Skip to main content
Dialogues in Clinical Neuroscience logoLink to Dialogues in Clinical Neuroscience
. 2026 Nov 16;29(1):1–12. doi: 10.1080/19585969.2026.2747472

Microstate-specific functional connectivity in alcohol use disorder using resting-state EEG

Hanjin Park a, So Young Yoo b, Hyunho Lee c, Minkyung Park c, Jung-Seok Choi c,✉, Dongil Chung a,*,✉
PMCID: PMC13776749  PMID: 43038569

Abstract

Introduction

Alcohol Use Disorder (AUD) is often described as a “disconnection syndrome”, reflecting neural dysfunction associated with chronic alcohol use. However, previous studies on functional connectivity of resting-state EEG (rsEEG) in AUD have reported inconsistent findings, possibly due to overlooking the brain’s dynamic nature.

Methods

We examined rsEEG in 26 young adults with AUD and 35 healthy controls (HC), using microstate analysis to characterise transient brain states and to estimate functional connectivity within each state.

Results

Although microstate map topologies were comparable between the two groups, the temporal dynamics of the AUD group were biased towards microstates C and D. Particularly within microstate C, the AUD group showed significantly greater functional connectivity across most frontal-occipital channel pairs, whereas significantly reduced functional connectivity was observed in the interhemispheric channel pairs. Moreover, compared with conventional functional connectivity estimates, microstate-specific functional connectivity yielded superior classification performance in distinguishing the two groups. In addition, the AUD group showed less stable hub-like properties in microstate C, indicating diminished network efficiency.

Conclusions

Our findings suggest that transient EEG properties, exemplified by microstate-specific functional connectivity, provide informative neural markers for characterising the impact of alcohol use on the brain.

Keywords: Alcohol use disorder, microstate, EEG, dynamic connectivity, neural marker

Introduction

Alcohol Use Disorder (AUD) is a chronic condition characterised by neural dysfunction, leading to alterations in brain structure and function (Andrews-Hanna et al. 2007; Chanraud et al. 2011; Dupuy and Chanraud 2016; Bottino et al. 2025). Chronic alcohol use, associated with impaired self-control and impulsivity behaviour, is often described as “disconnection syndrome”—a condition characterised by disrupted neural connectivity (Dupuy and Chanraud 2016; Shokri-Kojori et al. 2017; Vergara et al. 2017; Elton et al. 2021). In line with such notion, previous studies analysed synchronisation and phase locking value (PLV) of resting-state electroencephalography (EEG) in AUD and showed that their functional connectivity is impaired (de Bruin et al. 2004; Herrera-Díaz et al. 2016; Wang et al. 2025). Other neuroimaging studies corroborated these reports and suggested that disruptions in functional connectivity within the default mode network (DMN) may underlie the behavioural impairments observed in individuals with AUD (Andrews-Hanna et al. 2007; Chanraud et al. 2007; 2011; Bottino et al. 2025). Despite a large body of accumulated data and evidence, there have also been studies in AUD reporting findings that contradict these results, reporting either hyperconnectivity (Winterer et al. 2003) or no statistical differences in functional connectivity compared to healthy controls (HC) (de Bruin et al. 2004; Park et al. 2017; Sion et al. 2020). Such inconsistencies may be explained by the well-known large individual differences observed among individuals with AUD (Andrews-Hanna et al. 2007; Chanraud et al. 2011; Dupuy and Chanraud 2016; Bottino et al. 2025), although alternative possibilities have not yet been explored.

Many classical EEG approaches, including those used in various AUD studies, assume stationarity in EEG signals, meaning that statistical properties such as the mean and variance remain constant throughout the analysis window. In contrast to this statistical assumption, another line of research treats the EEG signal as a dynamical system that transitions among multiple transient brain states (Betzel et al. 2012; Khanna et al. 2015; Tarailis et al. 2024). Analyses of the state occurrences and transition properties have been linked to various neurodegenerative (Stevens and Kircher 1998; Nishida et al. 2013) and neuropsychiatric disorders (Strik et al. 1995; Kikuchi et al. 2011; Kim et al. 2024). Given that relationships among signals with dynamic properties may be over- or underestimated depending on their respective states (Stam 2005), previously reported inconsistencies in functional connectivity findings in AUD may stem from overlooking such dynamic state changes.

Here, we hypothesised that functional connectivity estimates that incorporate the dynamic properties of EEG signals would facilitate the identification of AUD-associated characteristics in resting-state EEG (rsEEG). Specifically, we propose that integrating microstate analysis, which classifies transient brain states with distinct scalp topographies, with the computation of functional connectivity within each state would provide a more stable tool for examining potential neural alterations in individuals with AUD. To test this possibility, we analysed rsEEGs recorded from 26 young adults with AUD and 35 HC, identified latent microstates to account for temporal dynamics in the signals, and computed functional connectivity associated with each state. Subsequently, both the temporal properties of microstates and the graph-theoretical network properties were compared between the two groups, and their contributions to classification accuracy were tested against those of conventional functional connectivity approaches.

Methods

Participants

Twenty-six young adults diagnosed with AUD (male/female = 19/7, age = 27.62 ± 5.39 years) and 35 healthy controls (male/female = 25/10, age = 24.94 ± 2.90 years) were recruited for the current study. Note that the age range of participants in the current study is 19–35, which contrasts with most previous AUD research that predominantly focuses on middle-aged populations (between 40 and 50) (Bottino et al. 2025). This age range was selected to reflect the earlier age at which individuals initiate alcohol consumption, as documented over the past two decades (Hingson et al. 2006), and to minimise potential confounding effects of comorbidity that become more common with ageing (Sullivan and Pfefferbaum 2019). AUD patients were recruited from the outpatient clinic at Seoul Metropolitan Government Seoul National University (SMG-SNU) Boramae Medical Centre, while HCs were recruited from local communities and universities through public announcements. AUD was diagnosed by a clinically experienced psychiatrist according to DSM-5 criteria, and the individual patient’s severity was assessed using the Alcohol Use Disorders Identification Test (AUDIT) (Saunders et al. 1993; Kim et al. 2008). HCs consumed no more than 14 standard drinks per week and no more than four standard drinks on any single occasion. HC participants had no prior history of AUD or psychiatric disorders. None of the participants had a history of intellectual disability, psychotic, neurological, or seizure disorders. Furthermore, all participants were right-handed, medication-naïve, and had an intelligence quotient (IQ) above 70.

