Neurochemical and functional reorganization of the cognitive-ear link underlies cognitive impairment in presbycusis

Abstract

Recent studies suggest that the interaction between presbycusis and cognitive impairment may be partially explained by the cognitive-ear link. However, the underlying neurophysiological mechanisms remain largely unknown. In this study, we combined magnetic resonance spectroscopy (MRS) and resting-state functional magnetic resonance imaging (fMRI) to investigate auditory gamma-aminobutyric acid (GABA) and glutamate (Glu) levels, intra- and inter-network functional connectivity, and their relationships with auditory and cognitive function in 51 presbycusis patients and 51 well-matched healthy controls. Our results confirmed reorganization of the cognitive-ear link in presbycusis, including decreased auditory GABA and Glu levels and aberrant functional connectivity involving auditory networks (AN) and cognitive-related networks, which were associated with reduced speech perception or cognitive impairment. Moreover, mediation analyses revealed that decreased auditory GABA levels and dysconnectivity between the AN and default mode network (DMN) mediated the association between hearing loss and impaired information processing speed in presbycusis. These findings highlight the importance of AN-DMN dysconnectivity in cognitive-ear link reorganization leading to cognitive impairment, and hearing loss may drive reorganization via decreased auditory GABA levels. Modulation of GABA neurotransmission may lead to new treatment strategies for cognitive impairment in presbycusis patients.

1. Introduction

Presbycusis, which is defined as binaural symmetrical progressive neurological deafness, is a form of hearing impairment that occurs with aging. Presbycusis is characterized by high-frequency hearing loss and degraded speech comprehension, particularly when in noisy environments (Gates and Mills, 2005). Hearing loss is a consequence of aging of the auditory organs and is usually related to physiological decline of the inner ear cells and auditory cortex (Liu and Yan, 2007; Peelle and Wingfield, 2016). However, impaired speech comprehension in presbycusis patients cannot be fully explained by the function of the auditory cortex (Panza et al., 2015). Accumulating evidence suggests that presbycusis is related to accelerated cognitive decline, incident cognitive impairment, and dementia (Lin et al., 2011a, 2013; Shen et al., 2018). Thus, in addition to impaired inner ear function, presbycusis patients show a central auditory processing disorder (CAPD) (Sardone et al., 2019). Compared to peripheral hearing loss, there is evidence of a stronger association between CAPD and cognitive decline (Sardone et al., 2020, 2021; Yuan et al., 2018). Based on such studies, a cognitive-ear link has been proposed suggesting that, besides the ear and auditory cortex, other cognitive-related cortices participate in auditory information processing (Lozupone et al., 2020). Accordingly, degraded auditory input in presbycusis may impact the intrinsic cognitive-ear link (Lozupone et al., 2020; Sardone et al., 2019). However, it is still largely unknown how hearing loss affects the cognitive-ear link and whether it eventually leads to cognitive impairment.

Previous magnetic resonance imaging (MRI) studies have shown that the auditory cortex and cognitive-related cortices are associated with cognitive impairment in presbycusis patients (Delano et al., 2020; Ren et al., 2018). Specifically, presbycusis was associated with gray matter (GM) volume decreases in the primary auditory cortex (i.e., Heschl’s gyrus (HG)) and other temporal lobe structures (Eckert et al., 2012; Peelle et al., 2011). Another structural MRI study using surface-based morphometry reported that presbycusis patients showed GM atrophy in several auditory cortical areas and nodes of the default mode network (DMN), including the bilateral precuneus and right posterior cingulate cortex (PCC) (Ren et al., 2018). An aberrant amplitude of low-frequency fluctuations (ALFF) in the HG, dorsolateral prefrontal cortex (dlPFC), frontal eye field, and key nodes of the DMN have also been observed in presbycusis patients and were associated with specific cognitive impairments, namely verbal learning and memory (Ren et al., 2021). Taken together, these findings suggest that presbycusis-related hearing loss may initiate a series of cascading reactions that lead to wide-ranging reorganization of the brain, rather than reorganization of a single cortical region.

By studying fluctuations in blood oxygenation level dependent (BOLD) signals in the resting state, functional MRI (fMRI) can be used to reflect neurophysiological relationships between spatially distant brain regions and to evaluate functional connectivity (FC) between brain regions (Biswal et al., 1995). Recent resting-state functional connectivity (rsFC) studies have found stronger FC between the dlPFC and posterodorsal auditory processing pathways, as well as weaker FC between the PCC and key nodes of the DMN, which were related to specific cognitive impairment in presbycusis patients (Ren et al., 2021). Although this approach can reveal aberrant rsFC between two brain regions, dysfunction of intra- and inter-network FC in the whole brain cannot be explored. To better understand the neuropathological mechanism of cognitive-ear link in presbycusis, studies at the brain network level are thus needed. Independent component analysis (ICA) is a data-driven method that divides MRI data into spatially independent components without a priori hypothesis. After removing interfering components such as heartbeat, respiration, and head movement, the remaining spatially specific components can form resting-state networks (RSNs). ICA can assess the synchronization and interactions among multiple networks, thus overcoming the shortcomings of time-dependent analysis (Calhoun et al., 2010). The reliability and validity of this approach has been verified in studies involving patients with various cognitive disorders (de Vos et al., 2018; Ruppert et al., 2021; Vettore et al., 2021).

As the main excitatory and inhibitory neurotransmitters, glutamate (Glu) and gamma-aminobutyric acid (GABA), respectively, play important roles in maintaining the balance of central auditory processing (Bartos et al., 2007). GABA also participates in neural plasticity and synchronous network oscillation, which are important for effective information processing and normal cognitive function (Bartos et al., 2007; Lalwani et al., 2019; Sumner et al., 2010). Notably, a magnetic resonance spectroscopy (MRS) study reported decreased auditory Glu levels in presbycusis patients compared with young healthy controls (Profant et al., 2013). However, GABA is difficult to measure using conventional MRS due to its relatively low level and because its resonance overlaps with other metabolites (Govindaraju et al., 2000). Mescher-Garwood point-resolved spectroscopy sequence (MEGA-PRESS), a MRS editing technique, can be used to separate GABA signals from other metabolites and effectively quantify GABA levels (Mescher et al., 1998). We previously reported decreased auditory GABA levels in presbycusis patients using MEGA-PRESS, which was closely associated with hearing loss (Gao et al., 2015). Another MEGA-PRESS study found that decreased GABA levels were related to cognitive decline in elderly people and speculated that GABA levels may play a significant role in neural dedifferentiation (Lalwani et al., 2019). Notably, generation of the BOLD fMRI signal depends on excitatory glutamatergic projection neurons, and GABAergic interneurons may indirectly regulate the BOLD signal via inhibitory feedback signals in microcircuits (Muthukumaraswamy et al., 2009). Importantly, there are key regulatory functions for glutamatergic and GABAergic systems in functionally connected cortical circuits. In a study of healthy participants, Stagg and her colleagues demonstrated that primary motor GABA levels were associated with FC strength in the motor network (Stagg, 2014). Thus, aberrant GABA and Glu levels in presbycusis patients measured by MRS are likely related to alterations in rsFC of brain networks.

