopes for tones at and well below the BFs (Yates et al., 1990; Heinz and Young, 2005), and compare them to Fig. 5 in Heinz and Young (2005). However, this attempt was unsuccessful (not shown). One of the major reasons we think was that off-BF RLFs are affected by neural inhibition in central auditory neurons, which rendered the method of Yates et al. (1990) invalid. But as the comparison between Fig. 3.17.B & C indicate, the degree and threshold shifts and distributions of relatively fine and broad tuning curves agreed between our VCN neuron population and the moderate/severe HL population ANFs in the previous study. Therefore, explanations other than 235 differences in hair cell damages need to be sought to explain the more rapid SOE in VCN neurons in traumatized ears. There are two possibilities, which are not necessarily mutually exclusive. First, there may be axonal sprouting and formation of new synapses after cochlear damage (Benson et al., 1997, Kim et al., 2004), which may lead to wider BF ranges a VCN neuron receives its input from. Secondly, the reduction of phase-nonlinearity of the BM response after trauma may lead to stronger across-BF synchrony and faster strengthening of this synchrony with increasing level (Carney, 1994). The abnormally strong synchrony may lead to more rapid SOE in VCN neurons, which receives convergent input from multiple AN fibers. 4.3. Central neural alterations following cochlear damage Normal functions and neural activities in the AN afferent fibers play important roles in maintaining normal neuronal structure and function of the CN neurons. Previous studies found that cochlear lesions or pathologies induced degeneration of afferent axons in the CN (Gentschev and Sotelo, 1973; Morest et al., 1997), decrease in cell size and density (e.g., Lesperance et al., 1995; Willott et al., 1987), abnormalities in synaptic protein synthesis (e.g., Luo et al., 1999, Suneja et al., 1998), and loss and regenerations of primary axons and synaptic endings (e.g., Benson et al., 1997; Kim et al., 2004). These changes were elicited by decrease or cessation of afferent activities and are more closely associated with IHC damage than with OHC damage (Lespserance et al., 1995). The electrical and synaptic properties can be also altered following cochlear damage (e.g., Oleskevich & Walmsley, 2002; Wang and Manis, 2005, 2006). There have been relatively few single-neuron studies on auditory brainstem physiology in hearing impaired animals. To our knowledge, this study was the first investigation of in-vivo single unit physiology in VCN of animals with permanent NIHL. 236 In the only previous single-unit neural recording in CN of hearing impaired animals known to us, Lonsbury-Martin and Martin (1981) showed that about two-thirds of VCN neurons in rhesus monkeys with temporary threshold shifts induced by noise exposure showed depressed driven discharge rates at all levels, while the remaining one-third showed higher-than-normal driven rates at high levels. Although the authors didn’t perform unit classification and their deafening protocol differs from ours, their observation is in principal consistent with the findings in the current study, which showed that different VCN neuronal types show different post-trauma alterations. Ma and Young (2006) demonstrated that putative DCN principal neurons show abnormal level-frequency response maps following a high-frequency acoustic trauma. These abnormal response maps can be described as widely tuned and predominantly excitatory, which are suggestive of formation of novel connections and changes in excitation-inhibition balances. In comparison, the physiological changes observed in exposed VCN neurons in the current study were not as dramatic. Except for a few neurons exhibiting extraordinary PSTH shapes which were allocated to the unusual category, most of the impaired neurons showed normal PSTH patterns not dissimilar from normal ones. Furthermore, as described in Section 3.5.2, except for few unusual type neurons, level-frequency response areas of exposed VCN neurons didn’t show significant differences from normal ones, except for threshold and tuning alterations accountable by changes in ANF tuning curves. Also, from the response maps, there was no compelling evidence for reduction of inhibition or formation of new synaptic inputs. These differences between VCN and DCN may be explained by the difference in synaptic and circuitry complexities in the two major divisions of ANF. The DCN contains an intricate neuropil made up 237 of many cell types and different sources of counterbalancing excitation and inhibition, and DCN also receives a variety of non-auditory inputs (Young and Oertel, 2004). In contrast, the inputs of VCN principal neurons mainly come from cochlear afferent fibers, and inhibition plays a relatively less important role in VCN than in DCN. The cellular signaling following changes in afferent input patterns may be more complicated in the dorsal division. Based on these findings, one may conclude that the complexity of post-trauma alterations is positively correlated with the complexity of the neural structure before damages. An in-vitro intracellular study in antero- and postero-VCN neuron of rats with bilaterally ablated cochleas (Francis and Manis, 2000) observed that the two electrophysiological response types, termed type-I and type-II, could be still found in the VCN about 2 weeks following deafferentiation. Type-I neurons show linear current-voltage (I-V) relationships and regular non-adapting firing in response to current steps. They correspond to stellate cells. Type-II neurons show nonlinear I-V relationships and fire transiently in response to current step. They correspond to bushy cells (Oertel, 1983). In the current study, Pri/PriN and various subtypes of chopper units were recorded from after acoustic trauma, which was consistent with that electrophysiological finding. Francis and Manis (2000) also showed that except for a few small but significant alterations, the electrical properties of the type-I and type-II neurons were not significantly affected by deafferentiation. This result could explain the resemblance of post-traumatic changes in ANF (Heinz and Young, 2004) and VCN Pri/PriN neurons following acoustic trauma. However, the fact that the chopper (stellate) neurons didn’t show shallower rate-level slopes as the fibers and Pri/PriN neurons did, couldn’t be explained by the minimally altered memrane electrical properties in type-I neurons. The reason may lie in other sub-cellular 238 level alterations, such as change in synaptic properties and neuronal circuitries (Kim et al., 2004). Benson et al. (1997) used immunochemical methods to study the effect of primary deafferentiation on synaptic formation in the CN of guinea pigs, and observed losses of synaptic terminals in AVCN and PVCN following cochlear ablation. These losses began to recover through synaptogensis after a 7-day period in the anterior AVCN, but not in the anterior PVCN. The inability of PVCN neurons to initiate regeneration of lost synapses after deafferentiation could be an explanation of the scarcity of recorded Onset-type neurons recorded from in this study. The active regeneration of synapses in the anterior AVCN triggered by deafferentiation could explain the generally unchanged response types in Pri/PriN and Chopper PSTH types, which correspond to AVCN cell types. Steeper rate-level slopes in certain AVCN response types, include choppers and unusual-type, could be explained by formation of new synaptic connections (Kim et al., 2004) and subsequent convergence of larger numbers of inputs on each neuron. As shown in Fig. 3.35, the effect of acoustic trauma to decrease the BF-tone rate-level slopes was slightly stronger in the VCN Pri/PriN neurons than in AN fibers. Although we cannot completely rule out the possibility that the noise overexposures on average caused more severe damages to inner hair cells, this phenomenon may also be explained by the decreased capability of bushy cells to integrate and relay inputs from afferent fibers. In DBA mice with early-onset age-related hearing loss, Wang and Manis (2005) found that miniature EPSCs at the endbulbs of Held underwent a decrease in amplitude and spontaneous release probability. Also, the spiking threshold with current injection was elevated (Wang and Manis, 2006). If similar changes 239 happened to primary endings on bushy cells in the acoustically traumatized cats, the resultant decreased synaptic transmission efficiency could explain the shallower rate-matching curves in VCN PL units. One of the most interesting observations of the present study is that primary-like and non-primary-like neurons in the VCN exhibit different alterations in rate-level responses after acoustic trauma. Previous studies (Lee et al., 2003; Redd et al., 2002) reported different synaptic and morphological abnormalities in bushy and stellate neurons in the VCNs of congenitally deaf white cats (DWC). Compared with normal-hearing controls, primary endings on spherical bushy cells in DWC contain fewer vesicles, and bushy cells show loss of intracellular cisternae and hypertrophy of post-synaptic densities. In contrast, stellate cells in VCN of deaf white cats show little synaptic abnormalities beyond slightly smaller-than-normal bouton synaptic endings. Although the etiology of noise-induced and congenital HL presumably differ and the results from the DWC studies involve development effects, these findings nonetheless suggest that different neuronal populations in the VCN may respond to diminished afferent inputs with different compensatory self-adjustments, which could explain the different responses to acoustic trauma observed in primary-like and non-primary-like VCN neurons in the current study. 