Microstates are transient patterns of scalp configurations of brain activity measured by electroencephalography (EEG) at rest. To date, replicating EEG microstates in magnetoencephalography (MEG) data remains challenging. In this study with 113 participants, we aimed to identify prototypical MEG microstates (mMS) at rest, explore their corresponding brain sources, and relate their temporal features to changes in brain activity during open-eyes (ROE) or closed-eyes resting state (RCE). Additionally, we examined their relationship with stimulus-related activity during an auditory Mismatch Negativity (MMN) task. Meta-criterion validation of individual recurrent scalp topographies of resting-state brain activity at the group level identified six mMS. Four mMSs showed a strong spatial correlation with canonical EEG microstates. Fitting mMSs to the MEG signals revealed that mMSs were associated with different brain sources (mMS A/mMS B: left/right occipito-parietal; mMS C: fronto-temporal; mMS D: centro-medial; mMS F/mMS G: left/right fronto-parietal) and that mMS time coverage differed significantly across experimental conditions. Increases in occipital alpha power in RCE relative to ROE correlated with greater mMS A and mMS B time coverage. In the MMN task, the lateralization of deviant detection was associated with mMS F and mMS G time coverage. These results suggest that the MEG signal can be effectively decomposed into microstates. Microstate source reconstruction and task-related modulations indicate that mMSs are associated with large-scale networks and localized activities. Thus, mMSs can provide insight into brain network dynamics and task- or stimulus-specific brain processes, offering a tool to study physiologic and dysfunctional brain activity.
Research Article
MEG microstates: an investigation of underlying brain sources and potential neurophysiological processes
https://doi.org/10.21203/rs.3.rs-4129107/v1
This work is licensed under a CC BY 4.0 License
Journal Publication
published 08 Aug, 2024
You are reading this latest preprint version
resting-state microstates
magnetoencephalography
mismatch negativity
occipital alpha activity
source-reconstructed brain activation
In electroencephalography (EEG), the spontaneous organization of discrete spatial configurations over time has been studied by defining microstates. Microstates are transient patterns of scalp configurations of brain activity, lasting from 50 to 150 milliseconds (Michel & Koenig, 2018). They have been hypothesized to derive from specific brain generators and functional networks, usually investigated by functional magnetic resonance imaging (fMRI). Neuroimaging revealed various resting-state networks associated with specific functions (e.g., somatomotor, visual, auditory, dorsal attention networks), including external stimulus processing or internal superior cognitive functions (Biswal et al., 1995; Deco et al., 2011; Seitzman et al., 2019). MEG provides superior source reconstruction compared to EEG and could enhance EEG microstate research with better-resolved spatial patterning (Hedrich et al., 2017). Despite extensive EEG research on microstates, replications of these results with MEG are still sparse.
Microstate configurations remain dominant for short periods and rapidly transition to another configuration, which in turn becomes dominant for a short duration. Thus, microstates are modeled such that only a single discrete global brain state occurs at any given time (Khanna et al., 2015). During the resting state, four dominant EEG microstates – conventionally labeled as A, B, C, and D - have been consistently described (Michel & Koenig, 2018). These microstates exhibit left anterior/right posterior (microstate A), right anterior/left posterior (microstate B), anterior to posterior (microstate C) dipole orientations, and a central maximum (microstate D) (see Fig. 1, panel 1). Although many studies have used four clusters as a fixed number, primarily to facilitate comparability across different experiments, cross-validation methods can identify a dataset-specific optimal number of clusters. This approach typically leads to substantial improvement in the explained variance of the signal (Michel & Koenig, 2018). Recently, a comprehensive systematic review conducted by Tarailis et al. (2023) updated the microstate classification with the introduction of three novel configurations alongside the four canonical microstates: a posterior variant of microstate C, as well as a left-lateralized and a right-lateralized microstate.
Attempts to understand the functional meaning of each microstate have included the definition of brain generators. In simultaneous EEG-fMRI signals, Britz et al. (2010) showed that the four canonical microstates are associated with four functional brain networks: microstate A with the auditory network, microstate B with the visual network, microstate C with parts of the cognitive control and default mode networks, and microstate D with the dorsal attention network. In a study of topographic source analysis of high-density EEG, Custo et al. (2017) partially replicated Britz et al.’s (2010) findings. They found that microstates A and B were associated with brain activity in regions involved in auditory and visual processing, respectively, and that microstate D sources were localized in the right inferior parietal lobe and the right middle and superior frontal gyri, which are spatially congruent with the dorsal attention network. However, at variance with Britz et al. (2010), Microstate C generators were estimated in the precuneus and the posterior cingulate cortex outside the cognitive control and the default mode network. Using source localization of EEG data, Pascual-Marqui et al. (2014) found substantial overlap between microstates. In addition to a shared hub in the posterior cingulate cortex, microstates A and B involved the left and right occipito-parietal cortex, and microstate C was attributed to sources in the anterior cingulate cortex. Inconsistently, an EEG-fMRI study associated microstate B with brain regions of the anterior default mode network and microstate A with areas of the sensorimotor network (Bréchet et al., 2019). The heterogeneity of source-based reconstructions of EEG microstate activity may be related to the modest spatial resolution of the EEG signal and to small sample sizes, which may have limited the reproducibility of the findings; the time scale of higher-resolution fMRI signals, on the other hand, may be too different from EEG to replicate the source reconstruction results.
An alternative approach to elucidate the functional role of EEG microstates has been to compare their spatial and temporal properties between resting with eyes open (ROE) and resting with eyes closed (RCE). Besides a reduction in total global explained variance in ROE compared to RCE (Seitzman, 2017; Zanesco et al., 2020), conflicting results have been obtained when examining differences between the temporal dynamics of specific microstates. Seitzman et al. (2017) reported a significant decrease in the duration of microstate D and an increase in the occurrence of microstate B in ROE compared to RCE. Instead, Zanesco et al. (2020) found an increase in the duration of microstate D but no change in the occurrence of microstate B. The association of microstate B with activity changes between resting-state conditions is further supported by the association of this microstate with alpha power (Croce et al., 2020), the frequency band that shows the most pronounced changes when transitioning from RCE to ROE (Adrian & Mathews, 1934; Berger, 1929).
