Afferent Loss, GABA, and Central Gain in Older Adults: Associations with Speech Recognition in Noise

Abstract

Deficits in auditory nerve (AN) function for older adults reduce afferent input to the cortex. The extent to which the cortex in older adults adapts to this loss of afferent input and the mechanisms underlying this adaptation are not well understood. We took a neural systems approach measuring AN and cortical evoked responses within 50 older and 27 younger human adults (59 female) to estimate central gain or increased cortical activity despite reduced AN activity. Relative to younger adults, older adults' AN response amplitudes were smaller, but cortical responses were not. We used the relationship between AN and cortical response amplitudes in younger adults to predict cortical response amplitudes for older adults from their AN responses. Central gain in older adults was thus defined as the difference between their observed cortical responses and those predicted from the parameter estimates of younger adults. In older adults, decreased afferent input contributed to lower cortical GABA levels, greater central gain, and poorer speech recognition in noise (SIN). These effects on SIN occur in addition to, and independent from, effects attributed to elevated hearing thresholds. Our results are consistent with animal models of central gain and suggest that reduced AN afferent input in some older adults may result in changes in cortical encoding and inhibitory neurotransmission, which contribute to reduced SIN. An advancement in our understanding of the changes that occur throughout the auditory system in response to the gradual loss of input with increasing age may provide potential therapeutic targets for intervention.

SIGNIFICANCE STATEMENT Age-related hearing loss is one of the most common chronic conditions of aging, yet little is known about how the cortex adapts to this loss of sensory input. We measured AN and cortical responses to the same stimulus in younger and older adults. In older adults we found hyperexcitability in cortical activity relative to concomitant declines in afferent input that are consistent with central gain. Lower levels of cortical GABA, an inhibitory neurotransmitter, were associated with greater central gain, which predicted poorer SIN. The results suggest that the cortex in older adults may adapt to attenuated sensory input by reducing inhibition to amplify the cortical response, but this amplification may lead to poorer SIN.

Introduction

Over the past 40 years a substantial body of evidence has demonstrated that in tandem with peripheral deficits profound changes occur in neural morphology and excitatory/inhibitory balance in the older adult cortex. Results from animal models show that cortical neurons can remap their receptive fields and rescale sensitivity (gain) following acute cochlear and auditory nerve (AN) deficits and that these changes can occur following hearing loss (Kotak et al., 2005; Sanes and Bao, 2009; Popescu and Polley, 2010; Engineer et al., 2011) and AN deafferentation (Kujawa and Liberman, 2009). Studies of sensory deprivation in hearing-impaired and deaf individuals have demonstrated similar changes throughout the auditory neuroaxis, during development and into adulthood (Dietrich et al., 2001; Husain et al., 2011; Boyen et al., 2013; Sharma et al., 2016). However, this adaptive plasticity in the cortex can include seemingly opposing effects across different experiments, animal models, and peripheral manipulations (Seki and Eggermont, 2003; Kotak et al., 2005; Sanes and Bao, 2009; Dong et al., 2010; Popescu and Polley, 2010; Engineer et al., 2011; Kraus et al., 2011; Salvi et al., 2016), and evidence suggests that plasticity associated with typically chronic conditions may not be well modeled by acute manipulations (Seybold et al., 2012). Although age-related hearing loss is one of the most common chronic conditions of aging, and our aging population is growing, little is known about how the brain adapts to this late-onset progressive decline and degradation of sensory input.

Decreased afferent input is hypothesized to contribute, in part, to age-related changes in inhibitory neurotransmission, specifically a loss of GABAergic (GABA) inhibition throughout the central auditory system, which in turn leads to increased cortical responses, a phenomenon known as central gain (Gutiérrez et al., 1994; Burianova et al., 2009; Caspary et al., 2013; Stebbings et al., 2016). Consistent with this central gain hypothesis, several studies have reported that older adults have cortical response magnitudes that are as large as or larger than those of younger adults (Herrmann et al., 2016; Dias et al., 2019; Anderson et al., 2020; Alain et al., 2022), despite potential decreases in afferent input. To date, several studies have attempted to characterize this relationship in humans by examining associations between average pure-tone thresholds [pure-tone average (PTA)] and estimates of GABA derived from magnetic resonance spectroscopy. However, the results are mixed (Chen et al., 2013; Profant et al., 2013; Gao et al., 2015; Lalwani et al., 2019). To systematically address these equivocal findings, we used a neural systems approach that uses neural activity from the AN to provide a more comprehensive estimate of differences in afferent input and their effects on the cortex between younger and older adults. We used these measures to model relationships between the AN and cortex to test the hypothesis that relative to a control cohort of younger adults, older adults will demonstrate larger responses at the cortex than that predicted by their AN activity. Moreover, we predicted that this central gain observed only in older adults would be associated with lower levels of GABA.

Evidence from animal models of peripheral deafferentation suggests that reduced inhibition (GABA) and enhanced cortical responses contribute to deficits in, rather than enhancement of, complex auditory processing (Resnik and Polley, 2021). In humans, emerging evidence suggests that lower levels of GABA also contribute to poorer speech recognition in noise (SIN) and decreased neural distinctiveness between cortical representations of speech versus music (Lalwani et al., 2019; Dobri and Ross, 2021). However, these studies did not account for differences between younger and older adults in peripheral afferent innervation, and it is unclear the extent to which reduced GABA levels were the result of general age-related decreases in inhibitory control or also related to a loss in sensory input. Our within-subject mediation approach was used to test the hypothesis that age-related loss in neural activity at the AN, coupled with lower GABA levels, contributes to increased responsiveness at the cortex (central gain), and poorer SIN.

Materials and Methods

Participants

Fifty older adults (age 55–86 years, mean age 66.4 years, 41 female) and 27 younger adults (age 18–29, mean age 23.5 years, 18 female) were recruited from the Charleston, South Carolina, 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 CNS disorders, use of neuroactive drugs, and contraindications for safe magnetic resonance imaging scanning. Younger participants were required to have pure-tone thresholds ≤20 dB hearing loss (HL) from 0.25 kHz 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 those of 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 1. As an estimate of hearing thresholds, the pure-tone threshold average (PTA) was calculated for each participant, representing the average hearing thresholds in the test ear from 0.5 kHz to 4 kHz. PTA in older adults ranged from 0 dB HL to 43.75 dB HL. Participants provided written informed consent before participating in this study approved by the Medical University of South Carolina Institutional Review Board.