4.4. The role of VCN in central overexcitability following cochlear damage A large body of previous literature has documented the central overexcitability following cochlear damage (See Tab. 1.1. for a brief summary). These studies were conducted in many different animals species (including rodents, carnivorae and primates), based on a variety of cochlear insults (acoustic trauma, ototoxic drugs, and cochlear ablation), on different measurement techniques (behavioral, evoke-response and single-unit recordings), and also 240 constitutes a mixture of different recovery times (permanent versus acute). But they invariantly observed neural overexcitability in certain central auditory structures, which attests the ubiquity and predictability of the central changes. These central changes were often akin to the behavioral traits of recruitment: although absolute thresholds are elevated after cochlear damage, the increase of response amplitudes with sound level or electrical stimulation intensity becomes significantly faster than normal controls. In many instances the steepening of the response amplitude-intensity functions was strong enough to cause the catching-up or even overtaking of the normal response amplitudes by the ones in the impaired ears. In human patients with SNHL, this type of overexcitability has also been demonstrated by functional neurogimaging (Morita et al., 2003; Langers et al., 2007). Despite the prevalence of central overexcitability, a consensus has not been reached as to the loci of the generation of this overexcitability. One of the aims of the current study is to elucidate whether VCN neurons show overexcitability following NIHL. Evidence from previous studies was equivocal. In a number of previous studies in cochlear damaged animals, evoked responses were measured in multiple stages of the pathway in the same animals. Some of those studies provided evidence that compared with higher auditory structures, the CN doesn’t exhibit clear overexcitability after hearing loss. Salvi et al. (2000) showed that evoked responses in the CN didn’t show overexcitability following carboplatin-induced selective IHC damage in the chinchilla, in the face of clear overexcitability in the IC-ER. Similarly, Szczepaniak and Moller (1993, 1996) recorded IC-ER, and distinguished between four peaks in the ER waveform Peak A and B reflects input from auditory structures caudal to IC, including CN and SOC; while Peak C 241 and D are generated by neuronal activities in IC per se. Following acute noise exposure, amplitudes of all peaks decrease. But given longer recovery time (> 1 hr), the amplitudes of peaks A and B recovered back to close-normal value, while peaks C and D show amplitudes greater than normal. Other studies (Popelar et al., 1987; Qiu et al., 2000) found overexcitability on the cortical level, but not on any subcortical levels, including the IC. Conversely, several previous studies also provided evidence for the presence of overexcitability in VCN following cochlear damage (Saunders et al., 1972, Lonsbury-Martin and Martin, 1981). For example, Saunders et al. (1972) observed “over-recruitment” effects in the ERs in both CN and IC approximately 1 week after acoustic trauma. Taken together, these studies suggest that although the VCN may well play a role in generating hyperactivity in higher parts of the auditory pathway after acoustic trauma, the up-regulation of excitability may not be strong enough to be observed in VCN per se in many preparations. It needs to be mentioned that a myriad of factors, including animals species, type of cochlear damage, time course of recovery and anesthesia state have strong influence of the results on these studies. But when considered as a whole, according to previous studies, the evidence for the absence of overexcitability in the CN overweighs the evidence for the converse. However, a distinction needs to be made between the absence of overexcitability and the absence of homeostatic regulation. The latter may happen in the absence of the former. Considering that AN show hypo-excitability on both single-unit (Heinz and Young, 2004; Liberman and Kiang, 1984) and ER (e.g., Popelar et al., 1987; Qiu et al., 2000; Salvi et al., 2000) levels following acoustic trauma, even similar-to-normal response-level relationship in the CN would suggest the presence of up-regulation of excitability in the structure. 242 The experimental paradigm used in the current study aims to mimic the most common cases of NIHL in humans. A sufficiently long (> 30 days) recovery time was given and neural recording was conducted under decerebrated unanesthetized states so that the temporary effects could disappear and neuronal plastic changes could stabilize. The results from the current study indicate that the question of overexcitability in the VCN needs to be answered on a neuron-type-by-neuron-type basis. On the single-neuron level, certain neuronal types (PL and Locker neurons) show post-traumatic trends of excitability that follows that of the ANF. In contrast, non-primary-like (chopper and unusual) neurons show signs of up-regulation. Another distinction needs to be made between all-BF and on-BF encoding. For all-BF encoding, essentially every neuronal types (except Lockers) exhibited recruitment-like behavior (Fig. 3.36). But this behavior could not be described as “over-excitability”, because it mainly results from broadened frequency tuning. For on-BF encoding, up-regulations of excitability were seen in chopper and unusual-type neurons, but not in PL ones. When all neuronal types were taken into account, the slopes of average RLFs were slightly, but not dramatically steeper in the CN after trauma. This result is consistent with the previous results which suggest the absence of strong overexcitability in CN following cochlear damages (Salvi et al., 1990, 2000; Szczepaniak and Moller, 1996; Qiu et al., 2000; Popelar et al., 1987). One natural question is in whether those ER measurements reflect on-BF or all-BF encoding. Different electrode placement (relatively to the isofrequency laminas) and different acoustic stimuli (clicks versus tone bursts) used in various studies may render the answer to this question ambiguous. This could be a contributing factor to the disagreement on whether CN shows overexcitability, as according to the result of the current study, those experiments in which 243 ERs are sampled in a relatively small BF region tends to see little or no overexcitability, whereas those with sampling across a wide BF region would see overexcitability. At the meantime, the results of the current study suggest that electrode placement into the region of VCN consisting of mainly bushy cells should see no overexcitability, whereas by recording ER from VCN regions consisting of mainly stellate and other non-bushy-type cells, overexcitability and steeper ER input-output functions can be seen. Based on the observations of the current study there are two possible ways the post-trauma alterations in the CN may lead to stronger and more prevalent hyperexcitability in the IC (Bock et al., 1982; Salvi et al., 1990, 2000; Sterbing and Schrott-Fischer, 2000). (1) The VCN principal neuronal types showing up-regulation of excitability (e.g., choppers) directly project to the IC (Cant & Benson, 2003), contributing to overexcitability in the IC. Meanwhile, those principal neuronal types showing no up-regulation of excitability (e.g., Pri/PriN neurons), send output to other sub-collicular nuclei (e.g., SOC and nuclei of the LL) which in turn project to the IC. Further up-regulation of excitability may happens at those intermediate stages (Gerken et al., 1979). (2) The recruitment-like overexcitability observed in all-BF encoding in the current study may underlie the overexcitability reported for on-BF encoding in IC and other more central structures, because convergence of inputs happens along the ascending pathway. In sum, results of the current study regarding excitability of VCN neurons following cochlear damage, when considered together with previous observations, suggest the hypothesis that cochlear-damage-induced overexcitability in high parts of the auditory pathway takes shape in a hierarchical bottom-up manner, in which up-regulation of excitability happen at each major auditory nuclei and accumulates along the ascending pathway. This hypothesis can explain the 244 positive correlation between the level of an auditory structure and the likelihood and strength of the cochlear-damage-induced over-excitability observed in the previous studies (Tab. 1.1). Furthermore, this study also shows that the VCN is the lowest stage where this cumulative adjustment process happens, although not all VCN neuronal types take part in this process. 