EEG brain networks modulation during sleep onset: the effects of aging

Abstract

The aim of the present study is to investigate differences in brain networks modulation during the pre- and post-sleep onset period, both within and between two groups of young and older individuals. Thirty-six healthy elderly and 40 young subjects participated. EEG signals were recorded during pre- and post-sleep onset periods and functional connectivity analysis, specifically focusing on the small world (SW) index, applied to EEG data (i.e., frequency bands) was examined. Significant differences in SW values were found between the pre-sleep and post-sleep onset phases in both young and older groups, with a reduction in the SW index in the theta band common to both groups. Additionally, an increase in the SW index in the beta band was exclusive to the elderly group during the post-sleep onset period, while an increase in the sigma band was exclusive to the young group. Furthermore, differences between the young and elderly groups were found during both phases, including a decrease in the SW index within the delta band, an increment in the sigma and beta bands in the elderly compared to the young group during the pre-sleep onset period, and a notable absence of sigma band modulation in the elderly group during the post-sleep onset condition. These findings provide insights into age-related changes in sleep-related brain network dynamics and their potential impact on sleep quality and cognitive functions, prompting interventions aimed at supporting healthy aging and addressing age-related cognitive decline.

Introduction

The process of human aging is complex, individualized and irreversible and involves all organs of the body, including—despite its remarkable resilience and plasticity—the brain [1–3].

The physiological aging process contributes to a great number of quantitative and qualitative modifications in the brain, starting from single synaptic density and plasticity, ion channels, neurotransmitters, neuronal and glial metabolism and extending up to brain morphology and size, secondary to neuronal loss and cortical/subcortical atrophy [4, 5].

As a consequence of these modifications, healthy brain aging impacts on sleep duration and quality. There is a growing amount of literature demonstrating the strict relationship between sleep nd cognitive functions (e.g., Wild et al. [6]), especially in aging population. Although this evidence may not always point to a strong or direct causal link, several longitudinal and cross-sectional studies suggest that sleep plays a role in cognitive health. Falck and colleagues [7] demonstrated that greater sleep efficiency, an objective measure of sleep quality, was associated with better cognitive performance in older adults, even when accounting for physical activity. Moreover, several studies found that changes in sleep patterns, such as reduced sleep efficiency and fragmented sleep, are commonly seen in individuals at risk for cognitive impairments, including dementia [8–10]. For example, Minchao Li and colleagues found that both short and long sleep durations were associated to lower cognitive function, but they also observed a consistent positive relationship between better self-reported sleep quality and higher cognitive scores [11] suggesting that both the quantity and quality of sleep can influence cognitive health. Findings provided by McSorley and colleagues lead to a more complex picture, showing that only objective measures of sleep disruption, rather than self-reported sleep quality, were associated with cognitive decline over a 5-year period [12]. In this view, although these findings should be interpreted with caution, it is reasonable to hypothesize that age-related changes in sleep could—at least in part—contribute to decline in fluid cognitive abilities, which is associated in the elderly with reductions in executive function, memory, attention, and processing speed [13].

Indeed, several polysomnographic (PSG) studies have previously demonstrated changes in sleep architecture during aging: advanced sleep timing, longer time taken to fall asleep, shorter overall sleep duration, increased sleep fragmentation, more fragile sleep, reduced amount of deeper NREM stages, i.e., slow wave sleep (SWS), increased time spent in lighter NREM stages 1 and 2, shorter and fewer NREM-REM sleep cycles [14].

Since its introduction, the best tool of choice to measure sleep staging characteristics has been polysomnography (PSG), which provides reliable measures of sleep quality and quantity, also allowing the identification of age-related peculiarities during both development and aging. In particular, the EEG data analyses have highlighted some of the key aging-related electrophysiological sleep disturbances, such as lower NREM delta activity, less time spent in the deeper stages, and drop in frequency and amplitude of sleep spindles and K-complexes [15]. Moreover, as the aging brain is characterized by a loss of synaptic contacts, neuronal assemblies and myelinated fibers connecting them in different cortical regions, there is a progressive degradation of the ability to integrate many different functions leading—among the other—to a decline in cognitive performances [4]. One of the most recent and advanced approaches to EEG data analysis is the brain functional connectivity exploration [16–18]. Indeed, the analysis of brain functional connectivity with graph theory has emerged as a new promising tool based on the fact that the brain is modeled as a network in which all functions—including cognitive ones—are the result of coordinated activity among distant brain areas cooperating in a dynamic and hierarchically settled network organization [19]. In the context of EEG signal analysis and graph theory, there are several indices used to study the properties of brain networks. One particularly interesting index is the small world (SW) parameter. It is derived from graph theory metrics such as clustering coefficient and characteristic path length. The SW provides insights into the balance between local clustering (i.e., the presence of densely connected neuronal groups) and global integration (i.e., efficient information transfer) within the brain functional connectivity network [20–22].

