Dynamic brain functional connectivity in age-related hearing loss during auditory selective spatial attention

Abstract

Age-related hearing loss (ARHL) is a common health problem that impairs auditory perception. However, the dynamic patterns of brain functional connectivity in ARHL during auditory spatial selective attention have not been thoroughly investigated. In this study, 32 older adults were recruited to investigate the dynamic brain functional connectivity in ARHL. First, an experimental paradigm for auditory spatial selective attention was designed, and neural electrical signals were recorded using electroencephalography. Then, a multilayer time-varying brain network was constructed based on multiple time windows, equally dividing each epoch signal to capture dynamic functional connectivity across time scales. Finally, the core layer brain network was identified by the multilayer time-varying brain network properties to investigate the changing patterns of network topology. Behavioral analysis revealed a significant negative correlation between the severity of hearing loss and auditory spatial selective attention performance. Multilayer time-varying brain network analysis revealed that worsening hearing loss was found to lead to increased inter-layer connectivity strength, decreased multilayer modularity and a higher participation coefficient. This suggests that the brain compensates by weakening the independence of local functional modules and enhancing cross-interaction. Core layer analysis further highlighted the critical role of the right parietal lobe in auditory spatial selective attention. It also suggested that connectivity between the right prefrontal and frontal lobes may play a compensatory role in ARHL. In conclusion, these findings provide important neuroscientific insights into the dynamic brain functional connectivity of ARHL, and potential biomarkers and time windows for the development of precision auditory rehabilitation strategies.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s11357-025-02069-8.

Introduction

Age-related hearing loss (ARHL) has become one of the most pressing health concerns among the elderly, with the aging of the global population. ARHL is typically characterized by progressive, symmetric, high-frequency hearing loss in both ears and a gradual decline in auditory perception and speech recognition [1]. In particular, ARHL often showed significant deficits in auditory spatial selective attention in complex auditory environments [2]. This refers to the ability to focus on the target source while suppressing distractor sources in complex auditory environments, commonly known as the “cocktail party effect” [3, 4]. This ability is critical for everyday functioning, particularly for efficiently identifying speakers in complex auditory environments, essential for effective social communication.

Despite substantial evidence that ARHL has deficits in auditory perception and speech recognition, research on the neural mechanisms underlying auditory spatial selective attention in ARHL remains limited [1, 5]. Existing psychoacoustic studies have shown that individuals with hearing loss have significant deficits in sound localization and spatial sensitivity in complex auditory environments [6, 7]. These deficits have been attributed primarily to high-frequency hearing loss and impaired processing of interaural time and intensity difference cues [8]. Further research has shown that the ability to process interaural time and intensity difference cues decreases as hearing loss worsens, resulting in reduced sound localization accuracy [9]. In addition, hearing loss may impair interference suppression and the efficient allocation of attentional resources [10]. Electrophysiological measures, particularly those using event-related potentials, have shown that older adults exhibit delayed and reduced neural responses in auditory spatial selective attention tasks compared to younger adults [11]. These deficits were most pronounced in the P2 and P3 components, as well as the mismatch negativity (MMN) component, indicating a reduced efficiency of the brain in processing complex auditory information [12]. Specifically, reductions in the amplitude of the P2 and P3 components indicated difficulties in target detection and allocation of attentional resources in older adults [13, 14]. In addition, reductions in MMN amplitude and prolonged latency suggested that the poor spatial discrimination abilities observed in individuals with hearing loss may be related to compensatory cortical changes [15]. While these studies have primarily focused on the temporal dimension of auditory attentional processing in individuals with hearing loss, they have largely neglected the spatial dimension, i.e., the relevant brain regions involved in auditory spatial selective attention.

Numerous studies using techniques such as functional magnetic resonance imaging have attempted to identify the relevant brain regions involved in spatial auditory processing [16, 17]. Research has shown that individuals with hearing loss exhibit decreased activation in the temporal cortex and increased activation in the frontal cortex compared to individuals with normal hearing [18]. These regional changes may indicate a compensatory effect of auditory stimuli on the redistribution of cortical processing. In addition, some studies have shown that increasing severity of hearing loss is associated with decreased activity in the parietal cortex, which manifests as a decline in performance on sound localization and selective attention tasks [19]. Other studies have found that older adults with hearing loss experience a faster rate of brain volume decline, particularly in the right temporal cortex, suggesting that cortical structural changes are related to the duration and degree of hearing loss [20]. While these studies have identified significant changes in relevant brain regions in ARHL, they have not elucidated the functional connectivity effects between these regions. Furthermore, several studies have shown that brain functional connectivity is extensively altered in ARHL, suggesting that hearing loss may disrupt and enhance functional connectivity between relevant brain regions [21, 22]. Specifically, individuals with hearing loss exhibited more communication pathways between relevant brain regions but lower information exchange efficiency than those with normal hearing. Meanwhile, the directional signal within the dorsal–ventral pathway was reversed [23]. Furthermore, research has found that connectivity strength in the dorsal attention network increases during tasks requiring attention and stimulus–response [24]. Moreover, worsening hearing loss was significantly correlated with reduced resting-state functional connectivity, with connectivity strength of the dorsal attention network gradually decreasing as the severity of hearing loss increased [25]. These changes in brain functional connectivity may result from the cross-modal reorganization of the auditory cortex in ARHL [26]. However, most studies rely on fixed time windows or resting-state analysis, which do not account for dynamic brain functional connectivity across different time scales.