Ethics

The study protocol was approved by the Institutional Review Board of the Seoul Metropolitan Government Seoul National University (SMG-SNU) Boramae Medical Centre, Seoul, Republic of Korea (IRB number: 16-2014-139), and all procedures were conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent before participation.

Clinical measures and psychological assessments

Demographic information, including age, sex, and years of education, was collected from all participants. The AUDIT (Saunders et al. 1993; Kim et al. 2008), a 10-item questionnaire that assesses hazardous drinking behaviour, was administered to evaluate the severity of alcohol consumption. Each item was rated on a four-point Likert scale (0 = never, 4 = very frequent), and total AUDIT scores range from 0 to 40. Given the strong association between AUD and altered psychosocial status, we administered additional psychological questionnaires and used their scores as measures of individual differences. Specifically, individuals’ dispositional sensitivity to rewards and punishments was assessed using the Behavioural Activation Scale (BAS) and Behavioural Inhibition Scale (BIS) (Carver and White 1994), which consist of 13 and 7 items rated on a four-point Likert scale, respectively. In addition, individuals’ stress levels and mental well-being were measured using the Psychosocial Well-being Index (PWI), a 45-item self-report questionnaire designed to evaluate physical and psychological well-being over the past several weeks (Kim 1999). The total BAS score ranges from 13 to 52, the total BIS score ranges from 7 to 28, and the PWI score ranges from 0 to 135.

EEG acquisition

The EEG data were digitised at a sampling frequency of 1000 Hz and amplified using SynAmps 2 (Compumedics, Abbotsford, Victoria, Australia) and the Neuroscan system (Scan 4.5, Compumedics). EEG signals were recorded with a 64-channel Quick-Cap (Neuroscan, Compumedics), which follows the modified 10–20 system for electrode placement. The recording took place in a sound-shielded room, and participants were instructed to keep their eyes closed for 5 min. The ground electrode was positioned between FPz and Fz, and a bipolar reference electrode was placed at the mastoid. Vertical and horizontal electrooculograms (EOGs) were recorded using electrodes positioned above and below the left eye to monitor ocular movements. The impedance of all electrodes was maintained below 5 kΩ throughout the recording. The collected EEG data were bandpass filtered between 0.1–100 Hz.

EEG preprocessing

Raw EEG data were preprocessed using MNE-Python (Gramfort et al. 2013). For analysis, 19 channels (FP1, FP2, F3, F7, FZ, F4, F8, C3, CZ, C4, P3, PZ, P4, O1, O2, T7, T8, P7, P8) were selected from the 64-channel recordings to improve computational efficiency and maximise global explained variance (GEV) in the subsequent microstate analysis (see below). To minimise potential edge artefacts, the first and last 30 s of the 5-minute recordings were excluded. Preprocessing steps included bandpass filtering (0.5–51 Hz), re-referencing to the common average, segmenting the data into 1-second epochs, and rejecting epochs with amplitudes exceeding 100 μV. Independent component analysis (ICA) was performed to remove artefacts; however, upon visual inspection, no clear noise components were identified. Finally, an alpha-band (8–13 Hz) bandpass filter was applied, as this frequency range exhibited the highest GEV.

Microstate analysis

To estimate microstates and their transitions, we conducted both individual- and group-level analyses. At the individual level, we first computed the global field potential (GFP) from the preprocessed EEG data and identified its local maxima. A modified k-means clustering algorithm (k = 4) was then applied to these GFP peaks, with 50 runs of cross-validation. Note that the number of microstate classes was fixed at four, consistent with the most commonly reported microstate patterns in the literature (Khanna et al. 2015; Tarailis et al. 2024). This procedure yielded four representative microstate clusters for each participant. At the group level, the individual-level microstate clusters were used as input to a modified k-means clustering algorithm to derive group-level microstate maps for the HC and AUD groups separately. With visual inspection, we confirmed that the microstate maps estimated for both groups corresponded well to the canonical microstates A, B, C, and D, i.e. the four prominent microstate classes typically reported in healthy individuals (Khanna et al. 2015; Tarailis et al. 2024). Subsequently, we computed cosine similarity to examine topological similarities between the two groups.

Microstate topographies can be reliably identified even with a relatively small number of electrodes (Khanna et al. 2015). To select a channel configuration that balanced computational efficiency and maximised GEV, we compared mean GEV values across four microstates using 64-, 32-, and 19-channel data, corresponding to three standard electrode montages spanning the standard and extended 10–20 systems. Mean GEV values in the AUD group were 0.56 ± 0.02, 0.58 ± 0.02, and 0.63 ± 0.01 for the 64, 32, and 19 channel configurations, respectively. Because only the 19-channel configuration achieved a mean GEV exceeding 0.60 (Tarailis et al. 2024), while also providing the greatest computational efficiency, the 19-channel data were used for all subsequent analyses.

The template microstate maps (A-D) were then back-fitted to each participant’s preprocessed EEG data using a winner-takes-all approach with 35 msec temporal smoothing (Poulsen et al. 2018). This procedure yielded individual-level microstate sequences representing the latent dynamic EEG states for each subject. To characterise these sequences, we calculated the GEV (range = 0–1), which quantifies how well each of the four template maps accounts for the original EEG data based on their spatial correlation (Khanna et al. 2015). In addition, we computed two temporal parameters: time coverage, defined as the proportion of the total recording time occupied by each microstate, and mean duration, defined as the average continuous time each microstate remained active (Khanna et al. 2015).