Resting-state microstate configurations have also been investigated by examining possible differences between groups or variations of temporal parameters when fitted to the signal during the performance of a cognitive task. Microstate A has been mainly associated with phonological operations (Korn et al., 2021; Diaz et al., 2014; Diaz et al., 2013) and, to a lesser extent, visual processing (Cui et al., 2021; Jabès et al., 2021; Zappasodi et al., 2019). However, increases in microstate A time coverage have been recently associated with arousal and alertness (Antonova et al., 2022; Ke et al., 2021; Jawinski et al., 2021). Microstate B appears robustly involved in visual processing and visuospatial attention (Sietzman et al., 2017). Microstate C functions may reflect mind-wandering and introspection (Tarailis et al., 2021). Microstate D appears to be related to executive function, working memory, and attention (Tarailis et al., 2023), although mixed interpretations have been reported in the literature (see Milz et al., 2016). These findings suggest that microstates A and B primarily involve sensory processing, whereas microstate C is associated with higher cognitive processes and microstate D with executive function and attention. However, results are heterogeneous, experimental paradigms differ, and methodological issues limit generalizations.
Compared to EEG, MEG provides more accurate spatial and less noisy measurements of brain activity, especially with planar gradiometers (Garcés et al., 2017). These MEG features can potentially improve microstates' spatial and temporal definition, allowing more precise identification of their corresponding cortical generators and functional roles. To date, the replication of EEG microstates in MEG data remains challenging. Coquelet et al. (2022) investigated microstates and Hidden Markov Model states in simultaneous MEG-EEG recordings. However, the four MEG microstates they obtained did not match EEG microstates regarding topographies and signal labeling, besides a poor temporal correspondence with Hidden Markov Model states. They concluded that microstates are not reproducible across imaging modalities. Tait and Zhang (2022) applied microstate analysis to MEG region of interest (ROI) time courses, generating ten microstates in the source space. These microstates were characterized by specific patterns of phase-synchronized connectivity within four bilateral networks (frontal, fronto-temporal, visual, and orbital microstates) and three pairs of symmetrical unilateral networks (left and right parietal, left and right temporal pole, left and right sensorimotor microstates). At the functional level, auditory stimuli resulted in more frequent identification of the frontotemporal network 100 ms after stimulus onset, suggesting a sensitivity of cortical microstates to event-related brain responses (Tait & Zhang, 2022). However, the described MEG studies differed from previous EEG research in many methodological aspects, such as source/sensor space, filtering, sampling rate, clustering algorithm, and temporal smoothing. These methodological differences may have contributed to the lack of successful replications of EEG microstate analysis when applied to MEG data.
The present study aimed to identify resting-state MEG microstates, following the methodology used to compute EEG microstates as closely as possible, and investigate their underlying brain sources and functional significance. To this end, we segmented the MEG signal by fitting resting-state MEG microstates to the MEG signal recorded during two resting-state conditions (ROE and RCE) and an auditory Mismatch Negativity task (MMN). For labeled time frames, we performed 1) source reconstruction, 2) computed the average activity in parcels of the brain surface corresponding to different functional networks, and 3) tested possible changes in MEG microstate time coverage during different resting-state and task conditions, exploring possible associations of MEG microstates with alpha power during RCE and stimulus processing in MMN.
Participants
The participants included in this study were a subset of individuals from the control groups reported by Valt et al. (2022) in their investigation of MMN in psychosis. Of the 135 control participants, only 113 individuals had usable resting-state recordings. Therefore, the sample size for the current study consisted of 113 healthy participants with a mean age of 26.5 years (SD 6.8), of whom 58.3% were female. Participants were required to be unmedicated and have no history of neurological or psychiatric disorders. Additionally, a trained psychologist conducted a Structured Clinical Interview for DSM-5 (First et al., 2017) to confirm the absence of any relevant psychiatric conditions. The Ethics Committee of the University Hospital of Bari approved the MEG protocol used in this study, and all participants gave informed consent for the MEG data collection.
Procedure
The experimental protocol included two 5-minute resting-state conditions and two tasks. The two resting-state conditions were one with eyes open (i.e., ROE) and one with eyes closed (i.e., RCE). The tasks were an auditory mismatch negativity (i.e., MMN) task with duration-deviant stimuli (Näätänen et al.., 2007) and a visual task designed to evoke a 10Hz visual steady-state potential. The visual task was not analyzed as part of this study.
During the resting-state conditions, participants were instructed to remain still throughout the recording session. They were explicitly advised not to engage in any particular thoughts, to avoid nodding off, and to avoid falling asleep. During the ROE, participants were required to keep their eyes open and focus on a fixation cross displayed on a screen, while during the RCE, participants were instructed to keep their eyes closed.
In the MMN task, participants were asked to watch a silent Tom & Jerry cartoon while a sequence of 667 auditory stimuli was presented to both ears through earphones. All the auditory stimuli had a frequency of 1000 Hz and an intensity of 80 dB, while the duration was 50 ms for 612 stimuli (standard) and 100 ms for 55 stimuli (deviant). Deviant and standard stimuli were pseudorandomly intermixed. The interstimulus interval between the tones was 500 ms.
MEG data recording and processing
MEG signals were acquired using a 306-channel Elekta Neuromag® TRIUX whole-head system (Elekta Neuromag Oy, Helsinki, Finland) with a recording sampling rate of 1000 Hz. The MEG signal was processed offline with Elekta MaxFilter™ software from Elekta Neuromag Oy (Helsinki, Finland) to remove external noise with temporal Signal-Space Separation (tSSS), as described by Taulu and Simola (2006), and to correct for head motion using data from five Head Position Identifier coils. After these preliminary preprocessing steps, the MEG signal was further processed using Fieldtrip (Oostenveld et al., 2011) to prepare the ROE data for MEG microstate identification or with Brainstorm (Tadel et al., 2011) to prepare the ROE, RCE, and MMN data for MEG microstate fitting for source reconstruction, network activity analysis, alpha power calculation, and MMN peak amplitude extraction. The processed data, converted to a BrainVision format, were then imported into the Cartool software (64-bit release version 4.10. Revision 7573; Brunet et al., 2011) for all the microstate analyses.
