Resting-State Brain Oscillations and Working Memory: The Role of EEG Coherence in Healthy Middle-Aged Individuals

Abstract

This study investigated whether resting-state EEG coherence in the alpha, beta, and theta frequency bands predicts working memory performance in healthy middle-aged adults (N = 27, aged 49–64). Unlike prior research focusing on young adults or clinical populations, we examined the relationship between EEG coherence during eyes-open rest and performance on a range of working memory tasks, including updating (n-back task), switching, Stroop, and complex operation span (OSPAN task). Hierarchical regression analyses revealed that demographic variables (age, education) were generally not significant predictors, except for education in the updating task. Inclusion of EEG coherence significantly increased explained variance: alpha, beta, and theta coherence predicted performance in the updating task, while alpha and beta coherence predicted outcomes in the OSPAN task. Specifically, higher alpha coherence was associated with better performance, whereas lower theta and beta coherence predicted superior outcomes, suggesting enhanced neural flexibility and efficient cognitive resource allocation. EEG coherence did not significantly predict performance in the switching or Stroop tasks, likely because these tasks rely more on rapid reactive responses and local neural activity not captured by resting-state synchronization. These findings indicate that resting-state EEG coherence may serve as a frequency-specific neurophysiological marker of working memory in middle age. Future research should explore longitudinal changes and potential interventions, such as neurofeedback, to modulate coherence and enhance cognitive function.

1. Introduction

Contemporary research emphasizes working memory (WM), defined by A. Baddeley (1992) as a central system that temporarily stores and actively manipulates information needed for tasks such as language comprehension, learning, reasoning, and decision-making. Unlike short-term memory, WM involves both holding and processing information, making it essential for everyday cognitive functioning. Probably the most prominent of WM models, i.e., A. D. Baddeley and Hitch (1974) model, suggest its tripartite nature. Two of its components, the phonological loop and the visuospatial sketchpad, most closely resemble traditional short-term storage systems. The phonological loop temporarily stores and rehearses speech-based information and plays a central role in language acquisition and everyday verbal processing. The visuospatial sketchpad maintains and manipulates visual–spatial information and supports tasks such as mental imagery, navigation, and object location. Coordinating these storage systems is the central executive, an attentional control system that allocates cognitive resources and supports complex tasks, such as planning and problem-solving; it is also vulnerable to neurological disorders like Alzheimer’s disease (A. Baddeley, 1992, 2010; A. Baddeley et al., 2021). Other influential accounts also highlight that WM comprises multiple executive components. In particular, the framework proposed by Miyake et al. (2000) conceptualizes executive aspects of WM through three core processes: updating, inhibition, and shifting (Banich, 2009). Updating involves replacing old information with new and is often measured with n-back tasks. Inhibition refers to suppressing automatic responses, as in the Stroop task. Shifting, or cognitive flexibility, supports adaptation to changing task demands and is typically assessed with tasks requiring alternation between rules. In addition to these executive components, overall WM capacity is commonly measured with the operation span task (OSPAN), which combines remembering letter sequences with solving simple math problems (Turner & Engle, 1989).

EEG Coherence and Its Role in Cognitive Processing

Neuroimaging research has further clarified the roles of regions such as the prefrontal and parietal cortices in supporting WM (Lara & Wallis, 2015). Cortical connectivity between brain regions can be assessed using EEG coherence. Coherence quantifies the synchronization between signals recorded at different scalp sites (Basharpoor et al., 2021). In frontal and parietal regions, coherence in the alpha, beta, and theta bands reflects communication between areas and is important for WM and overall cognitive functioning. High coherence indicates strong coordination, supporting specific cognitive processes such as memory, language, concept retrieval, and music processing (Vysata et al., 2014). For example, memory tasks in healthy individuals often show increased synchronization between regions actively engaged in coordinated processing, which is measurable via EEG coherence (Vysata et al., 2014). Investigating functional connections between brain regions, rather than focusing solely on localized power, is important given the cortex’s structure. The cortex supports extensive signal diffusion and integration (Braitenberg & Schüz, 1991). Most pyramidal cells project to distant regions within or across hemispheres, and their axons branch to transmit signals within a ~3 mm radius. Consequently, any cortical neuron can connect to nearly any other within two or three synapses, allowing efficient signal convergence and divergence. Thus, measuring interactions between regions provides richer information than analyzing isolated activity (Chen et al., 2025; Gasser et al., 2003; Gevins et al., 1995; Jaušovec & Jaušovec, 2012).