Auditory spatial selective attention was a dynamic process involving complex interactions and cooperation between multiple brain regions across different time scales [27]. In other words, cortical activation and inhibition were not static but fluctuated over time at different time scales. Therefore, research methods based on fixed time windows may fail to capture the complete dynamic characteristics of brain activity during such tasks, especially when attempting to reveal the dynamic brain functional connectivity over time. To address this limitation, many researchers have proposed the multilayer time-varying brain network analysis method to explore functional connectivity across time scales during resting-state and specific-task activity performance [28, 29]. This method models and analyzes the activation patterns and network connectivity of brain regions across different time scales. Thus, it can provide a more comprehensive understanding of the effects at various brain levels, especially concerning the dynamic interactions and information flow between multiple brain regions [30].

Therefore, this study employed the multilayer time-varying brain network analysis to investigate the dynamic patterns of brain functional connectivity in ARHL during auditory selective spatial attention. Specifically, auditory selective spatial attention tasks were designed in a complex auditory environment with multiple distractor sources and background noise. Neural electrical signals were recorded using the high-temporal-resolution electroencephalogram (EEG) and preprocessed into multiple epochs. Furthermore, a multilayer time-varying brain network was constructed by dividing each epoch into six equally spaced time windows to capture dynamic patterns of brain activity across time scales. The complexity and correlation of the network were further analyzed quantitatively, including multilayer modularity, participation coefficient, layer-layer correlation, and conditional probability. Finally, the core layer brain network was identified by the multilayer time-varying brain network properties to investigate the topology of brain functional connectivity. The overall research scheme is shown in Fig. 1.

Fig. 1.

Overall research scheme. a Experimental paradigm design. b Data acquisition. c Data preprocessing. d Multilayer time-varying network construction. e Multilayer time-varying network analysis. f Core layer brain network analysis

Material and method

Participants

A total of 32 participants ranging in age from 60 to 77 years (mean ± SD: 65.9 ± 3.6 years) were recruited for this study. The basic information of the participants can be found in the Supplementary Materials, and the pure-tone audiometry results at different frequencies are shown in Fig. 2. According to pure-tone audiometry at 0.5, 1, 2 and 4 kHz, the participants were divided into three groups: Normal hearing (10 participants, 6 females, with hearing thresholds of ≤25 dB), Mild ARHL (12 participants, 6 females, with hearing thresholds between 26 and 40 dB) and Moderate ARHL (10 participants, 5 females, with hearing thresholds ≥ 41 dB). All ARHL participants had symmetric hearing loss with an interaural threshold difference not exceeding 15 dB. Additionally, there were no significant differences in age between the three groups (F(2, 29) = 0.188, p = 0.188).

Fig. 2.

Pure tone audiometry results. a Left ear. b Right ear. Green, pink, and yellow represent normal, mild, and moderate, respectively. The shadow represents the standard error

All participants met the eligibility criteria through detailed questionnaires to reduce EEG variations caused by non-hearing related factors: (1) Right-handed, with no prior or current use of hearing aids or cochlear implants; (2) No cognitive impairment or decline, and no history of neurological disorders or psychiatric conditions; and (3) No vascular risk factors or diseases, and no use of medications affecting cerebral blood flow or neural activity. The study protocol was approved by the Ethics Review Committee of Tianjin University (Approval No. TJUE-2024–384), and all participants provided written informed consent prior to the experiment.

Experimental paradigm

The experimental paradigm for auditory spatial selective attention was designed as shown in Fig. 3a. The paradigm included four directions, i.e., 90° and 30° on either side of the participant’s midplane. The sound field environment was created using a semicircular loudspeaker array with a radius of 1 m. The participant’s name was used as a target source to maintain spatial attention, while three different names were used as distractor sources to create a complex environment. All target and distractor name stimuli were recorded by the same standard Mandarin speaker using a Sennheiser e845S microphone at a sampling rate of 48,000 Hz. In addition, white noise was selected as the background noise to be added to the auditory stimulus to achieve a signal-to-noise ratio level of 0 dB. Finally, these stimuli were normalized to the same root mean square amplitude and duration using Adobe Audition 2022 software.

Fig. 3.

Experimental paradigm for auditory spatial selective attention. a Diagram of the experimental paradigm. b Experimental procedure

The experiment consisted of 240 sets of auditory stimuli, divided into two blocks with a 10 min rest intervals, as shown in Fig. 3b. The sequence of the auditory stimuli was pseudorandomized and implemented using MATLAB 2021a and Psychtoolbox-3. During the experiment, four sound sources were played simultaneously from four separate loudspeakers, with strictly aligned start and end times. To prevent eye drift, a screen displaying a fixed central “ + ” symbol was placed in front of the participants. Furthermore, task prompts were displayed to guide the participants in determining the direction of the target sound, which they had to identify by pressing the corresponding button within 2000 ms of the stimulus ending. Following a 500 ms silence, another set of auditory stimuli was presented, looped until the end. The total testing time for each participant was approximately 25 min. Additionally, participants completed a short, 2–3 min training session to familiarize themselves with the task, and the experiment was conducted in a semi-anechoic chamber.