Microstate-specific functional connectivity

To assess microstate-specific functional connectivity, each individual’s EEG were segmented using a non-overlapping sliding time window. Specifically, we used a 230 msec time window to ensure coverage of multiple cycles of the target frequency within the signal-of-interest (here, 8–13 Hz), allowing for reliable functional connectivity estimation (Cohen 2014). Previous microstate studies have suggested that each state typically persists for approximately 80–120 msec before transitioning to the next state (Khanna et al. 2015), as reflected by mean duration measures. To preserve sensitivity to these relatively rapid temporal dynamics, we chose not to use a longer analysis window. Among the segmented time windows, we included only those in which a given microstate persisted for at least 210 msec, accounting for the 35-msec temporal smoothing—corresponding to less than half of a typical microstate duration—applied during back-fitting, to ensure stable microstate-specific functional connectivity estimation.

Subsequently, the Hilbert transform was applied to the EEG data from each selected window to obtain the instantaneous amplitude and phase. Coherence (COH) between channels x and y was defined as:

COHx,ym=|1N∑t=1NAx(t)Ay(t)ei(φxy(t))|(1N∑t=1NAx2(t))(1N∑t=1NAy2(t))

where m denotes a specific microstate, Ax(t) is the instantaneous amplitude of channel x at time t, φxy(t) is the instantaneous phase difference between channels x and y at time t, and N is the total number of time points. COH was computed between all possible channel pairs. To eliminate potential spurious COH, COH values lower than the median of all possible channel-pair COH values were set to zero (Rubinov and Sporns 2010). For each microstate-specific functional connectivity, COH values were averaged across all selected windows belonging to that microstate.

As a comparison, conventional functional connectivity was also estimated from the EEG data without distinguishing microstates. For each participant, the EEG data were segmented into 1-second epochs, and COH between all channel pairs was computed using the same procedure.

We hypothesised that examining functional connectivity within each microstate would better capture AUD-related neural characteristics than conventional connectivity measures. To evaluate the impact of distinguishing latent states in EEG functional connectivity estimation, we conducted classification analyses between the HC and AUD groups using a support vector machine (SVM) with fixed default parameters (C = 1, polynomial kernel with degree = 2, and automatic kernel scaling). Note that we neither performed hyperparameter tuning nor used a nested cross-validation framework, as the primary goal of the current study was to identify AUD-associated features in rsEEG rather than to optimise classification performance. For SVM classification, the upper-triangular elements of the functional connectivity matrices were vectorised (19 × 18/2 = 171 dimensions) and subsequently z-scored. Principal component analysis (PCA) was then applied, and the first four principal components were retained as input features to reduce the risk of overfitting. Participants’ group labels (HC or AUD) served as the output classes. For group-comparison analyses, z-scoring was performed across the entire dataset, including both HC and AUD groups. Classification accuracy (95% confidence interval) for the microstate-specific functional connectivity was evaluated using 5-fold cross-validation and was then compared with the classification accuracy obtained from the conventional functional connectivity.

In addition, sensitivity, specificity, and balanced accuracy were calculated, and permutation testing (1,000 iterations) was performed to evaluate the reliability of the classification accuracy. Specifically, we performed permutation testing by randomly shuffling the labels between the conventional and microstate C datasets, and between the conventional and microstate D datasets, followed by recalculation of the classification accuracies and their differences. This procedure was repeated 1,000 times to generate null distributions and assess the statistical significance of the observed accuracy differences.

For visualisation purposes, the first two principal component scores were used to project the data into a two-dimensional space. In addition, to assess the distinctiveness of functional connectivity patterns across four microstates, we applied PCA-clustering to microstate-specific functional connectivity features within each group.

Graph-theoretical network analyses

To assess the efficiency of information communication within each functional connectivity network, we conducted exploratory graph-theoretical network analyses using graph-theoretical measures, including network strength (STR), eigenvector centrality (EC), clustering coefficient (CC), betweenness centrality (BC), and closeness centrality (C) (see Supplementary text). Channel nodes with strength values exceeding one standard deviation above the mean across all nodes were identified as hub candidates. Candidate nodes exceeding the threshold in at least three out of the four graph-theoretical measures were identified as hubs (Supplementary text). Group comparisons of hub nodes were conducted qualitatively, such that for each microstate in each group, we assessed whether hub-like nodes were identified.

Statistical analyses

To investigate temporal dynamics across microstates within each group, we conducted repeated-measures analyses of variance (rmANOVAs) separately for each group and for each microstate parameter (i.e. GEV, time coverage, and mean duration), with microstate (A, B, C, and D) as a within-subject factor. Greenhouse-Geisser, Huynh-Feldt, and lower-bound corrections were applied to account for potential violations of the sphericity assumption. Post-hoc pairwise comparisons were conducted using Tukey’s Honestly Significant Difference (HSD) test to identify significant differences between microstates for each parameter within each group. Two-sample t-tests were conducted to examine group differences between the AUD and HC groups. Note that all between-group comparisons were controlled for individuals’ age and years of education. Multiple comparisons were corrected using the Benjamini-Hochberg false discovery rate (FDR) procedure (Benjamini and Hochberg 1995), and a corrected p-value of less than 0.05 was considered statistically significant. Pearson’s correlation coefficients were computed to investigate the relationship between psychological variables and microstate-specific functional connectivity values, controlling for individuals’ age and years of education. All statistical analysis was done using MATLAB R2021b (Mathworks).

Results

Demographics

The AUD and HC groups showed a comparable sex ratio (χ2(1) = 0.021, PFDR = 0.885), but differed significantly in age (t(59) = 2.12, PFDR = 0.019) and education (t(59) = −2.94, PFDR = 0.006) (Table S1). To control for age and education differences between the two groups, partial correlation analyses were subsequently conducted when examining group differences in microstate parameters, microstate-specific functional connectivity and psychological variables.