Figure 1.

Pure-tone audiograms. Pure-tone air conduction thresholds at audiometric frequencies (0.25 kHz to 8 kHz) for the right (test) ear for younger adults (red lines) and older adults (black lines). Pure-tone thresholds for younger adults were required to be ≤20 dB HL at each frequency. Pure-tone thresholds for older adults ranged from 0 to 65 dB HL from 0.25 kHz to 4 kHz.

Measures of AN and cortical function

To assess central gain, we examined the amplitude of the AN response and the amplitude of the cortical positive peak (P1) and negative peak (N1) auditory evoked response elicited in the same participants in response to a 100 µs rectangular pulse. AN responses were estimated from the amplitude of the N1 peak of the compound action potential (CAP), which represents summated activity from the AN. The scalp-measured P1-N1 cortical response amplitude was measured to quantify auditory cortical activity.

AN response acquisition and analysis

The N1 of the CAP was elicited by 100 µs rectangular pulses, alternating polarity, presented at 11.1/s through an ER3C Insert Earphone (Etymotic Technologies). Stimuli and stimulus triggers were created and controlled using RPvdsEx from Tucker Davis Technologies (TDT). Stimuli were presented at 100 dB pSPL. N1s were recorded in two 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 TDT headstage connected to the bipolar channels of a Neuroscan SynAmpsRT amplifier in AC mode with 2010× gain (Compumedics). Testing was done in an acoustically and electrically shielded room. Participants reclined in a chair and were encouraged to rest quietly for the duration of testing. Participants were allowed to sleep during CAP recording sessions only. Continuous neural activity was analyzed off-line in MATLAB (MathWorks) using the toolboxes EEGlab (Delorme and Makeig, 2004) and ERPLab (Lopez-Calderon and Luck, 2014). Continuous EEG signals were bandpass filtered between 0.150 kHz and 3 kHz. Stimulus triggers were sent by TDT RPvdsEx and were shifted to account for the 1 ms delay introduced by the earphones 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 ms to 0 ms prestimulus baseline (McClaskey et al., 2018). Trials were identified and rejected on the basis of a peak threshold deflection of 45 µV and by visual inspection. Epoched responses for the remaining trials were averaged. N1 peak selection was performed by two independent and experienced reviewers and assessed for repeatability across multiple runs. Although only the 100 dB pSPL condition is reported here to match cortical levels, CAP responses were collected from 70 to 110 dB pSPL to aid peak detection. The N1 peak-to-baseline amplitude and peak latency were measured in ERPlab using custom MATLAB functions.

Cortical P1-N1 acquisition and analysis

Cortical P1-N1 amplitudes were recorded from a 64-channel Neuroscan QuickCap (international 10–20 system) connected to a SynAmps RT 64-Channel Amplifier in AC mode with 2010× gain. Bipolar electrodes placed above and below the left eye recorded vertical electro-oculogram activity. Curry 8 software was used to record the EEG signal at a 1 kHz sampling rate. Neural activity was recorded while participants were passively listening to a 100 µs 100 dB pSPL alternating polarity click with a 2000 ms interstimulus interval (20 ms jitter). The click stimulus was identical to that presented for the CAP but at a slower rate. Both the stimulus and trigger were created, controlled, and presented via RPvdsEx from TDT. Sessions of 200 clicks were recorded from each participant. Continuous EEG data were processed off-line using a combination of EEGLab (Delorme and Makeig, 2004) and ERPLab (Lopez-Calderon and Luck, 2014). The recorded EEG data were downsampled to 0.5 kHz, bandpass filtered from 1 to 30 Hz, re-referenced to the average of all electrodes, and corrected for ocular artifacts using independent components analysis. Individual trials were then segmented into epochs around the click onset (−100 ms to +500 ms) and baseline corrected (−100 ms to 0 ms). Any epochs contaminated by peak-to-peak deflections in excess of 100 µV were rejected using an automatic artifact rejection algorithm. For each participant, epoched data were averaged across trials to compute the average waveform [event-related potential (ERP)]. ERPs were analyzed from a cluster of frontal-central electrodes: F1, FZ, F2, FC1, FCZ, FC2, C1, CZ, C2 (Tremblay et al., 2001; Narne and Vanaja, 2008; Harris et al., 2012; Dias et al., 2019). The latencies of the first prominent negative peak (C1/N50), first prominent P1, second prominent N1, second prominent positive peak (P2), and third prominent negative peak (N2) were recorded. These latencies were used to compute temporal intervals for the automatic detection of the P1, N1, and P2 peak amplitudes and latencies in each of our channels of interest using custom MATLAB scripts. P1 was defined as the maximum positive peak between C1/N50 (or from stimulus onset if no C1/N50 was detected) and N1. N1 was defined as the maximum negative peak between P1 and P2. P2 was defined as the maximum positive peak between N1 and N2. The P1-N1 amplitude was then calculated as the difference between the positive peak amplitude of P1 and the negative deflection of N1 amplitude to yield a single cortical peak amplitude for each participant. If a P1 or N1 peak was not found in a channel between their respective latency windows defined from the average waveform across all our channels of interest, then that channel was omitted from the computed average of each component characteristic (peak amplitude and peak latency). As with previous studies, this approach allowed for automatic peak picking across channels, reducing the risk of human error, while more accurately measuring components by not considering those channels that failed to exhibit an identifiable component, likely as a result of noise from poor impedance (Anderer et al., 1996; Dias et al., 2019).