Several studies have investigated brain networks and connectivity during resting-state conditions to explore the impact of aging. One notable finding is that healthy aging results in a shift of brain resting stage, background EEG activity from posterior to anterior regions, with strengthened inter-regional connections among frontal, parietal, and temporal brain areas [17, 23–25]. These modifications lead to a reorganization of the brain topology and a propensity for the network to adopt a different organization as ageing progresses [26, 27]. This implies that the brain network structure becomes less specialized, more disorganized, and—at the end—less efficient with increasing age. In fact, higher SW values in low-frequency bands in the elderly have been related to functional disconnection processes and a reduction in long-range communication between cortical areas, which may contribute to a decline in network efficiency. Conversely, lower SW values in higher frequency bands, like alpha, are often associated with a shift toward a less random network structure, as seen in Alzheimer’s disease, and impaired cortico-cortical interactions [16, 28–30].

Connectivity values, particularly in SW values, exhibit age-related differences, showing lower values in the slow EEG bands and higher values in the more rapid frequency bands in older respect to younger individuals [31]. Overall, these findings provide evidence that brain aging affects network organization and connectivity, which may have implications for understanding the neural mechanisms underlying age-related brain modification and decline of some functions including cognition.

Recently, the brain networks theory has been applied to EEG measures recorded during the sleep onset process in healthy young subjects, with the aim to describe modifications in the complex dynamic connectivity of brain regions occurring during the transition from wakefulness to sleep. Indeed, the brain undergoes local frequency-specific changes in the EEG pattern during the sleep onset [32]. Previous studies have examined the synchronization of brain rhythms and the modulation of brain network during sleep onset [33]. Vecchio and colleagues investigated the SW characteristics of brain networks, as reflected in EEG rhythms, during the wakefulness-to-sleep transition by comparing pre- and post-sleep onset in young adults. They showed that sleep onset is associated with changes in the organization of brain networks. Specifically, in the sigma EEG activity recorded from frontal electrodes, there is an increase in SW, indicating stronger connectivity and a less ordered network. On the other hand, in the delta and theta bands, there are lower SW values, suggesting a more ordered network and a functional disconnection in those areas of the brain characterized by slow wave accumulation. The same research group has analyzed the changes in brain networks derived from EEG analysis during pre- and post-sleep onset conditions after 40 h of sleep deprivation (SD) by means of graph analysis. SD led to significant alterations in the small world index of brain networks, including decreased values in the delta and theta bands during post-sleep onset and increased values in the sigma band, while pre-sleep onset conditions during a recovery night after sleep deprivation showed a decreased small world index in the beta band compared to a night of normal sleep [34].

Although older individuals exhibit a longer latency to fall asleep [14] and a recent study found aging-related peculiarities in the local EEG pattern of the sleep onset [35], until now and to the best of our knowledge there is no previous study on the modulation of brain networks analyzed not only by comparing pre- and post-sleep onset conditions but also by examining whether and how these pre-post differences are influenced by aging. Understanding the differences in network modulation between young and older individuals during the wakefulness-to-sleep transition has important implications for our knowledge of age-related changes in brain function and sleep regulation. Moreover, investigating the impact of aging on brain network dynamics during sleep onset may contribute to the development of targeted interventions and strategies to promote healthy sleep patterns in older populations and to better elucidate causal relationships between sleep changes and age-related cognitive decline.

Within this theoretical frame, the aim of the present study is to investigate the modulation differences in brain networks during the pre- and post-sleep onset periods both within and between young and older individuals. To achieve this, we used functional connectivity analysis, specifically focusing on the SW index, applied to EEG data. By employing functional connectivity analysis, we aimed to capture the underlying patterns of brain network dynamics and examine how they are influenced by the sleep onset process during aging.