In the present study, we applied MEGA-PRESS with macromolecule suppression and PRESS to examine auditory GABA and Glu levels in presbycusis patients and healthy controls. Subsequently, intra- and inter-network FC strength were investigated using an ICA-based analysis. We hypothesized that aberrant auditory metabolite levels and FC strength are involved in the reorganization of the cognitive-ear link in presbycusis. A hierarchical method was applied to test the hypothesis: (1) whether there are group differences in metabolite levels and FC strength; (2) whether metabolite levels or FC strength would be associated with hearing loss or cognitive impairment; (3) whether metabolite levels mediate the association between hearing loss and FC strength; (4) whether FC strength mediates the association between metabolite levels and cognitive impairment; and (5) whether metabolite levels and FC strength served as two mediators, mediating the association between hearing loss and cognitive impairment in presbycusis. The findings of this study will enhance our understanding of the reorganization of the cognitive-ear link at the level of RSNs and shed light on the neurochemical underpinnings of this reorganization in presbycusis patients.

2. Material and methods

2.1. Participants

Fifty-one presbycusis patients (28 males and 23 females; mean age 65.16 ± 2.43 years, range 60–70 years) and the same number of normal hearing (NH) controls (21 males and 30 females; mean age 64.67 ± 1.67 years, range 62–69 years) matched for age, gender, and education level were enrolled in this study (Table 1). All participants were of Han Chinese ethnicity, ≥ 60 years old, Mandarin speakers, and right-handed. The pure tone average (PTA) was used to evaluate hearing level at the thresholds of 0.5, 1, 2, and 4 kHz. PTA greater than a 25 dB hearing level in the better hearing ear was the inclusion criteria for the presbycusis group, while it was not more than a 25 dB hearing level for the NH group (Lin et al., 2011b). The exclusion criteria were as follows: (1) a diagnosis of asymmetrical or conductive hearing loss, Meniere’s disease, acoustic neuroma, tinnitus, or self-reported hyperacusis; (2) history of ototoxic drug use, otologic surgery, head injury or stroke, previous or current noise exposure, or hearing aid use; (3) neurological or mental illness; and (4) contraindications for MRI. This study was approved by the Shandong University Institutional Review Board and all participants provided written informed consent.

Table 1.

Demographic and clinical data of the NH group and presbycusis group.

Characteristics	NH	presbycusis	x2 /t value	df	p-value
	(n = 51)	(n = 51)
Age (years)	64.67 ± 1.67	65.16 ± 2.43	1.188	88.638	0.238a
Gender (male/female)	21/30	28/23	1.925	1	0.165b
Disease duration (years)	–	5.80 ± 4.90	–	–	–
Education (years)	11.43 ± 2.56	10.31 ± 4.37	1.578	80.663	0.119a
Anxiety	3.61 ± 3.38	3.04 ± 3.19	0.874	100	0.384a
Depression	3.33 ± 3.52	3.84 ± 3.89	−0.694	100	0.489a
Alcohol abuse (yes/no)	2/49	4/47	0.177	1	0.674b
Smoking (yes/no)	4/47	9/42	2.204	1	0.138b
Hyperlipemia (yes/no)	9/42	9/42	0.000	1	1.000b
Hypertension (yes/no)	20/31	27/24	1.933	1	0.164b
Diabetes (yes/no)	8/43	8/43	0.000	1	1.000b

Notes:Quantitative variables are expressed as mean ± standard deviation. Levels of anxiety and depression were assessed according to the Hospital Anxiety and Depression Scale (HADS).

Abbreviations:NH, normal hearing;df, degree of freedom.

^ap-value were obtained using two-samplet-tests (two-tailed).

^bp-value were obtained using Chi-square Test (two-tailed).

2.2. Auditory assessment

Prior to audiometry, an otoscopic examination was performed to remove cerumen and ensure the integrity of the tympanic membrane. Next, tympanometry was performed using a Madsen Electronics Zodiac 901 Middle Ear Analyzer to ensure the normal function of the middle ear. The pure tone threshold was assessed via a clinical audiometer (Madsen Electronics Midimate 622) coupled with TDH-39P telephonic headphones for each ear separately at frequencies of 0.125, 0.25, 0.5, 1, 2, 4, and 8 kHz. Speech detection was assessed using the speech reception threshold (SRT). The spondee words were assessed using automatic HOPE software under quiet conditions, following the SRT guidelines of the American Speech Hearing Association (Schlauch et al., 1996).

2.3. Neuropsychological assessment

Neuropsychological tests were performed in a specific order and took about 1 hour. First, the Montreal Cognitive Assessment (MoCA) (Nasreddine et al., 2005) was performed to assess general cognitive function. Next, the Auditory Verbal Learning Test (AVLT, Chinese version) (Zhao et al., 2012), Stroop Color-Word Interference Test (Stroop) (Perianez et al., 2021), Symbol Digit Modalities Test (SDMT) (Harand et al., 2018), Trail-Making Test (TMT) Part A (TMT-A), and Part B (TMT-B) (Wang et al., 2021) were performed to assess auditory learning and memory, working memory, attention, information processing speed and executive function, respectively. Each participant’s emotional state was evaluated using the Hospital Anxiety and Depression Scale (Zigmond and Snaith, 1983).