Proton magnetic resonance spectroscopy acquisition and processing

Structural magnetic resonance imaging (MRI) scans and proton magnetic resonance spectroscopy (¹H-MRS) data were acquired for a subset of participants with CAP and cortical responses (N = 65; 20 younger, mean age 23 years, 15 female; 45 older, mean age 66 years, 32 female). Siemens Trio and Prisma 3T scanners were used to collect images. The same 32-channel head coil was used with both scanners. A structural scan was first taken for voxel placement and tissue segmentation (TR = 5000 ms, TE = 2.98 ms, flip angle = 4°, 176 slices with a 256 × 256 matrix, slice thickness = 1.0 mm, and no slice gap). Following placement of saturation bands and shimming via FASTESTMAP, ¹H-MRS estimates of GABA were acquired. Because of a scanner upgrade and subsequent change in protocol, a subset of ¹H-MRS was acquired using a MEGA-PRESS sequence (N = 43, 14 younger, mean age 23 years, 11 female; 29 older, mean age 66 years, 21 female) with TR = 2000 ms, TE = 68 ms, number of averages = 150 (Mullins et al., 2014), and a subset using a HERMES sequence (N = 22, 6 younger, mean age 21 years, 4 females; 16 older, mean age 67 years, 11 females) with TR = 2000 ms, TE = 80 ms, number of averages = 320 (Chan et al., 2016). Unsuppressed water spectra were coacquired for each sequence and voxel location. ¹H-MRS data in all sequences were acquired from a 30 mm × 25 mm × 25 mm voxel in our region of interest in left auditory cortex (Fig. 2A). A second voxel was placed in a control region to test the specificity of our effects for auditory cortex, for MEGA-PRESS data a voxel was placed in posterior parietal cortex (Fig. 2B), and for HERMES data a voxel was placed in occipital cortex (Fig. 2C). Each voxel was placed by a trained technician and oriented to avoid any overlap with the skull or ventricles. The auditory cortex voxel was centered on Heschl's gyrus. The parietal control region was placed above the parietal-occipital fissure, and the visual cortex voxel was centered on primary visual cortex between the parietal-occipital and calcarine fissures. HERMES and MEGA-PRESS offer similar reliabilities for measures of GABA (Prisciandaro et al., 2020).

Figure 2.

¹H-MRS voxel placement and GABA spectrum. A–C, Sample auditory cortex (A), parietal lobe (B), and occipital lobe (C) voxel placement. Samples were selected from one younger (A) and two older participants (B, C). Note the difference in atrophy between participant A, B, and C and therefore voxel composition across participants. D, E, Example of fitted MEGA-PRESS (D) and HERMES (E) from the auditory voxel. The GABA-edited spectrum is shown in blue (plotted as a function of ppm). Overlaid in red is the model of best fit. Below the plot, the residual between these two is shown in black.

GABA values (MEGA-PRESS and HERMES) were processed using the Gannet MATLAB tool kit (Edden et al., 2014). Only metabolites with fitting uncertainties <20% were retained. Water was quantified from a Gaussian–Lorentzian fit to the nonwater suppressed data. Quality-control evaluation of ¹H-MRS spectra resulted in minor data loss, and the number of cases available for analyses was N = 59 (20 younger, mean age 23 years, 15 female; 39 older, mean age 65 years, 11 female). Within-voxel tissue fractions of gray matter (GM), white matter (WM), and CSF were calculated based on automated segmentation using the Statistical Parametric Mapping 12 toolbox for MATLAB (Wellcome Department of Cognitive Neurology) using a volume mask generated in Gannet. Metabolite concentrations (GABA) were normalized to unsuppressed water (metabolite/water). The amount of GABA in the MRS voxel depends in part on its tissue composition. The composition of tissue types is inevitably heterogeneous because of the large size of the MRS voxel, and this effect is even more pronounced in studies examining populations with atrophy because of age or illness. Each tissue type contains a different amount of GABA. GM, where the majority of GABAergic interneurons are found, contains more GABA than WM (Bhattacharyya et al., 2011). CSF contains negligible amounts of GABA. We analyzed two measures of GABA—the metabolite/water ratio corrected for within-voxel CSF fraction (Prescot and Renshaw, 2013), hereafter referred to as GABA_csf-corr, and nontissue corrected GABA level, hereafter referred to as GABA_raw, which is still referenced to water but not corrected for tissue concentrations within the voxel. GABA_csf-corr represents the abundance of intracellular and extracellular GABA relative to the cortical volume. Nontissue corrected GABA_raw provides information about the overall number of GABA-containing cells in the cortex.

Speech recognition in noise

We measured speech recognition in noise (SIN) using the Quick Speech-in-Noise Test (QuickSIN; Etymotic Research; Killion et al., 2004). QuickSIN was collected in a subset of participants with CAP N1 and cortical P1-N1 data (N = 72; 24 younger, 16 female; 48 older, 35 female). Of these, all 24 younger adults and 39 of the 48 older adults also had spectroscopy data. The QuickSIN materials include five lists of six sentences each, with each sentence containing five keywords, for a total of 30 keywords in each list. Background 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 binaurally 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 AA222 Audio Traveler. We computed the average number of keywords (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 reported as SNR loss, calculated as 25.5 minus total key words correct out of 30. Lower values of SNR loss represent better SIN.

Experimental design and statistical analysis

Statistical analyses were performed in R software using multivariate analysis, general linear modeling, and generalized linear mixed-effects modeling (lme4 package; Bates et al., 2015), and path analysis (lavaan package; Rosseel, 2012). Coefficient estimates (B) and standardized coefficients (β) are reported for the parameters of the tested models. The Benjamini–Hochberg procedure was used to correct for false discovery rate using p-adjust in R.

ERP analyses and central gain

We used a general linear model multivariate approach to test for the presence of central gain. We hypothesized that if central gain is present and results in part from reduced afferent input, then older adults would have AN response amplitudes that are smaller than those of younger adults but cortical responses that are similar to or larger than those of younger adults. Amplitudes of the N1 of the CAP were used to quantify AN activity. and amplitudes of the early obligatory potentials, the P1, the N1, and the combined P1-N1, were used to quantify cortical activity. Within our statistical models, AN and cortical response amplitudes were the outcome variables, and age group (younger or older) was the fixed factor predictor.