Methods

Participants

In the present study, two datasets [35, 36], respectively of 36 healthy elderly subjects (16 F; mean age = 68.40 years, standard error (SE) = 1.08 years; age range = 59–81 years) and 40 young subjects (20 F; mean age = 23.86 years, SE = 2.88 years; age range = 18–29 years), were analyzed. The inclusion criteria for the participants were: regular sleep–wake rhythms, no daytime napping habits, no excessive daytime sleepiness, no sleep, medical, neurological, or psychiatric disorders, no use of psychoactive, hypnotic drugs, or relevant medication that might affect the sleep EEG, absence of alcohol, and other substance addiction. These criteria were evaluated through a clinical interview. Participants were requested to maintain a regular sleep–wake rhythm during the week before the PSG recording; compliance was controlled through 1-week sleep log in both groups and actigraphic recordings (AMI Mini Motionlogger device) in young adults. The mini-mental state examination (MMSE) was used to perform a cognitive screening in the group of older adults to exclude cognitive decline (mean corrected MMSE ± SE: 27.04 ± 0.24).

The experimental procedures adhered to the Declaration of Helsinki guidelines and national regulations. All the participants provided informed consent after the study protocol was approved by the local Ethical Committee.

EEG recording

PSG was conducted in a room specifically designed to minimize external disturbances and maintain a controlled temperature, using an Esaote Biomedica VEGA 24 polygraph or Micromed system plus digital polygraph. Data were collected during an undisturbed night of sleep, starting according to the habitual individual sleep schedule and concluding after 7.5 h of accumulated sleep (monitored in real time by experienced sleep experts). EEG signals were recorded from 19 scalp electrodes using the unipolar derivation method based on the 10–20 system with a sampling rate of 128/256 Hz, referenced to the averaged mastoids in young adults and to the ground electrode (Fpz) and offline re-referenced to the averaged mastoids in older adults. A pulse oximeter was placed on the right index finger to monitor oxygen saturation and exclude sleep breathing disorders in the older individuals.

The submental EMG activity was recorded with a time constant of 0.03 s, while bipolar horizontal eye movements (EOG) were recorded using electrodes positioned approximately 1 cm away from the dominant eye’s medial and lateral canthi, with a time constant of 1 s. The impedance of these electrodes was consistently maintained below 5 kΩ.

For the present analyses, we considered the EEG signals recorded both 5 min prior to sleep onset (pre-sleep onset) and 5 min following the initiation of sleep (post-sleep onset), considering the first epoch of stage 2 NREM (i.e., appearance of the first sleep spindles or K-complex) as the sleep onset [36, 37].

The sleep stages were visually scored in 12-s epochs using the central EEG derivation at Cz, along with EMG and EOG, following the established criteria [38].

EEG preprocessing

Offline, the EEG data were processed using a customed MATLAB software, based on EEGLAB toolbox (Swartz Center for Computational Neurosciences, La Jolla, CA, USA). All the EEG records were down sampled with a frequency of 128 Hz and epochs of 2 s were extracted from the original EEG continuous data. Those with aberrant waveforms or artefactual activity were removed firstly by an expert visual inspection and then, by using the Infomax ICA algorithms [39–41].

Functional connectivity analysis

To investigate the functional connectivity of the brain, the exact low-resolution electromagnetic tomography (eLORETA) software was utilized [16, 17, 30]. The eLORETA algorithm offers a linear inverse solution for EEG data, eliminating localization errors and providing accurate source localization under ideal conditions, free from noise [42]. By aligning with the EEG potential distribution on the scalp, eLORETA software computed a discrete, three-dimensional (3D) distributed linear, minimum-norm, weighted, inverse solution. The inclusion of weights in eLORETA ensured precise localization of sources, crucial for testing point sources and generating images of current density with accurate localization, although with low spatial resolution due to high correlation between neighboring neuronal sources [43, 44].

For brain connectivity assessment, eLORETA software was employed to evaluate 84 regions, providing a topographic view of the entire brain with the center aligned to the available 42 Brodmann areas (BAs) in both the left and right cerebral hemispheres. These BAs include regions such as 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 17, 18, 19, 20, 21, 22, 23, 24, 25, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, and 47.

To investigate the electrical brain activity used for assessing cerebral-functional connectivity, regions of interest (ROIs) were established. The electric neuronal activities of all voxels within each cortical ROI were averaged to generate a single signal for each ROI, computed with eLORETA. For each hemisphere, the intracortical lagged linear coherence was evaluated among the eLORETA current density time series of the 84 ROIs. This analysis was performed using the “all nearest voxels” method [45], measuring coherence between all possible pairs of the 84 ROIs across the seven independent EEG frequency bands (delta (0.5–4.5 Hz), theta (5–7.5 Hz), alpha (8–11.5 Hz), sigma (12–15.5 Hz), and beta (16–24.5 Hz) for each participant and condition [34, 46].