Data were down-sampled to 250 Hz, and frequencies associated with the European power grid (notch filter: 50 Hz) and high frequencies (low-pass filter: 30 Hz) were removed. Artifacts related to eye movements and heartbeat were identified and corrected using Independent Component Analysis (ICA) trained on all the sensors. To ensure optimal signal quality for subsequent microstate identification, the ROE data were segmented into 2-second time windows. Subsequently, epochs containing potential artifacts were identified and rejected using Fieldtrip's automatic artifact rejection function (ft_artifact_zvalue). The resulting cleaned epochs (mean number of epochs 121.2 ± 14.8, corresponding to 242.3 ± 29.6 seconds of clean recording on average) were concatenated. For each sensor, gradiometer pairs were combined to obtain the planar gradient magnitude over both directions by computing the square root of the sum of the squares for each gradiometer pair – as reported by Coquelet et al. (2022).
MEG microstate identification
A spatial filter was first applied to smooth topographies and remove outliers (Michel & Brunet, 2019). The spatial filter determines and interpolates the maximum and minimum values from the six nearest neighboring sensors for each location. Maps were then extracted at peaks in the Global Field Power (GFP) of the data, which are time frames with the highest signal-to-noise ratio. K-means clustering was then applied to the resulting collection of maps.
Cartool’s default parameters for resting-state MEG analysis were used for k-means clustering, i.e., only positive values, no average reference, and the polarity of the data not being relevant. K-means clustering (see Fig. 1, Panel 2) was applied to individual ROE data to estimate the optimal set of maps for each subject (individual level) and then across all participants to estimate the optimal set of maps for the entire dataset (group-level). For both levels of clustering, the optimal number of clusters was determined automatically via a meta-criterion validation (Michel & Koenig, 2018). This cross-validation method determines the optimal k-number as the median output of different validated criteria (i.e., Gamma, Silhouettes, Davies and Bouldin, Point-Biserial, Dunn, Kraznowski-Lai Index, and the median value of these criteria). The maps obtained at the group level were the MEG microstate used for all subsequent analyses.
To assess the similarity between MEG microstates and the four canonical EEG maps, spatial correlation coefficients were computed between four putative canonical MEG microstates and the four canonical EEG microstates published by Tomescu et al. (2018). These EEG templates are publicly accessible on GitHub (https://github.com/ThomasKoenigBern/MS-emplate-Explorer/tree/main/Template%20Maps%202023-04-19). To ensure the comparability of the microstate maps, the EEG and MEG montages were averaged with Cartool. Additionally, microstate topographies were normalized to range from 0 to 1. Spatial correlation coefficients were then used to interpret similarities between MEG maps and EEG templates.
MEG microstate fitting
The MEG microstates identified from the ROE recordings were fitted to each individual recording (ROE, RCE, and MMN). This procedure assigned the microstate with the highest spatial correlation to each time frame (i.e., competitive fitting) only if the correlation between the data topography and the microstate topography was above the minimum threshold of 0.50. Consequently, some parts of the data may remain unlabeled. Temporal smoothing was applied (window half size = 3 time frames, Besag factor = 10). After the fitting procedure for each microstate, the proportional time coverage was calculated as the sum of the time points labeled by each microstate divided by the total number of time points in the data.
Source reconstruction of MEG microstates
The individual inverse source model was constructed as overlapping spheres with 15,002 vertices, with a noise covariance matrix derived from a 2-min empty room recording, which was acquired and processed with the same settings used while the participants were performing the tasks. After linear scaling, the ROE sensor-based signals were converted to source-based signals with standard Low-Resolution Brain Electromagnetic Tomography (sLORETA, Pascual-Marqui, 2002). The brain cortex model included constrained vertices extracted from the ICBM152 MRI template (Mazziotta et al., 2001). The source-based data corresponding to periods labeled by a given MEG microstate were averaged with each time frame weighted by the correlation between data topography and microstate topography. The average source activity of the whole signal was subtracted from the average source activity of each MEG microstate to isolate the specific activation associated with each MEG microstate.
In addition, for each MEG microstate, we averaged the source-estimated signal over labeled time frames in each of the 34 parcels (17 parcels for each hemisphere) of Schaefer’s brain atlas (Schaefer et al., 2018). This brain atlas parcellates the cerebral cortex into brain networks (e.g., Default Mode, Saliency, etc.) based on functional connectivity data from neuroimaging studies.
Alpha power computation
To assess the relationship between MEG microstate modulations between ROE and RCE and occipital alpha power enhancement when participants kept their eyes closed compared to when they kept their eyes open, the frequency power of activity at the right and left occipital sensors in ROE and RCE was calculated using Fast Fourier Transformation. The resulting frequency power spectra were then averaged over the selected sensors. To quantify the change in occipital alpha activity during RCE compared to ROE, the average frequency power of RCE was subtracted from the average frequency power of ROE. The amplitude of the most prominent negative peak within the 8 to 12 Hz frequency range was used as an index of the increase in occipital alpha activity during RCE.
MMN peak amplitude computation
To investigate potential variations in MEG microstate detection rates during the MMN time window, the number of segments labeled by a specific MEG microstate at a given time point from 150 ms to 300 ms after stimulus onset was calculated for 55 standard and 55 deviant auditory stimuli. Thus, the maximum detection rate in a time point was 55, meaning that a specific MEG microstate was identified in all the segments at that specific time point.
We then assessed whether MEG microstate labeling was sensitive to detecting MMN amplitude differences between the left and right auditory regions. For the MMN peak amplitude computation, the MMN source-based signal was divided into epochs starting 200 ms before stimulus onset and lasting 700 ms. The potentials of deviant stimuli and standard stimuli preceding a deviant stimulus were then normalized to z-scores based on the 200 ms pre-stimulus interval, converted to absolute values, and spatially smoothed before averaging. The MMN peak amplitude was determined as the most prominent negative peak between 150 and 300 ms in the subtraction of deviant from standard activity reconstructed to vertices of parcels of the Destrieux atlas (Destrieux et al., 2010). These parcels correspond to the left and right primary and secondary auditory regions, i.e., the superior temporal gyrus/planum temporale parcel and the temporal transverse sulcus parcel.
Statistical analysis on MEG microstates
For the network analysis, to determine whether there was an increase or decrease of microstate-related mean activity in specific network parcels of the brain surface, we computed a null distribution of activities in the brain parcels extracted from 1000 permutations of random assignments of microstates to the data, preserving the number and duration of the identified MEG microstates. Afterward, the association of an MEG microstate with a specific network was determined by comparing the normalized mean signal amplitude in a specific parcel with the corresponding normalized null distribution. Observations significantly different from the mean of the null distribution were determined according to the critical |Z-score| of 3.6. This Z-score corresponds to a Bonferroni corrected p-value of 0.0002 (α/number of comparisons = .05/(34*6)).