Alpha waves occur during relaxed yet alert states and are linked to better learning, memory, concentration, and stress reduction (Hima et al., 2020). Abnormal alpha coherence is associated with neurological deficits, including stroke, brain tumors, and schizophrenia (Dubovik et al., 2012; Martino et al., 2011; Hinkley et al., 2011). Beta waves dominate wakefulness and support logical thinking, problem-solving, attention, and information retention (Hima et al., 2020). Low-frequency beta (12–15 Hz, sensorimotor rhythm) aids proprioception, muscle control, emotional stability, and focus. Mid-frequency beta (15–20 Hz) supports externally oriented attention and executive functions, with excessive interhemispheric beta coherence potentially impairing cognitive flexibility (Basharpoor et al., 2021). Theta waves are linked to learning and memory, active in relaxed states that enhance information retention. Elevated theta activity is common in creative individuals, supporting problem-solving and intuitive performance (Hima et al., 2020; Koudelková et al., 2018). Increased long-range theta coherence between frontal and posterior regions correlates with higher WM demands and successful episodic encoding, particularly during tasks requiring information manipulation rather than simple recall (Nyhus & Curran, 2010; Sauseng et al., 2005).

Population Aging and Changes in EEG Patterns

Population aging is a global phenomenon impacting all aspects of society. Advances in medical technology, therapies, and pharmaceuticals have steadily increased life expectancy by about three months per year since 1840, suggesting many children born in developed countries in 2000 may reach 100 (Oeppen & Vaupel, 2002). Lifestyle changes—better nutrition, more physical activity, and reduced harmful habits like smoking—further reduce disease burden (Mathers & Loncar, 2006). Growing awareness of mental health, personal development, and preventive care also supports longer, higher-quality lives. Socioeconomic factors, including improved education, working conditions, and social security, enhance quality of life in old age. Each additional year of education reduces mortality risk by 2%, with 18 years lowering it by 34% (IHME-CHAIN Collaborators, 2024). As a result, a rapidly growing share of the global population now comprises older adults. While longer life expectancy offers many benefits, it also raises the risk of cognitive diseases such as dementia and age-related cognitive decline. Aging increases vulnerability to changes in brain function that can affect personality, social interactions, and daily functioning. Identifying strategies to mitigate cognitive decline and delay the onset of deficits is therefore essential (Jabès et al., 2021).

Normal aging, without cognitive impairment, is linked to a general decline in WM efficiency. Maintaining it is crucial for older adults’ independence, decision-making, and social functioning. Research on healthy adults shows that memory-related processes involve increased synchronization between coordinating brain regions. Age-related declines in interhemispheric coherence at rest are thought to reflect progressive brain atrophy and reduced synaptic connectivity. Terry and Katzman (2001) argued, in a theoretical extrapolation, that if humans were to live to approximately 130 years, the cumulative loss of synapses due to normal aging alone could result in cognitive performance resembling that observed in dementia, even in the absence of Alzheimer’s disease pathology. Age-related neuronal changes, including shorter and fewer dendrites, further reduce synaptic density, leading to decreased hemispheric synchronization and potential impacts on cognitive function (Vysata et al., 2014).

In the theta and alpha frequency bands, coherence clearly declines with age, reflecting reduced synchronization between brain regions in these ranges. In contrast, beta band coherence increases, possibly indicating compensatory mechanisms or changes in information processing (Vysata et al., 2014). This increase does not imply global synchronization like in alpha and theta waves but rather stronger coherence within a smaller, active network of regions. The decline in interhemispheric coherence, particularly in the theta and alpha bands, is partly due to reduced cortical connectivity. Aging-related decreases in connectivity stem from neural degeneration, fewer synapses, and weakened networks. This reflects physiological changes in the maturation and specialization of brain systems, as aging makes neural networks less flexible and coordinated activity—especially in theta and alpha bands critical for memory, learning, and information integration—more difficult (Vysata et al., 2014).

