Abstract
A common complaint of older adults is difficulty understanding speech, particularly in challenging listening conditions. Accumulating evidence suggests that these difficulties may reflect a loss and/or dysfunction of auditory nerve (AN) fibers. We used a novel approach to study age-related changes in AN structure and several measures of AN function, including neural synchrony, in 58 older adults and 42 younger adults. AN activity was measured in response to an auditory click (compound action potential; CAP), presented at stimulus levels ranging from 70 to 110 dB pSPL. Poorer AN function was observed for older than younger adults across CAP measures at higher but not lower stimulus levels. Associations across metrics and stimulus levels were consistent with age-related AN disengagement and AN dyssynchrony. High-resolution T2-weighted structural imaging revealed age-related differences in the density of cranial nerve VIII, with lower density in older adults with poorer neural synchrony. Individual differences in neural synchrony were the strongest predictor of speech recognition, such that poorer synchrony predicted poorer recognition of time-compressed speech and poorer speech recognition in noise for both younger and older adults. These results have broad clinical implications and are consistent with an interpretation that age-related atrophy at the level of the AN contributes to poorer neural synchrony and may explain some of the perceptual difficulties of older adults.
SIGNIFICANCE STATEMENT Differences in auditory nerve (AN) pathophysiology may contribute to the large variations in hearing and communication abilities of older adults. However, current diagnostics focus largely on the increase in detection thresholds, which is likely because of the absence of indirect measures of AN function in standard clinical test batteries. Using novel metrics of AN function, combined with estimates of AN structure and auditory function, we identified age-related differences across measures that we interpret to represent age-related reductions in AN engagement and poorer neural synchrony. Structure-function associations are consistent with an explanation of AN deficits that arise from age-related atrophy of the AN. Associations between neural synchrony and speech recognition suggest that individual and age-related deficits in neural synchrony contribute to speech recognition deficits.
Introduction
There is growing evidence that many of the functional difficulties associated with age-related hearing loss are not solely because of elevated thresholds and declines in the auditory periphery (cochlea) but are exacerbated by changes within the central nervous system, including the loss and/or inactivity (disengagement) of auditory nerve (AN) fibers. Age-related AN fiber loss/dysfunction exceeds cochlear hair cell loss (Makary et al., 2011; Wu et al., 2019). Further, synaptic loss and myelin abnormalities often occur before significant loss of AN fibers (Xing et al., 2012) and may affect AN function. Because of the inability to directly characterize AN function in humans, the impact of AN loss and/or dysfunction in older adults on suprathreshold auditory function, including speech recognition, is largely unknown.
The AN is biodiverse, with AN fibers often categorized by their spontaneous firing rate (SR; Liberman, 1978). Excessive noise exposure and aging are thought to disproportionately affect low-SR fibers (Schmiedt et al., 1996; Furman et al., 2013; Kujawa and Liberman, 2015). Compared with high-SR fibers, low-SR fibers have higher thresholds, larger dynamic ranges, better preservation of timing information, longer first-spike latencies, and slower conduction velocities (Liberman, 1978; Heil and Irvine, 1997; Bourien et al., 2014). In addition to engagement of low-SR fibers with increasing stimulus levels, neural synchrony increases (Heil and Irvine, 1997; Heil, 2004), contributing to larger AN response amplitudes at higher stimulus levels (Harris et al., 2018). These observations guided the development of a novel multimetric approach for characterizing specific patterns of AN function (Fig. 1; Harris et al., 2018).
Multimetric approach to characterizing AN function. A representative CAP waveform from one younger participant illustrating the four time-domain AN metrics, with time from 0 to 5 ms on the x-axis, and amplitude (µV) on the y-axis. Phase locking value (PLV) is shown below in the heat map, with time from −1 to 10 ms on the x-axis and frequency (Hz) on the y-axis. The following AN metrics are illustrated: peak amplitude (µV), peak latency (ms); onset latency (ms), half-width latency (ms); and PLV. The value of the PLV above baseline (green) is indicated by color (yellow, red). PLV was extracted as the peak within a 2-ms time window surrounding the peak of N1 and between 625 and 3120 Hz (black rectangle).
Multimetric approach to characterizing AN disengagement and AN dyssynchrony
AN function may be differentially affected by multiple underlying pathologies. A differential approach examining changes across suprathreshold compound action potential (CAP) metrics of peak amplitude, half-width latency of the response, and an estimate of neural synchrony may better characterize individual differences in AN function than a single measures of peak amplitude. For example, in a healthy population of AN fibers, with increasing stimulus level, recruitment of low-SR fibers increases and neural synchrony improves, both contributing to larger response amplitudes (Fig. 2). Associations across AN metrics (peak amplitude, half-width, and neural synchrony) at higher and lower stimulus levels are hypothesized to differentiate normal AN function from AN dysfunction that reflects reduced numbers of AN fibers activated (AN disengagement) and/or AN dyssynchrony (Fig. 3). We define AN disengagement as a decrease in the number of AN fibers activated because of a loss of AN fibers, reduced endocochlear potential, or cochlear synaptopathy. Larger peak amplitudes and longer half-width latencies may be interpreted as indicative of more diversity across AN fibers, with active high-SR and low-SR fibers resulting in larger and wider responses. We define AN dyssynchrony as a reduced synchrony in the summated activity of AN fibers across trials. AN dyssynchrony is hypothesized to result from a loss or disruption of myelin or cochlear synaptopathy. Hypothesized associations across metrics and the extent to which these patterns may reflect AN disengagement, AN dyssynchrony, or a combination of AN disengagement and AN dyssynchrony are provided in Figure 3.
Schematic depicting the changes in N1 peak amplitude and latency with increasing stimulus level in younger adults and older adults. The N1 is plotted as a positive response to demonstrate the predicted effects of increasing stimulus level (gray lines to black lines) in younger and older adults. Our previous study in younger adults (Harris et al., 2018) showed that, with increasing stimulus level, peak amplitude and half-width latency increased and peak latency decreased. We hypothesized that peak amplitude, half-width latency, and neural synchrony (calculated in the frequency domain so not shown) will increase less with increasing stimulus level in older than younger adults.