For the analysis of MEG microstate temporal features, we computed the time coverage as the percentage of signal labeled by a specific MEG microstate in ROE, RCE, and MMN. After a preliminary analysis of time coverage in ROE, data were contrasted between ROE and RCE and between ROE and MMN in two analyses of variance (ANOVA) that tested the factors Condition and Microstates. Significant main effects and interactions involving the factor Microstates were examined with microstate-specific post-hoc t-tests, with the α of .05 adjusted with Bonferroni’s correction to account for multiple comparisons.
In RCE, besides the ANOVA contrasting ROE and RCE, we correlated the differences in individual microstate time coverage between ROE and RCE with the differences in occipital alpha power between ROE and RCE to evaluate the sensitivity of MEG microstates to changes in occipital alpha activity.
In MMN, one ANOVA contrasted MEG microstate time coverages between ROE and MMN, while a second ANOVA investigated stimulus-related differences by considering Microstate and Stimulus (deviant vs. standard) as factors. This ANOVA focused on variations in the average frequency of MEG microstate detections during the 150–300 ms time window after the onset of standard or deviant stimuli. A significant interaction would imply microstate sensitivity to physical differences in the stimuli. Furthermore, we investigated the MEG microstate's sensitivity to the MMN laterality. To this end, we correlated the difference in MEG microstate detection rates between standard and deviant stimuli, termed MEG microstate mismatch negativity (mMS-MMN), with the discrepancy in MMN peak amplitude between the right and left auditory regions (MMN lateralization score). A positive correlation between mMS-MMN and MMN lateralization scores would indicate a MEG microstate sensitivity to increase MMN amplitudes over the right auditory region. In contrast, a negative correlation would indicate a microstate sensitivity to increase MMN amplitudes over the left auditory region.
MEG microstate identification
K-means clustering on ROE activity identified six distinct MEG microstates as the best fit to the data based on the meta-criterion (Fig. 2). Among these MEG microstates (magnetic microstates, mMS), four showed topographies similar to canonical EEG microstates (mMS2, mMS3, mMS4, mMS5). In contrast, two (mMS1 and mMS6) showed fronto-lateral right and left patterns, respectively. On average, the percentage of ROE time frames labelled by the six mMSs was 70% (SD = 10%, range 50%-96%), with an average time coverage for each mMS of approximately 10%.
Four putative canonical MEG microstates correlated with the four canonical EEG microstates labeled as A-D (as reported in Tomescu et al., 2018). Strong spatial correlations were present between microstate A and mMS5, r = 0.89, p < .001, between microstate B and mMS3, r = 0.89, p < .001, between microstate C and mMS2, r = 0.98, p < .001, and between microstate D and mMS4, r = 0.88, p < .001 (see Fig. 3 for all the correlations). Thus, this analysis showed that some MEG microstates were spatially similar to EEG microstates.
Based on topographies, the six MEG microstates were named as followed: mMS1 as mMS G, mMS2 as mMS C, mMS3 as mMS B, mMS4 as mMS D, mMS5 as mMS A, and mMS6 as mMS F, following the canonical nomenclature (Tarailis et al., 2023).
MEG microstate source reconstruction
Descriptively, source-based analyses of the six MEG microstates (Fig. 4) showed left and right occipito-parietal activity for mMS A and mMS B, respectively, a fronto-temporal activity for mMS C, a centro-medial activity for mMS D, and left and right fronto-parietal activity along the central sulcus for mMS F and mMS G, respectively.
Mean activity in time frames labeled by an mMS was compared with the mean activity of the normalized null distribution of randomly selected time frames to determine microstate-related activity changes in parcels of the brain surface that reflect distinct functional brain networks. Table 1 reports the networks that showed a significant difference from random activity. Interestingly, for each mMS, there were statistically significant increases as well as decreases in specific brain parcels compared to random (reported in Table 1 as + and -, respectively).
This analysis indicated that, overall, mMS G and mMS B showed greater activity in networks of the right hemisphere and less activity in networks of the left hemisphere. Conversely, mMS A and mMS F mainly presented increased activity in networks of the left hemisphere and decreased activity in networks of the right hemisphere. The pattern of activity was bilateral for mMS C and mMS D.
Interestingly, most networks that showed less activity in mMS F and mMS G presented greater activity in mMS A and mMS B, respectively. This fourfold symmetry (see colors in Table 1) applied to mMS A, mMS B, mMS B, and mMS G in the three Control networks, the Default Mode C network, the two Dorsal Attention networks, the two Salience/Ventral Attention networks, the two Somato-Motor networks, the Temporal-Parietal network, and the two Visual networks. In general, mMS F and mMS G captured mostly increases in ipsilateral fronto-central activity from Control A, Control B, Dorsal Attentional B, Saliency/Ventral Attention, and Somato-Motor networks, whereas mMS A and mMS B captured mostly increments of ipsilateral posterior and medial occipito-parietal activity from Control A, Default Mode C, Dorsal Attention, Somato-Motor B, Temporal-Parietal, and Visual networks. For mMS A and mMS B, there were also control-lateral activations in Control C and Visual Periphery. Networks that showed changes of activity not congruent with this fourfold pattern were the Default Mode B network, which showed a deactivation for mMS B and an activation for mMS F, and the Somato-Motor A network, which showed a deactivation for mMS F.
In contrast to the strong lateralization of mMS A, mMS B, mMS F, and mMS G, bilateral activity patterns characterized mMS C and mMS D. In the data labeled by mMS C, the bilateral Salience/Ventral B networks were the only ones to show greater activity, whereas less activity was recorded in the bilateral Control C network, the two Dorsal Attention networks, and the two Visual networks. The signal labeled by mMS D showed greater than random activity only in the bilateral Control C networks, the Dorsal Attention B networks, and the Somato-Motor A networks.
Table 1 Networks showing greater (+) or less (-) than random activity in parcels of the brain surfaces for source-reconstructed data labeled by an MEG microstate.
Note: Some cells are colored in green or red to show the fourfold symmetric pattern of activity for the four lateralized MEG microstates (i.e., mMS A, mMS B, mMS F, and mMS G). The gray color highlights the bilateral pattern for the two medial microstates (i.e., mMS C and mMS D).