Age-related changes in EEG coherence are most pronounced in the alpha and theta bands and least evident in beta. Lower-frequency waves (alpha, theta) involve large, highly coordinated networks that typically decline with age, whereas higher-frequency beta activity starts with lower initial coherence but may increase across the lifespan, possibly reflecting compensatory mechanisms. Beyond aging, coherence has been linked to cognitive performance. For example, Thatcher et al. (2005) found that EEG coherence, as an index of functional connectivity, predicts IQ better than power measures. Other studies, however, indicate that reduced coherence can also correlate with higher IQ, suggesting more differentiated and efficient neural processing (Silberstein et al., 2004). Studies also highlight that different frequency bands support distinct cognitive demands. Okuhata et al. (2009) found that although no band is specific to a single task, theta coherence is associated with dynamic integration, attention, and WM; alpha coherence reflects stimulus modality; and beta coherence relates to verbal processing and simple tasks, often showing reduced frontal-temporal connectivity during sequential processing. Task-specific patterns have been further demonstrated by Jaušovec and Jaušovec (2000). Essay writing elicited the highest lower-alpha coherence, particularly in the left hemisphere, consistent with the complex language, creativity, and planning processes it requires. Reading and figurative-reasoning tasks produced lower coherence and less interregional coordination, while divergent-thinking tasks showed intermediate patterns with smaller regional differences. In the upper-alpha band, task differences were even more pronounced: essay writing showed strong inter- and intrahemispheric cooperation, mainly in the left hemisphere; reading showed clear separation among frontal, temporal, parietal, and occipital regions; while figurative tasks followed a similar pattern with somewhat greater frontal cooperation. Divergent thinking tasks displayed complex coherence differences, including right hemisphere separations. Finally, Basharpoor et al. (2021) found that higher interhemispheric theta coherence predicts poorer executive functions, and Merrin et al. (1989) reported increased theta coherence in schizophrenia patients, consistent with impairments in executive function.

Despite the evidence of age-related changes in connectivity emerging well before older age, few studies have examined this relation in healthy middle-aged adults (e.g., Javaid et al., 2022; Kober et al., 2016). These reports show age-moderated alterations in EEG coherence during various memory tasks, suggesting that changes in neural connectivity emerge in middle adulthood rather than being solely a consequence of late-life decline. Given the limited and often conflicting research—most of which has focused on younger or clinical populations—this study investigates how EEG coherence in the alpha, beta, and theta bands predicts performance on WM tasks in healthy middle-aged adults. Following the framework outlined by Fleck et al. (2017), participants aged 49–64 were selected because this period represents a critical preclinical window for detecting early cognitive and neural changes associated with aging (Sperling et al., 2011). By focusing on the understudied age group, we aim to clarify how neural connectivity is related to cognitive performance in healthy aging. Furthermore, resting-state brain activity differs between eyes-open and eyes-closed conditions, with eyes-closed recordings reflecting internally oriented mental processes and eyes-open recordings capturing attentional engagement and external stimulus processing (Marx et al., 2004; Miraglia et al., 2016). Based on this rationale, the present study employed an eyes-open resting-state EEG paradigm to better characterize functional brain organization relevant to cognitive performance and functional connectivity in middle-aged adults.

Thus, the aim of this study was to investigate whether coherence in the alpha, beta and theta frequency bands is associated with performance across several WM tasks. We hypothesize that EEG coherence will significantly predict WM performance. More precisely, higher alpha band coherence is expected to relate to better performance on shifting, updating (NB), inhibition (Stroop), and working-memory capacity (OSPAN) tasks, while lower beta and theta band coherence is also expected to predict better performance, controlling for age and education. Beyond advancing understanding of these associations, evidence from aging research suggests that EEG coherence patterns are linked with broader cognitive function across the lifespan and may aid early identification of age-related cognitive changes (Lejko et al., 2020; Warren et al., 2025); if robustly replicated, resting-state coherence could therefore have practical utility as a non-invasive electrophysiological marker to inform cognitive training and early intervention strategies for individuals at risk of decline.

2. Materials and Methods

2.1. Participants and Procedure

The study included a convenience sample of 27 participants, with 8 men (29.6%) and 19 women (70.4%). Participants’ ages ranged from 49 to 64, with a mean age of 54.3 (SD = 4.11). Based on the highest level of education completed, the sample was as follows: 1 participant (3.7%) completed only elementary school, 10 (37%) completed high school, 1 (3.7%) completed BA studies, 13 (48.1%) completed MA studies, and 2 (7.4%) completed postgraduate studies or a doctoral degree. This study is part of a larger research project aimed at examining the effectiveness of cognitive training on the cognitive abilities of older adults.