A–C, Hypothesized patterns of AN dysfunction as revealed by differential changes in three AN metrics at a higher stimulus level are plotted in triangular format (dotted lines) relative to the same three metrics reflecting normal AN function (solid lines), with increasing values for each metric deviating from the center origin. A, AN disengagement. A disproportionate loss of low-SR fiber activity will result in smaller CAP peak amplitudes and narrower half-width (shorter half-width latency) with increasing stimulus level without affecting AN synchrony as shown by no change in PLV. B, AN dyssynchrony. AN dyssynchrony with preserved numbers of AN fibers will result in smaller CAP peak amplitudes, wider half-widths (longer half-width latency), and substantially reduced PLV because of jitter across trials. C, AN disengagement and AN dyssynchrony. A loss of AN fibers or synapses would result in smaller peak amplitudes and narrower half-width (shorter half-width latency; as in A). However, AN dyssynchrony can result in smaller peak amplitudes and wider half-widths (as in B); therefore, the combination of these two opposing factors is reflected in a modest predicted change in half-width and reduced PLV.
We applied this multimetric model along with a measure of AN structure to examine associations across metrics in a cohort of older adults, with and without elevated pure-tone thresholds. To date, aging effects on AN function have focused solely on AN response amplitudes, with smaller peak amplitudes for older adults compared with younger adults (Burkard and Sims, 2001; Konrad-Martin et al., 2012; Anderson, et al., 2021). AN disengagement may co-occur with AN dyssynchrony and may be associated with a loss of AN fibers or synapses (disengagement) and a loss or disruption of myelin (dyssynchrony), which may occur with increasing age (Xing et al., 2012). We hypothesized that these multiple age-related pathologies will be reflected not only in a reduction in response amplitudes, but across metrics (Fig. 3C), and are consistent with age-related differences in AN structure. Although the focus of the current study was on AN pathology and AN dysfunction that may occur in older as compared with younger adults, specific hypotheses for these multiple AN metrics can be generated for other clinical populations based on underlying pathologies with known effects on AN function.
Finally, we examined the extent to which age-related AN structural and functional declines predicted various measures of auditory function in older adults. Because of higher thresholds of low-SR fibers, age-related loss of these fibers may disproportionally affect suprathreshold measures of auditory function for older adults while leaving detection thresholds relatively unaffected (AN disengagement). Similarly, deficits in AN synchrony have been shown to affect complex auditory processing more than simple detection (Lopez-Poveda and Barrios, 2013; Carney, 2018; Resnik and Polley, 2021). Accordingly, we predicted that the ability to resolve small changes in frequency and rapid changes in speech, and understand speech in noise, relate to the SR diversity and synchrony of active AN fibers, which may explain the speech recognition problems experienced by older adults. Coupled with the inability to directly measure AN function in humans and the absence of indirect measures of AN function in standard clinical test batteries, the approach and findings presented here may have high impact in advancing our understanding of the effects of age-related AN structural changes and AN dysfunction on auditory functional declines, especially in older adults.
Materials and Methods
Participants
A total of 58 older adults (aged 56–84 years, mean age = 66.6 years, 40 female) and 42 younger adults (aged 19–30, mean age = 24.4 years, 28 female) were recruited from the Charleston, SC community. Inclusion criteria included English as a native language and a Mini-Mental Status Examination score of at least 27. Exclusion criteria included a history of head trauma, seizures, conductive hearing loss or active otologic disease, self-reported central nervous system disorders, and contraindications for safe magnetic resonance imaging scanning. Data from 25 out of the 42 younger adults in the current study have been reported previously (Harris et al., 2018). Younger participants were required to have pure-tone thresholds ≤20 dB HL from 0.25 to 8 kHz. Older adults were included if their hearing loss at or below 4 kHz did not exceed 65 dB HL. As a result, pure-tone thresholds varied in older adults; some had audiometric thresholds that were similar to younger adults, whereas others had mild-to-moderate sloping sensorineural hearing loss. Audiometric thresholds for the test ear (right ear) for both groups are shown in Figure 4. Participants provided written informed consent before participating in this Medical University of South Carolina Institutional Review Board-approved study.
Pure-tone audiograms. Pure-tone air conduction thresholds at audiometric frequencies (0.25–8 kHz) for the right (test) ear for younger adults (light gray solid lines) and older adults (dark gray solid lines). Pure-tone thresholds for younger adults were required to be ≤20 dB HL at each frequency. Pure-tone thresholds in older adults ranged from 0 to 65 dB HL from 0.25 to 4 kHz.
Measures of AN function
N1 acquisition and analysis
The N1 of the CAP was elicited by 100-µs rectangular pulses, alternating polarity, presented at 11.1/s to the right ear through an insert earphone (ER3C; Etymotic Research). N1s were recorded in blocks of 1100 trials (550 of each polarity). N1 responses were recorded using a tympanic membrane electrode (Sanibel) in the test (right) ear, an inverting electrode placed on the contralateral (left) mastoid, and a low forehead grounding electrode. Auditory brainstem responses (ABRs) were simultaneously recorded for reference in identifying Wave I/N1 using a high forehead active electrode, an inverting mastoid electrode (on the right mastoid), and a low forehead grounding electrode. All recordings were collected at a sampling rate of 20 kHz using a custom headstage (Tucker Davis Technologies; TDT) connected to the bipolar channels of a Neuroscan SynAmpsRT amplifier (Compumedics USA). N1s were recorded in response to stimuli ranging in level from 70 to 110 dB pSPL in 10-dB steps, with each level presented twice. Testing was done in an acoustically and electrically shielded room. Participants reclined in a chair and were encouraged to sleep or rest quietly for the duration of testing.
Continuous neural activity was analyzed offline in MATLAB (MathWorks) using EEGlab and the ERPlab toolbox. Continuous EEG signals were bandpass filtered between 150 and 3000 Hz. Stimulus triggers were shifted to account for the 1-ms delay introduced by the headphones and the 0.6-ms delay of the TDT digital-to-analog convertor. The filtered data were epoched from −2 to 10 ms and baseline corrected to a −2 to 0-ms prestimulus baseline (McClaskey et al., 2018). Trials were identified and rejected on the basis of a peak threshold detection of 45 µV and by visual inspection. Epoched responses for the remaining trials were averaged. Results reported here focus on the N1, or first negative peak of the CAP. N1 peak selection was performed by two independent reviewers and assessed for repeatability across multiple runs.
Onset latency, peak latency, peak amplitude, and half-width latency
We measured both onset latency and peak latency of the N1. The N1 peak-to-baseline amplitude and peak latency of the N1 were identified using visual overlay cursors on a computer monitor. Peak amplitude was measured in reference to the average baseline. The onset of the N1 was calculated in the ERPlab toolbox using the fractional peak latency function (Luck, 2004). The half-width latency of the response was calculated as the time (in milliseconds) from the onset of N1 to the peak latency of N1.