MEG microstates in RCE
The ANOVA that contrasted ROE and RCE revealed significant main effects for both Condition and Microstate, with a significant interaction qualifying these effects (Fig. 5). The significant main effect of Condition, F(1,112) = 4.91, p = .007, ηp2 = .04, indicated higher total signal labeling in RCE than ROE, t(112) = 3.13, p = .006, d = .30. The significant main effect of Microstate, F(3.2,358.5) = 8.21, p < .001, ηp2 = .04, indicated that the time coverage of mMS D was higher than the time coverage of mMS C, mMS F and mMS G, ts(112) > 3.92, ps < .001, ds > .37, and that mMS A and mMS B had also time coverage higher than mMS G, ts(112) > 3.71, ps < .003, ds > .35 (all other contrasts were not significant, ts(112) < 2.78, ps > .051, ds < .26). The significant Condition by Microstate interaction, F(3.7,411.5) = 25.39, p < .001, ηp2 = .13, revealed task-dependent modulations of the time coverages of the six MEG microstates. Planned contrasts showed (see Fig. 5) that, on the one hand, the time coverage of mMS C, mMS F and mMS G was significantly larger in ROE than RCE, ts(112) > 3.15, ps < .001, ds > .30. On the other hand, the time coverage of mMS A and mMS B was significantly larger in RCE than ROE, ts(112) > 6.31, ps < .001, ds > .59. The time coverage of mMS D did not differ significantly between the two resting-state conditions, t(112) = 0.85, p = .396.
The time coverage differences between RCE and ROE of mMS A, mMS B, and mMS C were significantly correlated with alpha power differences between RCE and ROE over the occipital sensors (see Fig. 6). Positive correlations were found for mMS A, r = .266, p = .005, and mMS B, r = .352, p < .001, indicating that increases of alpha power in RCE resulted in prolonged time coverage of these two posterior MEG microstates in RCE compared to ROE. In contrast, a negative correlation was found for mMS C, r = − .226, p = .017, suggesting that occipital alpha activity during RCE decreased the time coverage of this anterior microstate.
MEG microstates in MMN
The ANOVA that contrasted MEG microstates between ROE and MMN (see Fig. 7) revealed a significant main effect of Microstate, F(3.4,378.6) = 4.43, p = .003, ηp2 = .04, with a significant interaction between Microstate and Condition, F(3.4,385.0) = 3.11, p = .021, ηp2 = .03, but not significant main effect of Condition, F(1,112) = 0.93, p = .336. The significant main effect of Microstate indicated that the time coverage of mMS D was higher than the time coverage of mMS A, mMS B, and mMS G, ts(112) > 3.20, ps < .019, ds > .30 (all other contrasts were not significant, ts(112) < 2.78, ps > .068, ds < .26). The significant Condition by Microstate interaction suggested that mMS C had larger, t(112) = 2.91, p = .031, d = .21, but mMS D smaller time coverage in MMN than ROE, t(112) = 2.42, p = .017, d = .23 (all other contrasts, ts(112) < 1.53, p = .128), but both these two results did not survive correction for multiple comparisons.
Besides differing globally, MEG microstates showed also a sensitivity to the type of auditory stimulus. This sensitivity was indicated by the significant interaction between Stimulus and Microstate in the time-window 150–300 ms after stimulus onset, F(5,560) = 11.66, p < .001, ηp2 = .09. (Fig. 8). Compared to standard stimuli, deviant stimuli had a higher detection rate for mMS G, t(112) = 4.92, p < .001, d = 462, and a lower detection rate for mMS A, mMS B, and mMS D, ts(112) > 3.24, ps < .002, ds > .31.
Interestingly, there were also significant correlations between MMN-lateralization and changes in mMS detection rates for standard and deviant stimuli (Fig. 9). Specifically, mMS G was detected more frequently in deviant than in standard trials the more the MMN amplitude was lateralized to the right, r = − .34, p < .001; conversely, when the MMN mainly was lateralized to the left, there was an increase in the detection rates of mMS F, r = .39, p < .001.
In this study, we aimed to reproduce EEG microstates in a large sample of MEG data. Taking advantage of the superior spatial resolution of MEG, we explored the patterns of brain activity labeled by each microstate in the source space, focusing on functional brain networks. We then investigated the functional meaning of MEG microstates by examining their temporal dynamics across different experimental (RCE, ROE, and MMN) and stimulus conditions (standard and deviant auditory stimuli).
We found that six MEG microstates provided the best clustering for the MEG signal, encompassing approximately 70% of the data. Of the six identified mMS, four showed strong spatial correlation with the topographies of the canonical EEG microstates (A, B, C, and D). In addition to the four canonical EEG microstates, the identified MEG microstates included two lateralized topographies. These two mMS topographies are consistent with the extended set of configurations reported in the EEG microstate meta-analysis by Tarailis and colleagues (2023). The successful replication of EEG microstate topographies using MEG signals suggests the feasibility of identifying microstates within magnetic signals.
A previous MEG study by Coquelet et al. (2022) identified four distinct MEG microstates: two anterior lateralized (left and right) and two posterior lateralized (left and right) configurations. The variance in MEG microstates between Coquelet et al.'s (2022) study and the present investigation may be due to differences in clustering methodology. Coquelet et al. (2022) pre-specified the number of clusters at four, whereas we used a data-driven approach that determined six as the optimal number of clusters. Interestingly, as reported in the supplementary materials, setting k = 4 in our group-level clustering analysis, we obtained microstate topographies visually similar to those of Coquelet et al. (2022). Lateralized microstates are often not observed in EEG studies. The challenge of detecting lateralized microstates in the EEG signal may be related to the effect of brain volume conduction (Holsheimer & Feenstra, 1977). Therefore, the MEG signal, which is less susceptible to spatial distortions, might offer superior resolution for studying microstates.
Tait and Chang (2022) implemented a microstate approach to study the MEG signal but moved from scalp-level to source-level data. In contrast to the approach used by Tait and Chang (2022), which directly clustered source-level activity, in the present study, we first clustered scalp-level activity into MEG microstates and, subsequently, moved to source-level activity to reconstruct the underlying brain generators. After subtracting the shared activity, source reconstruction revealed distinct brain regions associated with the six different MEG microstates. At the network level, mMS A, mMS B, mMS F, and mMS G were associated with changes in activity in parcels of the brain surface corresponding to distinct functional networks with a fourfold symmetry along the left/right and fronto-central/occipito-parietal dimensions.