4.5. Neural correlates of loudness recruitment in VCN The previous study in ANFs of acoustically traumatized cats found no substantial correlates of loudness recruitment in the rate-level responses of ANF following acoustic trauma (Heinz and Young, 2004; Heinz et al., 2005). The reprocessing of the same data by pseudopopulation approaches clearly confirmed this conclusion. The current study is based on a similar animal model of NIHL as the previous ANF study. In contrast to the previous study, post-traumatic alterations that unequivocally showed qualitative and quantitative resemblance to recruitment were observed in limited conditions. One question which stems naturally from the basic aims of the current study is “does recruitment exist in the VCN”? Before answering this question, one has to carefully define the meaning of this question. Recruitment is a psychophysical phenomenon, which depends on the working of the whole organism, while VCN is only a part of the auditory system. Therefore the “existence of recruitment” in VCN means the presence of post-trauma physiological alterations that resemble psychophysical manifestations of recruitment, i.e., elevated thresholds and abnormally rapid increases of loudness with intensity above the threshold. In the current study, it was found that recruitment-like phenomena, according to its definition, existed in on-BF encoding of certain non-primary-like VCN neurons, and in all-BF encoding of almost all types of VCN neurons. 245 Given the existence of recruitment effects, one more important question is whether these phenomena are causal to the psychophysical characteristics of loudness recruitment. In order to answer this question, one needs a knowledge of how intensity is encoded and what is the correlate of loudness on a single-neuron level in the VCN and other parts of the auditory pathway. Unfortunately, this question has no widely accepted answer to date. A long-standing debate in the psychophysics of intensity and loudness is whether the intensity of a tonal stimulus is encoded in an on-BF or all-BF way. There is evidence for both sides. Opponents of the on-BF encoding argue that in the ANFs, the BF-tone RLFs show dynamic ranges of only 20-40 dB, significantly below the psychophysical dynamic range (> 100 dB) (Smith, 1988). Proponents of on-BF encoding argue that within a narrow BF channel, there are fibers with varying spontaneous firing rates, which show a dispersed distribution of thresholds. And moreover, low-SR high-threshold fibers tend to show incomplete sloping saturation. Therefore summed RLFs for BF-tone inside a narrow BF should exhibit a sufficiently large dynamic range (Sachs and Abbas, 1979; Winslow and Sachs, 1988; Viemeister, 1988). As our data show (Fig. 3.23), the situation is more complicated in the VCN due to non-monotonic (decreasing) portions of BF-tone RLFs at high levels. As Fig. 3.35.A, B, & C show, on-BF (0.4-octave BF window) average RLFs have dynamic ranges only as wide as 20-40 dB and show either flat saturation or drop of rates at high levels. Therefore our data appears to argue against on-BF encoding of tone levels in the VCN. In both AN and VCN neuronal populations, all-BF encoding shows nearly linear relationship between driven rates and levels in dB (Fig. 3.36, Panels E and F), with dynamic ranges at least 80 dB wide and without any signs of saturation at high levels. In the VCN, this reflects that excitatory responses overwhelm inhibitory ones on the population level, which has 246 been shown to be untrue in more rostral parts of the central auditory pathway due to stronger and more prevalent inhibition (e.g. in IC, Ehret and Merzenich, 1988). These observations support that at least in VCN, all-BF encoding is a potential way in which the brain solves the dynamic range problem. Also, unequivocal recruitment-like phenomena were observed in all-BF encoding in virtually all VCN neuronal types. These findings support the hypothesis that recruitment mainly results from faster-than-normal spread of excitation along the entire tonotopy (Kiang et al., 1970; Evans, 1975). Moreover, if one assumes that these phenomenon are causal to psychophysical recruitment (which is difficult to test), these findings lend support to the all-BF coding theory. However, psychophysical studies using band-stop-noise masking paradigms suggest that off-BF encoding is not required for maintaining a wide dynamic range (e.g., Viemeister, 1988), nor is it required for abnormally rapid loudness increment in recruitment (Moore et al., 1985). Nevertheless, there are counter arguments to the validity of these conclusions. First, psychophysical level discrimination thresholds at high intensities can be increased by adding flanking noise masks, which suggest that off-frequency responses (spread of excitation) do play a significant role in level discrimination, at least at high levels, in both normal and hearing-impaired listeners (Schroder et al., 1994). Secondly, one physiological study showed that in the CN, flanking noise masks do not completely mask the spread of influences to off-frequency BF bands (Palmer and Evans, 1982), which flies in the face of the basic assumption of those psychophysical arguments against all-BF encoding. Inhibitory sidebands are prevalent in the main types of VCN neurons and they are not eliminated by acoustic trauma (Section 3.5.2). With the presence of these inhibitory areas, 247 off-frequency VCN neurons can signal the increase in level of the tone by decrease in firing rates even if they are initially masked by one pass-band of a band-stop noise mask. We obtained an amount of data by using the band-stop noise masks similar to those used in psychophysical studies and confirmed this possibility (not shown due to limitedness of the amount of data). Therefore, it seems that neither side in the on-BF-versus-off-BF debate has overwhelming evidences to convince the other side. Thus we have to consider both possibilities when interpreting our findings. If levels are encoded in an on-BF way, then only certain non-primary-like VCN neurons, including chopper and unusual PSTH-type neurons, show possible correlates of loudness recruitment. However, if tone levels are encoded in the all-BF manner, then virtually all VCN neuronal types, including the overall type, show post-traumatic changes in their RLFs that may well underlie loudness recruitment. Given the different degrees of apparent recruitment shown by different VCN neuronal types, another important question that follows is which sub-populations of VCN neurons participate in level encoding. The globular bushy cells, which correspond to the PriN PSTH type project to the lateral superior olives, a pathway which is responsible for sound localization based on interaural level differences (ILD). Based on this fact, it is naïve, but compelling to suggest that the PriN neurons are important for level encoding. However, we didn’t find any evidence that PriN neurons show steepened rate-level slopes after acoustic trauma. Instead, they show similar shallower BF-tone rate-level slopes as the ANFs do (Fig. 3.23, Row B). Rhode and Smith (1986) described a type of Onset-Chopper neurons with unusually wide dynamic ranges for tonal stimuli (> 80 dB), which led them to suggest that these neurons might play important roles in level encoding. These neurons probably correspond to the radial stellate 248 cells (Doucet and Ryugo, 2006). A few similar units were recorded from in the current study. The wide dynamic ranges of these neurons mainly result from the change from onset to more sustained firing patterns with increasing tone level. Unfortunately, we didn’t sample any impaired onset units in the sub-edge regions impaired ears. Although we can by no means rule out the possibility of uneven sampling, neither can we dismiss the interesting possibility that firing patterns of onset neurons change to more sustained ones indistinguishable from PriN or Pri types following HL. If that is indeed the case, then the onset category should also be a unit type showing recruitment-like phenomenon. 