Phase-locking value (PLV)
Neural synchrony across single trials was assessed by estimating the intertrial coherence, also known as the PLV. PLV was calculated in EEGlab using the single-trial-level responses. PLV is the length of the vector that is formed by averaging the complex phase angles of each trial at each frequency, which are obtained via time-frequency decomposition. PLV measures are unitless and range from 0, indicating absence of synchronization across trials, to 1, indicating perfect synchronization. For n trials, Fk(f,t) is the spectral estimate of trial k at frequency f and time t, and PLV was calculated as the following (Delorme and Makeig, 2004):
Time-frequency decomposition was performed with Hanning FFT tapers via EEGlab's newtimef() function, using 16 linearly spaced frequencies from 625 to 2500 Hz, with a pad-ratio of 2, and a window size of 32 samples. A single estimate of PLV was obtained for each participant as the peak PLV across a 2-ms window surrounding the N1 peak response from 625 to 2500 Hz. For more details, see Figure 1.
AN structural images
MRI acquisition and analysis
Structural images of the brainstem were acquired for a subset of participants with CAP responses, N = 40 [16 younger (mean age = 24, 11 female), 24 older (mean age = 66, 20 female)]. A Siemens Prisma 3T scanner with a 32-channel head coil was used to collect images at the brainstem level of CN VIII with a Constructive Interference in Steady State (CISS) sequence. The term CN VIII will be used in place of AN as it encompasses both the vestibular and cochlear branches of CN VIII. The term CN VIII will therefore be used when discussing structural imaging results. The CISS images were acquired with the following parameters: 64 slices with a 320 × 320 matrix; TR = 8.56 ms; TE = 3.91 ms; flip angle = 50°; slice thickness = 0.5 mm; and a 20% distance factor.
CISS images are commonly collected clinically to visualize the cranial nerves because of their relatively high resolution. CSF is typically hyperintense and cranial nerves are typically hypointense voids in CISS images. We inverted the image contrast to visualize CN VIII as a hyperintense structure relative to surrounding CSF. This approach allowed for the development of a modified Calavieri measurement (Roberts et al., 1993) to characterize CN VIII morphology. A custom MATLAB script was used to display and overlay a grid on each image section where CN VIII was present in the cerebellopontine angle cistern. Two raters clicked on each intersection of the grid that fell within the space of CN VIII. The median contrast value was then calculated across axial sections of the nerve. Intra-rater and inter-rater reliability of this contrast measurement were α = 0.989 and α = 0.966, respectively. In addition, the CN VIII contrast measure was highly correlated (r = 0.924) with an estimate of the median CN VIII T1 relaxation rate (R1) from lower resolution (1 mm isotropic) MPRAGE images (Marques et al., 2010) for a subset of participants with these lower resolution T1 data. Thus, variation in CN VIII morphology could be reliably measured using this method, and the inverted CISS contrast was strongly positively associated with a quantitative measure of T1 relaxation. Because the CISS contrast measure does not differentiate tissue types, we considered the median contrast measure as an index of tissue density, with higher contrast indicating a larger nerve, as shown later in the Results. The extent to which these difference in older adults represent a loss/dysfunction in myelin, nerve fibers, or both, is currently unknown.
Despite the strong positive association between CN VIII contrast and R1, the median contrast of voxels representing CSF adjacent to the measured portion of CN VIII was collected as a control measure because variables such as scanner heating and motion artifact could have contributed to contrast differences in the CISS images. CSF was measured using the same Cavalieri approach where voxels representing CSF adjacent to CN VIII were identified at the intersection of the gridlines. Intra-rater and inter-rater reliability for this measure were α = 0.994 and α = 0.986, respectively.
Measures of auditory function
Signal transduction and auditory processing from the cochlea to the cortex is accomplished, in part, because of the aforementioned diversity of biophysical properties of AN fibers. Most notably, intensity is coded in the driven activation and sensitivity of auditory afferents, with subsets or groups of neurons responding at lower and higher thresholds and across a wide range in sensitivity to sound level across neurons (Costalupes et al., 1984; Liberman, 2017). Similarly, frequency resolution may be propagated from the cochlea to the AN by firing of subgroups (based on SR) of auditory neurons with distinct temporal discharge patterns (Mann and Kelley, 2011). This diversity across AN fiber subtypes and activity enables encoding of a wide range of intensities and frequencies and may help maintain hearing in complex environments (Lopez-Poveda and Barrios, 2013). We hypothesized that the ability to resolve small changes in frequency, rapid changes in speech, and speech from noise may relate to the diversity and synchrony of afferent fibers. To test the hypotheses that auditory function for younger and older adults varies systematically with AN function, we acquired three behavioral measures from a subset of participants with measures of AN function. These included frequency-modulation (FM) detection, N = 53 [19 younger (13 female), 34 older (24 female)], recognition of time-compressed speech, N = 70 [28 younger (19 female), 42 older (33 female], and speech recognition in noise, N = 76 [28 younger (20 female), 48 older (36 female)]. Although participants were encouraged to take part in all experiments, 68 participants with AN measures completed both measures of speech recognition, and 44 participants with AN measures completed both measures of speech recognition and FM detection. All testing of auditory function was performed in a sound attenuated booth.
