Introduction

Fragile X syndrome (FXS) is one of the most common causes of inherited intellectual disabilities (Verkerk et al. 1991). FXS is a trinucleotide repeat disorder in which the CGG repeat in the untranslated region of the Fragile X Messenger Ribonucleoprotein 1 (FMRP1) gene is abnormally expanded, which yields reduced production of Fragile X Messenger Ribonucleoprotein (FMRP) (Bagni et al. 2012; Verkerk et al. 1991). FMRP is an RNA-binding protein that is essential in regulating local protein synthesis and normal cell functions, and thus, it is critical for cognitive function (Bagni et al. 2013, Garber et al. 2008). The average intellectual quotient (IQ) in FXS is approximately 40–45 compared to the normed average of 100 in the general typically developing population (Sansone et al. 2014). Comorbidity with autism is common, and the reported prevalence rates range from 15 to 60% (Budimirovic et al. 2011). Anxiety and attention deficit disorders are also frequently observed in FXS (Boyle et al. 2010).

While there is currently no cure for FXS, treatments to help ameliorate behavioral problems are actively developed and tested, and EEG is considered an important tool for understanding the pathophysiology of FXS and the mechanisms underlying effects of treatment. Past studies have reported several differences in EEG characteristics of FXS individuals relative to control participants, including enhanced amplitudes of ERP responses (Ethridge et al. 2019, Knoth et al. 2012) and elevated gamma power combined with reduced phase-locking (Ethridge et al. 2017; Pedapati et al. 2022; Wang et al. 2017). The peak alpha frequency has been reported to be lower than control participants (Pedapati et al. 2022; Smith et al. 2021, Van der Molen et al. 2013) with reduced alpha-gamma coupling accompanied by increased theta-gamma coupling (Wang et al. 2017).

The current study examined EEG microstates at rest in a large sample of individuals with FXS and control participants. EEG microstates are defined based on the topography of the scalp electric field. Each microstate represents a brief period in which the scalp topography remains quasi-stable (40–120 ms) (Lehmann 1971), resulting from the transient co-activation of underlying neural generators. Consequently, each change in the scalp topography of a microstate signifies a change in the sets of neural generators (Lehmann et al. 1987; Vaughan 1982). Thus, EEG microstates capture non-stationary global activity patterns that reflect activities of distinct, widely distributed networks (Khanna et al. 2015, Michel et al. 2018). EEG microstate analysis can therefore complement past ERP and spectral studies that emphasize average signal characteristics over time. To our knowledge, this is the first study on EEG microstates in FXS.

Each microstate is considered to represent a specific global brain state that has functional relevance for cognition. For instance, microstates show orderly transitions that correspond to cognitive states that are experienced by participants (Pipinis et al. 2017; Zanesco et al. 2021a). Different microstates are observed during visual and verbal processing (Lehmann et al. 1998, 2010; Milz et al. 2016). Microstates observed during the pre-stimulus period predict performance on cognitive tasks, including accuracy and error (Britz et al. 2010a, Britz et al. 2014) and reaction time (Zanesco et al. 2020a). The prominence of certain microstates also has been reported to change by arousal levels or sleepiness (Comsa et al. 2019). The suggested neural sources for EEG microstates include key structures of resting-state fMRI networks, such as the default mode, attention, and salience networks (Bréchet et al. 2020; Britz et al. 2010b; Custo et al. 2017; Musso et al. 2010; Yuan et al. 2012). Further, EEG microstates have been shown to predict fluctuations in fMRI dynamic functional connectivity with 90% accuracy (Abreu et al. 2020).

Past studies have shown that microstate temporal parameters (e.g., duration and occurrence) are altered in several clinical conditions, including autism, schizophrenia, and dementia (Al Zoubi et al. 2019; D’Croz-Baron et al. 2019, Jia et al. 2019, Lehmann et al. 2005, Musaeus et al. 2019, Nagabhushan Kalburgi et al. 2020, Smailovic et al. 2019, Takarae et al. 2022). The temporal parameters of microstates can be interpreted neurophysiologically and may implicate specific neural system impairments. For instance, a shorter duration of a microstate may reflect premature termination or problem sustaining activities of certain neural generators, and reduced occurrence may indicate difficulty coordinating activities of a specific set of neural generators that give rise to a certain microstate (Khanna et al. 2015). Alterations in transition patterns between microstates may indicate difficulty invoking or over-relying specific encoded sequential activation of neural assemblies (Khanna et al. 2015). Further, variations in the microstate topography could imply subtle structural or functional disturbances within a specific network (Murray et al. 2008; Vaughan 1982).