4.6. Level discrimination and increased rate variability As reviewed in the Introduction section 1.2.1, increased internal noise has long been suggested as an explanation of normal or near normal sound intensity JNDs in recruiting ears, in the face of greater loudness-level slopes. However, few previous studies addressed this question. Heinz et al. (2005b) studied this question in the ANFs of the same animal model used in the current study and observed little differences between rate variability in response to tones before and after HL. Greater rate variability was observed in the VCN neurons after HL in the current study. However, it needs to be pointed that, not all impaired neurons show rate variability above the normal range. This is presumably due to the fact of different degrees of morphological, electrical and synaptic degenerations in different individual neurons after trauma. According to signal detection theory, the discriminability of two stimuli is determined by both the difference in mean and the variance of two signals. Therefore, the effect of increased variability is to decrease the discriminability of two levels and increase intensity limen. Therefore it seems that in impaired and recruiting ears, there are two or more counteracting 249 effects on level discrimination. The steepened single-neuron rate level functions in certain populations of neurons and abnormally fast spread of excitation increases the sensitivity to level differences; however the greater uncertainty between the response and the eliciting stimuli causes worse sensitivity to changes. This can explain the normal or near normal level difference limens reported in many studies on level discrimination in impaired ears (e.g., Florentine et al., 1993; Schroder et al., 1994; Neely & Allen, 1997). As for the origin of this variability, the increased rate uncertainties are expected from central neuronal degeneration and changes in synaptic properties (e.g., Morest et al., 1997; Lee et al., 2003; Wang and Manis, 2005) in the CN, which decreases the reliability of neural transmission. Following this line of reasoning, rate variability should be even larger in higher parts of the pathway, and the total effect at the cortical level immediately before cognition and motor output levels, should be more dramatic than that observed in the VCN in the current study. It also needs to be mentioned that mean discharge rate is only one aspect of neural responses, increased uncertainties in other aspects of the neural code, such as spike timing, are also very likely to be deteriorated following acoustic trauma. 4.7. Vowel level and spectral encoding With regard to encoding the level of a wide-band sound such as a vowel, it is well accepted that an increase in band-width of a flat-spectrum noise (beyond one critical band) leads to an increase in perceived loudness (Moore, 2004). The loudness of multi-tone complexes, which closely resemble vowels in spectrum, have been shown to be linearly related to the area of excitation and can be well modeled by a weighting function that weights different tone components at different frequencies linearly (Florentine et al., 1978; Cacace & Margolis, 1985; 250 Leibold et al., 2007). Based on these observations, it appears reasonable to assume that the level of a vowel is encoded by the discharge rates of neurons of all BFs. Results from vowel RLF analysis suggested that recruitment-like effects can be observed in vowel level encoding after acoustic trauma (Fig. 3.46). However, it needs to be mentioned that the grand average RLFs for the vowel generated by averaging across the feature alignments from F1 to T3 underestimates the summed firing rates in real-population situations, for two reasons, (1) neurons with BFs near the fundamental frequency (F0) also show a local maximum firing rate on the tonotopy (Miller, 1999a), but no feature alignment at F0 was used in our experiments; (2) as for tonal stimuli, neurons with BFs above the band of the vowel (> 5 kHz) also respond to the vowel at sufficiently high levels, because of spread of excitation. Therefore the grand average RLFs shown in Fig. 3.46 captures the response of only those neurons with BFs falling into the vowel frequency region at and above F1, and should be regarded as an “on-BF” coding scheme. If more playback rates were incorporated into the SMP method to simulate the response of neurons with wider BFs, we would be able to get the “all-BF” version. We expect the recruitment-like effect of HL to be stronger in that “all-BF” version, i.e., the rate-matching curve should show a steeper slope, due to broadened tuning in impaired ears. Accompanying the abnormal level encoding is a seriously deteriorated rate-place code for the spectrum, which is consistent with previous findings that in AN, spectral encoding of speech sounds significantly deteriorate after hearing losses (Miller et al., 1997). Threshold shifts can often account for a small part of the deterioration in speech perception performances in patients with SNHL. The abnormal rate-BF codes observed in this study, which mainly results from loss 251 of sharply tuned frequency tuning curves, can explain the additional deterioration in speech perception in SNHL (Moore, 2002). Synchrony code, i.e., a code based on phase locking to the components of the vowel, couldn’t be easily studied by the SMP method, because the SMP procedure altered the fundamental and harmonic frequencies, often bringing them out of the frequency range of phase-locking for VCN neurons. However, observation that the strength of phase locking to tones became weaker following acoustic trauma (Section 3.3.2) suggested that there should be deterioration in synchrony coding of the vowel (Blackburn and Sachs, 1990) as well, because phase locking to pure tones and more complex periodic stimuli are based on similar mechanisms, namely, the AC component of IHC receptor potentials and synaptic transmission between AN fibers and VCN neurons. Previous studies on synchrony code for the vowel /ε/ in impaired ears (e.g., Miller et al., 1997; Schilling et al., 1998) showed abnormal temporal-code representations. Given this pathological input from AN afferents, the compromised abilities of VCN neurons to follow the phase locking of AN fibers are expected to further exacerbate the temporal coding of speech sounds in impaired ears. 4.8. Abnormal rate-spectral encoding after acoustic trauma The previously observed properties of linear WFs in the linear-nonlinear weighting model (LNWM) were reproduced in the normal VCN neurons in the current study: (1) BF dependence of WF magnitudes and widths; (2) level dependence of WF magnitudes and widths; (3) the systematic relationship between WF excitation/inhibition patterns and VCN neuronal types; (4) the level dependence of goodness of fit. The current study demonstrated that the linear-nonlinear weight model can also be applied 252 to auditory neurons in impaired ears. However, it was also observed that goodness of fit was slightly worse in impaired neurons. This is presumably caused by stronger rate variability (internal noise) in impaired ears. Weight functions in impaired ears show several systematic differences from those neurons in normal ears: (1) smaller excitatory weights at and near BF; (2) larger excitatory bandwidths. These changes reflected the alterations in basic properties such as diminished driven firing rates and broadened tuning. Smaller magnitudes of weight functions lead to worse rate contrast in encoding sound spectrums. The increase in width of weight functions is associated with broadened psychophysical auditory filters. These effects, in synergy with greater rate variability, are expected to lead to significantly worsened rate-spectral representations in impaired ears, as exemplified by the abnormal rate-place codes of vowel spectrum. 