Consistent with the planned inclusion criteria, the AUD group showed significantly higher AUDIT scores than the HC group (t(59) = 8.63, PFDR < 0.001). Individuals with AUD are known to exhibit altered psychological characteristics, including heightened sensitivity to reward and punishment, and elevated stress levels (Andrews-Hanna et al. 2007; Chanraud et al. 2011; Dupuy and Chanraud 2016; Bottino et al. 2025). Consistent with the previous literature, the AUD group showed significantly greater sensitivity to punishment (i.e. BIS: t(59) = 2.46, PFDR = 0.017), greater sensitivity to reward (i.e. BAS: t(59) = 2.44, PFDR = 0.017), and higher PWI scores (t(59) = 4.08, PFDR < 0.001) than the HC group. Given that these measures capture psychological characteristics of AUD, we used them to examine individual variability within the AUD group.

Individuals with AUD showed altered temporal properties across distinct microstates

For each group, four distinct microstate map topologies (microstates A, B, C, and D) were identified (Figure 1). Cosine similarity analyses for each microstate indicated that the topologies were highly comparable between the two groups (A: 0.99, B: 0.97, C: 0.98, D: 0.95).

Figure 1.

Four brain microstate visualizations, network diagrams, and functional connectivity heatmaps for AUD and HC groups. The figure features two panels labeled 'a' for AUD (Alcohol Use Disorder) and 'b' for HC (Healthy Controls), with four rows for each microstate (A-D). Each section includes a topographical brain map, a circular electrode connectivity diagram, and a functional connectivity heatmap, showing color gradients from blue (low) to red (high). The connectivity patterns and heatmap distributions of each microstate exhibited distinct characteristics in the two groups.

Microstate map topologies corresponded closely with their respective functional connectivity patterns. Four distinct microstates were estimated separately for (a) individuals with AUD and (b) HC. For each group, functional connectivity patterns were subsequently computed for each microstate. (a, b) The middle plots illustrate the channel pairs showing the top 20% of the strongest functional connectivity within each microstate, with the colours and line thickness indicating connectivity strength. The right plots display all channel pairs.

We assessed temporal dynamics across the four microstates within each group. In the HC group, GEV differed significantly across microstates (F(3, 102) = 13.78, p < 0.001; Table S2). Specifically, microstate D showed significantly higher GEV than microstates A and B (p < 0.001), as well as microstate C (p = 0.024). In addition, microstate B exhibited higher GEV than microstate A (p = 0.021). Time coverage showed a significant main effect of microstate (F(3, 102) = 2.98, p = 0.034); however, this effect did not remain significant after applying corrections for sphericity (all p > 0.05). In addition, no significant differences in mean duration were observed across microstates (F(3, 102) = 2.98, p = 0.13). These results suggest that, despite differences in GEV, all four microstates occurred with comparable frequency and duration in the HC group.

In the AUD group, GEV differed significantly across the four microstates (F(3, 102) = 21.33, p < 0.001), but the pattern differed from that observed in the HC group. Specifically, microstates A and B showed comparable GEV (p = 0.29), as did microstates C and D (p = 0.75). In addition, microstates C and D each exhibited significantly higher GEV than microstates A and B (all p < 0.001). In the AUD group, time coverage and mean duration differed significantly across microstates (time coverage: F(3, 102) = 12.69, p < 0.001; mean duration: F(3, 102) = 9.27, p < 0.001). Time coverage of microstate C was comparable to that of microstate D (p = 0.22), while both microstates C and D showed higher time coverage than microstates A (C: p = 0.028; D: p = 0.001) and B (C: p = 0.008; D: p < 0.001). Similarly, the mean duration of microstate C and D were comparable (p = 0.30), whereas both microstates C and D showed longer mean durations than microstates A (C: p = 0.046; D: p = 0.012) and B (C: p = 0.005; D: p = 0.003). These results show that, in the AUD group, microstates C and D were more dominant than microstates A and B.

Direct comparisons between the two groups, controlling for participants’ age and education, revealed significant differences in time coverage for microstates A, B, and D (Table S2, Figure 2). Specifically, the HC group showed significantly greater time coverage for microstates A and B than the AUD group (A: t(59) = 2.93, PFDR = 0.009; B: t(59) = 3.41, PFDR = 0.005), whereas time coverage for microstate D was significantly lower in the HC group than in the AUD group (D: t(59) = −2.44, PFDR = 0.023). These results indicate that the AUD group exhibited dominant and less dominant microstates rather than equally distributed states, suggesting reduced temporal dynamics in their rsEEG.

Figure 2.

Three-panel violin plots compare GEV, time coverage, and mean duration for AUD (red) and HC (blue) across four microstates. The figure displays three vertical panels of violin plots illustrating the distributions for GEV (0-1), time coverage (%), and mean duration (ms) across four microstates (A, B, C, D) for AUD (red) and HC (blue). The top panel reveals higher GEV values for AUD in microstates C. Note that the difference was not statistically significant after controlling for age and education. The middle panel shows time coverage percentages, with HC having significantly higher values in A and B, and AUD significantly higher in D. The bottom panel depicts mean duration, where AUD shows higher values in microstates C and D. These mean duration differences were not statistically significant after controlling for age and education. Each violin reflects data density with overlaid points showing individual data values.

Individuals with AUD exhibited significantly reduced time coverage in microstates A and B compared to HC. After controlling for age and education, time coverage for microstates A and B was significantly lower in individuals with AUD than in HC, whereas time coverage for microstate D was significantly greater in the AUD group than in the HC group. Neither GEV nor mean duration differed significantly between groups. Split violin plots illustrate the distributions of each microstate parameter. Red and blue markers represent individuals with AUD and HC, respectively. Gray box plots indicate the median and the 25th and 75th percentiles. *PFDR < 0.05; **PFDR < 0.01.