Although mMS D was the most prominent MEG microstate in ROE, instructing participants to keep their eyes closed during resting state significantly increased mMS A and mMS B time coverage. Interestingly, the increase in mMS A and mMS B time coverage between ROE and RCE correlated with the increase in alpha activity registered by occipital sensors between ROE and RCE. The interpretation that mMS B reflects the activity in the occipital cortex is consistent with empirical findings that its corresponding EEG microstate, namely EEG microstate B, is related to activity in the visual network (Britz et al., 2010; Custo et al., 2017). However, the similar interpretation that mMS A also reflects occipital activity contrasts with previous reports that linked its corresponding EEG microstate, namely EEG microstate A, to auditory processing (Britz et al., 2010). Notably, the observation that the time coverage of mMS A and mMS B is higher in RCE and ROE further supports the hypothesis that these two posterior MEG microstates reflect activity in the occipital lobe. RCE shows greater occipital alpha power than ROE (Berger, 1928). Such modulation is thought to reflect the degree of inhibition of visual information processing (Hohaia et al., 2022). In the present experiment, changes of mMS A and mMS B correlated with modulations of occipital alpha power: participants who showed the greatest potentiation of alpha power from ROE to RCE presented the most substantial increment of mMS A and mMS B time coverage. Consistent with this observation, Croce et al. (2020) reported that EEG microstate B is associated with alpha activity. The larger time-coverage of mMS A and mMS B during RCE than ROE is at variance with the report of an opposite pattern of modulations for the EEG microstate B by Seitzman et al. (2017). However, these conflicting results may be due to the larger set of microstates extracted in our study. Actually, in Seitzman et al.’s study the overall explained variance is reduced during ROE compared to RCE, and the authors suggested that this variation may be due to the limited ability of the canonical four microstates to accurately describe resting brain activity.
Modulations of MEG microstates were also observed in the MMN task. The performance of this task induced an increase in the time coverage of the frontal MEG microstate, namely mMS C. However, the presentation of a deviant stimulus increased the detection rate of the right-lateralized MEG microstate, mMS G. Notably, the increase in detection rates of mMS G mainly occurred in participants with the largest MMN amplitude in the right auditory region. Participants with an MMN lateralization to the left showed increased detection rates for the left-lateralized MEG microstate, i.e., mMS F, instead. The two lateralized MEG microstates resembled the topographies of EEG microstates F and G, as reported by Tarailis et al. (2023). These two EEG microstates are not part of the four canonical EEG microstates, and the literature does not provide reliable information about their functional role. The sensitivity of these two MEG microstates to auditory deviant detection suggests an involvement in auditory processing. However, the transverse gyrus, i.e., the location of the primary and secondary auditory cortex, is not part of the reconstructed brain activity for mMS G and mMS F. The reconstructed brain activity was located along the right and the left central sulcus, respectively, suggesting an involvement in somatosensory perception or motor planning and execution. Accordingly, previous studies have linked the right-lateralized EEG microstate to activity in the sensorimotor network (Custo et al., 2017; Bréchet et al., 2019).
The analysis of activity in parcels corresponding to functional brain networks showed that these two MEG microstates are associated with increased activity in the Somato-Motor A network. This network also includes parts of the superior temporal gyrus corresponding to the auditory regions. The higher mMS G detection rate in the perception of deviant auditory stimuli might be linked to dipole localizations on the scalp rather than directly reflecting auditory processing. Accordingly, task-related modulations of brain activity resulted in increased detection of those microstates topographically closer to the computationally identified source of the activity and decreased detections of more distal ones. It is important to note that MEG microstates do not reflect the activity of a parcel corresponding to a specific brain network but clusters of spatially adjacent networks. Thus, microstates may relate more to dipole localization over time than brain area-specific activity patterns. Therefore, an open question for future investigation is to what extent microstates reflect a homology of recurring brain activity patterns with regional specificity or whether they model activity patterns across the scalp with heterogeneous sources and time courses.
The main limitation of this study is the lack of simultaneous EEG recording, a condition that would have allowed a direct comparison between EEG microstates and MEG microstates. Coquelet et al. (2022) found no discernible temporal correlation between the labeling of MEG and EEG data. However, the labeling was performed with EEG and MEG microstates that showed substantial differences in topography. With the successful replication of the topography of canonical EEG microstates, a direct comparison may reveal a greater degree of agreement between the results of these two methods.
In conclusion, six MEG microstates, which resembled the topographies of EEG microstates, were identified. MEG microstates were sensitive to global and local activity, such as occipital alpha increments during RCE or event-related auditory activity during a passive auditory MMN task. Task- and stimulus-specific modulations of MEG microstates indicate that microstates can detect spontaneous resting-state brain activity and are sensitive to changes in brain activity during a task, offering a tool for studying normal or dysfunctional brain functioning.
Financial support.
This project has received funding from the European Union within the MUR PNRR grant for young researchers (PRISM: Profiling the risk for schizophrenia with machine learning, CUP H97G22000320005, SOE_0000127) to C.V.; from the European Union within the MUR PNRR Extended Partnership initiative on Neuroscience and Neuropharmacology (Project no. PE00000006 CUP H93C22000660006 “MNESYS, A multiscale integrated approach to the study of the nervous system in health and disease”) to A.R., A.B., and G.P. ; from the Apulia regional government (Project: “Early Identification of Psychosis Risk”) to A.B. This work was partially supported by the project FAIR - Future AI Research (Project no. PE00000013 CUP H97G22000210007), spoke 6 – Symbiotic AI, under the MUR PNRR Extended Partnership program funded by the NextGenerationEU, awarded to C.B.
Competing Interests
A.T. has received lecture fees from Angelini. A.B. has received lecture fees from Otsuka, Janssen, and Lundbeck, as well as consultant fees from Biogen. G.P. has received lecture fees from Lundbeck. All other authors have no biomedical financial interests or potential conflicts of interest to report.