253 V. Conclusions 1) Exposure to a 2-kHz centered narrowband noise induced hearing losses at low frequencies in cats. The variability of the extents and frequency spans of threshold shifts among individual animals was significant. 2) Ventral cochlear nucleus (VCN) neurons were recorded from normal-hearing and acoustically traumatized cats with a decerebrated unanesthetized preparation. VCN neurons in exposed ears showed threshold elevation and broadened frequency tuning, akin to the post-trauma alterations seen in the tuning properties of auditory nerve fibers after acoustic trauma. 3) VCN neurons were classified according to their post-stimulus-time histograms (PSTHs) in response to BF tones. There was no evidence that VCN PSTH types were radically altered after acoustic trauma. The major PSTH types, including primary-like, primary-like-with-notch and different subtypes of chopper and onset types were found in both normal-hearing and noise-exposed ears. However, there was evidence for the decrease in the prevalence of onset PSTH types in the VCN following trauma. A few neurons falling into the unusual PSTH category showed abnormal properties unseen in normal VCN neurons. 4) Basic quantitative properties of discharge patterns in response to BF-tone bursts, including latencies and spiking regularities were largely unaltered in most neuronal types. However, there was evidence that phase locking to BF tones in VCN units with BFs below 4 kHz was weakened after acoustic trauma. 5) In the noise-exposed ears, the BF-tone rate-level slopes were shallower-than-normal in Pri and PriN neurons, but were steeper-than-normal in non-primary-like groups including choppers and unusual neurons. The BF-tone rate-level slopes of low-BF phase-locking neurons were 254 unaltered after trauma. 6) For the pure tone stimuli, recruitment-like phenomenon, i.e., abnormally rapid increase of driven firing rates with increases in sound level, can be observed in the on-BF encoding in non-primary-like neurons including chopper and unusual neurons. However, the increase of Pri / PriN on-BF rates with level for on-BF tones was significantly shallower than normal, akin to the post-trauma alterations in AN fibers. 7)For all-BF encoding of tone levels, all VCN neuronal types in the exposed ear except Locker units showed recruitment-like phenomena. An important contributing factor to these phenomena was abnormally rapid spread of excitation to off-frequency BF channels with increasing tone level, which was due to broadened tuning curves of VCN neurons in impaired ears. There was evidence that the post-trauma increase in the speed of spread of excitation was greater in VCN neurons compared to in AN fibers. The recruitment-like phenomenon was seen for the overall VCN neuronal type as well. There was no evidence supporting the hypothesis that the threshold distribution of a population of auditory neurons with similar BFs becomes more compact after trauma. 8) Based on the rate-matching analysis, it was demonstrated that the abnormal tonal rate-level functions in VCN non-primary-like neurons and in VCN all-BF encoding were qualitatively and quantitative consistent with recruitment, suggesting that VCN might be the most caudal central auditory structure exhibiting explicit neural correlates of loudness recruitment. 9) The rate-level and rate-spectral representation of the steady-state vowel /ε/ were studied by using the spectral manipulation procedure. Abnormal rate-spectral encoding was seen in VCN neurons after acoustic trauma. Responses were dominated by response to the first 255 formant and rate-place contrasts were significantly poorer than normal. The summed rate activities increased with vowel level more rapidly in the impaired ear than in the normal-hearing ear, suggesting the existence of recruitment for wideband and complex stimuli. 10) The rate-spectral encoding of VCN neurons in exposed ears were studied by the linear-nonlinear weighting model and the random spectral shape stimuli. Reduced excitatory weights, broadened excitatory regions and diminished inhibitory regions distinguished the weight functions of impaired VCN neurons from normal ones. 11) Variability of discharge rates of VCN neurons in response to tonal stimuli increased after acoustic trauma. These observations provide novel clues to the pathophysiological mechanisms of abnormal level and speech perception in people with sensorineural hearing loss. 256 References Adams, J.C. 1997. Projections from octopus cells of the posteroventral cochlear nucleus to the ventral nucleus of the lateral lemniscus in cat and human. Auditory Neurosci. 3, 335-350. Bandyopadhyay, S. 2007. Spectral and temporal processing in the dorsal cochlear nucleus. Doctoral thesis, The Johns Hopkins University, Baltimore, MD. Banks, M.I., Sachs, M.B. 1991. Regularity analysis in a compartmental model of chopper units in the anteroventral cochlear nucleus. J. Neurophysiol. 65, 606-629. Barbour, D.L., Wang, X. 2003. Auditory cortical responses elicited in awake primates by random spectral stimuli. J. Neurosci. 23, 7194-7206. Bellingham, M.C., Lim, R., Walmsley, B. 1998. Developmental changes in EPSC quantal size and quantal content at a central glutamatergic synapse in rat. J. Physiol. 511, 861-869. Benson, C.G., Gross, J.S., Suneja, S.K., Potashner, S.J. 1997. Synaptophysin immunoreactivity in the cochlear nucleus after unilateral cochlear or ossicular removal. Synapse 25, 243-257. Blackburn, C.C., Sachs, M.B. 1989. Classification of unit types in the anteroventral cochlear nucleus: PST historgrams and regularity analysis. J. Neurophysiol. 62, 1303-1329. Blackburn, C.C., Sachs, M.B. 1990. The representation of the steady-state vowel sound /eh/ in the discharge patterns of cat ventral cochlear nucleus neurons. J. Neurophysiol. 63, 1191-1212. Bock, G.R., Frank, M.P., Steel, K.P. 1982. Preservation of central auditory function in the deafness mouse. Brain Res. 239, 608-612. Boettcher, R.A., Salvi, R.J. 1993. Functional changes in the ventral cochlear nucleus following acute acoustic stimulation. 1993 94, 2123-2134. 257 Bourk, T.R. 1976. Electrical responses of neural units in the anteroventral cochlear nucleus of the cat, Massachusetts Institute of Technology, Cambridge, MA. Bourk, T.R., Mielcarz, J.P., Norris, B.E. 1981. Tonotopic organization of the anteroventral cochlear nucleus of the cat. Hear. Res. 4, 215-241. Buus, S., Florentine, M. 2001. Growth of loudness in listeners with cochlear hearing losses: recruitment reconsidered. J. Assoc. Res. Otolaryngol. 3, 120-139. Cacace, A.T., Margolis, R.H. 1985. On the loudness of complex stimuli and its relationship to cochlear excitation. J. Acoust. Soc. Am. 78, 1568-1573. Young, E.D., Calhoun, B.M. 2005. Nonlinear modeling of auditory-nerve rate responses to wideband stimuli. J. Neurophysiol. 94, 4441-4454. Cant, N.B., Gatson, K.C. 1982. Pathways connecting the right and left cochlear nuclei. J. Comp. Neurol. 212, 313-326. Cant, N.B., Morest, D.K. 1984. The structural basis for stimulus coding in the cochlear nucleus. In: Berlin, C.I., (Ed.), Hearing science: recent advances. College-Hill Press, London. Cant, N.B., Hyson, R.L. 1992. Projections from the lateral nucleus of the trapezoid body to the medial superio olivary nucleus in the gerbil. Hear. Res. 58, 26-34. Cant, N.B., Benson, C.G. 2003. Parallel auditory pathways: projection patterns of the different neuronal populations in the dorsal and ventral cochlear nuclei. Brain. Res. Bull. 60, 457-474. Carlyon, R.P., Moore, B.C.J. 1984. Intensity discrimination: a severe departure from Weber's law. J. Acoust. Soc. Am. 76, 1369-1376. Carney, L.H. 1994. Spatiotemporal encoding of sound level: models for normal encoding and 258 recruitment of loudness. Hear. Res. 76, 31-44. Chatterjee, M., Zwislocki, J.J. 1998. Cochlear mechanisms of frequency and intensity coding. II. Dynamics and code for loudness. Hear. Res. 124, 170-181. Clarey, J.C., Barone, P., Imig, T.J. 1991. Physiology of thalamus and cortex. In: Popper, A.N., Fay, R.R., (Eds.), The mammalian auditory pathway: neurophysiology. Springer-Verlag, New York. Colburn, H.S., Carney, L.H., Heinz, M.G. 2003. Quantifying the information in auditory-nerve responses for level discrimination. J. Assoc. Res. Otolaryngol. 04, 294-311. Davis, G.W., Bezprozvanny, I. 2001. Maintaining the stability of neural function: a homeostatic hypothesis. Annu. Rev. Physiol. 63, 847-869. Delgutte, B., Kiang, N.Y.-S. 1984. Speech coding in the auditory nerve: I. Vowel-like sounds. J. Acoust. Soc. Am. 75, 866-878. Deng, L., Geisler, C.D. 1987. A composite auditory model for processing speech sounds. J. Acoust. Soc. Am. 82, 2001-2012. Doucet, J.R., Ryugo, D.K. 1997. Projecions from the ventral cochlear nucleus to he dorsal cochlear nucleus in rats. J. Comp. Neurol. 