FM detection
FM signals were digitally generated using custom MATLAB software (Bidelman et al., 2015) and converted to an analog signal using a LynxTWO-B (Lynx Studio Technology) sound card (sampling rate, 44.1 kHz; 24-bit resolution). The generated signal was sent through a headphone buffer (TDT HB7) to the right ear of a pair of Sennheiser HDA 200 earphones. The carrier frequency was 500 Hz, and the modulation frequency was 5 Hz. The time waveform of the FM stimulus was mathematically described as the following:
Recognition of time-compressed speech
Time-compressed speech stimuli were a subset of the stimuli used by Gordon-Salant and Fitzgibbons (1993, 2001), constructed from 180 low-predictability sentences from the Revised Speech Perception in Noise test (R-SPIN; Bilger et al., 1984) but presented in quiet. Sentences were digitized (10-kHz sampling rate), time-compressed, and low-pass filtered (5 kHz; Gordon-Salant and Fitzgibbons, 1993, 2001). The speech rate of the original uncompressed sentences was ∼200 words per minute (Gordon-Salant and Fitzgibbons, 2004; Gordon-Salant et al., 2011). Following time compression, sentences were divided into presentation lists that differed in time compression: a 50-sentence 40% time-compressed list, a 50-sentence 50% time-compressed list, a 50-sentence 60% time-compressed list, and a 30-sentence practice list comprised of 10 sentences each with 40%, 50%, or 60% time compression. The percentage given for compression describes the percentage that each sentence is compressed in length. For example, a 60% time-compressed sentence is compressed by 60% of its original duration, which makes the new duration 40% of its original length. Time compression was achieved using the Global Duration option of WEDW software (Bunnell, 2005). This transformation largely preserves the spectral aspects of the acoustic signal but reduces the duration in which the phonetic information is presented (Gordon-Salant and Fitzgibbons, 1993, 1999, 2001; Gordon-Salant and Friedman, 2011). For the speech recognition task with these time-compressed sentences, each participant was fit with an insert earphone (ER3C, Etymotic Research) in their right ear and presented time-compressed sentences at an overall level of 90 dB SPL (Gordon-Salant and Fitzgibbons, 1993, 1999, 2001; Gordon-Salant et al., 2007). The participants' task was to verbally identify as many words as possible following the presentation of each sentence; participants were encouraged to guess if they were unsure of a word or phrase. Following practice with 30 sentences with varying time compressions, participants listened to the 40%, 50%, and 60% lists of sentences in random order (e.g., 40% list, 60% list, 50% list). Accuracy was coded as the percentage of key words correctly identified for each list of time-compressed sentences (Gordon-Salant et al., 2014; Dias et al., 2019). Time-compressed speech scores are presented as the average correct across the three compression rates.
Speech recognition in noise
Speech recognition in noise was measured using the Quick Speech-in-Noise test (QuickSIN; Etymotic Research; Killion et al., 2004). The QuickSIN materials include five lists of six sentences each, with each sentence containing 5 keywords, for a total of 30 keywords in each list. The noise was a four-talker babble. The six sentences in each list progressively decrease in signal-to-noise ratio (SNR) from 25 to 0 dB in 5-dB steps. Sentences were presented through TDH-39 headphones at a fixed level of 70 dB HL (with noise level varying according to SNR) using a combination of an Onkyo Compact Disk Player and an Interacoustics Audio Traveler (AA222). We computed the average number of keywords (out of five) correctly identified at each SNR (25, 20, 15, 10, 5, and 0 dB), and summed the averages, for a total possible correct score of 30. QuickSIN performance is typically reported as SNR loss (25.5 minus total key words correct out of 30). However, data presented here reports QuickSIN results as the percent correct keywords out of 30 across six SNRs. Percent correct and SNR loss are linearly related so any relationships observed in one are identical to relationships observed in the other. QuickSIN reported in percent correct was used to maintain consistency with percent-correct scores for recognition of time-compressed speech.
Data analyses
Statistical analyses were performed in R using general linear model and generalized linear mixed model analyses (lme4; Bates et al., 2015). Effect sizes (B) and standardized effect sizes (β) are reported for each comparison. The Benjamini–Hochberg procedure was used to correct for false discovery rate using p adjust in R.
AN functional analyses
We used linear mixed models to examine age-related changes across stimulus level for each AN metric. To reduce the number of variables entered into regression models we examined N1 responses at a lower level (80 dB pSPL) and a higher level (110 dB pSPL). Each AN metric (peak amplitude, peak latency, onset latency, half-width latency, and PLV) was tested individually as the outcome variable with stimulus level (80 or 110 dB pSPL) as a fixed factor and participant as a random factor. To conservatively avoid inflation of age-group difference effects, data from two younger participants (one female) were removed from all further analyses as peak amplitude and PLV were >2 SDs above the mean. Significant age group by stimulus level interactions were analyzed post hoc to test the effect of stimulus level in each age group separately, using one-tailed or two-tailed t-tests. The use of one-tailed or two-tailed tests was hypothesis driven and explained in later sections for each AN metric. Model testing was used to compare the extent to which including age and hearing thresholds improved model fit (χ2). Linear regression and correlation analyses were used to examine associations across metrics. The input/output (I/O) slope of peak amplitude was used to represent growth of the N1 between 80 and 110 dB pSPL and was the outcome variable, whereas the half-width latency and PLV were predictors. Pearson correlations were examined between peak amplitude, halfwidth, and PLV at the lower and higher stimulus level.
AN structural analysis
We used independent sample t-tests to identify age-related differences in CN VIII structure. Structure-function associations were examined using linear regression and model testing to examine associations between age group, AN function, and CN VIII structure. Estimates of CN VIII CSF were used as a control to account for variables such as motion artifact and scanner heating. Data from one older participant (one female) were removed from all further analysis as CN VIII density was >2 SD below the mean. We examined each AN metric as outcome variables. AN metrics were examined at the higher stimulus level where age-related changes were observed. CN VIII median contrast was entered as a predictor variable. Subsequent models included CN VIII CSF contrast values and age group entered as predictor variables. We used model testing to compare the total amount of variance explained in each of our models. Analyses of associations between AN structure and auditory functional measures could not be completed in the current study because of a limited number of younger participants who completed all measures.
Auditory functional analyses
We used linear regression and model testing to examine associations between age group, auditory detection thresholds, auditory functional measures (FM detection, time-compressed speech, speech recognition in noise), and AN metrics at higher stimulus levels. AN metrics at higher stimulus levels were theorized to reflect differences in AN engagement and AN synchrony, important for speech understanding. We examined each auditory functional measure as outcome variables. To identify effects of age, age group was entered in the model as the predictor variable. We then tested models with each of the AN metrics. Next, we entered pure-tone average thresholds (PTAs; average of thresholds at 0.5–8 kHz) as a predictor variable in the model. We used model testing to compare the total amount of variance explained in each of our models.
Results
Effects of age group and stimulus level on AN metrics
We first examined the extent to which stimulus level (80 vs 110 dB pSPL) predicted each of the AN metrics in our combined sample of younger and older adults. Results of the full models for each AN metric (stimulus level, age group, and age group × stimulus level) are provided in Table 1. Figure 5 shows the individual data for younger and older adults for each metric at 80 dB pSPL and 110 dB pSPL (referred to as lower stimulus level and higher stimulus level, respectively).