2.4. MR scanning

All participants were scanned using a 3 Tesla MR scanner (Philips, Achieva) with an eight-channel phased-array head coil. Structural data were obtained using a 3D T1-weighed sequence (repetition time (TR)/echo time (TE) = 8.1/3.7 ms, field of view (FOV) = 240 × 240 mm², slice thickness = 1 mm, 160 slices in total, isotropic voxels = 1 mm). A gradient EPI sequence was used to obtain the rs-fMRI data (TR/TE = 2000/35 ms, FOV = 240 × 240 mm², in-plane resolution = 3.75 × 3.75 mm², 35 slices with a 4 mm slice thickness, and 240 dynamics). We selected the bilateral auditory regions centered on the HG as the volumes of interest (VOIs) with a 40 × 30 × 20 mm³ size, following the recommendations of Abdul-Kareem et al. (Abdul-Kareem and Sluming, 2008) (Fig. 1). GABA-edited data were obtained using a macromolecules (MM)-suppressed MEGA-PRESS sequence (Harris et al., 2015) (TR/TE=2000/80 ms, bandwidth = 2000 Hz, “ON/OFF” editing pulses = 1.9/1.5 ppm, 320 averages), with the symmetrical-suppression method used to suppress the MM signal (Harris et al., 2015). Based on differences between the ON and OFF spectra, edited-GABA data can be obtained. The Glu data were obtained using the PRESS sequence with the same VOI (TR/TE=2000/35 ms, bandwidth = 2000 Hz, 64 averages). Chemical Shift Selective Suppression (CHESS) was used for water suppression. FASTMAP shimming of the VOI was automatically performed prior to each acquisition. Unsuppressed water data were obtained using a shorter measurement with 8 averages. The same session was acquired in a fixed order for each participant. Each scanning session included: (1) localization, (2) 3D T1-weighted images, (3) rs-fMRI, (4) MM-suppressed MEGA-PRESS sequence, (5) PRESS sequence, and (6) fluid-attenuated inversion recovery (FLAIR) sequence. It took approximately 42 min to complete the MR scans.

Fig. 1.

Representative MM-suppressed MEGA-PRESS and PRESS spectra are shown from the volumes of interests (40 × 30 × 20 mm³) centered on the Heschl’s gyrus in the left auditory region (A) and the right auditory region (B), respectively. The color bar represents the overlap of the volumes of interests across all participants. MM, macromolecules; MEGA-PRESS, Mescher–Garwood point-resolved spectroscopy sequence.

2.5. Functional data pre-processing

The functional data were pre-processed using the Data Processing & Analysis for Brain Imaging (DPABI) V5.1 toolbox (Yan et al., 2016). The processing steps were as follows: (1) exclusion of the first 10 volumes out of all 240 dynamics; (2) correction for slice-timing and realignment for head motion correction (motion > 2 mm in transition or > 2° in rotation were removed); (3) co-registration of the T1-weighted and functional data, followed by segmentation into GM, white matter (WM), and cerebrospinal fluid (CSF); (4) spatial normalization of the co-registered functional data to standard space; (5) application of nuisance covariate regression with the Friston 24 head motion parameters (Friston, 2015) and CSF signal, followed by smoothing with an isotropic Gaussian kernel (full-width by half-maximum (FWHM) = 4 mm); and (6) linear detrending and temporal filtering (0.01–0.08 Hz).

2.6. Independent component analysis

For the ICA analysis, the Group ICA of fMRI Toolbox (GIFT) software (http://icatb.sourceforge.net/) was applied as follows: (1) data reduction at the individual level was performed using a two-step principal component analysis and 40 independent components (ICs) were extracted for each participants; (2) ICs were extracted using an Infomax ICA algorithm to generate the final set of ICs; (3) the time courses and spatial maps were reconstructed according to the composition of each participant’s dataset and the results of data dimensionality reduction; and (4) all results were converted to z-scores for display and those that best matched the RSNs template were selected for further evaluation (Smith et al., 2009).

2.7. Functional connectivity analysis

Ultimately, 12 of the 40 ICs were selected for FC analysis. The ICs are spatially independent but exhibit significant dependencies in the temporal aspect. The FC maps were extracted and transformed to z-value maps by Fisher’s method before further analysis. Using age, gender, and education level as covariates, differences in intra-network FC between the two groups, as well as a comparison of all 12 × 12 inter-network FC matrices between the groups, were evaluated using a two-sample t-test. The false discovery rate (FDR) correction was applied for multiple comparisons (p < 0.05, cluster size > 5 voxels).

2.8. MRS data analysis

The MM-suppressed MEGA-PRESS data were processed using Gannet 3.1 (Edden et al., 2014) with a 3 Hz line broadening and Gaussian curve to fit GABA peaks at 3 ppm. The PRESS data were analyzed with LCModel (version 6.3–1 M) (Provencher, 1993) to estimate Glu levels. A simulated basis set which comes from the software developer was used for the LCModel analysis. The basis set included the following metabolites: alanine, aspartate, creatine, phosphocreatine, GABA, glucose, Glu, glutamine, glycerophosphocholine, phosphocholine, glutathione, myo-inositol, lactate, N-acetylaspartate, N-acetylaspartylglutamate, scyllo-inositol and taurine. Next, Gannet co-registered the VOIs to the 3D T1-weighted data and segmented the VOIs into GM, WM, and CSF. Spectra with fitting errors of GABA < 20% and Cramer-Rao lower bounds (CRLB) of Glu < 20% were chosen for further investigation. Lastly, the metabolite levels were corrected for T1 and T2 relaxation times and partial volume effects and the water-scaled levels of metabolites in institutional units (i.u.) were calculated using the following formula (Gasparovic et al., 2006; Mullins et al., 2014):

$$\begin{gathered} [GABA]=\frac{S_{GABA}}{R_{GABA}}\times \left[H_2\text{O}\right]\times VI{S}_W \\ \times \frac{f_{GM}\times R_{H2O_{-}GM}+f_{WM}\times R_{H2O\text{\_}WM}+f_{CSF}\times R_{H2O\text{\_}CSF}}{S_{H_2\text{O}}\times \left(1-f_{CSF}\right)\times k} \end{gathered}$$ (1)

$$\begin{gathered} [Glu]=\frac{S_{Glu}}{R_{Glu}}\times \left[H_{2}O\right]\times VI{S}_W \\ \times \frac{f_{GM}\times R_{H2O_{-}GM}+f_{WM}\times R_{H2O_{-}WM}+f_{CSF}\times R_{H2O\text{\_}CSF}}{S_{H_2\text{O}}\times \left(1-f_{CSF}\right)} \end{gathered}$$ (2)