Before the beginning of the study, participants signed an informed consent form explaining the purpose of the research, guaranteeing anonymity, and outlining the possibility of withdrawing without consequences, as well as access to results after completion. Participants then filled in the sociodemographic questionnaire, providing information on age, sex, education level, and current employment status. EEG data were collected using the Mobita 32-channel wireless EEG system (Biopac Systems Inc., Goleta, CA, USA), with electrodes placed according to the extended international 10/20 system: Fp1, Fpz, Fp2, F7, F3, Fz, F4, F8, FC5, FC1, FC2, FC6, T7, C3, Cz, C4, T8, TP9, CP5, CP1, CP2, CP6, TP10, P7, P3, Pz, P4, P8, PO, O1, Oz, and O2. Participants were fitted with an EEG cap, and electrodes were connected before receiving task instructions. The session began with 2 min of relaxed sitting with eyes closed, followed by 2 min of relaxed sitting with eyes open. After this baseline period, participants proceeded to complete the experimental tasks.

2.2. Tasks

N-Back Task (NB; Jaeggi et al., 2011) is a computerized task designed to assess the ability to update information in WM. The task was administered via the Unity platform. Participants were asked to press the spacebar on the computer keyboard whenever the stimulus displayed on the screen matched the one presented n steps back.

The task consists of six blocks, preceded by a practice session, with difficulty levels gradually increasing (1-back, 2-back, and 3-back). At the beginning of each block, the number of steps (n) is indicated. Each block includes 20 + n stimuli (1-back = 21, 2-back = 22, 3-back = 23 stimuli). Each stimulus is presented on the screen for 500 ms, followed by a 2500 ms pause. While the original version uses abstract stimuli, in this version participants were presented with eight different letters.

The final task score is calculated as the mean proportion of correct responses across the three levels. For each level, the proportion of correct responses is computed, and the three values are averaged to produce an overall n-back score. This provides a comprehensive measure of the ability to update information in WM. Higher scores indicate better capacity to track, retain, and update information, while lower scores reflect weaker updating abilities or a higher number of errors in the comparison and decision-making process.

Switching Task. In this task, words were presented in the center of the screen, and the task is to classify each word according to one of two criteria. The first criterion involved deciding whether the presented word represented a living or non-living entity, while the second criterion concerned its physical size, determining whether the object denoted by the word was larger or smaller than a soccer ball. Above each word, a symbol appeared, indicating the classification criterion: a heart for the living/non-living category and double arrows for the larger/smaller category. Participants were instructed to classify the words as quickly and accurately as possible according to the current criterion.

The task was administered using the OpenSesame software V3.3 and consisted of several phases. First, participants completed two separate blocks, one for each category, with each block containing 25 test stimuli and seven practice stimuli. Next, a mixed block followed, in which the classification criteria alternated from trial to trial depending on the symbol above the word. This block included a total of 64 stimuli, of which 16 were practice trials and excluded from the final analysis. In half of the remaining trials, the criterion changed from the previous trial (switch trials), while in the other half it remained the same (repeat trials).

The final outcome of the switching task is expressed as the switch-cost, which measures the difference in average reaction times (RT) between the two types of trials. It is calculated as the mean difference in RT for correctly answered trials where the criterion changed versus those where it stayed the same. In other words, it quantifies how much additional time participants needed to respond correctly when switching to a new criterion compared to continuing with the same one. Higher switch-cost values indicate greater difficulty in switching and thus lower cognitive flexibility, whereas lower values suggest better adaptation to new tasks and higher flexibility in adjusting to rule changes.

Stroop Task. The Stroop task is a classic measure of cognitive control and selective attention. In this task, participants were presented with words representing color names (red, yellow, blue, and green) displayed in different font colors. Additionally, neutral stimuli, such as a string of symbols (“*****”), were included and also presented in different colors. In some trials, the font color matched the meaning of the word (congruent stimuli), while in others it did not (incongruent stimuli). Participants’ task was to indicate as quickly and accurately as possible the color in which the word or neutral stimulus was written by pressing the corresponding key on the keyboard. The focus was on ignoring the word’s meaning and attending only to the text color.

Before the main task, participants completed a practice block to familiarize themselves with the procedure. This was followed by three experimental blocks with different proportions of congruent stimuli: 25% in the first block, 50% in the second, and 75% in the third. Neutral stimuli were evenly distributed within each block. Reaction time was not limited, allowing participants to respond at their own pace, but emphasis was placed on speed and accuracy.