AN metrics as predicted by stimulus level, age group, and the stimulus level × age group interaction
AN metrics as a function of stimulus level and age group. A, Group average CAP waveforms for younger (dotted gray line) and older adults (solid black line) in response to a 110 dB pSPL click. With increasing level, (B) peak amplitude increased (became more negative), (C–E) half-width latency increased, and onset and peak latencies decreased. F, PLVs increased with increasing level in younger adults, indicating improved synchrony of N1. Individual data points were randomly jittered around each stimulus level for clarity in display. Median values are marked by a solid line. Significant effects of age and intensity are described in Table 1.
Peak amplitude
In our model with only stimulus level, peak amplitude increased (became more negative) with increasing stimulus level [B = −0.12 (SE = 0.024), β = −0.56, t(96) = −4.99, p < 0.001]. Adding age group to the model significantly improved model fit [χ2(2) = 19.41, p < 0.001]. Adding PTA to the model did not significantly improve model fit [χ2(1) = 0.90, p = 0.34]. Peak amplitude still increased with increasing stimulus level and there was a significant interaction between age group and stimulus level, showing that the effect of increasing stimulus level on peak amplitude differed for younger and older adults. In a post hoc analysis with the two age groups modeled separately, stimulus level was a significant predictor of peak amplitude for both younger adults [B = −0.20 (SE = 0.043), t(38) = −4.69, p < 0.001] and older adults [B = −0.07 (SE = 0.026), t(57) = −2.53, p = 0.01], showing that peak amplitudes were larger at the higher than lower stimulus level for both age groups. Finally, we used one-tailed t-tests to explain the age-group-by-stimulus-level interaction in the full model: peak amplitudes were similar for younger and older adults (t(93) = 1.09, p = 0.14) at the lower level, but were significantly larger for younger than older adults (t(93) = 4.069, p < 0.001) at the higher level.
Onset latency and peak latency
With increasing stimulus level, onset and peak latency of the N1 shifts earlier in time. This earlier response is attributed to both shorter response latencies of AN fibers with increasing stimulus level and the broadening of peripheral auditory filters. However, several factors may contribute to differences in the timing of N1 and may have differential effects on onset and peak latency. In our previous study in younger adults (Harris et al., 2018), we found a larger decrease in onset latency relative to peak latency with increasing stimulus level, resulting in a wider response (longer half-width latency) at higher compared with lower stimulus levels. As described earlier, this pattern is consistent with recruitment of low-SR fibers with increasing stimulus level and, therefore, a more biodiverse group of AN fibers at higher than lower stimulus levels. However, the onset latency and the peak latency of the N1 may both be negatively affected by aging, hearing loss, and noise exposure, resulting in smaller decreases in onset and peak latency with increasing stimulus level (Møller and Jho, 1991; Lichtenhan and Chertoff, 2008; Konrad-Martin et al., 2012; Tagoe et al., 2014). Two-tailed t tests were employed for post hoc analyses to examine these opposing influences on onset and peak latency. In our combined sample of younger and older adults, both onset latency [B = −0.56 (SE = 0.03), β = −1.31, t(93) = −18.55, p < 0.001] and peak latency [B = −0.54 (SE = 0.03), β = −1.27, t(93) = −17.86, p < 0.001] decreased with increasing stimulus level. Adding age group did not significantly improve model fit for onset latency [χ2(2) = 5.04, p = 0.08] and the main effect of age group and the age group by stimulus level interaction failed to reach significance (Table 1). Adding PTA to the model did not significantly improve model fit [χ2(1) = 2.65, p = 0.10]. Post hoc analyses show that onset latency did not significantly differ between younger and older adults at the lower (p = 0.08) or higher stimulus levels (p = 0.66). In contrast, adding age group to the model significantly improved model fit for peak latency [χ2(2) = 6.26, p = 0.04]. Peak latency decreased with increasing stimulus level and there was a main effect of age-group. The age group by stimulus level interaction failed to reach significance (Table 1). Adding PTA to the model did not significantly improve model fit [χ2(1) = 2.85, p = 0.1]. Separate post hoc models in younger adults [B = −0.61 (SE = 0.05), β = −1.38, t(38) = −12.95, p < 0.001] and older adults [B = −0.49 (SE = 0.04), t(56) = −12.95, p < 0.001] both revealed significant effects of stimulus level on peak latency, suggesting that peak latencies were shorter at the higher than the lower stimulus level. We used two-tailed independent t tests to test for differences in peak latencies at 80 dB and 110 dB pSPL and found no significant differences in peak latency between younger and older adults (p = 0.05 and p = 0.49, respectively). These results suggest that both older and younger adults show shorter onset and peak latencies at higher than lower stimulus levels.
Half-width latency
In our combined sample of younger and older adults, half-width was larger (longer half-width latency in milliseconds) at the higher than lower stimulus level [B = 0.003 (SE = 0.001), β = 0.38, t(93) = 2.96, p = 0.004]. Adding age group did not significantly improve model fit [χ2(2) = 4.94, p = 0.084]. Adding PTA to the model did not significantly improve model fit [χ2(1) = 0.90, p = 0.34]. In this model, half-width was significantly larger at the higher than lower stimulus level and there was a significant age group by stimulus level interaction. Models including each age group separately showed a significant increase in half-width at the higher level as compared with the lower level in younger adults [B = 0.06 (SE = 0.01), β = 0. 68, t(36) = 4.25, p < 0.001] but not in older adults [B = 0.01 (SE = 0.01), β = 0.18), t(56) = 0.96, p = 0.34]. As described earlier, a more diverse population of AN fibers was hypothesized to result in a wider N1 response. However, poor neural synchrony may also contribute to a wider N1 response. Therefore, we used two-tailed independent t tests to test for differences in half-width latency at each stimulus level. At the lower level, half-width latency differences between younger and older adults were not significant (t(93) = 0.62, p = 0.54), but at the higher level, half-width latencies were wider for younger adults than older adults (t(93) = −1.97, p = 0.03). As noted earlier for onset and peak latencies, these results are consistent with an interpretation that, compared with older adults, younger adults recruit a more biodiverse group of AN fibers at higher than lower stimulus levels.