Second, we used separate linear regression models in younger and older adults to test associations between AN response amplitudes and cortical response amplitudes in both groups. To reduce the number of comparisons, we used the combined P1-N1 complex response (peak-to-peak amplitude deflection) as the outcome variable for cortical responses, although similar associations were identified with the P1 and N1 independently (data not reported). We predicted that in younger adults, larger AN response amplitudes would predict larger cortical response amplitudes. We used the results from the regression model for AN and cortical responses for the younger adults and the AN response amplitudes of the older adults to determine the extent to which cortical response amplitudes in older adults were larger than predicted by their AN response amplitudes. We calculated older adults' predicted cortical responses based on the parameter estimates from the regression model for younger adults using the predict.lm function in R. We then subtracted the predicted P1-N1 response amplitudes from the observed P1-N1 response amplitudes and used this as a metric of central gain for older adults (central gain equals observed P1-N1 amplitude minus predicted P1-N1 amplitude). Central gain can only be calculated in older adults because predicted P1-N1 amplitudes are based on the regression model from the younger adults. Using this predictive regression approach, we were able to compare the relationship between the AN and cortical activity in older adults relative to a reference or control group (younger adults) while also generating a metric of central gain. By calculating older adults' expected cortical responses based on their individual differences in afferent input, relative to a control group, this approach can directly test our hypothesis that decreased AN amplitudes contribute to central gain in older adults. Although previous studies have examined central gain at the level of the brainstem by examining ratios of central and peripheral components (Psatta and Matei, 1988; Grose et al., 2019), ratios make it difficult to determine where in the auditory system differences arise and the direction of effects. This approach is similar to work in our lab that assessed central gain in the brainstem of mice and older adults (Rumschlag et al., 2022)

We then examined the extent to which central gain for older adults was predicted by individual differences in PTA and GABA using linear regression and model testing. Separate regression models were used to test associations between central gain and GABA in auditory cortex and in our control region.

We used linear regression and model testing to examine associations between CAP amplitudes, GABA, central gain, PTA, and SIN in older adults. To identify effects of AN loss, AN amplitudes were entered in the model as the predictor variable. We then tested models with GABA entered. Next, we entered central gain as a predictor variable in the model. Finally, we entered PTA as a predictor variable. We used model testing to compare the total amount of variance explained in each of our models.

Building on this regression approach, we examined the extent to which inhibition (GABA) mediated the relationship between AN response amplitudes and central gain in older adults by conducting a path analyses in R using the lavaan package (Rosseel, 2012). Multiple mediation analyses allowed us to test the effect of one proposed mediator while accounting for the effects of other variables, or the extent to which a third variable (i.e., GABA) accounts for the relationship between two variables (i.e., AN responses and central gain). These mediation models consider the impact of an intervening variable (a mediator), M, which is posited to transmit the influence of an independent variable, X, onto an outcome, Y. To test our hypothesis, we first estimated the effect of afferent input (AN response amplitudes) on central gain (observed-predicted P1-N1 amplitude), and the effects of central gain and PTA on SIN (Model A). We then tested the extent to which GABA mediates the relationship between AN responses and central gain (Model B). We assessed the magnitude of effects using standardized β coefficients and evaluated the reliability of the mediation effect using percentile (nonparametric) 95% confidence intervals generated by a bootstrapping procedure with a resample rate of 10,000 (Preacher and Hayes, 2008; Hayes and Scharkow, 2013). We evaluated model fit using recommended fit indexes, that is, the chi-square exact test of model fit, the root mean square error of approximation (RMSEA), the standardized root mean square residual (SRMR), and the Comparative Fit Index (CFI) (Fairchild and McDaniel, 2017). We report the variance in central gain that is explained by the focal predictor (AN response amplitudes) and by the mediation path (GABA), and we report the variance in SIN that is explained by central gain and PTA. To support the hypothesis that GABA mediates the relationship between AN response amplitudes and central gain, the strength of the direct relationship between AN response amplitudes and central gain (AN response amplitudes→central gain) must be either reduced in magnitude or no longer significant when the GABA mediation path (AN response amplitudes→GABA→central gain) is added into the model. If the direct association between AN response amplitudes and central gain becomes nonsignificant on the addition of the GABA mediation term, then the relationship between the two variables is fully mediated by GABA so that the effect of X (AN response amplitudes) on Y (central gain) is conveyed entirely through the mediator M (GABA). If the direct path remains significant, the relationship is said to be partially mediated.

¹H-MRS analysis

We used independent samples t tests to assess differences in spectroscopy data collected with MEGA-PRESS versus HERMES. No significant differences in demographics (age, sex), fit measures, or metabolites were identified (p > 0.05), and including acquisition type did not improve model fit when examining relationships across metrics. Therefore, data from the two sequences were combined for subsequent analyses. We used independent samples t tests to identify age-group differences in fit error and unsuppressed water amplitude. Associations among GABA and age group, PTA, and SIN were examined using linear regression and model testing. Tissue composition (GM/GM plus WM) was entered as a covariate in each model.

SIN analyses across age group

We used linear regression and model testing to examine associations among age group, PTA, SIN, and GABA. To identify effects of age, age group was entered in the model as the predictor variable. We then tested models with GABA entered separately from each voxel location. Next, we entered PTA as a predictor variable in the model. We used model testing to compare the total amount of variance explained in each of our models.

Sex was included as a covariate in each model. However, sex was not a significant predictor of central gain, GABA, or SIN, and did not improve model fit (p > 0.05) and is therefore not reported here.

Results

AN and cortical responses in younger and older adults

We hypothesized that age-related loss of afferent activity contributes to central gain. We first examined the extent to which aging results in central gain using a multivariate approach that examines the effects of age group on AN and cortical activity while accounting for subject-level effects. AN activity was predicted to be reduced in older adults relative to younger adults, whereas cortical activity in older adults was predicted to be equivalent to or larger than that of younger adults. AN and cortical response waveforms are shown in Figure 3. Consistent with our prior reports and those of others (Burkard and Sims, 2001; Konrad-Martin et al., 2012; Anderson et al., 2021; Harris et al., 2021), age group was a significant predictor of AN response amplitude, with older adults exhibiting significantly smaller AN responses than younger adults (Table 1). In contrast, P1 amplitudes were significantly larger in older than younger adults, and N1 and P1-N1 cortical responses elicited by the same stimulus resulted in response amplitudes that were not significantly different between older and younger adults (Table 1). Pure-tone average was not a significant predictor of AN response amplitudes or of cortical response amplitudes, nor did it improve model fit (Table 1). Individual response amplitudes for the AN N1 and the cortical P1-N1 responses are provided in Figure 4.

Figure 3.

Group average waveforms. A, Group averaged CAP waveforms for younger (red line) and older (black line) adults in response to a 100 dB pSPL click measured from a tympanic membrane electrode. The N1 peak is labeled, indicating the summed population response from the auditory nerve. B, Group averaged cortical P1-N1-P2 waveforms for younger (red line) and older (black line) adults in response to a 100 dB pSPL click measured from a cluster of frontal-central scalp electrodes. The P1, N1, and P2 peaks are labeled. Despite decreases in the CAP N1 amplitude, cortical response amplitudes are similar or larger in older adults compared with younger adults.

Table 1.