Author Contribution
Christian Valt #Conceptualization, Data curation, Supervision, Data analysis, Methodology, Writing – original draftAngelantonio Tavella #Conceptualization, Data curation, Data analysis, Methodology, Writing – original draftCristina BerchioSupervision, Methodology, Data Analysis, Writing – review & editingLeonardo SportelliMethodology, Data Analysis, Writing – review & editingAntonio RampinoMethodology, Resources, Writing – review & editingAlessandro BertolinoConceptualization, Resources, Methodology, Supervision, Visualization, Writing – review & editingGiulio PergolaConceptualization, Resources, Methodology, Supervision, Visualization, Writing – review & editing# equal contribution as leading authors
- Adrian ED, Mathews BHC (1934) The Berger rhythm: Potential changes from the occipital lobe in man. Brain 57:355–385. https://doi.org/10.1093/brain/awp324
- Antonova E, Holding M, Suen HC, Sumich A, Maex R, Nehaniv C (2022) EEG microstates: Functional significance and short-term test-retest reliability. Neuroimage: Rep 2(2):100089. https://doi.org/10.1016/j.ynirp.2022.100089
- Berger H (1929) Ueber das Elektroenzephalogramm des Menschen. Archives Psychiatry 87:527–570. https://doi.org/10.1007/BF01797193
- Biswal B, Zerrin Yetkin F, Haughton VM, Hyde JS (1995) Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magn Reson Med 34(4):537–541. https://doi.org/10.1002/mrm.1910340409
- Bréchet L, Brunet D, Birot G, Gruetter R, Michel CM, Jorge J (2019) Capturing the spatiotemporal dynamics of self-generated, task-initiated thoughts with EEG and fMRI. NeuroImage 194:82–92. https://doi.org/10.1016/j.neuroimage.2019.03.029
- Britz J, Van De Ville D, Michel CM (2010) BOLD correlates of EEG topography reveal rapid resting-state network dynamics. NeuroImage 52(4):1162–1170. https://doi.org/10.1016/j.neuroimage.2010.02.052
- Brunet D, Murray MM, Michel CM (2011) Spatiotemporal analysis of multichannel EEG: CARTOOL. Computational intelligence and neuroscience, 2011, 1–15. https://doi.org/10.1155/2011/813870
- Coquelet N, De Tiège X, Roshchupkina L, Peigneux P, Goldman S, Woolrich M, Wens V (2022) Microstates and power envelope hidden Markov modeling probe bursting brain activity at different timescales. NeuroImage 247:118850. https://doi.org/10.1016/j.neuroimage.2021.118850
- Croce P, Quercia A, Costa S, Zappasodi F (2020) EEG microstates associated with intra-and inter-subject alpha variability. Sci Rep 10(1):2469. https://doi.org/10.1038/s41598-020-58787-w
- Cui R, Jiang J, Zeng L, Jiang L, Xia Z, Dong L, Yao D (2021) Action video gaming experience related to altered resting-state EEG temporal and spatial complexity. Front Hum Neurosci 15:640329. https://doi.org/10.3389/fnhum.2021.640329
- Custo A, Van De Ville D, Wells WM, Tomescu MI, Brunet D, Michel CM (2017) Electroencephalographic resting-state networks: source localization of microstates. Brain Connect 7(10):671–682. https://doi.org/10.1089/brain.2016.0476
- Deco G, Jirsa VK, McIntosh AR (2011) Emerging concepts for the dynamical organization of resting-state activity in the brain. Nat Rev Neurosci 12(1):43–56. https://doi.org/10.1038/nrn2961
- Destrieux C, Fischl B, Dale A, Halgren E (2010) Automatic parcellation of human cortical gyri and sulci using standard anatomical nomenclature. NeuroImage 53(1):1–15. https://doi.org/10.1016/j.neuroimage.2010.06.010
- Diaz BA, Van Der Sluis S, Benjamins JS, Stoffers D, Hardstone R, Mansvelder HD, Linkenkaer-Hansen K (2014) The ARSQ 2.0 reveals age and personality effects on mind-wandering experiences. Front Psychol 5:271. https://doi.org/10.3389/fpsyg.2014.00271
- Diaz BA, Van Der Sluis S, Moens S, Benjamins JS, Migliorati F, Stoffers D, Linkenkaer-Hansen K (2013) The Amsterdam Resting-State Questionnaire reveals multiple phenotypes of resting-state cognition. Front Hum Neurosci 7:446. https://doi.org/10.3389/fnhum.2013.00446
- First MB, Skodol AE, Bender DS, Oldham JM (2017) User's guide for the Structured Clinical Interview for the DSM-5® Alternative Model for Personality Disorders (SCID-5-AMPD). American Psychiatric Pub
- Garcés P, López-Sanz D, Maestú F, Pereda E (2017) Choice of magnetometers and gradiometers after signal space separation. Sensors 17(12):2926. https://doi.org/10.3390/s17122926
- Hedrich T, Pellegrino G, Kobayashi E, Lina JM, Grova C (2017) Comparison of the spatial resolution of source imaging techniques in high-density EEG and MEG. NeuroImage 157:531–544. https://doi.org/10.1016/j.neuroimage.2017.06.022
- Hohaia W, Saurels BW, Johnston A, Yarrow K, Arnold DH (2022) Occipital alpha-band brain waves when the eyes are closed are shaped by ongoing visual processes. Sci Rep 12(1):1194. https://doi.org/10.1038/s41598-022-05289-6
- Holsheimer J, Feenstra BWA (1977) Volume conduction and EEG measurements within the brain: a quantitative approach to the influence of electrical spread on the linear relationship of activity measured at different locations. Electroencephalogr Clin Neurophysiol 43(1):52–58. https://doi.org/10.1016/0013-4694(77)90194-8
- Jabès A, Klencklen G, Ruggeri P, Michel CM, Lavenex B, P., Lavenex P (2021) Resting-state EEG microstates parallel age‐related differences in allocentric spatial working memory performance. Brain Topogr 34:442–460. https://doi.org/10.1007/s10548-021-00835-3
- Jawinski P, Markett S, Sander C, Huang J, Ulke C, Hegerl U, Hensch T (2021) The big five personality traits and brain arousal in the resting state. Brain Sci 11(10):1272. https://doi.org/10.3390/brainsci11101272