where S_GABA, S_Glu, and S_H2O are the GABA, Glu, and water signal integrals, respectively, as determined using Gannet (Edden et al., 2014) and LCModel (Provencher, 1993); [H₂O] is the concentration of pure water (55,550 mmol/kg); VISw is the water visibility of MR (0.65); and f_GM, f_WM, and f_CSF are the fractions of water attributable to GM, WM, and CSF, respectively (Gasparovic et al., 2006). k is the editing efficiency of MEGA-PRESS (0.5). The relaxation attenuation factors are calculated using the equation ${\text{R}}_{\text{H}2{\text{O}}_{-}y}=\exp \left[-\text{TE}/\text{T}{2}_{{\text{W}}_{-}y}\right]\left(1-\exp \left[-\text{TR}/{\text{T1}}_{{\text{W}}_{-}y}\right]\right)$, where ${\text{T1}}_{{\text{W}}_{-}y}$ and $\text{T}{2}_{{\text{W}}_{-}y}$ are the T1 and T2 relaxation times of water in compartment y (GM, WM, or CSF). Similarly, R_GABA and R_Glu are the relaxation attenuation factors for GABA and Glu, respectively. The following relaxation times were used in this study: GM water: T1 = 1331 ms, T2 = 110 ms; WM water: T1 = 832 ms, T2 = 79.2 ms; CSF: T1 = 3817 ms, T2 = 503 ms (Lu et al., 2005; Piechnik et al., 2009; Wansapura et al., 1999); GABA: T1 = 1310 ms, T2 = 88 ms (Edden et al., 2012; Puts et al., 2013); Glu: T1 = 1270 ms, and T2 = 181 ms (Ganji et al., 2012; Mlynarik et al., 2001).

2.9. Statistical analysis

A two-sample t-test was used to compare age, education level, anxiety/depression scores, MRS data, hearing assessment, and cognitive function scores between groups. Group comparisons for gender, alcohol abuse, smoking, hyperlipemia, hypertension, and diabetes were performed using the Chi-square test. Multiple stepwise regression analyses were then performed on the presbycusis group and NH groups, with cognitive function scores or hearing assessment as the dependent variables and demographic (age, gender, and education level) and metabolite levels or FC strength as the independent variables. Note that, only clinical performance and imaging measurements that exhibited significant group differences were selected. In the regression analyses, Bonferroni correction (p < 0.05/N, N = numbers of separate analyses) was applied for multiple comparisons. Then, we applied partial correlation analysis to explore the relationships between intra/inter-network FC and metabolite levels in the presbycusis and NH groups, with age, gender, and education level as covariates. Note that, only intra/inter-network FC and metabolite levels that are significant predictors of clinical performance were selected. In the correlation analyses, Bonferroni correction (p < 0.05/N, N = numbers of separate analyses) for multiple comparisons.

Significant correlations between GABA levels, FC strength, hearing loss, and cognitive impairment were found in presbycusis patients (see Results for further details). Thus, we then examined (1) GABA level as a potential mediator of the relationship between hearing loss and FC strength; and (2) FC strength as a potential mediator of the association between GABA level and cognitive impairment. A simple mediation model using the PROCESS Macro in SPSS (ver 4.0, model 4) was used for the above mediation analysis (Hayes, 2018), with age, gender, and education level as covariates. Lastly, we examined GABA level and FC strength as potential mediators of the association between hearing loss and cognitive impairment using a two-step serial mediation model in the PROCESS Macro in SPSS (ver 4.0, model 6) (Hayes, 2018), with age, gender, and education level as covariates. Indirect effects and the 95% confidence interval (CI) were determined using a bootstrapping method with 5000 samples. Indirect effects were considered significant if the 95% CI did not contain zero (Hayes, 2018).

3. Results

3.1. Comparison of participant characteristics

The patient and NH groups did not differ significantly in age, gender, education level, anxiety/depression scores, AVLT scores, hyperlipemia, etc. (all p > 0.05, Table 1 and 2). Compared with the NH group, presbycusis patients had poorer cognitive function, as shown by MoCA, SDMT, Stroop, TMT-A, and TMT-B test scores (all p < 0.01, Table 2), as well as poorer auditory function, such as PTA and SRT (all p < 0.001, Table 2). All participants had normal middle ear function with a type A tympanometry curve. The average pure tone threshold of bilateral ears at frequencies ranging from 125 to 8000 Hz is shown in Fig. 2.

Table 2.

Auditory and cognitive function data of the NH group and presbycusis group.

Characteristics	NH	presbycusis	t- value	df	p-value
	(n = 51)	(n = 51)
MoCA	26.67 ± 2.90	23.53 ± 4.90	3.936	81.127	<0.001***
AVLT	51.71 ± 13.08	48.22 ± 9.96	1.517	100	0.133
TMT-A (s)	59.90 ± 26.41	77.25 ± 38.42	−2.658	100	0.009**
TMT-B (s)	159.63 ± 62.87	212.16± 93.84	−3.321	87.360	0.001**
SDMT	35.33 ± 10.99	24.80 ± 12.23	4.574	100	<0.001***
Stroop (s)	132.61 ± 27.21	153.25± 43.60	−2.869	83.814	0.005**
PTA (dB/HL)	10.83 ± 3.50	38.33 ± 12.23	−15.444	58.127	<0.001***
SRT (dB/HL)	11.01 ± 3.97	38.34 ± 13.56	−13.818	58.498	<0.001***

Notes:The data are presented as means ± standard deviations.

Abbreviations:NH, normal hearing; dB/HL, decibel hearing level; s, second; df, degree of freedom; MoCA, Montreal Cognitive Assessment; AVLT, Auditory Verbal Learning Test; TMT, Trail-Making Test; SDMT, Symbol Digit Modalities Test; PTA, pure tone average; SRT, speech reception threshold.

^**p < 0.01.

^***p < 0.001.

Fig. 2.

The average pure tone threshold of bilateral ears at frequencies of 125–8000 Hz in the presbycusis group and normal hearing group.