The final outcome was calculated as a measure of inhibitory control, specifically the difference in average RTs between correctly answered incongruent and congruent stimuli. Lower scores indicate better inhibitory ability.

Operation Span Task (OSPAN; Turner & Engle, 1989; Unsworth et al., 2005) is one of the most commonly used tasks for assessing WM capacity. Participants were required to simultaneously solve simple mathematical problems and remember a sequence of letters. The task was conducted on a computer, and before the main task, participants completed several practice sessions: first for letter memorization, then for solving mathematical problems, and finally for the combination of both tasks.

During the task, letters were presented one at a time for 1000 ms each. After each letter, participants solved a simple math problem consisting of two operations and a proposed solution, responding by selecting “correct” or “incorrect” on the screen. After each sequence of letters, participants selected the letters previously shown from a table in the exact order they appeared. Sequence lengths varied from two to seven letters, with different difficulty levels presented in random order.

The total score was calculated as an absolute score, i.e., the sum of points for sequences in which all letters were correctly recalled in the presented order. The maximum possible score was 28, with higher scores indicating greater WM capacity. As a validity criterion, participants had to achieve at least 85% accuracy on the math problems. If a participant did not reach the minimum 85% accuracy, their total score was excluded from analysis. This ensures that participants did not neglect the processing component of the task and focus only on memorizing letters. Previous research (Unsworth et al., 2005) has demonstrated high reliability of this task, with Cronbach’s α = 0.78.

3. Results

3.1. EEG Data Processing and Behavioral Data Analysis

EEG data were preprocessed and analyzed using MATLAB (version R2020b) with the specialized FieldTrip toolbox (Oostenveld et al., 2011). In the initial preprocessing stage, a high-pass filter at 1 Hz and a low-pass filter at 40 Hz were applied to remove low-frequency noise and high-frequency signals. EEG signals were then re-referenced to the average of all electrodes to reduce the impact of local signal differences and improve comparability across scalp regions.

Data were segmented into non-overlapping two-second epochs, allowing the analysis of stable patterns of brain activity. Artifact rejection was performed in multiple steps. First, epochs containing gross artifacts (e.g., high-amplitude noise, movement-related disturbances, or electrode malfunctions) were identified through visual inspection and excluded from further analysis. Second, independent component analysis (ICA) was applied to the epoched data. Independent components were visually inspected based on their time courses, scalp topographies, and power spectra, and components clearly associated with non-neural artifacts (e.g., eye blinks, eye movements, or cardiac activity) were removed. Following ICA correction, the data were subjected to a final visual inspection to ensure that residual artifacts were minimized.

Next, spectral power of the EEG signals was computed using a multitaper fast Fourier transform (FFT) with a frequency resolution of 0.5 Hz. To account for individual differences, such as skull thickness or scalp conductivity, relative power was calculated—the ratio of power within a specific frequency band to the total power in the 1–30 Hz range. This standardization allows for meaningful comparisons across participants.

The central measure in this study was EEG coherence, which reflects the functional connectivity between two brain regions. Specifically, coherence quantifies the synchronization of EEG signals between two electrodes within a frequency band. High coherence indicates that two brain regions oscillate in similar rhythms (i.e., similar phase and frequency), suggesting stronger functional connectivity (Vysata et al., 2014). Coherence was calculated from the Fourier-transformed signals by analyzing the phase and frequency alignment between electrodes. Values range from 0 (no connectivity) to 1 (perfect synchronization). In this study, coherence was analyzed within the theta (4–8 Hz), alpha (8–13 Hz), and beta (13–30 Hz) bands.

Finally, performance on cognitive tasks—the n-back (NB) task, switching task, Stroop task, and Operation Span (OSPAN) task—was analyzed. Descriptive statistics, including means, standard deviations, and correlations among variables, were computed. Statistical analyses were performed using Jamovi (version 2.6).

3.2. Correlation Matrix and Descriptive Statistics

Table 1. Correlation matrix and descriptive statistics for the variables used in the model (N = 27).