EEG data acquisition and preprocessing

EEG data were recorded using the Neuroscan system (Compumedics, Australia), which employs a 64-channel configuration based on the international 10–20 system. The recorded EEG signals were amplified with the Neuroscan SynAmps2. The M1 and M2 electrodes were not used because the Ref electrode served as the reference. Moreover, the CB1 and CB2 electrodes were excluded because they were not relevant for cerebellar localization. During the experiment, the impedance of the remaining electrodes was kept below 10 kΩ, with a sampling rate of 1000 Hz and an online bandpass filter set to 0.1–150 Hz, along with a 50 Hz notch filter.

Raw EEG data were preprocessed using the open-source EEGLAB toolbox for MATLAB 2021b software. First, a 0.1–30 Hz bandpass filter was applied to reduce low-frequency drift and high-frequency noise. The data were then downsampled to 256 Hz for further processing. Independent component analysis was applied to remove ocular artifacts caused by eye movements or blinks. Furthermore, the data were segmented into 700 ms epochs, with a 100 ms pre-stimulus and 600 ms post-stimulus period, and baseline correction was performed using the first 100 ms. In addition, epochs with a maximum-minimum difference greater than 100 μV were excluded. Finally, the data were re-referenced to the average of all electrodes to improve the signal-to-noise ratio of the EEG signals.

Multilayer time-varying brain network construction

The phase-locking value (PLV) has been widely used to analyze the phase synchronization relationship between nonlinear and nonstationary EEG signals [31, 32]. Therefore, this study employed the PLV to quantify the phase synchronization relationship between EEG signals from different channels, serving as a connectivity weight for the multilayer time-varying brain network [33].

$$\begin{array}{c}\left(P,L,V\right|=\left|e^{i\mathrm{\Delta }\varphi rel\left(t\right)}\right|=\left|\frac{1}{N},{\sum }_{n=1}^N,e^{i\mathrm{\Delta }\varphi rel\left(t_n\right)}\right|\\ \left|=\right)\sqrt{cos\mathrm{\Delta }\varphi re{l}^2+sin\mathrm{\Delta }\varphi rel{\left(t\right)}^2}\end{array}$$ 1

where N is the number of trials, $\varnothing$(t, k) and ψ(t, k) are the instantaneous phases of channels $\varnothing$ and ψ at time t during the k-th trial. Moreover, some studies have shown that auditory selective spatial attention is closely related to the alpha band [19, 34]. Therefore, the PLV in the alpha band (8–12 Hz) was computed between different channels.

In addition, a sparse set of 21 canonical electrodes was selected as the nodes of the brain network due to the sensitivity of the PLV index to common sources [27]. Specifically, the node selection strategy optimizes the electrode spacing (≥3 cm) to effectively reduce volume conduction interference between adjacent electrodes. The final selection of electrodes included FP1, FPz, FP2, F7, F3, Fz, F4, F8, T7, C3, Cz, C4, T8, P7, P3, Pz, P4, P8, O1, Oz, and O2.

Each 600 ms epoch was equally divided into six-time windows (W1, W2, W3, W4, W5, and W6), with each time window lasting 100 ms, as described in previous research [32]. For each time window, the cross- and same-time connectivity matrices, i.e., inter- and intra-layer connectivity matrices, were constructed as shown in Fig. 4. Specifically, the PLV of EEG pairs within epochs was calculated separately to construct the intra-layer connectivity matrix, resulting in six matrices (21 × 21 × 6). Moreover, the PLV of EEG pairs between epochs was calculated to construct the inter-layer connectivity matrix, resulting in thirty matrices (21 × 21 × 30). Finally, the super-connectivity matrix (126 × 126), representing the multilayer time-varying brain network, was constructed by placing the six intra-layer connectivity matrices along the diagonal and the thirty inter-layer connectivity matrices along the off-diagonal.

Fig. 4.

The diagram of the multilayer time-varying brain network construction

Multilayer time-varying and core layer brain networks analysis

The multilayer time-varying brain network was represented by the super-connectivity matrix, with the M-layer denoted as follows [35]:

$$A=diag\left(A_t\right)+(H\otimes B)$$ 2

$$\mathit{diag}\left(A_t\right)=\left[\begin{array}{cccc}{A}_1& 0& \cdots & 0\\ 0& A_2& \cdots & 0\\ ⋮& ⋮& \ddots & ⋮\\ 0& 0& \cdots & A_M\end{array}\right]$$ 3

$$H=\left[\begin{array}{cccc}0& H_{12}& \cdots & H_{1M}\\ H_{21}& 0& \cdots & H_{2M}\\ ⋮& ⋮& \ddots & ⋮\\ H_{M1}& H_{M2}& \cdots & 0\end{array}\right]$$ 4

where $\otimes$ represents the Kronecker product of the matrix. The term $\text{diag}\left(A_t\right)$ represents a block diagonal matrix with the matrices $A_t$ on the diagonal and matrices consisting of zeros on the non-diagonals. The term $(H\otimes B)$ represents the inter-layer connectivity between the layers. The matrix $B$ = $E$ (where $E$ is a matrix with all elements equal to 1) indicates that any node in one layer can be connected to any node in another layer [36]. The term $H_{\mathit{MN}}$ represents the interaction matrix between the layers.