Neural synchrony
PLV values were higher at higher than lower stimulus levels in the model that included both younger and older adults [B = 0.0005 (SE = 0.0001), β = 0.44, t(94) = 3.86, p < 0.001]. Adding age group to the model significantly improved model fit [χ2(2) = 25.30, p < 0.001]. Adding PTA to the model did not significantly improve model fit [χ2(1) = 0.35, p = 0.55]. PLV increased with increasing stimulus level and there was a significant interaction of age group and stimulus level (Table 1), showing that the effects of increasing stimulus level on PLV differ for younger and older adults. Post hoc analyses revealed a significant effect of stimulus level on PLV for younger adults [B = 0.011 (SE = 0.0021), β = 0.75, t(38) = 5.13, p < 0.001] but not for older adults [B = 0.0001 (SE = 0.0001), β = 0.15, t(57) = 0.78, p = 0.44]. To test the hypothesis that younger adults exhibited stronger PLV than older adults we used one-tailed t tests. Results of one-tailed t tests indicate that PLV values were similar for younger and older adults at the lower stimulus level (t(93) = −0.55, p = 0.29), but PLV values were higher for younger adults than older adults at the higher stimulus level (t(93) = −3.71, p < 0.001). These results are consistent with an interpretation that neural synchrony increases with increasing stimulus level in younger but not older adults. The lack of association between pure-tone thresholds and suprathreshold AN metrics suggests that these metrics are sensitive to age-related differences in AN function not evident in the audiogram.
Effects of age group, half-width latency, and PLV on peak amplitude
Age-related changes with increasing stimulus level were observed for three variables, peak amplitude, as reported previously (Burkard and Sims, 2001; Konrad-Martin et al., 2012), half-width, and PLV. As described earlier and illustrated in Figures 2 and 3, age-related AN dysfunction is hypothesized to arise from patterns of AN disengagement and AN dyssynchrony. Based on this interpretation, measures of AN function would be poorer for older than younger adults at the higher but not lower stimulus level, consistent with the recruitment of low-SR fibers and increased neural synchrony at higher levels. Similarly, we theorized that age-related deficits in peak amplitude would be driven by changes in both half-width (reflecting recruitment of a more diverse population of AN fibers) and PLV (reflecting AN synchrony). We used linear regression to examine the extent to which peak amplitude differences were associated with age group, halfwidth, and PLV at lower and higher stimulus levels (Table 2). At lower stimulus levels, peak amplitudes were driven, in part, by differences in PLV. At higher stimulus levels, smaller peak amplitudes were associated with age, and both shorter half-width latency, and decreased PLV. These results are broadly consistent with our hypothesis shown in Figure 3C of AN disengagement and AN dyssynchrony. Furthermore, to compare actual results with the schematics in Figure 3, AN metrics of peak amplitude, half-width latency, and PLV were normalized, scaled, and plotted in triangular format at 80 dB pSPL (Fig. 6A) and 110 dB pSPL (Fig. 6B). At the higher stimulus level (Fig. 6B), decreases in peak amplitude, half-width latency, and PLV values appear consistent with an interpretation of AN disengagement and AN dysynchrony for older adults, as illustrated in Figure 3C. Correlations were examined across metrics at the lower and higher stimulus levels. At the lower level, peak amplitude and PLV were significantly negatively correlated (r = −0.39, p < 0.001), however, half-width latency was not significantly correlated with the other metrics (p > 0.05). At the higher stimulus level, peak amplitude was significantly negatively correlated with PLV (r = −0.61, p < 0.001) and half-width (r = –0.50, p < 0.001), and PLV and half-width were significantly correlated (r = 0.37, p < 0.001).
Associations between peak amplitude, age group, half-width, and PLV
AN metrics at 80 dB pSPL and 110 dB pSPL for younger and older adults. Individual data for three AN metrics (peak amplitude, half-width latency, PLV) at 80 and 110 dB pSPL were z-transformed and then scaled to a range of 0–2. Means for each metric are plotted such that larger peak amplitudes, wider half-width responses (longer half-width latencies), and higher PLV values were plotted further from the origin of their scales (as in Fig. 3A–C). Significant differences in the three metrics between younger and older adults were observed at the higher stimulus level (B) but not at the lower stimulus level (A). Moreover, the results for older adults are consistent with the theorized patterns of age-related AN disengagement and AN dyssynchrony and their combined effects on AN function as described in Figure 3C.
Associations of AN function and AN structure (CN VIII density)
Similar to findings in human temporal bones showing an age-related loss of AN fibers and disruption in myelin (Makary et al., 2011; Wu et al., 2019), we predicted that older adults would have lower CN VIII densities, and lower density would relate to age-related deficits in AN function. Independent sample t tests demonstrated that the density of CN VIII was greater in younger than older adults (t(37) = 2.57, p = 0.007), which occurred after controlling for cerebrospinal fluid (CSF) contrast (t(37) = 2.24, p = 0.016).
We next used linear regression and model testing in the subset of participants with CN VIII data to examine the extent to which age-group differences and individual differences in CN VIII density predicted differences in AN function. CN VIII density was found to be associated with PLV at higher stimulus levels but not with peak amplitude or half-width latency (p > 0.05). For further analysis of the association with PLV, model testing revealed that age group remained a significant predictor of PLV in this smaller sample (multiple R2 = 0.17, F(1,36) = 7.3, p = 0.01), as it was for the larger sample reported earlier. Adding measures of CN VIII density to the model significantly improved model fit (change R2 = 0.10, F(1,35) = 4.84, p = 0.03); CN VIII density was a significant predictor of PLV [B = 0.17 (SE = 0.08), β = 0.34, t(35) = 2.20, p = 0.03], whereas age group was no longer a significant predictor of PLV in this model (p = 0.09). Adding the CSF measure to the model did not significantly improve model fit (change R2 = 0.001, F(1,34) = 0.005, p = 0.95). The scatterplot in Figure 7 shows that younger and older adults with higher CN VIII densities had higher PLV values, which also shows that CN VIII density accounted for significant variation in PLV that was independent of age.
Association of CN VIII density and PLV. A, Inverted CISS sections from an average of subject images. CSF is typically hyperintense and cranial nerves are typically hypointense voids in CISS images. Image contrast was inverted to visualize CN VIII as a hyperintense structure relative to surrounding CSF. A custom MATLAB script was used to display and overlay a grid on each image section where CN VIII was present in the cerebellopontine angle cistern (green). The cochlea is outlined in purple. B, Elevated CN VIII contrast predicts higher PLV. Standardized estimates of CN VIII, after regressing out CSF median values, are plotted against PLV measured at 110 dB pSPL for older adults (gray circles) and younger adults (white triangles). Younger and older adults with higher CN VIII density had higher PLV values at a higher stimulus level, consistent with an interpretation of better neural synchrony.