3.2. Comparison of auditory region MRS data

No significant group differences in the GM or WM fractions were found within the bilateral auditory regions (all p > 0.05, Table 3). Some GABA and Glu measurements were excluded from analysis due to fitting error and CRLB, as described in the methods (GABA: left, 2 control and 2 patients; right: 2 controls. Glu: left, 2 patients). No statistically significant difference was found in fitting error or CRLB between groups within the bilateral auditory regions (all p > 0.05, Table 3). Compared with the NH group, presbycusis patients had decreased GABA levels in the right auditory region (1.10 ± 0.22 i.u. vs 1.29 ± 0.39 i.u., p = 0.005); however, no significant differences were found in the left auditory region (1.13 ± 0.39 i.u. vs 1.16 ± 0.24 i.u., p = 0.626). Furthermore, decreased Glu levels were found in bilateral auditory regions in patients with presbycusis (left: 6.87 ± 1.03 i.u. vs. 7.32 ± 0.53 i.u., p = 0.008, right: 7.32 ± 0.86 vs 7.68 ± 0.68 i.u., p = 0.019, Table 3).

Table 3.

MRS data of the NH group and presbycusis group.

Characteristics	NH	presbycusis	t-value	df	p-value
Left auditory region
GABA levels (i.u.)	1.16±0.24	1.13±0.39	0.488	96	0.626
GABA fitting errors	9.23±2.79	9.44±3.06	−0.352	96	0.726
GMF (%)	52.24±5.06	51.99±3.71	0.279	91.81	0.781
WMF (%)	32.01±8.33	30.71±5.53	0.927	86.924	0.357
Glu levels (i.u.)	7.32±0.53	6.87±1.03	2.748	71.132	0.008**
Glu CRLB	5.75±0.74	5.90±1.16	−0.781	81.302	0.437
Right auditory region
GABA levels (i.u.)	1.29±0.39	1.10±0.22	2.914	76.187	0.005**
GABA fitting errors	9.54±2.99	9.90±3.16	−0.593	98	0.554
GMF (%)	53.29±4.87	52.82±3.80	0.544	100	0.588
WMF (%)	31.27±8.04	30.86±4.94	0.310	82.998	0.757
Glu levels (i.u.)	7.68±0.68	7.32±0.86	2.377	100	0.019*
Glu CRLB	5.55±0.58	5.78±1.21	−1.258	71.754	0.213

Notes:The data are presented as mean ± standard deviations.

Abbreviations:MRS, magnetic resonance spectroscopy; i.u., institutional units; NH, normal hearing; GABA, gamma-aminobutyric acid; Glu, glutamate; GMF, gray matter fractions; WMF, white matter fractions; CRLB, Cramer-Rao lower bound. df, degree of freedom.

^*p < 0.05.

^**p < 0.01.

3.3. Resting-state networks

No participants were excluded due to head motion. Twelve ICs were acquired using the group ICA approach. The ICs belong to the following 12 RSNs (Fig. 3): the anterior default mode network (aDMN; IC 1), posterior default mode network (pDMN; IC 20), default mode network (DMN; IC 38), visual network (VN; IC 13), auditory network (AN; IC 10), sensorimotor network (SMN; IC 28), left executive control network (lECN; IC 35), right executive control network (rECN; IC 36), basal ganglia network (BGN; IC 34), salience network I (SN I; IC 5), salience network II (SN II; IC 37), and dorsal attention network (DAN; IC 33).

Fig. 3.

Spatial distribution of 12 intrinsic resting-state networks determined by independent component analysis. The colormaps represent the z-values. aDMN, anterior default mode network; pDMN, posterior default mode network; DMN, default mode network; VN, visual network; AN, auditory network; SMN, sensorimotor network; lECN, left executive control network; rECN, right executive control network; BGN, basal ganglia network; SN I, salience network I; SN II, salience network II; DAN, dorsal attention network.

3.4. Comparison of intra-network FC

Intra-network FC differences were observed for the pDMN, DMN, lECN, and rECN. Compared with the NH group, presbycusis patients exhibited significantly decreased FC of the bilateral precuneus within the pDMN and DMN and decreased FC of the bilateral PCC within pDMN. They also showed significantly decreased FC of the left superior frontal cortex (SFC) and left inferior parietal lobule (IPL) within the lECN, and decreased FC of the right dlPFC and right SFC within the rECN (FDR corrected, p< 0.05, cluster size > 5 voxels), as shown in Table 4 and Fig. 4.

Table 4.

Brain regions with significant differences in the intra-network functional connectivity between the presbycusis group and NH group.

RSNs	Brain region	Brodmann area	MNI coordinates			T value	Cluster size
			x	y	z
pDMN	L posterior cingulate cortex	23	3	−21	24	6.2247	15
pDMN	R posterior cingulate cortex	23	3	−21	24	6.2247	15
pDMN	L precuneus	7	−15	−69	36	5.4475	97
pDMN	R precuneus	7	−15	−69	36	5.4475	52
DMN	L precuneus	7	−18	−54	15	7.2241	14
DMN	R precuneus	7	21	−51	15	7.7276	18
lECN	L superior frontal cortex	10	0	27	36	4.49	10
lECN	L inferior parietal lobule	39	−36	−57	42	4.968	20
rECN	R superior frontal cortex	10	21	24	54	5.03	11
rECN	R dorsolateral prefrontal cortex	9	39	27	36	4.3899	12

Notes:FDR corrected p < 0.05, cluster size > 5 voxels.

Abbreviations:NH, normal hearing; RSNs, resting-state networks; MNI, Montreal Neurological Institute; DMN, default mode network; pDMN, posterior default mode network; lECN, left executive control network; rECN, right executive control network; L, left; R, right.

Fig. 4.

Group difference distributions of intra-network FC between the presbycusis group and normal hearing group. Significantly decreased FC within pDMN, lECN, rECN and DMN are found in the presbycusis group (p ⟨ 0.05, cluster size ⟩ 5 voxels, FDR corrected). FC, functional connectivity; pDMN, posterior default mode network; lECN, left executive control network; rECN, right executive control network; DMN, default mode network.

3.5. Comparison of inter-network FC

Compared with the NH group, presbycusis patients showed significantly decreased inter-network FC between the AN and DMN, SN I and aDMN, SN I and SN II, SN I and VN, SN I and BGN, but significantly increased FC between the aDMN and SN II (FDR corrected, p < 0.05, Fig. 5). Additionally, presbycusis patients showed a trend of increased inter-network FC between the DMN and SMN, aDMN and pDMN (Fig. 5).

Fig. 5.

Group difference of inter-network FC between the presbycusis group and normal hearing group. The lines connect the arc pairs represent significant differences in the FC between corresponding resting-state networks pairs (p < 0.05, FDR corrected). Red lines denote a significant decrease of the inter-network FC; green lines denote significantly increased inter-network FC; green dotted lines denote increased trend of inter-network FC in the presbycusis group. FC, functional connectivity.