385, 245-264. Doucet, J.R., Ryugo, D.K. 2006. Structural and functional classes of multipolar cells in the ventral cochlear nucleus. Anat. Rec. Pt. A 288A, 331-344. Efron, B., Tibshrani, R.J. 1993. An introduction to bootstrap Chapman & Hall. Ehret, G., Merzenich, M.M. 1988. Neuronal discharge rate is unsuitable for encoding sound intensity at the inferior-colliculus level. Hear. Res. 35, 1-8. Ertel, J.E., Fowlkes, E.B. 1976. Some algorithms for linear spline and piecewise multiple linear 259 regression. J. Am. Stat. Assoc. 71, 640-648. Evans, E.F. 1975. The sharpening of cochlear frequency selectivity in the normal and abnormal cochlea. Audiology 14, 419-442. Fant, C.G.M. 1960. Acoustic theory of speech production The Hague, Mouton. Ferragamo, M.J., Golding, N.L., Gardner, S.M., Oertel, D. 1998. Golgi cells in the superficial granule cell domain overlying the ventral cochlear nucleus: morphology and electrophysiology in slices. J. Comp. Neurol. 400, 519-528. Ferragamo, M.J., Oertel, D. 2002. Octopus cells of the mammalian ventral cochlear nucleus sense the rate of depolarization. J. Neurophysiol. 87, 2262-2270. Florentine, M., Buus, S., Bonding, P. 1978. Loudness of complex sounds as a function of the standard stimulus and the number of components. J. Acoust. Soc. Am. 64, 1036-1040. Florentine, M., Reed, C.M., Rabinowitz, W.M., Braida, L.D., Durlach, N.I., Buus, S. 1993. Intensity discrimination in listeners with sensorineural hearing loss. J. Acoustic. Soc. Am. 34, 2575-2585. Francis, H.W., Manis, P.B. 2000. Effects of deafferetiation on the electrophysiology of ventral cochlear nucleus neurons. Hear. Res. 149, 91-105. Franck, K.R. 1994. Anditory-nerve responses to speech sounds in partially deafened cats. Master's thesis, The Johns Hopkins University, Baltimore, MD. Furman, A.C., Avissar, M., Saunders, J.C. 2006. The effect of intense sound exposure on phase locking in the check (Gallus domesticus) cochlear nerve. Eur. J. Neurosci. 24, 2003-2010. Gaese, B.H., Ostwald, J. 2001. Anesthesia changes frequency tuning of neurons in the rat primary auditory cortex J. Neurophysiol. 86, 1062-1066. 260 Gardner, S.M., Trussell, L.O., Oertel, D. 1999. Time course and permeation of synaptic AMPA receptors in cochlear nuclear neurons correlate with input. J. Neurosci. 19, 8721-8729. Gentschev, T., Sotelo, C. 1973. Degenerative patterns in the ventral cochlear nucleus of the rat after primary deafferentiation. An ultrastructural study. . Brain Res. 62, 37-60. Gerken, G.M. 1979. Central denervation hypersensitivity in the auditory system of the cat. J. Acoust. Soc. Am. 66, 721-727. Goldberg, J.M., Brown, P.B. 1969. Response of binaural neurons of dog superior olivary complex to dichotic tonal stimuli: some physiological mechanisms of sound locailzation. J. Neurophysiol. 32, 613-636. Harrison, R.V. 1981. Rate-versus-intensity functions and related AP responses in normal and pathological guinea pig and human cochleas. J. Acoust. Soc. Am. 70, 1036-1044. Heil, P., Rajan, R., Irvine, D.R.F. 1994. Topographic representation of tone intensity along the isofrequency axis of cat primary auditory cortex. Hear. Res. 76, 188-202. Heinz, M.G., Young, E.D. 2004. Response growth with sound level in auditory-nerve fibers after noise-induced hearing loss. J. Neurophysiol. 91, 784-795. Heinz, M.G., Issa, J.B., Young, E.D. 2004. Level coding in he auditory nerve following noise-induced hearing loss, Assoc. for Res. in Otolaryngol. Abstracts. pp. 324. Heinz, M.G., Issa, J.B., Young, E.D. 2005. Auditory-nerve rate responses are inconsistent with common hypotheses for the neural correlates of loudness recruitment. JARO 6, 91-105. Heinz, M.G., Scepanovic, D., Issa, J.B., Sachs, M.B., Young, E.D. 2005. Normal and impaired level encoding: effects of noise-induced hearing loss on auditory-nerve responses. In: de Cheveigne, A., McAdams, S., Collet, L., (Eds.), Auditory signal processing: physiology, 261 psychoacoustics and models. Springer-Verlag, New York. Hellman, R.P., Meiselman, C.H. 1990. Loudness relations for individuals and groups in normal and impaired hearing. J. Acoust. Soc. Am. 88, 2596-2606. Hellman, R.P., Meiselman, C.H. 1993. Rate of loudness growth of pure tones in normal and impaired hearing. J. Acoust. Soc. Am. 93, 966-975. Hellman, R.P. 1994. Relation between the growth of loudness and high-frequency excitation. J. Acoust. Soc. Am. 96, 2655-2663. Huizing, H.C. 1948. The symptom of recruitment and speech intelligibility. Acta. Oto-laryngol. 36, 346-355. Illing, R.-B., Kraus, K.S., Meidinger, M.A. 2005. Reconnecting neuronal networks in the auditory brainstem following unilateral deafening. Hear. Res. 206. Johnson, D.H. 1980. The relationship between spike rate and synchrony in response of auditory nerve fibers to single tones. J. Acoust. Soc. Am. 68, 1115-1122. Kiang, N.Y.-S., Watanabe, T., Thomas, E.C., Clark, L.F. 1965. Discharge patterns of single fibers in the cat's auditory nerve MIT Press, Cambridge, MA. Kiang, N.Y.S., Moxon, E.C., Levine, R.A. 1970. Auditory nerve activity in cats with normal and abnormal cochleas. In: Wolstenholme, G.E.W., Knight, T., (Eds.), Sensorineural hearing loss. Churchill, London. pp. 241-273. Kim, J.J., Gross, J.S., Morest, D.K., Potashner, S.J. 2004. Quantitative study of degeneration and new growth of axons and synaptic endings in the chinchilla cochlear nucleus after acoustic overstimulation. J. Neurosci. Res. 77, 827-842. Kuwada, S., Batra, R., Stanford, T.R. 1989. Monaural and binaural response properties of neurons 262 in the inferior colliculus of the rabbit: effects of sodium pentobarbital. J. Neurophysiol. 61, 269-282. Lai, Y.C., Winslow, R.L., Sachs, M.B. 1994. A model of selective processing of auditory-nerve inputs by stellate cells of the antero-ventral cochlear nucleus. J. Comput. Neurosci. 1, 167-194. Langers, D., Backes, W., van Dijk, P. 2006. Brain activation in relation to sound intensity and loudness. In: Kollmeier, B., Klump, G., Hohmann, V., Langermann, U., Mauermann, M., Uppenkamp, S., Verhey, J., (Eds.), 14th International Symposium on Hearing, Cloppenburg, Germany. Langers, D.R.M., van Dijk, P., Schoenmaker, E.S., Bakes, W.H. 2007. fMRI activation in relation to sound intensity and loudness. NeuroImage 35, 709-718. Le Prell, G.S., Sachs, M.B., May, B.J. 1996. Representation of vowel-like spectra by discharge rate responses of auditory-nerve fibes. Aud. Neurosci. 2, 275-288. Lee, D.J., Cahill, H.B., Ryugo, D.K. 2003. Effects of congenital deafness in the cochlear nuclei of Shaker-2 mice: an ultrastructural analysis of synapse morphology in the endbulbs of Held. J. Neurocytol. 32, 229-243. Leibold, L.J., Tan, H., Khaddam, S., Jesteadt, W. 2007. Contributions of individual components to the overall loudness of a multitone complex. J. Acoust. Soc. Am. 121, 2822-2831. Lesperance, M.M., Helfert, R.H., Altschuler, R.A. 1995. Deafness induced cell size changes in rostral AVCN of the guinea pig. Hear. Res. 86, 77-81. Liberman, M.C. 1978. Auditory-nerve response from cats raised in a low-noise chamber. J. Acoust. Soc. Am. 63, 442-455. 263 Liberman, M.C., Kiang, N.Y.-S. 1978. Acoustic trauma in cats. Cochlear pathology and auditory-nerve activity. Acta Otolaryngol. Suppl. 358, 1-63. Liberman, M.C. 1984. Single-neuron labeling and chronic cochlear pathology. I. Threshold shift and characteristic-frequency shift. Hear. Res. 16, 33-41. Liberman, M.C., Dodds, L.W. 1984. Single-neuron labeling and chronic cochlear pathology. II. Stereocilia damage and alterations of spontaneous discharge rate s. Hear. Res. 16, 43-53. Liberman, M.C., Kiang, N.Y.-S. 1984. Single-neuron labeling and chronic cochlear pathology. IV. Stereocilia damage and alterations in rate- and phase-level functions. Hear. Res. 1984, 75-90. Liberman, M.C., Dodds, L.W. 1984b. Single-neuron labeling and chronic cochlear pathology. III. Stereocilia damage and alterations of thresholds tuning curves. Hear. Res. 16, 55-74. Lonsbury-Martin, B.L., Martin, G.K. 1981. Effects of moderately intense sound on auditory sensitivity in rhesus monkeys: behavioral and neural observations. J. Neurophysiol. 46, 563-586. Luo, L., Ryan, A.F., Saint Marie, R.L. 1999. Cochlear ablation alters acoustically induced c-fos mRNA expression in the adult rat auditory brainstem. J. Comp. Neurol. 404, 271-283. Ma, W.-L., Young, E.D. 2006. Loss of inhibition, and decreased spontaneous rates in dorsal cochlear nucleus following acoustic trauma. Hear. Res. 216-217, 176-188. Manis, P.B., Marx, S.O. 1991. Outward currents in isolated ventral cochlear nucleus neurons. J. Neurosci. 