Therefore, the overall intra-layer $⟨A⟩$ and inter-layer $⟨H⟩$ connectivity strengths in the multilayer time-varying brain network were defined by the average of all elements in the diagonal and the off-diagonal blocks of the super-connectivity matrix, respectively [37]:

$$⟨A⟩=\frac{1}{M}\sum _{i=1}^M{A}_i$$ 5

$$⟨H⟩=\frac{1}{M(M-1)}\sum _{m=1}^M\sum _{n=1,n\ne m}^M⟨H_{\mathit{mn}}B⟩$$ 6

The multilayer modularity (MD) was used to quantify the segregation degree between different layers, which characterizes the degree of change in network clustering clusters over a period of time, as follows [38]:

$$\begin{array}{c}MD(\gamma ,\omega )=\frac{1}{2\mu }{\sum }_{\mathit{ijlm}}\left[\left(W_l,-,{\gamma }_l,\frac{k_i^l{k}_j^l}{2{\sum }_l{k}_i^l}\right),\delta ,\left(W_{\mathit{il}},,,W_{\mathit{jl}}\right),+\right)\\ \left(\delta ,(i,j),{\omega }_{\mathit{jml}}\right]\delta \left(W_{\mathit{il}},,,W_{\mathit{jm}}\right)\end{array}$$ 7

where $\mu$ is the overall connection weight in the multilayer time-varying brain network, and ${\gamma }_l$ is the topological resolution parameter of the layer $l$. ${\omega }_{\mathit{jml}}$ is the temporal connection parameter of node $j$ between the $l$-th and $m$-th layers. $k_i^l$ is the connection strength of the node $i$ at the $l$-th layer. When node $i$ and $j$ are in the same module, the values of $\delta \left(W_{\mathit{il}},,,W_{\mathit{jl}}\right)$ and $\delta \left(W_{\mathit{il}},,,W_{\mathit{jm}}\right)$ are 1, otherwise they are 0.

The participant coefficient (PC) was used to quantify the integration degree of the multilayer time-varying brain network. It indicates whether the connections of a given node are distributed uniformly across the different layers. The PC of the node $i$ is defined as follows [39]:

$${\mathit{PC}}_i=\frac{M}{M-1}\left[1,-,\sum _{l=1}^M,{\left(\frac{k_i^l}{o_i}\right)}^2\right]$$ 8

where $o_i$ is the overall connection strength of the node $i$ across $M$ layers. If all connections of the node $i$ are in the same layer, ${\mathit{PC}}_i$=0. Conversely, if the node $i$ has the same number of connection in each of the $M$ layers, ${\mathit{PC}}_i$=1. The $\mathit{PC}$ of the overall multilayer time-varying brain network is defined as the average of ${\mathit{PC}}_i$ for all nodes.

The layer-layer correlation (LLC) and conditional probability (CP) was used to measure the correlation between different layers, which is defined as follows [28] [38]:

$$LLC\left(W_{m,ij},,,W_{n,ij}\right)=\frac{{\sum }_{\mathit{ij}}(W_{m,ij}-\overline{W_{m,ij}})(W_{n,ij}-\overline{W_{n,ij}})}{\sqrt{\sum _{\mathit{ij}}(W_{m,ij}-\overline{W_{m,ij}}{)}^2}\times \sqrt{\sum _{\mathit{ij}}(W_{n,ij}-\overline{W_{n,ij}}{)}^2}}$$ 9

$$CP\left(W_{m,ij},\mid ,W_{n,ij}\right)=\frac{{\sum }_{\mathit{ij}}{W}_{m,ij}{W}_{n,ij}}{{\sum }_{\mathit{ij}}{W}_{n,ij}}$$ 10

Furthermore, the core layer brain network was identified by the layer-layer correlation and conditional probability of the multilayer time-varying brain network. Specifically, the layer-layer correlation and conditional probability between different groups (normal, mild, and moderate ARHL groups) were statistically analyzed, and the layer with the most significant differences was identified as the core layer brain network. BrainNetViewer toolbox was used to visualize the topology of the core layer brain network, and the connectivity strength was used to quantitatively analyze the coupling between different brain regions [40].

Statistical analysis

The Shapiro–Wilk test was initially conducted to assess the normality of the data. Both the behavioral data and the network properties were found to follow a normal distribution (p > 0.05). Therefore, a one-way repeated measures analysis of variance (ANOVA) was performed to statistically analyze the behavioral data and network properties across the different groups, i.e., normal, mild, and moderate groups. Post-hoc independent samples t-tests between different groups were then performed for each significant ANOVA. Results were considered significant if the p-value was less than 0.05, with the p-value corrected for multiple comparisons using the Bonferroni method. All statistical analyses were performed with GraphPad Prism 10 software.