3.6. Relationships between metabolite levels or FC strength and cognitive function

In the presbycusis group, GABA levels in the right auditory region were negatively correlated with TMT-A scores (β = −44.492, 95% CI [−76.637, −12.348], p = 0.008, R² = 0.577). The intra-network FC of the left SFC was positively correlated with SDMT scores (β = 4.213, 95% CI [1.918, 6.508], p = 0.001, R² = 0.571), and FC between the AN and DMN was negatively correlated with TMT-A scores (β = −41.399, 95% CI [−68.491, −14.307], p = 0.003, R² = 0.614). As there are five neuropsychological assessments in total leading to five comparisons, we applied Bonferroni correction (p < 0.05/5; regarding five separate regression analyses) for multiple comparisons. No relationships were observed between metabolite levels or FC strength and cognitive function in the NH group.

3.7. Relationships between metabolite levels or FC strength and hearing assessment

In the presbycusis group, GABA levels in the right auditory region were negatively correlated with PTA scores (β = −24.428, 95% CI [−36.233, −12.623], p < 0.001, R² = 0.353), and SRT (β = −23.059, 95% CI [−37.985, −8.132], p = 0.003, R² = 0.17). The inter-network FC between the aDMN and SN II was positively correlated with SRT (β = 21.357, 95% CI [8.08, 34.634], p = 0.002, R² = 0.176), and FC between the AN and DMN were negatively correlated with PTA scores (β = −14.684, 95% CI [−25.489, −3.88], p = 0.009, R² = 0.279). As there are two hearing assessments in total leading to two comparisons, we applied Bonferroni correction (p < 0.05/2; regarding two separate regression analyses) for multiple comparisons. No relationships were observed between metabolite levels or FC strength and hearing assessment in the NH group.

3.8. Relationship between auditory metabolite levels and FC strength

GABA levels in the right auditory region, intra-network FC of the left SFC, inter-network FC between the aDMN and SN II, and inter-network FC between the DMN and AN were selected for further analysis. In the presbycusis group, GABA levels in the right auditory region were positively correlated with FC between the AN and DMN (r = 0.436, p = 0.002, df = 46). As there are three intra/inter-network FC and one metabolite levels in total leading to three comparisons, we applied Bonferroni correction (p < 0.05/3; regarding three separate analyses) for multiple comparisons. No relationships were found between auditory metabolite levels and FC strength in the NH group.

3.9. Interactions among GABA level, FC strength, and clinical scores

The simple mediation model (A) revealed a significant total effect (c = −0.0082, p = 0.0194), suggesting that PTA predicts FC between AN and DMN in presbycusis patients (Fig. 6A). In the path analysis, GABA levels in the right auditory region were negatively related to PTA (a = −0.0106, p = 0.0002) but positively related to FC between the AN and DMN (b = 0.4185, p = 0.0246). After controlling for GABA levels, the direct effect of PTA on FC was no longer significant (c’ = −0.0038, p = 0.3201). The bootstrap procedure revealed a significant indirect effect of GABA levels in the right auditory region (a × b = −0.0044, 95% CI [−0.0097, −0.0006]).

Fig. 6.

Mediation models of the associations among hearing loss, GABA levels, FC, and cognitive impairment in presbycusis group. Mediation effects of the right auditory GABA levels on the association between the PTA and inter-network FC of AN-DMN (A). Mediation effects of the AN-DMN FC on the association between the right auditory GABA levels and TMT-A scores (B). GABA, gamma-aminobutyric acid; PTA, pure tone average; DMN, default mode network; AN, auditory network; TMT-A, Trail-Making Test-A; FC, functional connectivity.

Next, simple mediation model (B) revealed a significant total effect (c = −41.9633, p = 0.0113), suggesting that GABA levels in the right auditory region predict TMT-A scores in presbycusis patients (Fig. 6B). In the path analysis, FC between AN and DMN was positively related to GABA levels (a = 0.5103, p = 0.0019) but negatively related to TMT-A scores (b = −32.2995, p = 0.0310). After controlling for FC, the direct effect of GABA levels on TMT-A scores was no longer significant (c’ = −25.4821, p = 0.1400). The bootstrap procedure revealed a signif- icant indirect effect of FC between the AN and DMN (a × b = −16.4824, 95% CI [−37.4419, −3.1110]).

Finally, the serial mediation model revealed a significant total effect (c = 0.8404, p = 0.0151), suggesting that PTA predicts TMT-A scores in presbycusis patients (Fig. 7). In the path analysis, PTA was negatively related to GABA levels in the right auditory region (a1 = −0.0106, p = 0.0002) and negatively related to FC between the AN and DMN (a2 = −0.0038, p = 0.3201). The right auditory GABA levels were positively related to AN-DMN FC (a3 = 0.4185, p = 0.0246) and negatively related to TMT-A scores (b1 = −16.4544, p = 0.3832). The AN-DMN FC was negatively related to TMT-A scores (b2 = −29.8366, p = 0.0473). After controlling for GABA levels and AN-DMN FC, the direct effect of PTA on TMT-A scores was no longer statistically significant (c’ = −0.4223, p = 0.2619). The total indirect effect was significant (a1×b1+ a2×b2 + a1×a3×b2 = 0.4180, 95% CI [0.0449, 0.8951]), with a significant serial indirect effect observed from PTA via GABA levels and AN-DMN FC to TMT-A scores (a1×a3×b2 = 0.1319, 95% CI [0.0079, 0.4006]).

Fig. 7.

A serial mediation model including the right auditory GABA levels and AN-DMN FC as mediators of the associations between hearing loss and cognitive impairment in presbycusis group. GABA, gamma-aminobutyric acid; PTA, pure tone average; FC, functional connectivity; DMN, default mode network; AN, auditory network; TMT-A, Trail-Making Test-A.

4. Discussion

To our knowledge, this is the first study to combine edited MRS and rs-fMRI data to investigate the neurochemical profile of the auditory region and the functional architecture of RSNs in presbycusis. The main findings include: (1) decreased GABA levels in the right auditory region and decreased Glu levels in bilateral auditory regions; (2) decreased intra-network FC within the DMN and executive control network, decreased inter-network FC between the auditory network and DMN, which was positively correlated with information processing speed, and increased FC between the aDMN and SN II, which was positively correlated with speech perception ability; and (3) a mediating effect of right auditory GABA levels and auditory network-DMN FC on the association between hearing loss and impaired information processing speed in presbycusis. These findings shed light on the neurochemical underpinnings of cognitive-ear link reorganization at the large-scale brain network level in presbycusis.