11, 2865-2880. May, B.J., Sachs, M.B. 1992. Dynamic range of neural rate responses in the ventral cochlear nucleus of awake cats. J. Neurophysiol. 68, 1589-1602. 264 May, B.J., LePrell, G.S., Sachs, M.B. 1998. Vowel representations in the ventral cochlear nucleus of the cat: effects of level, background noise, and behavioral state. J. Neurophysiol. 79, 1755-1767. Melcher, J.R. 1993. The cellular generators of brainstem auditory evoked potential, Massachusetts Institute of Technology, Cambridge, MA. Michler, S.A., Illing, R.-B. 2002. Acoustic trauma induces reemergence of the growth- and plasticity-associated protein GAP-43 in the rat auditory brainstem. J. Comp. Neurol. 451, 250-266. Miller, R.L., Schilling, J.R., Franck, K.R., Young, E.D. 1997. Effects of acoustic trauma on the representation of the vowel /epsilon/ in cat auditory nerve fibers. J. Acoust. Soc. Am. 101, 3602-3616. Miller, R.L., Calhoun, B.M., Young, E.D. 1999a. Contrast enhancement improves the representation of /eh/-like vowels in hearing-impaired auditory nerve. J. Acoust. Soc. Am. 106, 2693-2708. Miskolczy-Fodor, F. 1960. Relationship between loudness and duration of tonal pulses. III. Response in cases of abnormal loudness function. J. Acoust. Soc. Am. 32, 486-492. Moore, B.C.J., Glasberg, B.R., Hess, R.F., Birchall, J.P. 1985. Effects of flanking noise on the rate of growth of loudness of tones in normal and recruiting ears. J. Acoust. Soc. Am. 77, 1505-1513. Moore, B.C.J., Wojtczak, M., Vickers, D.A. 1996. Effects of loudness recruitment on the perception of amplitude modulation. J. Acoust. Soc. Am. 100, 481-489. Moore, B.C.J. 2002. Perceptual consequences of cochlear damage Oxford University Press, 265 Oxford. Moore, B.C.J. 2004. Testing the concept of softness imperception: loudness near threshold for hearing-impaired ears. J. Acoust. Soc. Am. 115, 3103-3111. Morest, D.K., Kim, J., Bohne, B.A. 1997. Neuronal and transneuronal degeneration of auditory axons in the brainstem after cochlear lesions in the chinchilla: cochleotopic and non-cochleotopic patterns. Hear. Res. 103, 151-168. Morita, T., Naito, Y., Nagamine, T., Fujiki, N., Shibasaki, H., Ito, J. 2003. Enhanced activation of the auditory cortex in patients with inner-ear hearing impairment: a magnetoencephalographic study. Clin. Neurophysiol. 114, 851-859. Neely, S.T., Allen, J.B. 1997. Relation between the rate of growth of loudness and the intensity DL. In: Jesteadt, W., (Ed.), Modeling sensorineural hearing loss. Erlbaum, Mahwah, NJ. pp. 213-222. Ngan, E.M., May, B.J. 2001. Relationship between the auditory brainstem response and auditory nerve thresholds in cats with hearing loss. Hear. Res. 156, 44-52. Norena, A.J., Tomita, M., Eggermont, J.J. 2003. Neural changes in cat auditory cortex after a transient pure-tone trauma. J. Neurophysiol. 90, 2387-2401. Norena, A.J., Eggermont, J.J. 2005. Enriched acoustic environment after noise trauma reduces hearing loss and prevents cortical map reorganization. J. Neurosci. 25, 699-705. Oertel, D. 1983. Synaptic responses and electrical properties of cells in brain slices of mouse anteroventral cochlear nucleus. J. Neurosci. 3, 2043-2053. Oleskevich, S., Walmsley, B. 2002. Synaptic transimission in the auditory brainstem of normal and congenitally deaf mice. J. Physiol. 540, 447-455. 266 Osen, K.K. 1970. Course and termination of primary afferents in the cochlear nuclei of the cat. An experimental study. Arch. Ital. Biol. 108, 21-51. Ostapoff, O.P., Morest, D.K. 1991. Synaptic organization of globular bushy cells in the ventral cochlear nucleus of the cat: a quantitative study. . J. Comp. Neurol. 314, 598-613. Palmer, A.R., Evans, E.F. 1982. Intensity coding in the auditory periphery of the cat: responses of cochlear nerve and cochlear nucleus neurons to signals in the presence of bandstop masking noise. Hear. Res. 7, 305-323. Palmer, A.R., Moorjani, P.A. 1993. Responses to speech signals in the normal and pathological peripheral auditory system. Prog. Brain Res. 97, 107-115. Patuzzi, R., Sellick, P.M. 1983. A comparison between basilar membrane and inner hair cell receptor potential input-output functions in the guinea pig cochlea. J. Acoust. Soc. Am. 74, 1734-1741. Pfingst, B.E., O'Connor, T.A. 1981. Characteristics of neurons in auditory cortex of monkeys performing a simple auditory task. J. Neurophysiol. 45, 16-34. Pickles, J.O. 1979. Psychophysical frequency resolution in the cat as determined by simultaneous masking and its relation to auditory-nerve resolution. J. Acoust. Soc. Am. 66, 1725-1732. Pickles, J.O. 1988. An Introduction to the Physiology of Hearing Academic Press, London. Popelar, J., Syka, J., Berndt, H. 1987. Effect of noise on auditory evoked responses in awake guinea pigs. Hearing Res. 26, 239-247. Pugh, J.E., Jr., Moody, D.B., Anderson, D.J. 1979. Electrocochleography and experimentally induced loudness recruitment. Arch. Otorhinolaryngol. 224, 241-255. Qiu, C., Salvi, R.J., Ding, D., Burkard, R. 2000. Inner hair cell loss leads to enhanced response 267 amplitudes in auditory cortex of unanesthetized chinchillas: evidence for increased system gain. Hear. Res. 139, 153-171. Rajan, R., Irvine, D.R., Wise, L.Z., P., H. 1993. Effect of unilateral partial cochlear lesion in adult cats on the representation of lesioned and unlesioned cochleas in primary auditory cortex. J. Comp. Neurol. 338, 17-49. Ramachandran, R., Davis, K.A., May, B.J. 1999. Single-unit responses in the inferior colliculus of decerebrate cats. I. Classification based on frequency response maps. J. Neurophysiol. 82, 152-163. Redd, E.E., Cahill, H.B., Pongstapron, T., Ryugo, D.K. 2002. The effects of congenital deafness on auditory nerve synapses: Type I and Type II multipolar cells in the anteroventral cochlear nucleus of cats. J. Assoc. Res. Otolaryngol. 3, 403-417. Rees, A., Palmer, A.R. 1988. Rate-intensity functions and their modification by broadband noise for neurons in guinea pig inferior colliculus. J. Acoust. Soc. Am. 83, 1488-1498. Relkin, E.M., Doucet, J.R. 1997. Is loudness simply proportional to the auditory nerve spike count? J. Acoust. Soc. Am. 101, 2735-2740. Rhode, W.S., Smith, P.H., Oertel, D. 1983. Physiological response properties of cells labeled intracellularly with horseradish peroxidase in cat dorsal cochlear nucleus. J. Comp. Neurol. 213, 426-447. Rhode, W.S., Oertel, D., Smith, P.H. 1983. Physiological response properties of cells labeled intracellularly with horseradish peroxidase in cat ventral cochlear nucleus. J. Comp. Neurol. 213, 448-463. Rhode, W.S., Smith, P.H. 1986. Encoding time and intensity in the ventral cochlear nucleus of the 268 cat. J. Neurophysiol. 56, 261-286. Rhode, W.S., Greenberg, S. 1991. Physiology of the cochlear nuclei. In: Popper, A.N., Fay, R.R., (Eds.), The mammalian auditory pathway: neurophysiology. Springer, New York. Rice, J.J., May, B.J., Spirou, G.A., Young, E.D. 1992. Pinna-based spectral cues for sound localization in cat. Hear. Res. 58, 132-152. Robles, L., Ruggero, M.A. 2001. Mechanics of the mammalian cochlea. Physiol. Rev. 81, 1305-1352. Rouiller, E., de Ribaupierre, Y., Morel, A., de Ribaupierre, F. 1983. Intensity functions of single unit responses to tone in the medial geniculate body of cat. Hear. Res. 11, 235-247. Ryugo, D.K., Sento, S. 1991. Synaptic connections of auditory nerve in cats: relationship between endbulbs of Held and spherical bushy cells. J. Comp. Neurol. 305, 35-48. Sachs, M.B., Abbas, P.J. 1974. Rate versus level functions for auditory nerve fibers in cats: tone-burst stimuli. J. Acoust. Soc. Am. 56, 1835-1847. Sachs, M.B., Young, E.D. 1979. Encoding of steady-state vowels in the auditory nerve: representation in terms of discharge rates. J. Acoust. Soc. Am. 66, 470-479. Salvi, R.J., Hamernik, R.P., Henderson, D. 1983. Response patterns of auditory nerve fibers during temporary threshold shift. Hear. Res. 10, 37-67. Salvi, R.J., Henderson, D., Hamernik, R.P., Ahroon, W.A. 1983. Neural correlates of sensorineural hearing loss. Ear Hear. 4, 115-129. Salvi, R.J., Saunders, J.C., Gratton, M.A., Arehole, S., Powers, N. 1990. Enhanced evoked response amplitudes in the inferior colliculus of the chinchilla following acoustic trauma. Hearing Res. 50, 245-258. 269 Salvi, R.J., Wang, J., Ding, D. 2000. Auditory plasticity and hyperactivity following cochlear damage. Hear. Res. 147, 261-274. Saunders, J.C., Bock, G.R., James, R., Chen, C.-S. 1972. Effects of priming for audiogenic seizure on auditory evoked responses in the cochlear nucleus and inferior colliculus of BALB/c mice. Exp. Neurol. 37, 388-394. Schaette, R., Kempter, R. 2006. Development of tinnitus-related neuronal hyperactivity through homeostatic plasticity after hearing loss: a computational model. Eur. J. Neurosci. 