The lagged linear connectivity in a specific frequency band (ω) was calculated using the equation below, which takes into account the covariance and variance of the time series [45]:

$${\text{LagR}}_{\text{xyw}}^2=\frac{{\left[\text{lmCov},(,x,,,y,)\right]}^2}{\text{Var}(x)\times \text{Var}(y)-{\left[\text{ReCov},(,x,,,y,)\right]}^2}$$

The lagged linear connectivity provides a measure of true physiological cerebral connectivity, independent of low spatial resolution and volume conduction. The resulting lagged linear connectivity values for each frequency band between all pairs of ROIs were used as weights for constructing the networks in the subsequent graph analysis.

Graph analysis

A network is a mathematical representation of a complex system, consisting of nodes (vertices) and links (edges) connecting pairs of nodes. In brain networks, nodes typically represent discrete neuronal assemblies in specific brain regions, while links can represent anatomical, functional, or effective connections [47], depending on the dataset [48]. Anatomical connections typically correspond to white matter fiber tracts connecting gray matter brain regions (such as cortical areas or subcortical relays), while functional connections reflect temporal correlations in activity and can occur between anatomically unconnected regions that functionally cooperate through indirect connections.

In this study, weighted and undirected networks were constructed. The vertices of the network represent the estimated cortical sources in the Brodmann areas (BAs), and the edges are weighted based on the lagged linear values between each pair of vertices. The Brain Connectivity Toolbox (BCT, http://www.brain-connectivity-toolbox.net/) was utilized, along with home-made Matlab scripts, for the graph analysis.

To assess the balance between local connectedness and global integration within the functional weighted brain networks, the small world index ($\text{SW}$) was computed. SW organization lies between that of random networks, which have short overall path lengths but low levels of local clustering, and regular networks, which exhibit high levels of clustering but longer path lengths [30]. This implies that nodes in small world networks are connected through relatively few intermediate steps, while maintaining few direct connections [17]. $\text{SW}$ index is defined as the ratio between the normalized clustering coefficient ($C$) and the normalized characteristic path length ($L$) for each frequency band.

The $C$ quantifies the ratio of existing connections between neighboring nodes to the maximum possible number of edges between neighbors. It can be expressed as the number of triangles in which a vertex participates, normalized by the maximum number of such triangles [49–52]. In the current study, the weighted $C$ was computed by replacing the number of triangles with the sum of triangle intensities. The weights associated with the links ($i,j$) are represented by $w_{\text{ij}}$, assuming normalized weights ranging from 0 to 1 for all $i$ and $j$. The weighted $C$ is computed by taking the geometric mean of triangles around each vertex $i$ [49, 50]:

$$C=\frac{1}{n}\sum _{i\in N}\frac{{2t_i}^W}{k_i\left(k_i,-,1\right)}$$

with

$$t_i^w=\frac{1}{2}\sum _{j,hϵN}\left(w_{\text{ij}},w_{\text{ih}},w_{\text{jh}}\right)_{}^{_{}^1\left./,_3^{}\right)}$$

The characteristic path length ($L$) is defined as the shortest weighted path length between two nodes $i$ and $j$. It is computed as the sum of weighted shortest paths over all pairs of nodes in the graph [49, 50]:

$$L=\frac{1}{n}\sum _{iϵN}\frac{{\sum }_{jϵN,j\ne i}{d}_{\text{ij}}^W}{n-1}$$

where

$$d_{\text{ij}}^w=\sum a_{\text{uv}}ϵg_{i\leftrightarrow j}^{w}f(w_{\text{uv}})$$

The function f represents a mapping from weight to length, and $g_i^w\leftrightarrow j$ denotes the shortest weighted path between nodes $i$ and $j$.

Individual normalized relative measures were obtained by dividing the characteristic path length and clustering coefficient values by their average across all bands for each subject.

Finally, the $\text{SW}$ index, obtained as the ratio between normalized $C$ and $L$, was evaluated for all subjects in both pre- and post-sleep onset conditions in all the EEG frequency bands [33].

Statistical analysis

In line with the study’s objective, a mixed-design analysis of variance (ANOVA) was performed to test the $\text{SW}$ index modulation among the factors time (pre-sleep onset, post-sleep onset), frequency band (delta, theta, alpha, sigma, beta), and group (young, elderly) as non-repeated factor, utilizing the software package Statistica (StatSoft Inc.).

The analyses were conducted with a statistical cut-off level of p < 0.05. The normality of the data was tested using the Kolmogorov–Smirnov test, and the hypothesis of Gaussianity could not be rejected. ANOVA was chosen since it is known to be robust for the departure of normality and homoscedasticity of data being treated. Greenhouse and Geisser correction was used for the protection against the violation of the sphericity assumption in the mixed ANOVA design.