Associations of AN function and auditory function
As described earlier, age-related differences in auditory function may be explained, in part, by AN dysfunction (decreases in peak amplitude, half-width latency, and PLV values at the higher stimulus level), which is consistent with interpretations of age-related AN disengagement and dyssynchrony (Fig. 3C). The model that provided the best fit for each measure of auditory function is shown in Table 3. When a significant interaction with age group was indicated, we performed post hoc analyses examining associations for younger and older groups separately.
Associations between neural synchrony and three measures of auditory function
FM detection
Our earlier results (He et al., 1998, 2007; Harris et al., 2008) revealed that age-related differences in FM detection thresholds were larger at lower modulation frequencies than higher frequencies, which could not be explained by differences in detection thresholds. Consistent with our previous studies, age group was a significant predictor of FM detection thresholds (Table 3), that is, FM detection thresholds were lower in younger than older adults. These age-related differences in FM detection were not explained by age-related differences in PTA or by AN metrics, as each metric tested (peak amplitude, half-width latency, and PLV) failed to improve model fit (p > 0.05). These results confirm that FM detection is poorer for older than younger adults, but this change in auditory function cannot be explained by age-group differences in AN function.
Recognition of time-compressed speech
Recognition scores of time-compressed speech (averaged across compression rate) were higher for younger than older adults, as demonstrated when age group was initially entered in the model (multiple R2 = 0.20, p < 0.001). Each AN metric was tested to determine the extent to which individual differences in AN function explained these results. The inclusion of PLV in the model (but not the other AN metrics) significantly improved model fit (change R2 = 0. 09, F(1,37) = 23.27, p < 0.001). As shown in Table 3, PLV was a significant predictor of recognition of time-compressed speech and the effects of age group were no longer significant. There was not a significant age group by PLV interaction. After accounting for age group, PTA was not a significant predictor of recognition of time-compressed speech and including PTA did not improve model fit (change R2 = 0.01, F(1,64) = 1.93, p = 0.29). Similarly, PTA was not a significant predictor of recognition of time-compressed speech when the analysis was restricted to older adults (p = 0.32). These results are consistent with an interpretation that reduced neural synchrony, rather than age per se, independently contributes to individual differences in the recognition of time-compressed speech.
Recognition of speech in noise
Consistent with previous results (Sheft et al., 2012; McClaskey et al., 2019), age was not a significant predictor of QuickSIN scores (multiple R2 = 0.02, F(1,73) = 1.16, p = 0.28). Like time-compressed speech, PLV (and no other AN metric) was a significant predictor of QuickSIN scores, that is, higher PLV values were associated with better recognition (Fig. 8; Table 3). When entered into our model with PLV, PTA was not a significant predictor of QuickSIN scores and did not significantly improve model fit (change R2 = 0.03, F(1,73) = 2.62, p = 0.11). However, when the analysis was restricted to older adults, PTA was a modest yet significant predictor of QuickSIN scores (r(37) = 0.31, p = 0.03), with higher thresholds associated with poorer scores. Our results are consistent with an interpretation that reduced neural synchrony, rather than age per se, contributes to individual differences in speech recognition in noise and that elevated pure-tone thresholds, and associated decreased audibility, in older adults are an additional contributing factor.
Associations of PLV and two measures of speech recognition. PLV is plotted against recognition of time-compressed speech (averaged across compression rate; A) and sentence recognition in noise (QuickSIN; B) for older adults (gray circles) and younger adults (white triangles). Younger and older adults with higher PLV values at a higher stimulus level had better speech recognition for both tasks (r = 0.25, p = 0.03).
Discussion
Loss or dysfunction of AN fibers has long been hypothesized as a contributor to auditory dysfunction in older adults. Here, we present three novel findings that advance our understanding of neural presbyacusis. First, we expanded on our previous work in younger adults (Harris et al., 2018) and mice (McClaskey et al., 2020) to show age-related and level-dependent differences in measures of AN function, most notably smaller and narrower CAP waveforms and lower PLV values at a higher stimulus level for older adults, consistent with the interpretation of AN disengagement (e.g., AN fiber loss and/or inactivity) and AN dyssynchrony in older adults (Figs. 3C, 6B). Our results replicate and expand on previous findings of an age-related decrease in the amplitude of Wave I of the ABR (Burkard and Sims, 2001; Konrad-Martin et al., 2012; Anderson et al., 2021) and further suggest that deficits in neural synchrony may contribute to these previously reported differences in response amplitudes. Second, we show that lower PLV values were associated with CN VIII density, where higher density was associated with higher PLV for younger and older adults. Third, we demonstrate that individual differences in PLV were predictive of recognition of time-compressed speech and speech recognition in noise for younger and older adults, consistent with predictive models of AN degeneration and speech recognition (Lopez-Poveda and Barrios, 2013; Lopez-Poveda, 2014). Together, the results indicate that diminished AN PLV, due in part to differences in AN structure, contributes to auditory perception difficulties in older adults.
Age-related AN disengagement and AN dyssynchrony
In our previous studies in humans and mice (Harris et al., 2018; McClaskey et al., 2020), we demonstrated that measures of AN function are sensitive to individual differences in the activity of low-SR (high threshold) AN fibers. While debate remains surrounding the extent to which low-SR fibers contribute to N1 responses (Bourien et al., 2014), studies in mice and humans have shown that low-SR fibers contribute to the N1 response at high levels (Sergeyenko et al., 2013; Mehraei et al., 2016). Here, differences in AN metrics were observed between older and younger adults at higher but not lower intensity levels, consistent with an interpretation of age-related AN disengagement, specifically of low-SR fibers. Moreover, longer half-width latencies were observed at higher stimulus levels for younger but not older adults, consistent with engagement of a more diverse population of AN fibers with increasing stimulus level. In addition to AN engagement, we demonstrated that AN metrics at the higher stimulus level were associated with PLV, suggesting that age-related loss of neural synchrony contributes to age-related deficits in AN function. Taken together, age-related changes across metrics and stimulus levels suggest that age-related deficits in AN function may result from a loss or inactivity of AN fibers, and a decrease in neural synchrony. Comparing across multiple metrics at a lower and higher stimulus level revealed a clearer interpretation of the underlying AN pathology associated with AN dysfunction.