Associations between age group and response amplitudes

Models	β	SE	t	p
N1 auditory nerve amplitude
Age group	0.117	0.048	2.453	0.000***
PTA	−0.023	0.029	−0.778	0.439
P1 cortical response amplitude
Age group	0.340	0.100	3.408	0.001***
PTA	−0.049	0.104	−0.472	0.638
N1 cortical response amplitude
Age group	0.004	0.146	0.027	0.978
PTA	0.080	0.090	0.889	0.377
P1-N1 cortical response amplitude
Age group	0.275	0.166	1.658	0.101
PTA	−0.048	0.103	−0.472	0.638

***p ≤ 0.001.

Figure 4.

Auditory nerve (N1) and cortical (P1-N1) peak amplitudes. A, B, Peak amplitudes for the N1 of the CAP response (A) and the cortical P1-N1 response (B) for younger (red) and older (black) adults. For comparison, CAP amplitudes are plotted with negative values up, going from 0 to −1 so that larger responses are plotted in the positive direction in both A and B. Median values are marked by a solid line. CAP N1 peak amplitudes were significantly smaller in older adults than in younger adults (***p < 0.001). P1-N1 response amplitudes were not significantly different for older and younger adults.

In younger adults, larger AN responses predicted larger P1-N1 cortical responses (B = −1.42, SE = 0.56, β = 0.45, t₍₂₅₎ = 2.54, p = 0.017; Fig. 5A), but this was not the case in older adults (B = −0.28, SE = 0.67, β = −0.061, t₍₄₈₎ = −0.42, p = 0.68; Fig. 5B). Figure 6A plots the predicted P1-N1 response as a function of the observed P1-N1 response for older adults. Observed P1-N1 response amplitudes for most older adults were significantly larger than the predicted P1-N1 amplitude (paired t test, t₍₅₉₎ = −4.187, p < 0.001; Fig. 6B).

Figure 5.

CAP N1 amplitudes predict cortical P1-N1 amplitudes in younger but not older adults. A, B, CAP N1 amplitudes plotted relative to cortical P1-N1 amplitude in younger (A) and older (B) adults. CAP N1 amplitudes are plotted with increasing values (more negative) from left to right. Larger CAP N1 amplitudes were predictive of larger cortical P1-N1 amplitudes in younger but not older adults. Solid black line in A represents the significant association in younger adults.

Figure 6.

Observed cortical P1-N1 responses in older adults were larger than predicted. Significant associations were observed between the CAP N1 and P1-N1 amplitudes in younger adults (Fig. 4), and we used that model from younger adults to predict cortical responses in older adults based on their CAP amplitudes. A, Observed P1-N1 response amplitudes plotted against predicted P1-N1 amplitudes for older adults. Data points to the right of the line represent individuals with larger-than-predicted P1-N1 amplitudes. B, Mean observed and predicted amplitudes. Observed P1-N1 response amplitudes for older adults were significantly larger than amplitudes predicted from responses of younger adults (p < 0.001).

Factors that predict central gain in older adults

Our first model revealed that PTA was not a significant predictor of central gain in older adults (B = 0.01, SE = 0.01, β = −0.14, t₍₃₈₎ = −0.99, p = 0.33). Consistent with our previous study (Harris et al., 2021), individual differences in N1 amplitude were also not associated with PTA in older adults (r₍₃₈₎ = −0.05, p = 0.75). In our second model, GABA_csf-corr was the dependent variable. To control for differences in atrophy within the voxel, the amount of GM within the voxel was also included as a dependent variable. GABA_csf-corr was a significant predictor of central gain, with lower levels of GABA_csf-corr associated with more central gain (Fig. 7, Table 2). This association was specific to GABA_csf-corr in Heschel's gyrus and not associated with GABA_csf-corr in our control region (B = 0.16, SE = 0.35, β = −0.09, t₍₃₈₎ = −0.47, p = 0.64).

Figure 7.

Lower levels of GABA_csf-corr in auditory cortex predict greater central gain. Central gain, defined as the observed P1-N1 amplitude minus the predicted P1-N1 amplitude in older adults, plotted as a function of GABA_csf-corr levels in auditory cortex. Solid line represents the significant association between GABA and central gain.

Table 2.

Associations between central gain and GABA in auditory cortex

Model	B	β	SE	t	p
Model: predicting central gain
GABA	−0.792	−0.403	0.337	−2.345	0.026*
Fraction gray matter	3.791	0.233	2.793	1.357	0.185

*p < 0.05.

Increased central gain is associated with poorer speech recognition in noise

We used linear regression and model testing to determine the extent to which AN response amplitudes, GABA, central gain, and PTA predicted SIN. Models are provided in Table 3. In our first model we found that AN response amplitudes were not a significant predictor of SNR loss in older adults. When GABA was added to the model, it did not significantly predict SNR loss in older adults and did not improve model fit [Χ²(1) = 0.22, p = 0.64]. Central gain was found to be a significant predictor of SNR loss (Fig. 8), and significantly improved model fit [Χ²(1) = 10.97, p = 0.012]. PTA was not a significant predictor of SIN in older adults and did not improve model fit [Χ²(1) = 1.84, p = 0.19], and central gain remained significant. These results suggest that greater central gain in older adults, as indicated by their larger cortical responses relative to reduced afferent input, may contribute to poorer SIN independently of hearing sensitivity.

Table 3.

Associations between speech recognition in noise, AN amplitudes, GABA, central gain, and PTA

Model	B	β	SE	t	p
Model 1
CAP N1 amplitude	0.003	0.000	1.835	0.001	0.999
Model 2
CAP N1 amplitude	−0.195	−0.020	1.919	−0.102	0.920
GABA auditory region	−0.279	−0.079	0.706	−0.395	0.696
Fraction GM	−1.749	−0.081	4.258	−0.411	0.685
Model 3
CAP N1 amplitude	−1.644	−0.169	1.785	−0.921	0.366
GABA auditory region	0.378	0.107	0.671	0.563	0.578
Fraction GM	−3.665	−0.171	3.852	−0.952	0.350
Central gain	0.914	0.547	0.327	2.796	0.010*
Model 4
CAP N1 amplitude	−1.378	−0.142	1.765	−0.781	0.443
GABA auditory region	0.150	0.042	0.680	0.220	0.827
Fraction GM	−3.753	−0.175	3.786	−0.991	0.331
Central gain	0.894	0.535	0.322	2.781	0.010*
PTA	0.046	0.244	0.034	1.372	0.183

*p = 0.01.