Individuals with AUD showed distinctive microstate-specific functional connectivity patterns compared with HC

Next, we computed functional connectivity between pairs of EEG channels for each microstate and for each group. Each ‘microstate-specific functional connectivity’ pattern matched its corresponding microstate map topology (Figure 1). In both groups, the top 20% of the strongest functional connectivity in microstate A was between channels in the right frontal and left occipital regions; in microstate B, between the left frontal and right occipital regions; and in microstates C and D, between frontal and occipital regions. Subsequently, we conducted PCA-based clustering on functional connectivity across all microstates, which confirmed that microstate-specific functional connectivity patterns were quantitatively separable for both groups (Fig. S1). These patterns indicate that, as hypothesised, each microstate is associated with a distinct functional connectivity profile.

Comparisons of functional connectivity between the two groups revealed significant functional connectivity differences in microstates C and D (PFDR < 0.05), but not in microstates A or B (Figure 3). In microstate C, a large proportion of frontal-occipital channel pairs showed significantly greater functional connectivity (i.e. hyperconnectivity) in the AUD group than in the HC group, whereas interhemispheric channel pairs showed significant hypoconnectivity. By contrast, microstate D showed the opposite pattern of group differences in functional connectivity. Specifically, the AUD group showed hypoconnectivity in frontal-occipital channel pairs relative to the HC group, whereas interhemispheric channel pairs exhibited hyperconnectivity. Note that the number of windows used to estimate functional connectivity for microstates C and D did not significantly differ between groups (Table S3), suggesting that the observed connectivity differences were unlikely to be driven by unequal numbers of windows available for connectivity estimation. Nevertheless, it remains possible that true group differences specific to microstates A and B went undetected due to group differences in the number of windows available for connectivity estimation, warranting further investigation.

Figure 3.

Four scatter plots for Microstates A, B, C, D with classification boundaries and EEG network diagrams for C and D that showed significantly differences between the two groups. The figure comprises two panels. Panel (a) features four scatter plots for Microstates A, B, C, and D, showing red circles (AUD) and blue triangles (HC) with dashed lines indicating classification boundaries. Microstate A has 62.44% ± 5.26% accuracy, Microstate B shows 78.72% ± 3.76%, Microstate C has 88.46% ± 4.11%, and Microstate D achieves 95.00% ± 4.00%. Panel (b) presents EEG connectivity diagrams for Microstates C and D that showed significantly different values between the two groups. Red lines indicate hypoconnectivity in the AUD group compared to the HC group, while blue lines indicate hyperconnectivity in the AUD group.

Individuals with AUD showed distinctive functional connectivity patterns specific to microstates C and D compared to HC. (a) Functional connectivity patterns for each microstate were used to distinguish the two groups using a SVM. Each microstate-specific functional connectivity measure yielded classification accuracies exceeding the empirical chance level (57%, based on the group sizes). Specifically, connectivity patterns derived from microstates C and D achieved higher classification accuracies (88.46 ± 4.11% and 95.00 ± 4.00%) than those derived from microstates A and B (62.44 ± 5.26% and 78.72 ± 3.76%). For illustrative purposes, functional connectivity values in each group were plotted against the first two principal components. The colour scale (0–1) indicates the relative number of windows available for microstate-specific connectivity estimation in each individual, with darker colours representing a greater number of windows and lighter colours representing fewer windows. The dashed lines indicate the decision boundaries estimated using PCA-based clustering. (b) A set of functional connectivity specific to microstates C and D, estimated between frontal and parietal regions, was significantly different between the groups (PFDR < 0.05). Red lines indicate hypoconnectivity in the AUD group compared to the HC group, while blue lines indicate hyperconnectivity in the AUD group.

To examine whether EEG measures in the AUD group were associated with psychological status, we computed partial correlations (controlling for age and years of education) between microstate-specific functional connectivity measures that showed significant group differences and individual characteristics, including AUDIT scores, BIS and BAS scores, and PWI scores. In microstate C, functional connectivity between the left temporal region and multiple other regions (F8-C3, T7-C4, C3-P3, and C3-Pz) showed significant negative correlations with PWI scores (PFDR < 0.05; Figure 4). The other channel pair, F4-O1, showed significant positive correlations between functional connectivity and PWI scores (PFDR = 0.015; Figure 4). No significant correlations were found with the other psychological variables. In addition, microstate D-specific functional connectivity did not show significant correlations with any of the assessed psychological variables. These significant associations suggest that microstate C-specific functional connectivity is linked to AUD-related symptom severity.

Figure 4.

The significant correlations between microstate C-specific functional connectivity values and AUD-related symptoms. The figure features a circular EEG electrode map indicating positive (blue) and negative (red) correlations between electrode pairs, highlighting relationships such as F8-C3 and F4-O1. Five scatter plots display the correlation between PWI scores and functional connectivity for pairs: F8-C3, T7-C4, C3-P3, C3-Pz, and F4-O1. Each plot includes data points, a regression line, correlation coefficients (r), and FDR-corrected p-values (P_FDR), illustrating varying correlation trends. F8-C3 shows r = -0.746, P_FDR = 0.001; F4-O1 shows r = 0.584, P_FDR = 0.015.

Microstate C-specific functional connectivity in individuals with AUD is significantly associated with their AUD-related symptoms. Partial correlations were computed between functional connectivity in individuals with AUD and their AUD-related symptoms, controlling for age and education. Functional connectivity between the right frontotemporal or left parietal regions and the left temporal regions showed significant negative associations with PWI scores, whereas the functional connectivity between the right frontal and left parietal channels showed a positive association. Each marker represents an individual participant and black lines overlaid on the scatter plots denote the regression lines. The colour scale (0–1) indicates the relative number of windows available for microstate-specific connectivity estimation in each individual, with darker colours representing a greater number of windows and lighter colours representing fewer windows.

To directly test our hypothesis that microstate-specific functional connectivity offers a more stable approach for distinguishing HC and AUD groups, we used a SVM and compared its performance with that based on conventional functional connectivity measures. The SVM based on conventional functional connectivity measures showed limited discriminative ability (accuracy = 55.74 ± 8.17%; Fig. S2), which did not exceed the empirical chance level (57%), estimated based on the imbalance in group sizes. In contrast, microstate-specific functional connectivity achieved substantially better group separation across all microstates (Figure 3a). Notably, functional connectivity associated with microstates C and D yielded high classification accuracies of 88.46 ± 4.11% and 95.08 ± 4.00%, respectively.