Results

Behavioral performance

The behavioral performance of the normal, mild, and moderate groups was quantitatively analyzed during auditory selective spatial attention tasks. Figure 5a shows that behavioral accuracy decreased with increasing severity of hearing loss: 77.3% ± 13.6 for the normal group, 64.8% ± 18.1 for the mild group, and 46.0% ± 16.1 for the moderate group. Furthermore, ANOVA revealed a main effect of hearing loss on accuracy (F(2, 29) = 9.483, p < 0.001, ${\eta }^2$ = 0.395). Bonferroni’s post-hoc analysis showed that behavioral accuracy was significantly lower in the moderate group than in the normal and mild groups (t(29) = 4.322, p < 0.001, and t(29) = 2.714, p = 0.033). However, an insignificant difference was found between the normal and mild groups (t(29) = 1.800, p = 0.274). In addition, the correlation between behavioral accuracy and hearing loss was analyzed as shown in Fig. 5b. A significant negative correlation was observed between hearing loss and behavioral accuracy (R =  −0.630, p < 0.001), indicating that the more severe the hearing loss, the more difficult it is to effectively detect and identify the target source.

Fig. 5.

Behavioral performance and correlation. a Behavioral accuracy in the normal, mild, and moderate groups. b Correlation between hearing loss and behavioral accuracy. Green, pink, and yellow represent the normal, mild, and moderate groups, respectively. ^***p < 0.001 and ^*p < 0.05

Multilayer time-varying brain network

Figure 6a shows the distribution of the 21 canonical electrodes and illustrates how these electrodes were used to construct the multilayer time-varying brain network. The diagonal subnetworks represent intra-layer connectivity (e.g., W1-W1), while the non-diagonal subnetworks represent inter-layer connectivity (e.g., W1-W6). Figure 6b shows the multilayer time-varying brain network for the normal, mild, and moderate groups, representing the average for all participants in each group. It can be observed that the multilayer time-varying brain network shows a high degree of structuring across the three groups, reflecting the dynamic stability of the brain functional connectivity network during the tasks. In particular, the intra-layer connectivity strength was consistently stronger than the inter-layer connectivity strength in all groups. Moreover, the inter-layer connectivity strength decreased as the distance between layers increased in all groups, i.e., the connectivity strength between W1 and W6 was the lowest across all conditions.

Fig. 6.

Multilayer time-varying brain network and construction diagram. a The distribution of 21 canonical electrodes and the multilayer time-varying brain network construction diagram. b Multilayer time-varying brain networks in the normal, mild, and moderate groups

Strength and correlation of intra- and inter-layer connectivity

The intra- and inter-layer connectivity strengths for the normal, mild, and moderate groups were quantitatively analyzed. Figure 7a shows that there was an insignificant change in intra-layer connectivity strength between the normal, mild, and moderate groups (F(2, 29) = 0.191, p = 0.823, ${\eta }^2$ = 0.013). Figure 7b shows that the inter-layer connectivity strength increased with increasing severity of hearing loss. ANOVA revealed a main effect of hearing loss on the inter-layer connectivity strength (F(2, 29) = 7.181, p = 0.003, ${\eta }^2$ = 0.331). Bonferroni’s post-hoc analysis revealed that the connectivity strength was significantly stronger in the moderate group than in the normal and mild groups (t(29) = 3.575, p = 0.004, and t(29) = 2.940, p = 0.019). The correlation between intra- and inter-layer connectivity strength was further elucidated for the normal, mild, and moderate groups, as shown in Fig. 7c, d, and e. As shown in the figure, a positive correlation between intra- and inter-layer connectivity strength was observed in three groups (Normal: R = 0.228, p = 0.526; Mild: R = 0.512, p = 0.089; Moderate: R = 0.647, p = 0.043). Especially in the moderate group, a significant positive correlation suggested that the stronger the intra-layer connectivity strength, the stronger the inter-layer connectivity strength. In addition, the slope of the fitted curve increased with increasing severity of hearing loss, with values of 0.029, 0.193, and 0.492 for the normal, mild, and moderate groups, respectively.

Fig. 7.

The strength and correlation of intra- and inter-layer connectivity in the normal, mild, and moderate groups. a Intra-layer connectivity strength, b inter-layer connectivity strength, c, d, and e are the correlations between intra-layer $⟨A⟩$ and inter-layer $⟨H⟩$ connectivity strength in the normal, mild, and moderate groups, respectively. ^**p < 0.01 and ^*p < 0.05

Properties of the multilayer time-varying brain network

Figure 8a and b shows the multilayer modularity and participation coefficient for the normal, mild, and moderate groups. Specifically, as the severity of hearing loss increased, the multilayer modularity decreased while the participation coefficient increased. Furthermore, ANOVA revealed a main effect of hearing loss on both multilayer modularity (F(2, 29) = 9.205, p < 0.001, ${\eta }^2$ = 0.388) and participation coefficient (F(2, 29) = 6.624, p = 0.004, ${\eta }^2$ = 0.314). Bonferroni’s post-hoc analysis showed that the multilayer modularity of the moderate group was significantly lower than that of the normal and mild groups (t(29) = 4.161, p < 0.001, and t(29) = 3.067, p = 0.014). The participation coefficient of the moderate group was significantly higher than that of the normal group (t(29) = 3.609, p = 0.003). However, insignificant differences were found between the normal and mild groups in either multilayer modularity (t(29) = 1.278, p = 0.634) or participation coefficient (t(29) = 1.480, p = 0.449).