Figure 8.

Central gain predicts older adults' speech recognition in noise (QuickSIN). QuickSIN scores are represented as SNR loss, calculated as SNR loss = 25.5 – [number of keywords correctly identified of 30 total], with lower values of SNR loss indicating better speech recognition in noise. SNR loss was predicted by central gain in older adults, calculated as the difference between the observed P1-N1 and the predicted P1-N1 amplitude, with more central gain contributing to poorer SIN. Vertical gray line indicates central gain of zero. Central gain values greater than zero represent larger-than-predicted cortical response amplitudes. Solid black line indicates a significant relationship across variables (Table 2).

The pattern of results presented thus far are consistent with our hypothesis that declines in the AN in older adults may lead to changes in GABA, which in turn increase cortical response amplitudes and contribute to poorer SIN. These relationships for older adults were tested together in a path analysis represented in Figure 9, the parameters of which are reported in Table 4. The first model (Model A) illustrates how decreases in AN amplitudes for older adults may contribute to increased central gain (P1-N1 observed minus P1-N1 predicted). The chi-square test of exact model fit suggests that the model-implied variance-covariance matrix adequately reproduced the data, χ²(2) = 0.733, p = 0.693. Additional indexes also met the criteria for acceptable model fit (RMSEA = 0.000, CFI = 1.000). This suggests that increased central gain contributes to poorer SIN (i.e., higher SNR loss). We also included PTA in this model, but this association did not reach significance (p = 0.08). We next tested whether changes in GABA mediated the relationship between AN amplitudes and central gain (Model B). The chi-square test of exact model fit suggested that the model-implied variance–covariance matrix adequately reproduced the data [χ²(5) = 3.905, p = 0.563]. Additional indexes also met the criteria for acceptable model fit (RMSEA = 0.000, CFI = 1.000). We present Model A with the caveat that AN amplitudes were used to calculate predicted cortical responses and that predicted cortical responses were used to calculate central gain. Importantly, regression analyses found that predicted responses were significantly smaller than observed responses, and AN responses did not predict observed cortical amplitudes. Model A found that the differences between observed and predicted cortical responses, our measure of central gain, were predicted by AN responses. The relationship between AN amplitudes and central gain found in Model A may be explained by the nonindependence of their values (AN responses were used to calculate central gain). However, Model B found that this relationship becomes nonsignificant when GABA is added to Model A as a mediator between AN responses and central gain. This is important because it establishes that the relationship between AN responses and central gain estimates is explained by GABA, a factor that is calculated independent of both AN responses and the observed and predicted cortical responses used to compute central gain. Significant associations between AN amplitudes and GABA, GABA and central gain, and the now-nonsignificant association between AN amplitudes and central gain (p = 0.20), are consistent with our hypothesis that individual differences in GABA mediate the relationship between AN responses and central gain.

Figure 9.

A, Path analysis representing how smaller AN responses in older adults are associated with increased central gain, and increased central gain results in poorer QuickSIN performance (SNR Loss). B, Lower levels of auditory cortex GABA mediate the relationship between AN response and central gain. When GABA is added into the model, the association between AN responses and central gain is no longer significant, suggesting that the relationship between AN responses and central gain is fully mediated by GABA. Associations between individual differences in PTA and SNR loss in older adults did not reach significance (p = 0.08). Numbers describing relationships are reported as standardized coefficients (β). Solid lines represent significant associations. Associations with dotted lines were not significant. The parameters of the full model are reported in Table 4.

Table 4.

Parameter estimates for path analysis testing

Model	B	SE	β	z	p
Model A
CAP amplitude→central gain	2.081	0.829	0.347	2.734	0.006**
Central gain→SNR loss	0.682	0.287	0.420	3.246	0.001***
PTA→SNR loss	0.071	0.040	0.256	1.740	0.082
Model B
CAP amplitude→GABA	−0.963	0.426	−0.330	−2.485	0.013**
GABA→central gain	−0.889	0.401	−0.434	−2.586	0.010**
Central gain→SNR loss	0.682	0.279	0.420	3.224	0.001***
CAP amplitude→central gain	1.209	0.939	0.202	1.269	0.205
PTA→SNR loss	0.071	0.041	0.257	1.733	0.083

**p ≤ 0.01,

***p ≤ 0.001.

GABA: effects of age group, cortical region, and PTA in younger and older adults

Next, we examined changes in GABA across the sample of younger and older adults. Prior studies of GABA in auditory cortex and associations with age and hearing loss have been mixed, with some studies reporting a significant effect of age (Profant et al., 2013; Gao et al., 2015) and others reporting no significant differences between groups (Profant et al., 2013; Dobri and Ross, 2021; Lalwani et al., 2021). Before examining effects of age group on GABA we examined several factors that may affect GABA fit, including fit error and unsuppressed water amplitude and water line width. These values did not differ between younger and older adults (p > 0.05). We also examined age differences in tissue composition in each voxel, which also affects GABA, and found that older adults showed greater atrophy within both Heschl's gyrus (t₍₅₈₎ = 6.32, p < 0.001) and our control regions (t₍₅₈₎ =3.55, p < 0.001), calculated as an increase in the CSF fraction of the voxel (decrease in GM and WM). This indicates that older adults exhibited significant atrophy within both auditory cortex and our control regions relative to younger adults.

Similar to prior studies (Profant et al., 2013; Dobri and Ross, 2021), we examined the effects of age on GABA as both nontissue-corrected GABA_raw and CSF-corrected GABA (GABA_csf-corr). These analyses are important because of age differences in tissue composition and atrophy. We found a significant effect of age on GABA_raw levels in our auditory cortex region, although after including the total fraction of GM and WM within the voxel in the model as a control, GABA_raw was no longer a significant predictor (Table 5). This suggests that age-related atrophy resulted in lower levels of GABA_raw in older adults. Estimates of GABA_csf-corr account for these differences in atrophy, and therefore there was not a significant effect of age on GABA_csf-corr (B = 2.22, SE = 3.62, β = 0.11, t₍₅₈₎ = 0.61, p = 0.54). These results highlight two factors that cooccur in auditory cortex with age, atrophy and a loss of GABA.

Table 5.