Balanced accuracies (i.e. (sensitivity + specificity)/2) for microstates C and D were 87.53% ([sensitivity, specificity] = [94.29%, 80.77%]) and 94.23% ([100%, 88.46%]), respectively. In contrast, microstates A and B showed lower balanced accuracies of 62.14% ([74.29%, 50.00%]) and 73.08% ([82.86%, 73.08%]), respectively. Permutation testing confirmed significance for microstates C, D, and B (p < 0.001), but not for A (p = 0.17). Furthermore, to evaluate whether the proposed approach yielded statistically robust improvements in classification performance compared with the conventional approach, we performed permutation testing (1,000 iterations) on the accuracy differences between the two approaches (microstate-specific vs conventional functional connectivity). Microstate A did not show a statistically significant improvement over the conventional approach (Δaccuracy = 6.70%, permutation test, p = 0.24). In contrast, microstates B, C, and D demonstrated statistically robust improvements in classification accuracy, with increases of 22.98% (p = 0.009), 32.72% (p = 0.002), and 39.34% (p < 0.001), respectively. These results indicate that our proposed approach more effectively captures characteristic differences in functional connectivity associated with AUD.

To evaluate the impact of altered connectivity in AUD, we computed network properties of microstate-specific functional connectivity in each group (Supplementary text). In HC, across all microstates except microstate A, channel nodes identified as having hub-like properties based on network strength exhibited consistent hub characteristics across four different graph-theoretical measures (Fig. S3b). In contrast, the putative hub nodes in the AUD network failed to demonstrate hub-like topological properties (Fig. S3a). These results indicate that the functional connectivity network in the AUD group is more locally connected. This result suggests that individuals with AUD have a less efficient network structure.

Discussion

In the current study, we investigated functional connectivity in individuals with AUD while accounting for transient state changes in their resting-state EEG. Although the microstate map topologies were comparable between groups, the temporal dynamics across the four microstates showed significant differences. Moreover, compared to functional connectivity computed using a conventional approach, functional connectivity specific to microstates C and D demonstrated superior discriminative power in distinguishing the AUD group from the HC group. These results are in line with our hypothesis that failing to account for dynamic state changes contributes to inconsistencies in functional connectivity measures, and they highlight the potential utility of microstate-specific functional connectivity for investigating AUD.

The microstate time series has been known to vary across neuropsychiatric disorders (Lehmann et al. 2005; Kikuchi et al. 2011; Kim et al. 2024). In contrast to the classical assumption of stationarity in EEG, distinct topologies of electric potentials can be defined, whose transient occurrences and transitions capture global representations of the brain’s dynamic states and their abnormalities [for review, see 16]. Our data showed that the AUD and HC groups did not differ in the topological maps of the identified microstates, but rather differed in their occurrence patterns. Increased dominance of microstates C and D over other microstates indicates diminished network dynamics and reduced network efficiency in AUD. Prior research using both EEG and functional neuroimaging has suggested that microstates C and D are associated with the default mode network (DMN) and executive control network (ECN), respectively, rather than low-level sensory networks [for review, see 17]. Although the present findings were derived solely from rsEEG data, the predominance of these transient states in individuals with AUD may reflect persistent self-referential processing accompanied by compensatory recruitment of executive control systems, a profile commonly reported in addictive disorders (Goldstein and Volkow 2011; Zilverstand et al. 2018). However, given the methodological differences between the current study and previous work, interpretations regarding the functional significance of these dominant microstates remain speculative and warrant further investigation to directly evaluate their validity.

By definition, functional connectivity metrics quantify the degree of coherence between signals from two sources across trials or epochs, under the assumption that the underlying brain state remains stationary within the estimation window. However, when the assumption of stationarity is violated, connectivity estimates may become biased or less interpretable (Stam 2005). In the current study, we found that both the AUD and HC groups exhibited distinct microstates that dynamically transitioned from one state to another, giving rise to non-stationary temporal dynamics in resting-state EEG. Under such conditions, as noted above, estimating connectivity without accounting for temporal state dynamics may have introduced bias. Instead, we suggest that estimating connectivity separately for each microstate better accounts for the characteristics of connectivity measures and EEG signals, potentially leading to more reliable estimates. That is, microstate-specific functional connectivity may capture more informative neural features than conventional connectivity approaches, particularly with respect to transient and dynamic changes associated with transitions between brain states.

In the current study, functional connectivity analysis combined with microstate segmentation revealed distinct connectivity alterations in individuals with AUD between microstates C and D. Specifically, functional connectivity during microstate C was significantly reduced and negatively correlated with PWI scores in the AUD group, consistent with the notion of a disconnection syndrome (Dupuy and Chanraud 2016). Consistent with previous literature across different neural measure modalities emphasising the importance of dynamic connectivity analyses (Hutchison et al. 2013; Baker et al. 2014; Michel and Koenig 2018), our microstate-specific connectivity results were more informative in distinguishing individuals with AUD from HC. Furthermore, these findings suggest that inconsistencies reported in previous AUD studies may, in part, reflect the lack of separation among distinct microstates. It is worth noting that, in addition to local connectivity alterations (i.e. functional connectivity between channels), the global functional network in the AUD group was less efficient, as evidenced by the absence of stable hub-like properties. While our results point to a potential association between alcohol-related impairments in local, microstate-specific connectivity and reduced global network efficiency, the mechanistic pathways linking these effects are not yet understood, underscoring an important direction for future investigation.