Moreover, the post hoc analysis was performed using Duncan’s or Bonferroni test and a 0.05 significance level. In addition to p-values, effect sizes were calculated to assess the magnitude of the differences observed in the data. Two metrics were used: Cohen’s d and η² (eta-squared). Cohen’s d is a measure of standardized mean differences, commonly interpreted as small (0.2), medium (0.5), or large (0.8), while η², which represents the proportion of variance explained by the factor in question.

Additionally, a one-way analysis of variance (ANOVA) with the factors group (young, elderly) was conducted with a significance level of 0.05 using Statistica v.8 (StatSoft Inc).

Finally, to determine whether prolonged time to fall asleep is associated with specific changes in network dynamics, a correlation analysis was conducted between SW metrics across the seven EEG frequency bands and sleep latency in both young and elderly groups (Bonferroni corrected p < 0.05).

Results

The ANOVA results showed a statistically significant interaction (F(4,292) = 9.2264, P < 0.000001, Cohen’s d = 0.3329, η² = 0.112) among all variables: time (pre-sleep onset, post-sleep onset), frequency band (delta, theta, alpha, sigma, beta), and group (young, elderly).

In particular, the Duncan’s test post hoc analysis demonstrated significant differences in $\text{SW}$ values between the pre- and post-sleep onset conditions in both the elderly and young groups (Fig. 1). More specifically, in the elderly subjects, a decrease of the $\text{SW}$ index was observed in the theta band (p = 0.003012) in the post-sleep onset condition compared to the pre-sleep onset condition, while it increased in the beta band (p = 0.004311) in post compared to pre-sleep onset condition. Conversely, in the young subjects, a decrease of the $\text{SW}$ index was observed in the delta (p = 0.003098) and theta band (p = 0.000113), while an increase was observed in the sigma band (p = 0.000002) during the post-sleep onset compared to the pre-sleep onset.

Fig. 1.

Small world index values among the frequency bands of interest (delta (δ), theta (θ), alpha (α), sigma (σ), beta (β)) for the pre- and post-sleep onset conditions in elderly and young group. Significant values of $\text{SW}$ index between groups are reported with a single asterisk (*) that signifies values of SW index between conditions are reported with a number sign (#)

Moreover, Duncan’s post hoc comparisons revealed significant differences between the young and elderly groups during the pre- and post-sleep onset phases (Fig. 1). Specifically, during the pre-sleep onset condition, the elderly group exhibited a decrease in the $\text{SW}$ index within the delta band (p = 0.000004), whereas increase was observed in the sigma (p = 0.000001) and beta (p = 0.000003) bands compared to the young group. After the onset of sleep, the differences between the two groups persisted in the delta frequency band (p = 0.000011), where lower values were observed in the elderly group, and in the beta frequency band (p = 0.000002), where higher values were found in the elderly compared to the young. No significant difference was observed in the sigma band. A summary of the p-values of the within and between analysis have been reported, respectively, in Table 1 and Table 2.

Table 1.

Within-group statistical analysis (pre- vs. post-sleep onset). This table compares the small world (SW) properties within different frequency bands between pre and post sleep onset in two age groups: elderly and young. Statistical significance is represented by p-values, with values below 0.05 indicating significant differences. A dash (“-”) means no statistical differences were found (p > 0.05)

	Within statistical analysis (pre- vs. post-sleep onset)
	Elderly	Young
Delta	-	p = 0.003098
Theta	p = 0.003012	p = 0.000113
Alpha	-	-
Sigma	-	p = 0.000002
Beta	p = 0.004311	-

Table 2.

Between-group statistical analysis (elderly vs. young). This table shows the statistical comparison between elderly and young individuals in terms of small world (SW) properties across different frequency bands both pre- and post-sleep onset. Statistical significance is represented by p-values, with values below 0.05 indicating significant differences. A dash (“-”) means no statistical differences were found (p > 0.05)

	Between statistical analysis (elderly vs young)
	Pre-sleep onset	Post-sleep onset
Delta	p = 0.000004	p = 0.000011
Theta	-	-
Alpha	-	-
Sigma	p = 0.000001	-
Beta	p = 0.000003	p = 0.000002

Finally, a sensitivity analysis was conducted. ANOVAs were performed for each frequency band, with sleep onset as a within-participant effect and age group as a between-participant effect. The results of these analyses confirmed the findings reported above.

A qualitative representation of brain network architecture during aging in pre- and post-sleep onset phases in the significant EEG frequency bands (delta, sigma, and beta) is reported in Fig. 2.