Age-group difference in GABA

Model	B	β	SE	t	p
Model 1: age group
Age group	−0.245	−0.255	0.122	−2.011	0.049*
Model 2: age group and GM
Age group	0.053	0.055	0.127	0.416	0.679
Fraction GM	5.145	0.569	1.201	4.28	0.000***

*p < 0.05,

**p < 0.01,

***p < 0.001.

In our control region, despite significant differences in tissue composition, both GABA_raw (B = 0.05, SE = 0.11, β = 0.06, t₍₅₈) = 0.46, p = 0.65) and GABA_csf-corr (B = 0.08, SE = 0.12, β = 0.08, t₍₅₈₎ = 0.65, p = 0.52) did not differ across age groups. GABA_csf-corr levels were significantly higher in our control region than in auditory cortex (t₍₅₈₎ = −2.12, p = 0.036), even after accounting for differences in SNR within the voxel and water line width across regions. Together, these results show that older adults exhibited significant atrophy within both auditory cortex and our control regions relative to younger adults but significantly lower GABA_raw levels only in the auditory voxel.

We examined associations between GABA_csf-corr and PTA within older adults and across our sample of younger and older adults. PTA was not a significant predictor of GABA_csf-corr in auditory cortex in our sample of older adults (B = 5.77, SE = 3.10, β = 0.33, t₍₃₈₎ = 1.86, p = 0.07). PTA was not a significant predictor of GABA_csf-corr in our control region (B = 2.22, SE = 3.62, β = 0.11, t₍₃₈₎ = 0.61, p = 0.54). We next examined whether GABA_csf-corr was predicted by PTA across our sample of younger and older adults. We included GABA_csf-corr as the dependent variable, and PTA and age group as predictors. Again, PTA was not a significant predictor of GABA_csf-corr (B = −005, SE = 0.01, β = −0.09, t₍₅₈₎ = −0.47, p = 0.64).

SIN: associations with age group, AN responses, GABA and PTA

There is emerging evidence that GABA levels may predict SIN (Dobri and Ross, 2021), and as shown earlier in older adults, this association may be driven by associations with central gain. However, we predicted that based on the role of GABA in signal-in-noise detection in animal models (Resnik and Polley, 2021), and in neural distinctiveness in humans (Lalwani et al., 2019), GABA levels may be predictive of SIN across the sample. We used linear regression and model testing to determine the extent to which age group, AN amplitudes, GABA, tissue composition (GM/GM plus WM), and PTA predicted SIN. Models are provided in Table 6. Age group (Model 1) was not a significant predictor of SIN. AN amplitudes were also not a significant predictor of SIN (Model 2), nor did they improve model fit [Χ²(1) = 1.00, p = 0.410], therefore AN amplitudes were removed from subsequent models. GABA was a significant predictor of SIN, with lower levels of auditory cortex GABA_csf-corr predicting poorer SIN (Model 3). We then added PTA, which was found to be a significant predictor of SIN (Model 4). Both PTA and GABA_csf-corr remained significant predictors of SIN (Fig. 10, Model 4), and improved model fit [Χ²(1) =4.57, p = 0.038]. GABA_csf-corr from our control region was not a significant predictor of SIN (B = 0.23, SE = 0.53, β = 0.06, t₍₅₈₎ = 0.43, p = 0.67), and adding GABA_csf-corr from the control region to the above models did not improve model fit [Χ²₍₁₎ = 0.56, p = 0.46]. These results suggest that auditory cortex GABA and pure-tone thresholds are each unique predictors of SIN in both younger and older adults.

Table 6.

Associations between SIN, age group, GABA+, and PTA

Model	B	β	SE	t	p
Model 1
Age group	0.394	0.144	0.375	1.051	0.298
Model 2
Age group	0.412	0.159	0.38	1.086	0.283
CAP N1 amplitude	0.91	0.136	0.974	0.934	0.355
Model 3
Age group	0.67	0.245	0.426	1.574	0.122
GABA auditory region	−0.918	−0.314	0.397	−2.313	0.025*
Fraction GM	6.124	0.237	4.123	1.485	0.144
Model 4
Age group	−0.095	−0.035	0.545	−0.174	0.863
GABA auditory region	−0.857	−0.293	0.384	−2.228	0.031*
Fraction GM	7.015	0.272	4.005	1.751	0.086
PTA	0.046	0.407	0.021	2.138	0.038*

*p < 0.05.

Figure 10.

Auditory GABA+ and PTA predicted SNR loss. A, B, Lower levels of GABA+ in auditory cortex predicted poorer speech recognition in noise (higher SNR loss) in older (gray circles) and younger (red circles) adults (A). Adding PTA to the model significantly improved model fit, and higher PTA thresholds were associated with poorer speech recognition in noise (B). Statistics are provided in Table 5. Solid lines represent significant associations across variables.

Discussion

The underlying mechanisms and perceptual consequences of sensory-driven plasticity in older adults are largely unknown. Evidence from animal models suggests that both aging and hearing loss result in a slow peripheral deafferentation that propagates throughout the auditory system and contributes to changes in excitation and inhibition at the level of auditory cortex (Caspary et al., 2008; Chambers et al., 2016; Herrmann and Butler, 2021). Translating this to older humans, the current study (1) demonstrates central gain in older adults by comparing neural responses from both the AN and cortex; (2) identifies a potential underlying mechanism contributing to this gain, specifically lower levels of GABA; and (3) demonstrates that increasing central gain and lower GABA levels are associated with poorer SIN.

Central gain: decreased afferent innervation and inhibition contribute to larger cortical responses

Our results are consistent with animal models of central gain and suggest that individual differences in AN afferent input may contribute to changes in cortical encoding. Consistent with our prior work and that of others (Burkard and Sims, 2001; McClaskey et al., 2018; Anderson et al., 2021; Harris et al., 2021), we demonstrated a robust decrease in AN response amplitudes for suprathreshold stimulus levels that occurs independently of differences in PTA, suggesting decreased afferent innervation in older adults. Prior studies have examined central gain in the periphery by comparing wave I of the ABR to wave V, generated in the midbrain, and report decreased wave I amplitudes relative to wave V amplitudes in older adults (Grose et al., 2019; Rumschlag et al., 2022). Extending these results to the cortex, we observed larger-than-predicted cortical responses relative to AN responses in many older adults. The presence of central gain, although significant, was not observed in all older adults (Fig. 6A) and may be driven by changes in cortical levels of inhibition (Fig. 9).