The current study has the following limitations. First, the demographic characteristics of the AUD and HC groups were not fully matched; in particular, the AUD group was significantly older and had fewer years of education than the HC group. Although age and education were controlled for in all between-group statistical analyses, the possibility of residual confounding cannot be fully excluded when comparing groups with unmatched demographic characteristics. Future studies with better-matched samples are needed to confirm the replicability of these findings. Second, the sample sizes of our AUD and HC groups were relatively small to modest. Given the high dimensionality of the EEG-derived measures, we applied dimensionality reduction and corrected for multiple comparisons to enhance the robustness of the findings. Nevertheless, the possibility of overfitting cannot be completely ruled out given the limited sample size. Future studies with larger sample sizes would be important for examining individual differences and for evaluating microstate-specific functional connectivity measures as potential neural markers of alcohol-related characteristics. Third, interpretations of the observed between-group differences in microstate-specific functional connectivity remain speculative because the present study relied solely on rsEEG, unlike prior studies that combined EEG and functional neuroimaging. Fourth, due to the cross-sectional nature of the study, we cannot determine whether alcohol misuse and overuse contributed to the observed group differences or whether individuals with pre-existing vulnerabilities in network organisation were more likely to develop alcohol use disorder. To address this limitation, future studies could adopt longitudinal designs or causal neuromodulation approaches to examine how changes in alcohol dependence are associated with corresponding alterations in EEG features over time.

Nevertheless, our findings suggest that EEG microstate-specific functional connectivity provides informative neural markers for characterising the neural impact of alcohol use in young adults. The proposed approach to examining dynamic networks and their graph-theoretical characteristics delineated neural patterns that converge with prior empirical reports of brain’s structural and cognitive deficits in AUD (Shokri-Kojori et al. 2017; Rao and Topiwala 2020), helping to distinguish AUD-related alterations from other mixed patterns. In sum, these findings contribute to understanding how uncontrolled alcohol use is related to both local and global brain network organisation, and they highlight the potential utility of transient EEG properties as a tool for monitoring the severity of alcohol addiction.

Supplementary Material

ParkAUD_MS_supple_v6.docx
TDCN_A_2747472_SM2909.docx (405.4KB, docx)

Funding Statement

This work was supported in part by the National Research Foundation of Korea (NRF) (RS-2024-00420674 to D.C. and J.-S.C.; RS-2025-02263045 to D.C.).

Disclosure statement

The authors declare no competing financial or non-financial interests.

Data availability statement

The data and codes used in this study can be found in online repository: https://github.com/dongilchung/microstate-conn-alcohol.