	M	SD	1	2	3	4	5	6	7	8	9
1 Age	54.3	4.11	-
2 Education *	4.19	1.14	0.405 *	-
3 N-back	0.852	0.06	0.235	0.588 **	-
4 Stroop	192	165	−0.241	−0.084	−0.142	-
5 Switching	197	299	−0.083	−0.052	−0.260	0.048	-
6 OSPAN	32.3	10.8	0.086	0.150	0.484 *	−0.413 *	−0.126	-
7 Coht	0.530	0.133	0.074	−0.06	−0.285	0.302	−0.296	−0.193	-
8 Coha	0.482	0.112	0.128	0.332	0.327	−0.007	−0.015	−0.01	0.396 *	-
9 Cohb	0.388	0.0804	−0.115	0.169	−0.071	0.184	0.368	−0.493 *	0.001	0.501 *	-

Legend: coht—theta band coherence; coha—alpha band coherence; cohb—beta band coherence. * Highest completed education level (1—primary school; 2—secondary school; 3—vocational school; 4—university degree; 5—master’s degree or doctorate); * p < 0.05, ** p < 0.01.

shows the correlation matrix and descriptive statistics for the variables used in the regression analyses.

3.3. Hierarchical Regression Analyses

To address the main research question, a series of hierarchical regression analyses were conducted to examine whether coherence in the alpha, beta, and theta bands predicted performance on working memory tasks, above and beyond the variance explained by age and education. Hierarchical regression was selected because it allows for the sequential entry of predictors, enabling an explicit test of the incremental contribution of EEG coherence measures after accounting for well-established demographic covariates. Age and education were entered in the first step as control variables due to their known associations with cognitive performance, while coherence measures were entered in subsequent steps to assess their unique predictive value (

Table 2. Results of the hierarchical regression analyses, with age and education entered in the first block and EEG coherence measures entered in the second block as predictors, and performance on various working memory tasks as the criterion variable.

	n-back		OSPAN		Stroop		Switching
	Block1	Block 2	Block1	Block 2	Block1	Block 2	Block1	Block 2
	β	β	β	β	β	β	β	β
Age	−0.015	−0.054	−0.036	−0.169	−0.225	−0.21	−0.208	−0.09
Education	0.562 **	0.417 *	0.109	0.1	0.001	0.086	0.018	−0.089
coh_alfa		0.626 **		0.513 *		−0.353		0.075
coh_beta		−0.461 *		−0.793 **		0.321		0.41
coh_theta		−0.504 **		−0.358		0.463		−0.265
R2	0.309 *	0.602 **	0.01	0.463 *	0.05	0.241	0.041	0.254
ΔR2		0.293 *		0.453 *		0.191		0.214

Legend: coh_alpha—coherence in the alpha band; coh_beta—coherence in the beta band; coh_theta—coherence in the theta band; * p < 0.05, ** p < 0.01.

).

For all analyses, residual normality was assessed using the Shapiro–Wilk test, which showed no deviations from normality, satisfying the assumption for regression analysis. Visual inspection of residual plots confirmed the assumption of linearity, and VIF coefficients (VIF < 10, Tolerance > 0.1) indicated no multicollinearity among predictors. As shown in Table 2., for the n-back task, demographic variables accounted for 30.9% of the variance in the first step (R² = 0.309; F(2,22) = 4.93, p = 0.017), with education emerging as a significant predictor; higher education was associated with better performance, whereas age was not significant. In the second step, adding EEG coherence variables explained an additional 29.3% of variance (ΔR² = 0.293; F(3,19) = 4.67, p = 0.013), increasing the total explained variance to 60.2% (R² = 0.602; F(5,19) = 5.76, p = 0.002). After controlling for demographics, all three EEG coherence variables were significant. Alpha coherence positively predicted updating performance, whereas beta and theta coherence were negatively associated, indicating that lower beta and theta coherence corresponded to better task performance.

For the Operation Span task, demographic variables had negligible contribution to the total variance in the first step (1%; F(2,21) = 0.106, p = 0.9), whereas adding EEG coherence explained an additional 45.3% of the variance in the second step (ΔR² = 0.453; F(3,18) = 5.05, p = 0.01), resulting in a total of 46.3% variance explained (R² = 0.463; F(5,18) = 3.1, p = 0.034). Here, alpha and beta coherence emerged as significant predictors. Alpha coherence was positively associated with task performance, while beta coherence was negatively associated. This indicates that participants with higher alpha coherence and lower beta coherence achieved better OSPAN results.