Fig. 8.

The properties of the multilayer time-varying network in the normal, mild, and moderate groups and their correlation with hearing loss. a Multilayer modularity, b participation coefficient, c correlation between multilayer modularity and hearing loss, and d correlation between participation coefficient and hearing loss. Green, pink, and yellow represent normal, mild, and moderate groups, respectively. ^***p < 0.001, ^**p < 0.01, and ^*p < 0.05

Figure 8c and d shows the correlations between multilayer modularity and participation coefficient, respectively, and hearing loss. As can be seen in the figures, a significant negative correlation was found between multilayer modularity and hearing loss (R =  −0.619, p < 0.001). This indicates that the more severe the hearing loss, the lower the brain modularity and the more disorganized the network module partitioning. In contrast, the participation coefficient showed a significant positive correlation with hearing loss (R = 0.551, p = 0.001), suggesting that more information flows across modules in the brain with increasing severity of hearing loss.

Figure 9a shows the layer-layer correlation matrix, where a high correlation was observed between each layer, indicating a strong dependence and similarity between different layers. From the within-group perspective, the correlation generally decreased with increasing distance between different layers in most cases, indicating that the greater the distance, the lower the dependence and similarity. From the inter-group perspective, the correlation also decreased with increasing severity of hearing loss, suggesting impaired auditory processing, significant changes in network organization, and reduced similarity between different layers. Statistical analysis further revealed that LLC (W1, W3), LLC (W1, W4), LLC (W2, W4), and LLC (W3, W4) in the moderate group were significantly lower than those in the normal group (for all t(29) ≥ 2.566, p ≤ 0.047). In addition, LLC (W2, W3) and LLC (W2, W4) were significantly lower in the moderate group than in the mild group (for all t(29) ≥ 2.733, p ≤ 0.032).

Fig. 9.

Layer-layer correlation and conditional probability of the multilayer time-varying network in the normal, mild, and moderate groups. a Layer-layer correlation and b conditional probability. ^**p < 0.01 and ^*p < 0.05 between normal and moderate groups, ^#p < 0.05 between mild and moderate groups. Significance symbols were marked only in the lower triangular matrix due to the symmetry matrix for the layer-layer correlation

Figure 9b shows the conditional probability matrix, with a high probability observed between each layer, suggesting strong integration and stability between different layers. From the within-group perspective, the conditional probability generally decreased with increasing distance between different layers in most cases, indicating that the greater the distance, the lower the integration and stability. From the inter-group perspective, the conditional probability decreased with increasing severity of hearing loss, suggesting that individuals with normal hearing maintain better stability in task-related states. Statistical analysis further revealed that CP (W1, W4), CP (W2, W4), and CP (W3, W4) were significantly lower in the moderate group than in the normal group (for all t(29) ≥ 2.702, p ≤ 0.034). In addition, CP (W2, W4) was significantly lower in the moderate group than in the mild group (for all t(29) = 3.018, p = 0.016).

Core layer brain network

From the analysis of layer-layer correlation and conditional probability of the multilayer time-varying brain network, the fourth time window (400–500 ms) exhibited the most significant differences among the three groups. Therefore, the fourth time window was identified as the core layer of the multilayer time-varying brain network, and its topology and connectivity strength were further analyzed quantitatively. Figure 10a, b, and c shows that the hubs of the core layer brain network in all three groups were mainly located in the frontal cortex (FZ electrode). The connectivity strength between the right parietal and frontal lobes weakened with increasing severity of hearing loss, resulting in fewer modular structures in the right parietal. In contrast, the connectivity strength between the right frontal and prefrontal lobes strengthened with increasing severity of hearing loss, resulting in more modular structures in the right frontal lobe. Furthermore, the connectivity strength of the right frontal and right parietal lobes was quantitatively analyzed for the normal, mild, and moderate groups as shown in Fig. 10d. For the right frontal lobe, connectivity strength increased with increasing severity of hearing loss, but insignificant differences were found between the three groups (F(2, 29) = 0.032, p = 0.968, ${\eta }^2$ = 0.002). For the right parietal lobe, connectivity strength decreased with increasing severity of hearing loss. ANOVA revealed a significant main effect of hearing loss on connectivity strength (F(2, 29) = 3.738, p = 0.036, ${\eta }^2$ = 0.205). Bonferroni’s post-hoc analysis showed that connectivity strength was significantly lower in the moderate group than in the normal group (t(29) = 2.730, p = 0.032).

Fig. 10.