Fig. 2.

Brain networks architecture during aging in pre- and post-sleep onset phases in the significant EEG frequency bands (delta, sigma and beta)

No significant statistical differences were found in sleep onset latency between younger and older adults (F(1, 73) = 1.7594, p = 0.18884).

However, while the overall sleep latency did not differ significantly, further analysis revealed distinct age-related patterns when examining the relationship between SW and sleep latency. In the pre-sleep onset, a significant negative correlation was observed between sigma SW and sleep latency in the elderly group (Fig. 3, r = − 0.3930, p = 0.0177), indicating that individuals with higher sigma SW experienced shorter sleep latencies. In contrast, this correlation was not present in the younger group (Fig. 3, r = − 0.2037, p = 0.2135).

Fig. 3.

Scatterplots of the significant correlation between sigma SW and sleep latency in younger and older adults in the pre-sleep onset phase. The figure also reports the 95% confidence interval

In the post-sleep onset phase, the pattern reversed: a significant negative correlation was found in the younger group (Fig. 4, r = − 0.3799, p = 0.0177), where higher sigma SW was associated with shorter sleep latency, while no significant correlation was detected in the older group (Fig. 4, r = − 0.1917, p = 0.2626).

Fig. 4.

Scatterplots of the significant correlation between sigma SW and sleep latency in younger and older adults in the post-sleep onset phase. The figure also reports the 95% confidence interval

Discussion

The aim of the current study was to evaluate the modulation of brain networks during the pre- and post-sleep onset period both within and between young and older individuals. We utilized functional connectivity analysis, focusing on the graph theory metrics and, especially on the small world (SW) index, which provided a synthesis of integration and segregation networks characteristics.

Our first finding revealed significant differences in SW values between the pre-sleep onset and post-sleep onset phases within the young and older groups. Specifically, both groups showed a reduction in the SW index in the theta band during the post-sleep onset compared to the pre-sleep onset condition. This finding is consistent with several studies—with different methods and none of these by brain connectivity tools—demonstrating the presence of the mentioned EEG band modulations during different sleep phases [53–55]. In particular, the modulation of the theta EEG rhythm across different sleep stages is a well-studied phenomenon, especially during non-rapid eye movement (NREM) and rapid eye movement (REM) sleep [56]. Specifically, during NREM sleep, notably stage 1 and stage 2, theta rhythms may manifest episodic bursts which are associated with the transition from wakefulness to sleep [57, 58]. Additionally, some studies have reported a trend towards more ordered brain network configurations in lower EEG frequency bands, characterized by increased theta phase synchrony and reduced theta SW values during sleep [32, 59–61]. These findings support our observations in the theta band modulations between pre- and post-sleep onset, suggesting that these dynamic changes in brain network organization could be attributed to both disconnection from external sensory input and engagement in internal cognitive processes including memory consolidation of the daily events [62]. Moreover, this interpretation is consistent with the synaptic homeostasis hypothesis of sleep [63], which proposes that sleep plays a vital role in restoring a balance in synaptic strength. Since wakefulness tends to enhance synaptic strength, maintaining this heightened state becomes impractical due to energy demands and space constraints. In this context, sleep is believed to renormalize synaptic strength to a sustainable baseline level, supporting optimal memory and cognitive performance [64].

Furthermore, here we observed an increase of the SW index in the beta band during the post-sleep respect to pre-sleep onset phase solely within the elderly group, whereas an increase in the sigma band during the post-sleep onset period was observed exclusively in the young group. The increased SW values in the beta band, as observed in the elderly, represents a tendency of the brain network towards a more random organization, reflecting a less organized response during the transition from wakefulness to sleep in aging individuals. This beta modulation may be associated with age-related changes in sleep architecture, including shifts in the duration and composition of sleep stages, possibly due to the documented decline in functional connectivity observed in the aging brain within the EEG beta band [35] and to the loss of effective communication between different brain areas [65] during the transition to sleep. The modulation of the beta frequency dynamics observed solely in the elderly during the transition from wakefulness to early sleep, sparks an exploration into potential mechanisms and implications. The proposition that beta EEG frequencies support cognitive processes [66] raises questions about the specific neurophysiological processes at play. Several studies, such as the one by Carrier et al. [67], have reported that this modulation in the beta activity during NREM sleep is specific of the elderly condition. The authors hypothesized that the increase in EEG beta activity across the night in middle-aged subjects led to increased cortical activity during sleep, potentially making them more vulnerable to sleep disruption. This relative increase in EEG beta activity in older subjects during sleep may reflect higher amount of cortical activity pushing towards arousal, possibly contributing to their inability to maintain sleep and preceding the deterioration of subjective sleep quality. This suggestion aligns with studies indicating that older individuals encounter more difficulties than younger subjects in adapting to challenges to the sleep–wake cycle, such as shift work and jet lag [68–71]. It also supports the idea that less organized brain networks in the beta band in the elderly reflect a less organized response in the time period of transition between pre and post sleep onset.