In the regression models for the Stroop and Switching tasks, the total variance of the criterion was not statistically significant (Stroop: F(5,19) = 1.209, p = 0.343; Switching: F(5,19) = 1.296, p = 0.307), indicating that neither demographic variables nor EEG coherence significantly contributed to prediction of task performance, and the overall models were non-significant.

4. Discussion

The present study investigated whether resting-state EEG coherence in the alpha, beta, and theta frequency bands predicts WM performance in middle-aged adults. Unlike prior research, which has typically examined EEG activity during task performance in younger adults (Jaušovec & Jaušovec, 2000), we focused on eyes-open resting-state activity and its relationship to later WM outcomes in healthy older adults. Our sample of healthy adults, aged 49–64, allowed us to explore neural predictors in a population for whom early cognitive aging processes are particularly relevant. Age-related changes in brain oscillations and decline in WM are well documented, with reductions in alpha and theta power and coherence associated with broader cognitive decline (Klimesch, 1999; Hedden & Gabrieli, 2004), making this age group especially important for understanding these patterns. Furthermore, we used the eyes-open resting state condition; compared to the eyes-closed condition, this state shows higher global network efficiency, suggesting better information transfer and greater readiness to process information from the external environment. In addition, spontaneous brain activity may be higher when the eyes are open, possibly due to increased alertness and mind-wandering (Y. Wang et al., 2022).

Hierarchical regression analyses revealed that demographic variables explained 30.9% of the variance in n-back performance, with education emerging as the only significant predictor, reflecting the well-established role of cognitive reserve in supporting executive functions (Boller et al., 2017; Stern, 2009). Age was not significant, likely due to the relatively narrow age range. Adding EEG coherence increased explained variance to 60.2%, with all three frequency bands contributing significantly. Alpha coherence positively predicted the n-back performance, consistent with its role in inhibiting irrelevant information and supporting efficient WM updating (Klimesch, 1999). The n-back task places strong demands on continuous updating and suppression of no-longer-relevant representations. In this context, higher alpha coherence may reflect more efficient top-down inhibitory control, allowing participants to maintain task-relevant information while minimizing interference from previous stimuli. Conversely, theta and beta coherence negatively predicted performance, suggesting that elevated theta may reflect increased cognitive load (Cavanagh & Frank, 2014) or might indicate a compensatory or less efficient control state, which becomes detrimental when rapid updating is required, while higher beta coherence may indicate neural rigidity that limits cognitive flexibility (Hanslmayr et al., 2016).

For the OSPAN task, demographic variables did not account for significant variance, whereas EEG coherence contributed 46.3%, with alpha and beta bands emerging as key predictors. Greater alpha coherence was associated with better performance, likely reflecting enhanced attentional control and efficient allocation of WM resources (Jensen & Mazaheri, 2010; Klimesch, 2012). Higher beta coherence predicted lower performance, possibly indicating cognitive overload that impairs flexibility and task-switching (Engel & Fries, 2010). Theta coherence was not significant, suggesting its role may be task-specific or less critical in dual-processing WM tasks. Theta-related control processes appear to be more strongly engaged during dynamic updating than during the stable maintenance of representations.

In contrast, neither EEG coherence nor demographic variables significantly predicted performance on the Stroop or switching tasks, indicating that resting-state synchronization may be less relevant for tasks emphasizing response inhibition rather than continuous WM updating. Stroop and task-switching tasks primarily engage discrete, short-lived control processes that are recruited on a stimulus-by-stimulus basis, rather than requiring the continuous maintenance and updating of task-relevant representations across trials, as in n-back and OSPAN tasks. In this middle-aged sample with largely preserved executive functioning, such demands may be relatively modest, rendering these tasks less sensitive to baseline differences in functional connectivity. These findings suggest that neural synchronization supports certain executive functions, particularly updating and switching, but may not uniformly predict all cognitive control processes.

Overall, these results highlight the potential of resting-state EEG coherence as a biomarker of WM efficiency in middle-aged adults. Higher alpha and lower beta coherence appear to support more effective cognitive performance, emphasizing the importance of functional neural connectivity in maintaining WM. Future longitudinal research is needed to examine changes in resting-state coherence across aging and their implications for cognitive decline. The field would benefit from future studies incorporating methodological advancements in design, techniques and measures. For example, randomized controlled cognitive training with pre-, post-, and follow-up EEG assessments would serve in investigating whether training-induced changes in EEG coherence mediate improvements in WM and whether these effects differ across younger, middle-aged, and older adult cohorts. In addition, multimodal neuroimaging that combines EEG with fMRI or MEG during WM tasks could overcome the spatial limitations of EEG and clarify how coherence patterns map onto specific brain networks. Finally, exploring advanced connectivity estimation methods (e.g., coherence potentials or multivariate causal estimators) may increase sensitivity to dynamic functional interactions and strengthen links between connectivity and behavioral outcomes (Puthanmadam Subramaniyam & Thiagarajan, 2025).