The topology and connectivity strength of the core layer brain network in the normal, mild, and moderate groups. a, b, and c are the topology of the normal, mild, and moderate groups, respectively, and d connectivity strength of the right frontal and right parietal lobes. ^*p < 0.05. Note that the network topology sparsity is 0.2 to improve clarity

Discussion

This study investigated the dynamic brain functional connectivity of ARHL during auditory selective spatial attention using the multilayer time-varying network. The results revealed that the effects patterns of hearing loss on auditory selective spatial attention are reflected in behavioral performance and significant dynamic changes in brain functional connectivity. Therefore, this study provides a comprehensive analysis of brain functional connectivity across time scales, offering potential biomarkers and time windows for the development of precise auditory rehabilitation strategies.

Change patterns of behavioral performance in ARHL

Behavioral results indicated that behavioral performance on the auditory selective spatial attention task was better in the normal group than in the mild and moderate groups, with accuracy showing a decreasing trend across these groups. This finding is consistent with previous studies, some of which have suggested that hearing loss may lead to difficulties in auditory processing, thereby affecting task performance [9, 41]. Further correlation analysis revealed a significant negative relationship between hearing loss and behavioral accuracy, supporting that greater hearing loss is associated with poorer task performance. This may be due to reduced perceptual ability and difficulty in allocating attentional resources caused by hearing loss [2]. However, it is noteworthy that there was no significant difference in behavioral performance between the mild and normal groups. This may be attributed to the minimal interference of mild hearing loss on auditory processing, with compensatory mechanisms in the brain potentially mitigating its effects [18, 26].

Change patterns of multilayer time-varying brain network in ARHL

An innovative aspect of this study is the construction of a multilayer time-varying brain network to capture the dynamic patterns of brain functional connectivity in ARHL. This method enables the simultaneous characterization of intra-layer connectivity, which represents the connectivity between brain regions on the same time scale, and inter-layer connectivity, which reflects dynamic connectivity across time scales [30]. This study found that the inter-layer connectivity strength increased significantly with increasing severity of hearing loss, whereas the intra-layer connectivity strength remained relatively stable. Previous research has highlighted the critical role of the temporal scales in attentional processes, demonstrating that individuals with normal hearing show rapid increases in attentional regulation over time, whereas individuals with hearing loss show slower increases [2, 42]. The physiological mechanism underlying this phenomenon may be rooted in the brain’s neuroplastic compensatory response to degraded auditory input. Specifically, hearing loss may cause the brain to strengthen connectivity across time scales to compensate for the lack of auditory information. Particularly in the moderate group, the strengthening of the inter-layer connectivity may reflect the brain's adaptation process to hearing loss, which maintains the ability to perform tasks by strengthening connectivity across time scales.

This study also revealed that with increasing severity of hearing loss, the multilayer modularity of the multilayer time-varying brain network significantly decreased, while the participation coefficient significantly increased. Physiologically, the reduced modularity and increased participation coefficient may reflect a shift from “specialized processing” to “global integration” in the brain’s information processing mode [28, 43]. Normally, modular networks enable efficient parallel processing of specific tasks, but under auditory deprivation, the brain breaks down modular boundaries to recruit non-auditory brain regions for cross-module collaboration. For example, previous research has shown that several neurological disorders, such as depression and stroke, can disrupt the modular structure of brain networks, resulting in an increased reliance on cross-module information exchange [44, 45]. A similar phenomenon was observed for ARHL in the present study. In contrast, the moderate group showed a higher participation coefficient, suggesting that information flow in their brains was more frequent and widespread during auditory selective spatial attention tasks. This is similar to related research suggesting that individuals with hearing loss exhibit more communication pathways between relevant brain regions [23]. This structural reorganization of the network may represent a compensatory mechanism by which the brain enhances cross-module information exchange to mitigate the auditory information deficit caused by hearing loss [38].

In summary, the aforementioned changes in the properties of the multilayered time-varying brain network can serve as objective neurophysiological biomarkers to evaluate the effectiveness of rehabilitation interventions. Additionally, for moderate ARHL patients who exhibit obvious compensatory network adjustments, targeted auditory training programs can be designed to enhance the integration of cross-temporal scale information and cross-module communication. For example, training tasks involving multi-temporal auditory signal processing can help strengthen the brain’s compensatory capacity, thereby improving the efficiency of auditory spatial selective attention in complex acoustic environments.

Change patterns of layer-layer correlation and conditional probability in ARHL

This study found that the layer-layer correlation decreased significantly with increasing severity of hearing loss, especially over longer time intervals. From a physiological perspective, the reduced layer-layer correlation reflects impaired temporal coherence of neural oscillations. Auditory spatial attention relies on synchronized neural activity across different time windows to integrate sequential sensory information [29]. Specifically, the relatively high correlation in the normal and mild groups indicates that the brain network structure is relatively stable across different time scales and maintains a more coherent flow of information. This may contribute to their better performance on the auditory spatial selective attention task. However, the correlation in the moderate group was significantly reduced, indicating a severe disruption of temporal information transfer. This may reflect the brain’s compensatory mechanism of adjusting its network structure to mitigate the effects of hearing loss [5, 46]. The conditional probability results provided further insight into the stability and dependencies between different time windows, i.e., the conditional probability decreased significantly with increasing severity of hearing loss. The further decrease in conditional probability suggests a weakened dependency between adjacent time windows, indicating that the brain can no longer reliably predict subsequent neural activity patterns, resulting in inefficient information transmission.