143, 103-109. Schilling, J.R., Miller, R.L., Sachs, M.B., Young, E.D. 1998. Frequency shaped amplification changes the neural representation of speech with noise-induced hearing loss. Hear. Res. 117, 57-70. Schmiedt, R.A., Zwislocki, J.J. 1980. Effects of hair cell lesion on responses of cochlear nerve fibers. II. Single- and two-tone intensity functions in relation to tuning curves. J. Neurophysiol. 43, 1390-1405. Schroder, A.C., Viemeister, N.F., Nelson, D.A. 1994. Intensity discrmination in normal-hearing and hearing-impaired listeners. J. Acoustic. Soc. Am. 96, 2683-2693. Schroder, A.C., Viemeister, N.F., Nelson, D.A. 1994. Intensity discrimination in normal-hearing and hearing-impaired listeners. J. Acoust. Soc. Am. 96, 2683-2693. Schofield, B.R., Cant, N.B. 1996 Projections from the ventral cochlear nucleus to the inferior colliculus and the contralateral cochlear nucleus in guinea pigs. Hear. Res. 102, 1-14. Sellick, P.M., Patuzzi, R., Johnstone, B.M. 1982. Measurement of basilar membrane motion in the guinea pig using the Mossbauer technique. J. Acoust. Soc. Am. 72, 131-141. 270 Sewell, W.F. 1984. Furosemide selectively reduces one component in rate-level functions from auditory-nerve fibers. Hear. Res. 15, 69-72. Shofner, W.P., Young, E.D. 1985. Excitatory/inhibitiory responses in the cochlear nucleus: relationships to discharge patterns and responses to electrical stimulation of the auditory nerve. J. Neurophysiol. 54, 917-939. Shofner, W.P., Young, E.D. 1985. Excitatory/inhibitiory responses in the cochlear nucleus: relationships to discharge patterns and responses to electrical stimulation of the auditory nerve. J. Neurophysiol. 54, 917-939. Smith, R.L. 1988. Encoding sound intensity by auditory neurons. In: Edelman, G.M., Gall, W.E., Cowan, W.M., (Eds.), Auditory function: neurobiological bases of hearing. John Wiley & Sons, New York. pp. 243-274. Sokolich, W.G. 1977. Improved acoustic system for auditory research. J. Acoust. Soc. Am. Suppl. 62, S12. Spirou, G.A., Davis, K.A., Nelken, I., Young, E.D. 1999. Spectral integration by type II interneurons in dorsal cochlear nucleus. J. Neurophysiol. 82, 648-663. Stebbins, W.C. 1966. Auditory reaction time and the derivation of equal loudness contours for the monkey. J. Exp. Anal. Behav. 9, 135-142. Sterbing, S.J., Schrott-Fischer, A. 2002. Electrophysiological characteristics of inferior colliculus neurons in mutant mice with hereditary absence of cochlear outer hair cells. Hearing Res. 170, 179-189. Suga, N. 1977. Amplitude spectrum representation in the Dopper-shifted-CF processing area of 271 the auditory cortex of the mustache bat. Science 196, 64-67. Sumner, C.J., Tucci, D.L., Shore, S.E. 2005. Responses of ventral cochlear nucleus neurons to contralateral sound after conductive hearing loss. J. Neurophysiol. 94, 4234-4243. Suneja, S.K., Benson, C.G., Potashner, S.J. 1998. Glycine receptors in adult guinea pig brain stem auditory nuclei: regulation after unilateral cochlear ablation. Exp. Neurol. 154, 473-488. Syka, J., Rybalko, N., Popelar, J. 1994. Enhancement of the auditory cortex evoked responses in awake guinea pigs after noise exposure. Hearing Res. 78, 158-168. Syka, J. 2002. Plastic changes in the central auditory system after hearing loss, restoration of function, and during learning. Physiol. Rev. 82, 601-636. Szczepaniak, W.S., Moller, A.R. 1996. Evidence of neuronal plasticity within the inferior colliculus after noise exposure: a study of evoked potentials in the rat. Electroencephalogr. Clin. Neurophysiol. 100, 158-164. Tolbert, L.P., Morest, D.K. 1982. The neuronal architecture of the anteroventral cochlear nucleus of the cat in the region of the cochlear nerve root: electron microscopy. Neuroscience 7, 3030-3053. Turrigiano, G.G. 1999. Homeostatic plasticity in neuronal networks: the more things change, the more they stay the same. Trend Neurosci. 22, 221-222. Vale, C., Sanes, D.H. 2002. The effect of bilateral deafness on excitatory and inhibitiory synaptic strength in the inferior colliculus. Euro. J. Neurosci. 16, 2394-2404. Viemeister, N.F. 1988. Intensity coding and the dynamic range problem. Hear. Res. 34, 267-294. Wang, Y., Manis, P.B. 2005. Synaptic transmission at the cochlear nucleus endbulb synapse during age-related hearing loss in mice. J. Neurophysiol. 94, 1814-1824. 272 Wang, Y., Manis, P.B. 2006. Temporal coding by cochlear nucleus bushy cells in DBA/2J mice with early onset hearing loss. J. Assoc. Res. Otolaryngol. 7, 412-424. Westerman, L.A., Smith, R.L. 1988. A diffusion model of the transient response of the cochlear inner hair cell synapse. J. Acoust. Soc. Am. 83, 2266-2276. Winslow, R.L., Sachs, M.B. 1988. Single-tone intensity discrimination based on auditory-nerve rate responses in backgrounds of quiet, noise, and with stimuluation of crossed the olivocochlear bundle. Hear. Res. 35, 165-189. Willott, J.F., Jackson, L.M., Hunter, K.P. 1987. Morphometric study of the anteroventral cochlear nucleus of two mouse models of presbycusis. J. Comp. Neurol. 260, 472-480. Willott, J.F., Bross, L.S. 1990. Morphology of the octopus cell area of the cochlear nucleus in young and aging C57BL/6J and CBA/J mice. J. Comp. Neurol. 300, 61-81. Wong, J.C., Miller, R.L., Calhoun, B.M., Sachs, M.B., Young, E.D. 1998. Effects of high sound levels on responses to the vowel /epsilon/ in cat auditory nerve. Hear. Res. 123, 61-77. Yates, G.K., Winter, I.M., Robertson, D. 1990. Basilar membrane nonlinearity determines auditory nerve rate-intensity functions and cochlear dynamic range. Hear. Res. 45, 203-219. Young, E.D., Brownell, W.E. 1976. Responses to tones and noise of single cells in dorsal cochlear nucleus of unanesthetized cats. J. Neurophysiol. 39, 282-300. Young, E.D., Sachs, M.B. 1979. Representation of steady-state vowels in the temporal aspects of the discharge patterns of populations of auditory-nerve fibers. J. Acoust. Soc. Am. 66, 1381-1403. Young, E.D. 1988. Regularity and latency of units in ventral cochlear nucleus: implications for 273 unit classification and generation of response properties. J. Neurophysiol. 60, 1-29. Young, E.D., Davis, K.A. 2002. Circuitry and functions of the dorsal cochlear nucleus. In: Oertel, D., Fay, R.R., Ropper, A.N., (Eds.), Integrative functions in the mammalian auditory pathway. Springer. pp. 160-206. Young, E.D., Oertel, D. 2004. Cochlear nucleus. In: Shepherd, G.M., (Ed.), The synaptic organization of the brain. Oxford University Press, New York. pp. 125-163. Yu, J.J., Young, E.D. 2000. Linear and nonlinear pathways of spectral information transmission in the cochlear nucleus. PNAS 97, 11780-11786. Yu, J.J. 2003. Spectral information encoding in the cochlear nucleus and inferior colliculus: a study based on the random spectral shape method. Doctoral thesis, The Johns Hopkins University, Baltimore, MD. Zeng, F.-G., Turner, C.W. 1991. Binaural loudness matches in unilaterally impaired listeners. Quart. J. Exp. Psychol. 43A, 565-583. Zhang, M., Zwislocki, J.J. 1995. OHC response recruitment and its correlation with loudness recruitment. . Hear. Res. 85, 1-10. Zilany, M.S.A., Bruce, I.C. 2006. Modeling auditory-nerve responses for high sound pressure levels in the normal and impaired auditory periphery. J. Acoust. Soc. Am. 120, 1446-1466. Zurita, P., Villa, A.E., de Ribaupierre, Y., de Ribaupierre, F., Rouiller, E.M. 1994. Changes of single unit activity in the cat's auditory thalamus and cortex associated to different anesthetic conditions. Neurosci. Res. 19, 303-316. 274 Biography Shanqing Cai was born in Shanghai, China in 1983. He attended Shanghai Cao Yang No.2 High School from 1994 to 2001 and graduated as the valedictorian in 2001. From 2001 to 2005, he studied at the Department of Biomedical Engineering at Tsinghua University, Beijing. Shanqing Cai’s Bachelor’s thesis was on the psychophysics and image processing techniques of the proposed electronic visual prosthesis for restoring functional vision to the blind, and he was the co-author of a few journal papers and conference abstracts on the related topics. He received a Bachelor’s degree in engineering with high honors in the year 2005. From 2005 to 2007, Shanqing Cai studied at the Master’s degree program at the Department of Biomedical Engineering at Johns Hopkins University. After finishing his studies at the Johns Hopkins University, Shanqing Cai is going to study at the Speech and Hearing Bioscience and Technology Ph.D. program at the Harvard-MIT Division of Health Science and Technology at Massachusetts Institute of Technology, to study the neural mechanisms of normal and impaired speech production. 275