Furthermore, our study showed an increase of the SW index in the sigma band during the post-sleep compared to the pre-sleep onset period exclusively in the young group. This prevalence has been previously reported [33], and it has been interpreted in terms of increased synchronized firing of neurons during sleep period characterized by EEG spindles which is known to be crucial for long-term synaptic potentiation, strengthening of interneuronal connection finally helping memory consolidation during sleep [72]. Moreover, the lack of modulation of the SW index in the sigma band in the older group may be associated to a characteristic decline in sleep spindle activity within the sigma band in this age, resulting in lower amplitude, duration, and density of sigma activity during sleep. Indeed, sleep spindles are known to be essential for synaptic plasticity, declarative memory consolidation, and motor learning [72, 73]. Their decline with aging may have significant implications for cognitive functions related to memory and learning in aging populations and could be associated with the breakdown of long-range connectivity pathways [74]. Accordingly, this interpretation implies that age-related alterations in the structural integrity of cortico-cortical and subcortical white matter tracts may contribute to changes in the way neural networks synchronize their firing and communicate, ultimately impacting on the absence of brain networks modulation in the sigma band in the elderly, as we reported.

Besides, when we looked at the differences between groups in pre-sleep onset and post-sleep onset phases, we found that during the pre-sleep onset period, the elderly group exhibited a decrease in the SW index in the delta band compared to young adults, while increases were observed in the sigma and beta bands. During the post-sleep onset condition, the observed modulations in the delta and beta bands are preserved between the young and elderly groups, while the modulation in the sigma band is notably absent between them.

In the pre- and post-sleep onset period, the observed decrease in the SW index within the delta and theta bands may be attributed to the well-established influence of aging on sleep-related brain activity, specifically in the slow wave rhythms [75–78]. The progressive decline in the EEG spectral power of slow wave activity during sleep with age suggests an age-related alteration in the dynamics of slow wave generation. Moreover, alterations in the delta activity at sleep onset might reflect changes in the average synaptic strength and efficacy of synapses in specific cortical areas following prolonged wakefulness [79]. The correlation between the amplitude and slope of EEG slow waves and the synchrony, strength, and efficacy of synaptic connections among cortical neurons hints at a potential link between age-related changes in synaptic dynamics [80], and here it is reflected in the brain networks activity, in particular in a more ordered brain network architecture, as observed in the decrement of the SW index in the delta band. In fact, as previously demonstrated by Massimini and colleagues, there is a breakdown of transcallosal and long-range effective connectivity during NREM sleep early in the night, with a rapid downgrading of cortical areas interacting. They demonstrated that in this state, the TMS-EEG response failed to trigger a sustained, long-range pattern of activation, while solely maintaining a localized response to the transcranial stimulation [81], suggesting that the mechanisms underlying the generation of slow waves may also be responsible for blocking the emergence of specific long-range responses during NREM sleep. Accordingly, this reduction in the diffusion of information within the networks could be due to more ordered brain network characteristics in slow waves. Furthermore, while young adults reduce both delta and theta SW, the elderly reduce only theta SW during sleep. Disruptions in this balance, at least in the elderly [82], may ultimately impact on sleep consolidation and lead to disruptions in the continuity of cognitive processes [81, 83–86].

Moreover, the observed decrease in the SW index within the delta band in the elderly compared to the young group could be explained as a potential disruption in the homeostatic regulation of slow EEG activity among older adults particularly during the first sleep stages [87]. The changes in waveform shape associated with aging, suggesting a reduction in the synchronized firing of neurons responsible for generating sleep oscillations, provide additional insight into the potential mechanisms contributing to the observed brain network organization as expressed by the reduction of the SW index in the delta band in the elderly respect to the young age group.