Limitations and Future Directions

Despite promising findings, several limitations to this study should be acknowledged. The sample was small (N = 27) and convenience-based, with an unbalanced gender distribution (70% women), as well as educational attainment (with approximately 50% of participants holding a master’s degree or higher), which limits statistical power and generalizability. In particular, small samples may increase the risk of overestimating effect sizes and reduce the stability of regression estimates. Although EEG is non-invasive and offers excellent temporal resolution, it is susceptible to artifacts from muscle activity, eye movements, and environmental noise, and scalp-recorded signals cannot fully isolate activity from specific cortical regions. Additionally, data collection was conducted by multiple experimenters across different testing rooms, which—despite standardized procedures—may have introduced unwanted variability. Future research should aim to increase sample size, improve participant representativeness, and further standardize experimental conditions to enhance reliability and external validity.

Clinicians and cognitive trainers might leverage EEG coherence information in several ways—for example, as part of baseline assessment, to monitor intervention or training progress, or in combination with neurofeedback protocols. EEG coherence has already been used in WM training studies to characterize functional connectivity changes associated with performance (see Lin et al., 2024; Nouchi et al., 2021). EEG coherence measures could help identify individuals who are more likely to benefit from specific WM training paradigms, thereby supporting personalized intervention planning. Tracking changes in coherence over time could serve as an objective metric to monitor response to cognitive training and adjust the intensity or focus of the intervention if progress is limited (Laptinskaya et al., 2020). Neurofeedback interventions present a promising avenue for exploring causal links between EEG coherence and WM performance. By training individuals to modulate alpha, beta, and theta coherence, it may be possible to enhance cognitive outcomes. Combining EEG-based neurofeedback with electrophysiological recordings could provide deeper insights into the neural mechanisms underlying working memory and inform targeted interventions for middle-aged and older adults (e.g., J. R. Wang & Hsieh, 2013; Jiang et al., 2022). In particular, well-powered longitudinal studies that incorporate neurofeedback training would be critical for clarifying the long-term effects of modulating resting-state EEG coherence on working memory performance and broader trajectories of cognitive aging.

5. Conclusions

This study examined whether resting-state EEG coherence in the alpha, beta, and theta frequency bands predicts WM performance in healthy middle-aged adults. Unlike prior research focusing on young adults or clinical populations, we explored the link between resting-state coherence and cognitive performance in a sample aged 49–64. Hierarchical regression analyses revealed that including EEG coherence significantly increased the explained variance. Specifically, alpha, beta, and theta coherence predicted performance in the updating task, while alpha and beta coherence predicted outcomes in the complex operation span task. In contrast, resting-state EEG coherence did not significantly predict performance in the switching or Stroop tasks, likely because these tasks rely more on rapid reactive responses and local neural activity not captured by resting-state synchronization.

These findings suggest that resting-state EEG coherence may serve as a neurophysiological marker of WM, with frequency-specific contributions: alpha coherence supports efficient updating, while lower theta and beta coherence may reflect greater cognitive flexibility and attentional regulation. Importantly, these results extend previous research by demonstrating these effects in middle-aged adults, a group that has been underrepresented in EEG coherence research on cognition but is critically situated in the trajectory of cognitive aging. This aligns well with growing evidence that resting EEG coherence and other functional connectivity indices can reflect underlying neural health and may differentiate individuals with varying cognitive profiles prior to overt clinical impairments, suggesting their utility as early indicators of age-related cognitive changes.

In conclusion, resting-state EEG coherence shows promise as a predictor of cognitive performance in middle age adulthood, highlighting its potential as a biomarker for WM and a target for cognitive interventions. This perspective highlights the potential of EEG coherence to contribute to early identification of subtle cognitive decline and to the monitoring of interventions aimed at sustaining cognitive health across the lifespan. Further research in larger and longitudinal cohorts is needed to clarify its practical implications and inform strategies for promoting cognitive health across the lifespan.