Further statistical analysis showed that the differences between the three groups were most significant in the fourth (400–500 ms) time window. Previous studies have shown that in noisy environments, older adults have a lower N400 component amplitude than younger adults, suggesting a reduced delay in attentional control and speech processing in older adults [47]. Of note, the latency of the N400 components was approximately 400–500 ms. This phenomenon may lead to greater differences in the correlation and conditional probability of this windows. Therefore, in this study, the fourth layer was identified as the core layer of the multilayer time-varying network. This core layer provides a precise temporal target for the design of auditory rehabilitation strategies. Targeted training can be developed to modulate neural activity within this specific time window, alleviating the disruption to information transfer caused by hearing loss and improving the patient’s ability to process auditory spatial information more effectively.

Change patterns of the core layer brain network in ARHL

The core layer brain network analysis revealed that the connectivity patterns of the right frontal and right parietal lobes were changed in ARHL. Specifically, as the severity of hearing loss increased, the right parietal-frontal lobe connectivity strength weakened, while the right frontal-prefrontal connectivity strength increased. Previous studies have shown that parietal lobe activation is strongly associated with auditory attention, particularly target localization and selection [48, 49]. Thus, the increasing severity of hearing loss resulted in a progressive decrease in the right parietal lobe connectivity strength. These findings further highlighted the critical role of the right parietal lobe in auditory spatial selective attention. In addition, some studies have shown that older adults have significant late positive complexes and P2 components in the frontal lobe, possibly due to the compensatory effects of hearing loss [13, 47]. However, this compensatory effect was insignificant, possibly due to the mild to moderate levels of ARHL in this study. In summary, from a physiological perspective, the weakened right parietal connectivity may be attributed to reduced synaptic plasticity in the parietal lobe caused by long-term auditory deprivation. Conversely, the enhanced frontal-prefrontal connectivity reflects the recruitment of the prefrontal cortex for compensatory control.

These region-specific connectivity changes are important guiding significance for the development of personalized hearing-care strategies. On the one hand, for patients with weakened right parietal-frontal connectivity, a training program combining spatial attention and target localization can be implemented, thereby improving the patient’s ability to identify target sound directions in complex environments. On the other hand, for enhanced frontal-prefrontal connectivity reflecting compensatory effects, training can be designed to strengthen this mechanism further, maximizing the brain’s potential to compensate for deficits in auditory information processing.

Limitations and prospects

Although this study provides valuable insights into the dynamic brain functional connectivity patterns in ARHL during auditory spatial selective attention, some limitations exist. Firstly, the sample comprised only individuals with mild-to-moderate hearing loss, and was relatively small (although statistically valid). Future studies should incorporate a larger sample size and include participants with severe hearing loss to thoroughly investigate the dynamic functional connectivity patterns in ARHL. Secondly, this study focused on the dynamic patterns caused by hearing loss without considering factors such as age and cognition (although no difference). In future studies, these covariates will be included as fixed effects in linear mixed models or as covariates in regression analysis to further evaluate the potential interactions between these covariates and the group factor. Finally, we strictly excluded participants with confirmed vascular risk or related diseases during recruitment to minimize confounding effects on the EEG data. Future research will involve performing neuroimaging assessments to explicitly screen for subclinical cerebral small vessel disease and conducting sensitivity analyses to validate the robustness of the findings regarding ARHL-related dynamic network changes.

Conclusion

This study investigated the dynamic patterns of brain functional connectivity in ARHL during auditory spatial selective attention using the multilayer time-varying network. The findings demonstrated that ARHL exhibited significantly decreased performance on tasks with increasing severity of hearing loss. Concurrently, substantial change patterns were observed in the multilayer time-varying and core layer brain networks. Specifically, the higher the severity of hearing loss, the lower the coordination of local brain regions and the increased intensity of cross brain information exchange. Moreover, the right parietal lobe was found to play a critical role in auditory spatial selective attention, while connectivity between the right prefrontal and frontal lobes may serve a compensatory effect in ARHL. In summary, the dynamic brain network measures identified in this study deepen our understanding of the neural mechanisms underlying ARHL and provide potential biomarkers and time windows for the development of precision auditory rehabilitation strategies.

Data availability

The original data supporting the results of this study are available from the corresponding authors.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contribution

Hongxing Liu: conceptualization, methodology, writing—original draft. Yanru Bai: conceptualization, writing review, and editing. Mingkun Guo: methodology. Ran Zhao: formal analysis. Jianing Zhu: data curation. Guangjian Ni: conceptualization, methodology, writing review, and editing

Funding

This work was supported by projects from the National Natural Science Foundation of China (No. 824B2056), the National Key Research and Development Program of China (No. 2023YFF1203500) and the seed Foundation of Tianjin University (2025XJ1-0006).

Declarations

Ethics approval

The study protocol was approved by the Ethics Review Committee of Tianjin University (Approval No. TJUE-2024–384), and all participants provided written informed consent prior to the experiment.

Conflict of interest

The authors declare no competing interests.