Lack of association with pure-tone thresholds

Changes in AN activity, cortical response amplitudes (P1, N1, and P1-N1), central gain, and GABA were not associated with individual differences in PTA. It is now well established that individual differences in PTA are not associated with suprathreshold AN responses in older adults (Burkard and Sims, 2001; Konrad-Martin et al., 2012; Grose et al., 2019; Harris et al., 2021) and that deficits in AN activity occur in older adults with a range of pure-tone thresholds. To date, associations between PTA and GABA+ in auditory cortex have been equivocal, with one study showing that increased hearing thresholds were associated with lower levels of GABA in older adults (Gao et al., 2015), whereas other studies show no associations (Profant et al., 2013; Dobri and Ross, 2021; Lalwani et al., 2021). The current study examined this association in a relatively large sample of older adults with a wide range of PTAs and found no association between GABA and PTA in older adults or when examining associations across age groups. Central gain was also not associated with PTA. Together, these results suggest that standard clinical assessments of hearing thresholds may not be indicative of age-related changes in afferent innervation and sensory-driven plasticity.

Effects of age group on GABA

Consistent with results from the current study, prior studies often demonstrate an age-related reduction in GABA when examining GABA concentration that has not been corrected for atrophy and tissue composition within the voxel, suggesting that age-related reductions in GABA are in part because of age-related atrophy (Lalwani et al., 2019; Dobri and Ross, 2021). However, age-related atrophy was evident in both our auditory and control regions, yet only GABA from auditory cortex showed an age-related decrease. Effects of age on GABA in sensory cortices appears to be an area where future study is needed as levels of GABA in visual cortex have been reported to decrease (Simmonite et al., 2019; Chamberlain et al., 2021) and increase (Pitchaimuthu et al., 2017) with increasing age depending on the study, similar to findings in auditory cortex discussed above.

Relationship to SIN

We used path analyses to test the hypothesis that central gain, driven by a loss of afferent input and decreased GABA, contributes to poorer SIN in older adults. By examining central gain and GABA within the same participants, our results bridge work from animal models that suggest that central gain in the auditory system may disrupt signal-in-noise encoding (Resnik and Polley, 2021). Across the sample of older and younger adults, GABA levels but not AN amplitudes were related to SIN performance. Although the current study examined AN response amplitudes, our prior work has reported that AN neural synchrony contributes to changes in SIN performance (Harris et al., 2021). Together, results suggest that AN neural synchrony may have an impact on SIN, and decreases in afferent input may indirectly affect SIN when coupled with cortical changes in GABA and central gain in older adults. Moreover, associations between lower levels of cortical GABA and poorer SIN observed across the sample are consistent with previous studies reporting associations between lower GABA levels and poorer SIN and decreased neural distinctiveness (more similar neural activity across types of auditory stimuli; Lalwani et al., 2019; Dobri and Ross, 2021). These associations occur independent of the effects of PTA on SIN, suggesting that individual differences in GABA not related to age or hearing loss contribute to SIN. There is accumulating evidence in both animals and humans that decreased cortical inhibition and hyperactivity may negatively affect spectral, spatial, and temporal processing, contributing to deficits in SIN (Gleich and Strutz, 2011; Millman et al., 2017; Goossens et al., 2018; Herrmann and Butler, 2021; Resnik and Polley, 2021). Future studies are needed to identify the extent to which central gain in older adults, and individual differences in GABA in general, are associated with deficits in signal encoding and temporal processing.

Potential target for intervention

Understanding associations among sensory loss, cortical inhibition, neural markers, and behavioral performance may lead to targeted intervention strategies and biomarkers for intervention. For example, in the cognitive domain, greater brain signal variability is associated with better cognitive performance and higher levels of cortical GABA (Shew et al., 2011; Nugent et al., 2015; Sadeh and Clopath, 2021). In older adults with poor neuronal variability, pharmaceutically boosting levels of GABA increased neural variability (Lalwani et al., 2021). However, altering GABA may have widespread effects. Although the results presented here focus on auditory processing, there is emerging evidence that even moderate hearing loss increases cross-modal plasticity (Campbell and Sharma, 2014; Glick and Sharma, 2020). With hearing loss, many of the reorganized neurons (cross-modal plasticity) are multisensory neurons and respond to both auditory and visual cues (Meredith et al., 2012; Schormans et al., 2017). Moreover, although lower levels of GABA are associated with poorer unimodal perception (auditory, visual, and somatosensory) and cognitive processing (Gleich et al., 2003; Leventhal et al., 2003; Caspary et al., 2008; Gleich and Strutz, 2011; Quetscher et al., 2015; Balz et al., 2016; Ding et al., 2017; Pitchaimuthu et al., 2017; Cisneros-Franco et al., 2018; Leonte et al., 2018; Teichert et al., 2018; Cassady et al., 2019; Lalwani et al., 2019, 2021; Simmonite et al., 2019; Chamberlain et al., 2021; Dobri and Ross, 2021; Maes et al., 2021), they may also contribute to increased cross-modal plasticity and multisensory integration (Desgent and Ptito, 2012; Mao and Pallas, 2013). Therefore, it is still largely unknown how modifying GABA in auditory cortices of older adults may affect SIN, particularly if considering cross-modal audiovisual (AV) SIN. These unknowns are further confounded by the fact that alteration at the cortical level may alter central gain but will not restore afferent input. However, our results suggest that poorer SIN was associated with larger central gain, which reflects amplified cortical activity beyond that predicted by reduced AN function, and may therefore be amenable to intervention.

Summary and conclusions

In summary, these results demonstrate the following: (1) differences in the relationships between peripheral and central auditory responses between younger and older adults are consistent with central gain in older adults, (2) lower levels of GABA in older adults mediate the relationship between AN and central gain, (3) GABA levels do not appear to be driven by PTA or age group, (4) lower GABA levels in auditory cortex in older adults may be dependent on increased cortical atrophy, and (5) increased central gain in older adults, lower levels of GABA, and increased PTA are associated with poorer SIN. Together, the results suggest that sensory-driven plasticity in older adults may be maladaptive in relation to auditory SIN and highlights a potential factor that contributes to the variation in SIN not predicted by PTA. Several questions remain, including the auditory factors that underlie associations between enhanced cortical responses, lower levels of GABA, and poorer SIN in older adults, and the extent to which altering GABA may ameliorate these deficits and contribute to improved SIN.