4.1. Altered auditory GABA and Glu levels in presbycusis

As the main inhibitory neurotransmitter, GABA plays a critical role in acoustic information processing (Caspary et al., 1995). In the present study, GABA levels in the right auditory region were significantly decreased in the presbycusis group; however, no group differences were found in the left auditory region. We previously demonstrated lower GABA+ levels in the bilateral auditory regions in presbycusis patients using MEGA-PRESS, which is partially consistent with the current results (Gao et al., 2015). One possible reason for the discrepancy is that the detected GABA signal contains an important contribution of macromolecules using MEGA-PRESS (Harris et al., 2015). To the best of our knowledge, this is the first and largest study using MEGA-PRESS with macromolecular suppression to study auditory “pure GABA” levels in presbycusis patients. Our findings reflect dysfunctional GABAergic neurotransmission in the auditory cortex, which is consistent with the findings of several animal models of presbycusis, including decreased GABA release, reduced GABA neuron numbers, decreased GABA levels, and reduced glutamic acid decarboxylase levels in the auditory cortex (Burianova et al., 2009; Caspary et al., 2013; Syka, 2010).

A previous MRS study found that presbycusis patients had reduced Glu levels in bilateral auditory regions compared with young controls (Profant et al., 2013). However, the study lacked age-matched NH controls and the altered auditory Glu levels could be mainly due to age and not presbycusis. In the present study, we found decreased bilateral auditory Glu levels in presbycusis patients compared with age-matched controls, indicating reduced glutamatergic excitation in the central auditory system. Interestingly, a genome-wide association study reported that variation in the glutamate metabotropic receptor 7 (GRM7) gene may lead to the development of presbycusis by changing the mechanism of glutamate excitotoxicity susceptibility (Friedman et al., 2009). Our findings may reflect reorganization of the auditory cortex at the neurochemical level after partial hearing deprivation.

4.2. Decreased intra-network FC in presbycusis

We observed significantly decreased FC involving the bilateral precuneus and PCC within the DMN in presbycusis patients. The precuneus is a heterogenous cortical region with long- and short-range fibers to higher association cortical and subcortical structures (Tanglay et al., 2022); its connection to the auditory cortex through middle longitudinal fasciculus indicates that it may be involved in sound perception (Kalyvas et al., 2020). In addition, part of the precuneus serving as a distinct hub of the DMN can promote the integration of resting state process and regulate attention states and cognition (Tanglay et al., 2022; Utevsky et al., 2014). As another node of the DMN, the PCC participates in memory and cognitive processing and plays an important role in coordinating auditory stimulation under non-optimal conditions (Leech and Sharp, 2014). In line with our findings, decreased cortical volume and aberrant ALFF have been reported in the precuneus and PCC of presbycusis patients (Ren et al., 2018, 2021).

Furthermore, we found decreased intra-network FC of the ECN in presbycusis, specifically involving the dlPFC (multisensory integration and top-down regulation processes), SFC (speech perception and auditory processing), and IPL (speech processing) (Du et al., 2014; Morrone, 2010; Romanski et al., 1999; Simonyan and Fuertinger, 2015). These findings may be reflected as speech perception difficulties in presbycusis. Notably, we found that FC of the left SFC in presbycusis patients was positively correlated with SDMT scores, which reflects attention. Presbycusis-related hearing loss may thus trigger cognitive compensation via top-down modulatory control, which further affects coordination within the network.

4.3. Altered inter-network FC in presbycusis

The SN, anterior cingulate cortex (ACC) and anterior insula (AI) as core hubs, performs a regulatory function in DMN and central executive networks by integrating the internal and external environmental stimulation, including stimuli from the auditory pathway (Menon and Uddin, 2010). In our study, inter-network FC between SN I (ACC as the core hub) and visual network, basal ganglia network and aDMN were deceased. The ACC has a wide range of fiber connections to the limbic system and frontal and parietal cortex, and plays important roles in learning and memory (Luan et al., 2018). The ACC also receives highly processed auditory information from the rostral superior temporal gyrus and prefrontal cortex, which is mainly responsible for balancing the bottom-up and top-down regulation of auditory information (Crottaz-Herbette and Menon, 2006; Diekhof et al., 2009). The imbalance between SN I and other functional networks may reflect a pattern of large-scale reorganization in presbycusis in which the SN I plays an important coordinating role.

Although no difference in FC was found within the auditory network, decreased FC between the auditory network and DMN was observed in presbycusis patients. Furthermore, this decrease was negatively correlated with TMT-A scores, which reflect information processing speed. We also found increased FC between the aDMN and SN II (AI as the core hub), which was positively correlated with speech reception threshold of presbycusis patients. Presbycusis is often accompanied by degraded speech perception, especially in noisy environments; in such cases, more cognitive resources may be recruited to support speech perception (Cardin, 2016; Pauquet et al., 2021; Wayne and Johnsrude, 2015). Thus, increased FC between the aDMN and SN II may reflect a kind of compensation to facilitate speech reception. Possibly due to the compensation effect, increased FC between cognitive-related networks may compromise the interaction between auditory and cognitive-related networks such as DMN.

In summary, we found decreased GABA and Glu levels in the auditory region and aberrant intra- and inter-network FC after partial auditory deprivation, suggesting that presbycusis patients experience neurochemical reorganization in the auditory cortex and large-scale reorganization at the functional network level. Importantly, such reorganization was related to reduced speech perception or cognitive impairment. In our opinion, these results support the sensory-deprivation hypothesis, which suggest long-term sensory deprivation due to presbycusis may lead to neural deafferentation in the auditory region and then cortical reorganization involving cognitive-related brain regions that support speech perception, which may ultimately lead to cognitive decline (Griffiths et al., 2020; Jafari et al., 2021; Wayne and Johnsrude, 2015).