Additionally, in the pre- and post-sleep onset period age-related brain network changes were also observed in the faster EEG frequencies, namely the beta bands. One plausible explanation revolves around age-induced modifications in the connectivity patterns of neural networks, a phenomenon supported by various studies highlighting alterations in beta band connectivity during aging [88, 89]. The beta band, known for its involvement in memory tasks and its role in facilitating communication among distant cortical regions engaged in sensory processing, has a crucial role in age-related cognitive decline. In fact, disruptions in beta band connectivity could potentially impair cognitive function with aging. Alterations in network indices in high-frequency bands may indicate diminishing efficiency in neural communication with advancing age, as evidenced by shifts in EEG patterns [16]. This alteration implies a disruption in the delicate balance between localized specialization and broader integration within the aging brain. Consequently, the processing time may prolong, thereby contributing to the cognitive decline commonly associated with aging. This notion aligns with the broader conceptualization of brain aging as a disconnection process, supported by the well-documented deceleration of brain activity and decline in functional connectivity typically observed with age [16].

Another perspective to consider is the role of beta waves in motor control and movement planning. Beta activity is typically associated with states of wakefulness and activity: the higher SW in the beta band during the pre-sleep onset period in older individuals might be linked to and demonstrate the less coordinated transition from an alert and active state to a more relaxed and restful one [90, 91].

It is worth noting that the modulations observed in sleep network patterns in the elderly could also be attributed to the documented increase in excitability of the aging brain, as demonstrated in numerous studies utilizing TMS-EEG [86, 92–97]). This hyperexcitability might serve as a recruitment mechanism aiming at recruit “silent synapses/networks” for compensating the reduced efficiency of certain neurobiological processes associated with aging. It may be an adaptive response to the loss of synapses and neural connections commonly observed with aging.

The observation of higher SW values in the elderly after sleep onset, which indicates a more random network, points to a disruption in the ordered communication between different brain regions. It reflects in a more random network contributing to a decrease in the overall efficiency of information processing during sleep in the elderly. Several studies have demonstrated a modulation, specifically an increase, in the beta band in the post-sleep onset period in elderly people [67, 76, 98]. This beta band modulation might be viewed as a response to the increased randomness in network dynamics: the brain may enhance beta oscillations, possibly aiming to sustain some level of functional connectivity and mitigate the effects of the disrupted regularity. The interplay between these observations may also be influenced by cognitive processes occurring during sleep.

Finally, age-related differences in sleep onset latency and the relationship between SW network properties was explored in younger and older adults. Despite expectations of age-related increases in sleep latency, our results revealed no significant statistical differences between the two age groups, suggesting that sleep latency may be more resistant to the effects of aging compared to other sleep architecture parameters [99].

However, distinct age-related patterns emerged when examining the link between SW network properties and sleep latency. In the pre-sleep onset phase, older adults showed a significant negative correlation between sigma SW and sleep latency, indicating that more efficient SW network configurations helped them fall asleep faster. This suggests a compensatory mechanism in older adults, aligning with the concept of cognitive reserve [100]. In contrast, this relationship was absent in younger adults, likely due to their generally stable sleep architecture. In the post-sleep onset phase, younger adults exhibited a negative correlation between sigma SW and sleep latency, meaning efficient SW networks supported shorter sleep latency after sleep began. This highlights the role of SW network efficiency in maintaining and deepening sleep in younger adults [101]. No such correlation was found in older adults during this phase, suggesting age-related declines in the brain’s ability to sustain efficient network organization, contributing to reduced sleep stability with aging [102].

In conclusion, our current findings, which reveal significant differences between the young and elderly groups during both the pre-sleep onset and post-sleep onset phases and a correlation between network properties and sleep latency, introduce a novel dimension to the insights garnered from the previous study, significantly advancing our understanding of sleep-related network dynamics. These results highlight the importance of brain network organization in sleep processes and suggest that SW network properties may serve as a potential target for interventions aimed at improving sleep, particularly in older populations. Further investigations into the relationship between the EEG specific frequency bands, graph theory indexes, and sleep-related cognitive phases can contribute to a more nuanced understanding of age-related variations in sleep dynamics and their cognitive implications. Indeed, unravelling the complexities of sleep-related processes, the interplay of brain network organization during sleep and their vulnerability to aging could be essential for developing targeted interventions to support healthy aging and lays the groundwork for informed approaches to further research in this field, aiming to provide valuable insights into potential strategies for preserving cognitive health and promoting overall well-being in aging individuals.

Author contribution

FM: conceptualization, methodology, software, writing—original draft preparation; AC: software, writing—reviewing and editing; FV: supervision, writing—reviewing and editing; SS: writing—reviewing and editing; MG: writing—reviewing and editing; LDG: writing—reviewing and editing; PMR: conceptualization, supervision, methodology, writing—reviewing and editing.

Data availability

The data that support findings of this study are available on request from the corresponding author.

Declarations

Competing interests

The authors declare no competing interests.