Lack of associations with pure-tone thresholds
Despite significant variability in hearing thresholds of older adults, measures of AN dysfunction were not associated with differences in hearing thresholds. These results add to a growing literature demonstrating age-related AN dysfunction in older adults with normal hearing (Burkard and Sims, 2001; McClaskey et al., 2018) and with hearing loss (Konrad-Martin et al., 2012). Individual differences in younger adults suggest a wide range of AN function (Fig. 5), consistent with developmental and/or environmental impacts on AN function. Findings from animal models suggest that aging accelerates AN damage associated with environmental exposures (Fernandez et al., 2015). Taken together, similar associations across measures of AN function for younger and older adults, coupled with age-related deficits, suggest that aging may exaggerate developmental or environmental differences in AN function. However, longitudinal studies are still warranted.
Structure-function associations
Magnetic resonance imaging of the AN is used routinely in clinical practice, but structure-function assessments have been limited to cochlear implant users. In the current study, CN VIII density was lower in older compared with younger adults. The limited studies that have assessed the effects of normal aging on AN volume have yielded mixed results (Kang et al., 2012; Nakamichi et al., 2013; Özdemir and Kavak, 2019) and did not assess associations with AN function. We found that lower PLV was associated with lower CN VIII density, which may reflect the changes in myelin that occur with increasing age, or atrophy or loss of AN fibers. The modest but significant effect observed here was obtained using data from a non-quantitative acquisition (CISS), and thus additional study with quantitative scans may be necessary to understand the structural bases for the CN VIII density associations with age and PLV.
Relationship to auditory function
Most studies assessing AN function and auditory function in humans have focused on noise-induced cochlear synaptopathy, specifically examining associations between AN function and speech recognition in normal hearing adults. These studies often yield mixed results (Bramhall et al., 2019). In contrast, the current study focused on neural presbyacusis, or age-related changes in the structure and function of the AN. Aging in animal models and human temporal bones has been associated with exacerbated synaptopathy (Fernandez et al., 2015) but also a loss of AN fibers and changes in myelin (Xing et al., 2012). Although hair cell loss can lead to a loss of afferent AN fibers, changes in the AN often precede cochlear changes (Makary et al., 2011; Wu et al., 2019). A partial loss or dysfunction of afferent AN fibers is expected to affect auditory function by limiting the bandwidth of sensory transmission and reducing the redundancy of afferent encoding. These factors are not expected to affect simple processes like detection thresholds, but instead are associated with suprathreshold perceptual deficits when redundant encoding and accurate timing across AN fibers is crucial (Resnik and Polley, 2021).
Previously, we suggested that age-related changes in FM detection occurring at lower but not higher frequencies may be dependent on phase locking (He et al., 2007; Harris et al., 2008). Both FM detection and several AN metrics were poorer for older than younger adults, yet AN metrics did not explain age-related increases in FM detection thresholds. This lack of significant associations may be driven by several factors. Although eliciting the N1 by a click stimulus results in robust AN activation, such stimuli may not be optimal for a comparison with FM detection. Associations between AN function and FM detection may be found if the CAP was instead evoked by a low frequency tone-burst more similar to that used in the FM detection task. Studies examining associations between similar FM or amplitude modulation (AM) detection and possible noise-induced cochlear synaptopathy have yielded conflicting results (Bharadwaj et al., 2015; Prendergast et al., 2017; Yeend et al., 2017), however these prior studies did not measure AN function directly. Grose et al. (2019) reported age-related deficits in Wave I amplitude of the ABR, and AM detection and spectral modulation detection thresholds but not in the same participants. Although older adults consistently show decreases in low-frequency FM and AM detection than younger adults, the contribution of age-related AN dysfunction remains unknown.
Stronger neural phase locking of the AN as shown by higher values of PLV predicted better time-compressed speech recognition and speech recognition in noise in younger and older adults. AN dysfunction and speech recognition in older adults are well established, yet few studies have examined their associations. Studies of cochlear synaptopathy have included adults across the lifespan, and those that include older adults (55+ years of age) are more likely to report significant associations between AN function and speech recognition (Bramhall et al., 2015; Grant et al., 2020; Mepani et al., 2020) than studies restricted to younger and middle-aged adults (Guest et al., 2018; Bramhall et al., 2019). The apparent discrepancy across studies of AN function and speech recognition have been attributed to methodological differences (Grant et al., 2020). Estimates of neural synchrony in the current study, and previous work (Harris et al., 2018), were predictive of speech recognition in both younger and older adults, suggesting that PLV-based estimates of neural synchrony may be sensitive to individual differences in AN function and speech recognition across age groups. Large individual differences in speech recognition during difficult listening conditions are routinely reported in cohorts of young adults. Our results suggest that individual differences in PLV may contribute, in part, to these differences in speech recognition. These results add to a growing literature describing age-related deficits in phase locking to both sustained speech and non-speech stimuli at the level of the brainstem and cortex (Presacco et al., 2015; Harris and Dubno, 2017; Roque et al., 2019; Anderson et al., 2021)
In summary, we have previously reported results of a differential, multimetric approach for analyzing AN function in mice (McClaskey et al., 2020) and in younger adults (Harris et al., 2018), and now in the current study have expanded these results to comparisons of younger adults and older adults with normal and impaired hearing. Our results suggest that age-related and individual differences in suprathreshold AN function may arise from both a loss or disengagement of synapses/fibers and a decrease in neural synchrony. Changes in AN function were associated with deficits in AN structure. Age-related and individual differences in neural synchrony, as assessed by PLV, were associated with two measures of speech recognition. Current and future research will continue to test these hypotheses in animal models with known pathology and translate those findings to clinical populations. Translating these methods to clinical use may provide a novel method for identifying AN deficits that contribute to speech recognition difficulties in individuals with normal and impaired hearing.
Footnotes
This work was supported in part by National Institutes of Health (NIH) National Institute on Deafness and Other Communication Disorders Grants R01 DC 014467, R01 DC 017619, P50 DC 000422, and T32 DC 014435. The project also received support from the South Carolina Clinical and Translational Research Institute with an academic home at the Medical University of South Carolina, NIH/National Center for Research Resources (NCRR) Grant UL1 RR 029882. This investigation was conducted in a facility constructed with support from the NIH NCRR Research Facilities Improvement Program Grant C06 RR 014516. We thank the participants of our study. We also thank Dr. Sandra Gordan-Salant for providing the time-compressed speech stimuli, Brendan J. Balken for his assistance with data collection, and Dr. Richard Schmiedt for his advice on previous versions of this manuscript.
The authors declare no competing financial interests.
- Correspondence should be addressed to Kelly C. Harris at harriskc{at}musc.edu