4.4. Reorganization of the cognitive-ear link in presbycusis

The cognitive-ear link provides a theoretical basis for exploring the neural mechanisms underlying cognitive impairment in presbycusis; however, its reorganization patterns and relationship to cognitive impairment were largely unknown prior to this study. The serial mediation model analysis revealed that GABA levels in the right auditory region and FC between the auditory network and DMN mediated the association between hearing loss and cognitive impairment. Previous our study has found that GABA levels in the right auditory region are closely related to hearing loss in presbycusis (Gao et al., 2015). Furthermore, a recent animal study demonstrated that progressive hearing loss can trig- ger extensive alteration of the GABAergic system in the auditory cortex, suggesting a causal relationship (Beckmann et al., 2020). These findings support the results of our mediation model, i.e., that hearing loss leads to decreased GABA levels in the right auditory region. Moreover, decreased GABA levels may influence synaptic plasticity in the auditory cortex (Caspary et al., 2008). Thus, the model findings suggest that hearing loss may drive reorganization of the cognition-ear link by affecting right auditory GABA levels in presbycusis patients.

Additionally, the laterality of GABA levels in the serial mediation model may be related to the functional and anatomical asymmetry between the left and right auditory cortices (Tervaniemi and Hugdahl, 2003). A magnetic mismatch negativity study found that under noise conditions, the involvement of the left auditory cortex in speech recognition was considerably decreased, while the involvement of the right auditory cortex significantly increased, which indicates a relative lateralization toward the right hemisphere in supporting speech-in-noise understanding (Shtyrov et al., 1998). Importantly, speech-in-noise understanding deficit is the main characteristic of presbycusis patients (Gates and Mills, 2005). Moreover, a recent MRS study found speech-in-noise understanding in older adults was obviously correlated with the GABA levels in the right auditory cortex, but not in the left auditory cortex (Dobri and Ross, 2021). Thus, the GABAergic system of the right auditory cortex in presbycusis may be more involved in the cognitive-ear link than that of the left auditory cortex.

Furthermore, the simple mediation analysis (A) demonstrated that GABA levels in the right auditory region mediated the association between hearing loss and FC between the auditory network and DMN. Evidence indicated that GABAergic cells play a critical role in the synchronous activity between distant brain regions by means of local GABAergic interneurons modulating long-range excitatory connections, or long-range GABAergic projection (Caputi et al., 2013). A prior MRI study in healthy participants has shown that when GABAergic activity is enhanced by the non-benzodiazepine hypnotic zolpidem, resting state functional connectivity is increased in widespread brain networks, including auditory network (Licata et al., 2013). Consistent with our results, a previous MRS study found that ACC-caudate FC was positively related to GABA levels in the ACC in borderline personality disorder patients (Wang et al., 2017). According to the mediation models, hearing loss may trigger alteration in the GABAergic system in the auditory cortex, thereby leading to reduced FC between the auditory network and DMN in presbycusis. These findings indicate a potential neural pathway for reorganization of the cognitive-ear link.

Synchronized gamma oscillations, which appear to be a fundamental mode supporting information processing speed (Fries et al., 2007), are regulated and organized by GABAergic inhibition (Cardin et al., 2009; Whittington and Traub, 2003). Previous MRS studies have found that cortical GABA levels were obviously correlated with gamma oscillations in human brain (Chen et al., 2014; Muthukumaraswamy et al., 2009). Notably, our simple mediation analysis (B) demonstrated that FC between the auditory network and DMN mediated the relationship between GABA levels in the right auditory region and information processing speed. Based on cognitive-ear hypothesis (Lozupone et al., 2020), FC between the auditory network and DMN may represent the interaction between auditory and cognitive processing. Our findings suggest that this interaction plays an important role in the relationship between GABA levels and cognitive function. Moreover, the serial mediation model analysis revealed that hearing loss affected this interaction via a bottom-up route, such that decreased interaction eventually led to information processing speed decline in presbycusis patients. The auditory region has indirect connections with DMN via core hubs of salience network including the insula and ACC (Menon and Uddin, 2010). Moreover, there are direct and indirect connections between the auditory region and medial temporal lobe (MTL) (Munoz-Lopez et al., 2010), which is an important subsystem in the DMN (Andrews-Hanna, 2012; Chen et al., 2020). A number of task-based fMRI studies have found MTL subsystem plays a critical role in auditory cognition (Kumar et al., 2014, 2016), and increased activity of MTL subsystem was found in response to degraded speech stimuli in adults (Bishop and Miller, 2009). Previous studies have shown that the DMN is responsible for the integration of primary perception and advanced cognitive processing, which appears to be affected by sensory deprivation and is involved in information processing speed decline (Miraglia et al., 2020; Savini et al., 2019; Xing et al., 2020). Taken together, these findings suggest that reorganization of the cognitive-ear link may lead to cognitive impairment in presbycusis and that reduced connectivity between the auditory network and DMN may play a pivotal role in this process.

4.5. Limitations

Several limitations of this study must be noted. First, as a cross-sectional study, our results cannot determine causal relationships between the correlated measures. Longitudinal studies are needed to evaluate such relationships. Second, due to the lack of a control VOI in the MRS analysis, it is difficult to determine how unique the observed correlations are relative to neurotransmitters levels in other brain regions. Third, most GABA is located in two pools within neurons, namely the cytoplasm and presynaptic vesicles (Kwon et al., 2014). However, MRS can only detect total GABA in a specified area, it cannot distinguish these individual GABA pools (Stagg et al., 2011). Thus, our research on GABAergic neurotransmission as a target for presbycusis therapy may be limited.

5. Conclusion

This study demonstrated decreased auditory GABA and Glu levels and wide-ranging reorganization of intra- and inter-network FC involving auditory and cognitive networks in presbycusis. Notably, the pattern of reorganization was associated with reduced speech perception or cognitive impairment, providing evidence for the sensory deprivation hypothesis. Importantly, decreased auditory GABA levels and reduced AN-DMN connectivity mediated the association between hearing loss and impaired information processing speed in presbycusis patients, revealing a potential neural pathway of cognitive-ear link reorganization. In this pathway, dysconnectivity of the AN-DMN, suggesting dysfunctional interactions between auditory and cognitive processing, played a pivotal role in cognitive-ear link reorganization leading to cognitive impairment. Moreover, hearing loss may drive reorganization of the cognitive-ear link via decreased auditory GABA levels, suggesting a novel treatment target for cognitive impairment in presbycusis patients.

Data Availability

Data will be made available on request.

Data and code availability statement

The data and code that support the findings of this study are available from the corresponding author upon reasonable request.