References

  1. Andrews-Hanna JR, et al. 2007. Disruption of large-scale brain systems in advanced aging. Neuron. 56(5):924–935. 10.1016/j.neuron.2007.10.038 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Baker AP, et al. 2014. Fast transient networks in spontaneous human brain activity. Elife. 3:e01867. 10.7554/eLife.01867 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Benjamini Y, Hochberg Y.. 1995. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B (Methodol). 57(1):289–300. 10.1111/j.2517-6161.1995.tb02031.x [DOI] [Google Scholar]
  4. Betzel RF, et al. 2012. Synchronization dynamics and evidence for a repertoire of network states in resting EEG. Front Comput Neurosci. 6:74. 10.3389/fncom.2012.00074 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Bottino M, et al. 2025. Relating functional connectivity and alcohol use disorder: a systematic review and derivation of relevance maps for regions and connections. Hum Brain Mapp. 46(2):e70156. 10.1002/hbm.70156 [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Carver CS, White TL.. 1994. Behavioral inhibition, behavioral activation, and affective responses to impending reward and punishment: the BIS/BAS scales. J Pers Soc Psychol. 67(2):319–333. 10.1037/0022-3514.67.2.319 [DOI] [Google Scholar]
  7. Chanraud S, et al. 2007. Brain morphometry and cognitive performance in detoxified alcohol-dependents with preserved psychosocial functioning. Neuropsychopharmacology. 32(2):429–438. 10.1038/sj.npp.1301219 [DOI] [PubMed] [Google Scholar]
  8. Chanraud S, et al. 2011. Disruption of functional connectivity of the default-mode network in alcoholism. Cereb Cortex. 21(10):2272–2281. 10.1093/cercor/bhq297 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Cohen MX. 2014. Analyzing neural time series data: theory and practice. MIT Press. [Google Scholar]
  10. de Bruin EA, et al. 2004. Abnormal EEG synchronisation in heavily drinking students. Clin Neurophysiol. 115(9):2048–2055. 10.1016/j.clinph.2004.04.010 [DOI] [PubMed] [Google Scholar]
  11. Dupuy M, Chanraud S.. 2016. Imaging the addicted brain: alcohol. Int Rev Neurobiol. 129:1–31. 10.1016/bs.irn.2016.04.003 [DOI] [PubMed] [Google Scholar]
  12. Elton A, Garbutt JC, Boettiger CA.. 2021. Risk and resilience for alcohol use disorder revealed in brain functional connectivity. Neuroimage Clin. 32:102801. 10.1016/j.nicl.2021.102801 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Goldstein RZ, Volkow ND.. 2011. Dysfunction of the prefrontal cortex in addiction: neuroimaging findings and clinical implications. Nat Rev Neurosci. 12(11):652–669. 10.1038/nrn3119 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Gramfort A, et al. 2013. MEG and EEG data analysis with MNE-Python. Front Neuroinf. 7:267. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Herrera-Díaz A, et al. 2016. Functional connectivity and quantitative EEG in women with alcohol use disorders: a resting-state study. Brain Topogr. 29(3):368–381. 10.1007/s10548-015-0467-x [DOI] [PubMed] [Google Scholar]
  16. Hingson RW, Heeren T, Winter MR.. 2006. Age at drinking onset and alcohol dependence: age at onset, duration, and severity. Arch Pediatr Adolesc Med. 160(7):739–746. 10.1001/archpedi.160.7.739 [DOI] [PubMed] [Google Scholar]
  17. Hutchison RM, et al. 2013. Dynamic functional connectivity: promise, issues, and interpretations. Neuroimage. 80:360–378. 10.1016/j.neuroimage.2013.05.079 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Khanna A, et al. 2015. Microstates in resting-state EEG: current status and future directions. Neurosci Biobehav Rev. 49:105–113. 10.1016/j.neubiorev.2014.12.010 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Kikuchi M, et al. 2011. EEG microstate analysis in drug-naive patients with panic disorder. PLoS One. 6(7):e22912. 10.1371/journal.pone.0022912 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Kim JH. 1999. The reliability and validity test of psychosocial well-being index (PWI). J Korean Acad Nurs. 29(2):304–313. 10.4040/jkan.1999.29.2.304 [DOI] [Google Scholar]
  21. Kim JS, et al. 2024. Resting-state EEG microstate analysis of internet gaming disorder and alcohol use disorder. Dialogues Clin Neurosci. 26(1):89–102. 10.1080/19585969.2024.2432913 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Kim SS, et al. 2008. Psychometric properties of the alcohol use disorders identification test: a Korean version. Arch Psychiatr Nurs. 22(4):190–199. 10.1016/j.apnu.2007.07.005 [DOI] [PubMed] [Google Scholar]
  23. Lehmann D, et al. 2005. EEG microstate duration and syntax in acute, medication-naive, first-episode schizophrenia: a multi-center study. Psychiatry Res. 138(2):141–156. 10.1016/j.pscychresns.2004.05.007 [DOI] [PubMed] [Google Scholar]
  24. Michel CM, Koenig T.. 2018. EEG microstates as a tool for studying the temporal dynamics of whole-brain neuronal networks: a review. Neuroimage. 180(Pt B):577–593. 10.1016/j.neuroimage.2017.11.062 [DOI] [PubMed] [Google Scholar]
  25. Nishida K, et al. 2013. EEG microstates associated with salience and frontoparietal networks in frontotemporal dementia, schizophrenia and Alzheimer’s disease. Clin Neurophysiol. 124(6):1106–1114. 10.1016/j.clinph.2013.01.005 [DOI] [PubMed] [Google Scholar]
  26. Park SM, et al. 2017. Neural connectivity in Internet gaming disorder and alcohol use disorder: a resting-state EEG coherence study. Sci Rep. 7(1):1333. 10.1038/s41598-017-01419-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Poulsen AT, et al. 2018. Microstate EEGlab toolbox: an introductory guide. BioRxiv. :289850. [Google Scholar]
  28. Rao R, Topiwala A.. 2020. Alcohol use disorders and the brain. Addiction. 115(8):1580–1589. 10.1111/add.15023 [DOI] [PubMed] [Google Scholar]
  29. Rubinov M, Sporns O.. 2010. Complex network measures of brain connectivity: uses and interpretations. Neuroimage. 52(3):1059–1069. 10.1016/j.neuroimage.2009.10.003 [DOI] [PubMed] [Google Scholar]
  30. Saunders JB, II., et al. 1993. Development of the alcohol use disorders identification test (AUDIT): WHO collaborative project on early detection of persons with harmful alcohol consumption‐. Addiction. 88(6):791–804. 10.1111/j.1360-0443.1993.tb02093.x [DOI] [PubMed] [Google Scholar]
  31. Shokri-Kojori E, et al. 2017. Alcohol affects brain functional connectivity and its coupling with behavior: greater effects in male heavy drinkers. Mol Psychiatry. 22(8):1185–1195. 10.1038/mp.2016.25 [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Sion A, et al. 2020. Resting‐state connectivity and network parameter analysis in alcohol‐dependent males. A simultaneous EEG‐MEG study. J Neurosci Res. 98(10):1857–1876. 10.1002/jnr.24673 [DOI] [PubMed] [Google Scholar]
  33. Stam CJ. 2005. Nonlinear dynamical analysis of EEG and MEG: review of an emerging field. Clin Neurophysiol. 116(10):2266–2301. 10.1016/j.clinph.2005.06.011 [DOI] [PubMed] [Google Scholar]
  34. Stevens A, Kircher T.. 1998. Cognitive decline unlike normal aging is associated with alterations of EEG temporo-spatial characteristics. Eur Arch Psychiatry Clin Neurosci. 248(5):259–266. 10.1007/s004060050047 [DOI] [PubMed] [Google Scholar]
  35. Strik W, et al. 1995. Larger topographical variance and decreased duration of brain electric microstates in depression. J Neural Transm Gen Sect. 99(1-3):213–222. 10.1007/BF01271480 [DOI] [PubMed] [Google Scholar]
  36. Sullivan EV, Pfefferbaum A.. 2019. Brain-behavior relations and effects of aging and common comorbidities in alcohol use disorder: a review. Neuropsychology. 33(6):760–780. 10.1037/neu0000557 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Tarailis P, et al. 2024. The functional aspects of resting EEG microstates: a systematic review. Brain Topogr. 37(2):181–217. 10.1007/s10548-023-00958-9 [DOI] [PubMed] [Google Scholar]
  38. Vergara VM, et al. 2017. Alterations of resting state functional network connectivity in the brain of nicotine and alcohol users. Neuroimage. 151:45–54. 10.1016/j.neuroimage.2016.11.012 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Wang S, et al. 2025. Alcohol cue processing alters delta-band functional connectivity and theta-band network topology in alcohol use disorder: a task-based EEG study. BMC Psychiatry. 25(1):1206. 10.1186/s12888-025-07629-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Winterer G, et al. 2003. EEG phenotype in alcoholism: increased coherence in the depressive subtype. Acta Psychiatr Scand. 108(1):51–60. 10.1034/j.1600-0447.2003.00060.x [DOI] [PubMed] [Google Scholar]
  41. Zilverstand A, et al. 2018. Neuroimaging impaired response inhibition and salience attribution in human drug addiction: a systematic review. Neuron. 98(5):886–903. 10.1016/j.neuron.2018.03.048 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

ParkAUD_MS_supple_v6.docx
TDCN_A_2747472_SM2909.docx (405.4KB, docx)

Data Availability Statement

The data and codes used in this study can be found in online repository: https://github.com/dongilchung/microstate-conn-alcohol.


Articles from Dialogues in Clinical Neuroscience are provided here courtesy of Taylor & Francis

RESOURCES