- Ke M, Li J, Wang L (2021) Alteration in resting-state EEG microstates following 24 hours of total sleep deprivation in healthy young male subjects. Front Hum Neurosci 15:636252. https://doi.org/10.3389/fnhum.2021.636252
- Khanna A, Pascual-Leone A, Michel CM, Farzan F (2015) Microstates in resting-state EEG: current status and future directions. Neurosci Biobehavioral Reviews 49:105–113. https://doi.org/10.1016/j.neubiorev.2014.12.010
- Korn U, Krylova M, Heck KL, Häußinger FB, Stark RS, Alizadeh S, Munk MH (2021) EEG-Microstates Reflect Auditory Distraction After Attentive Audiovisual Perception Recruitment of Cognitive Control Networks. Front Syst Neurosci 15:751226. https://doi.org/10.3389/fnsys.2021.751226
- Mazziotta J, Toga A, Evans A, Fox P, Lancaster J, Zilles K, Mazoyer B (2001) A probabilistic atlas and reference system for the human brain: International Consortium for Brain Mapping (ICBM). Philosophical Trans Royal Soc Lond Ser B: Biol Sci 356(1412):1293–1322. https://doi.org/10.1098/rstb.2001.0915
- Michel CM, Brunet D (2019) EEG source imaging: a practical review of the analysis steps. Front Neurol 10:325. https://doi.org/10.3389/fneur.2019.00325
- Michel CM, Koenig T (2018) EEG microstates as a tool for studying the temporal dynamics of whole-brain neuronal networks: a review. NeuroImage 180:577–593. https://doi.org/10.1016/j.neuroimage.2017.11.062
- Milz P, Faber PL, Lehmann D, Koenig T, Kochi K, Pascual-Marqui RD (2016) The functional significance of EEG microstates—Associations with modalities of thinking. NeuroImage 125:643–656. https://doi.org/10.1016/j.neuroimage.2015.08.023
- Näätänen R, Paavilainen P, Rinne T, Alho K (2007) The mismatch negativity (MMN) in basic research of central auditory processing: a review https://doi.org/10.1016/j.clinph.2007.04.026. Clinical neurophysiology, 118(12), 2544–2590
- Oostenveld R, Fries P, Maris E, Schoffelen JM (2011) FieldTrip: open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Computational intelligence and neuroscience, 2011, 1–9. https://doi.org/10.1155/2011/156869
- Pascual-Marqui RD (2002) Standardized low-resolution brain electromagnetic tomography (sLORETA): technical details. Methods Find Exp Clin Pharmacol 24(Suppl D):5–12
- Pascual-Marqui RD, Lehmann D, Faber P, Milz P, Kochi K, Yoshimura M, Kinoshita T (2014) The resting microstate networks (RMN): cortical distributions, dynamics, and frequency specific information flow. arXiv preprint arXiv:1411.1949
- Seitzman BA, Abell M, Bartley SC, Erickson MA, Bolbecker AR, Hetrick WP (2017) Cognitive manipulation of brain electric microstates. NeuroImage 146:533–543. https://doi.org/10.1016/j.neuroimage.2016.10.002
- Seitzman BA, Snyder AZ, Leuthardt EC, Shimony JS (2019) The state of resting state networks. Top Magn Reson imaging: TMRI 28(4):189. https://doi.org/10.1097/rmr.0000000000000214
- Schaefer A, Kong R, Gordon EM, Laumann TO, Zuo XN, Holmes AJ, Yeo BT (2018) Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity MRI. Cereb Cortex 28(9):3095–3114. https://doi.org/10.1093/cercor/bhx179
- Tadel F, Baillet S, Mosher JC, Pantazis D, Leahy RM (2011) Brainstorm: a user-friendly application for MEG/EEG analysis. Computational intelligence and neuroscience, 2011, 1–13. https://doi.org/10.1155/2011/879716
- Tait L, Zhang J (2022) MEG cortical microstates: spatiotemporal characteristics, dynamic functional connectivity and stimulus-evoked responses. NeuroImage 251:119006. https://doi.org/10.1016/j.neuroimage.2022.119006
- Tarailis P, Koenig T, Michel CM, Griškova-Bulanova I (2023) The functional aspects of resting EEG microstates: A Systematic Review. Brain topography, 1–37. https://doi.org/10.1007/s10548-023-00958-9
- Tarailis P, Šimkutė D, Koenig T, Griškova-Bulanova I (2021) Relationship between spatiotemporal dynamics of the brain at rest and self-reported spontaneous thoughts: an EEG microstate approach. J personalized Med 11(11):1216. https://doi.org/10.3390/jpm11111216
- Taulu S, Simola J (2006) Spatiotemporal signal space separation method for rejecting nearby interference in MEG measurements. Phys Med Biol 51(7):1759. https://doi.org/10.1088/0031-9155/51/7/008
- Tomescu MI, Rihs TA, Rochas V, Hardmeier M, Britz J, Allali G, Michel CM (2018) From swing to cane: sex differences of EEG resting-state temporal patterns during maturation and aging. Dev Cogn Neurosci 31:58–66. https://doi.org/10.1016/j.dcn.2018.04.011
- Valt C, Quarto T, Tavella A, Romanelli F, Fazio L, Arcara G, Bertolino A (2023) Reduced magnetic mismatch negativity: a shared deficit in psychosis and related risk. Psychol Med 53(13):6037–6045. https://doi.org/10.1017/s003329172200321x
- Wendel K, Väisänen O, Malmivuo J, Gencer NG, Vanrumste B, Durka P (2009) … & Grave de Peralta Menendez, R. EEG/MEG source imaging: methods, challenges, and open issues. Computational intelligence and neuroscience, 2009. https://doi.org/10.1155/2009/656092
- Zanesco AP, King BG, Skwara AC, Saron CD (2020) Within and between-person correlates of the temporal dynamics of resting EEG microstates. NeuroImage 211:116631. https://doi.org/10.1016/j.neuroimage.2020.116631
- Zappasodi F, Perrucci MG, Saggino A, Croce P, Mercuri P, Romanelli R, Ebisch SJ (2019) EEG microstates distinguish between cognitive components of fluid reasoning. NeuroImage 189:560–573. https://doi.org/10.1016/j.neuroimage.2019.01.067
Competing interest reported. A.T. has received lecture fees from Angelini. A.B. has received lecture fees from Otsuka, Janssen, and Lundbeck, as well as consultant fees from Biogen. G.P. has received lecture fees from Lundbeck. All other authors have no biomedical financial interests or potential conflicts of interest to report.