Temporal parameters of microstates also distinguish between demographic characteristics, such as age and sex (Koenig et al. 2002; Zanesco et al. 2020b). Sex differences may be particularly important when considering FXS because FMRP1 is located on the X chromosome. The inheritance pattern follows the X chromosome linked dominance model, and the affected X chromosome is passed from the mother onto the offspring (Bartholomay et al. 2019). Because females have two X chromosomes, the unaffected X chromosome from the father allows some production of FMRP, and this results in milder symptoms for females with FXS. Accordingly, milder, intermediate level alterations in EEG characteristics have been reported in females (Pedapati et al. 2022; Smith et al. 2021). It is, however, not clear if the milder alteration may have network specificity, affecting only specific microstates.

We present a study of resting EEG microstates in a large sample of individuals with FXS and control participants. We examined group differences in temporal characteristics of EEG microstates and compared microstate topographies and transition patterns. We further examined relations between microstate characteristics and clinical/demographic characteristics, such as IQ and sex. Finally, we examined EEG spectral characteristics to connect our work with prior studies of FXS examining oscillatory properties of resting EEG.

Materials and Methods

Participants

Resting-state EEG data from 141 participants (70 FXS participants (32 females, 38 males) and 71 typically developing control (TDC) participants (30 females, 41 males) were included in this study. The following analyses were performed using publicly available data (https://zenodo.org/record/7149276), and additional data, such as chronological ages, were provided by the research group that conducted the original work. Details regarding recruitment and preprocessing are provided in published work (Pedapati et al. 2022; Smith et al. 2021). Briefly, all FXS participants were recruited through a federally established regional FXS center at a tertiary hospital in the USA and have full-mutation FXS that was confirmed by Southern Blot and PCR analyses (Pedapati et al. 2022; Smith et al. 2021). Only eyes-open resting EEG was collected during passive viewing of a silent movie that was standardized across participants (Pedapati et al. 2022; Smith et al. 2021). Six participants (4 FXS and 2 TDC participants) were excluded from the original sample because of excessive residual artifacts, and two participants (1 FXS and 1 TDC) were excluded because they had less than 2 min of usable data after preprocessing. The remaining 65 participants with FXS and 68 TDC participants were included in the further analysis. Spectral decomposition was performed with the downloaded data before additional preprocessing steps were applied for microstate analysis.

The total duration of the preprocessed data was similar between groups, with means(sd) of 257.88(38.14) and 263.97(30.14) seconds for the FXS and TDC groups respectively, t(120) = 1.01, n.s. Groups did not significantly differ based on chronological age or sex ratio but differed significantly by non-verbal IQs (NVIQ). Table 1 summarizes the clinical and demographic information for FXS and TDC groups.

Table 1 Clinical and demographic characteristics of participants included in the microstate analysis
Full size table

Microstate Analysis

Topographic clustering was employed to determine the optimal number and configuration of microstate topographies. Procedures and parameters for the first- and second-level clustering closely followed published work (Takarae et al. 2022; Zanesco et al. 2020b, 2021b).

Additional Preprocessing for Microstate Analysis

The downloaded data were low-pass filtered at 40 Hz, and the channel-wise mean of the entire data period was used to demean the signal for each channel. Then, Cartool (Brunet et al. 2011) was used to spatially smooth individual EEG using inter-septile weighted means (Michel et al. 2019).

First-level Clustering

Topographical maps were extracted at peaks of the global field power (GFP) during the first-level clustering using Cartool (Brunet et al. 2011). GFP peaks reflect moments of high neural synchronization, which provides the optimal characterization of the momentary, quasi-stable activation of large-scale networks (Zanesco 2020). Topographical maps at the GFP peaks were then entered into a modified k-means clustering for each participant. The optimal number of clusters was selected through an iterative process using the following method. A subset of k (varied from 1 to 12) maps was randomly selected from all topographical maps in the participant data as initial centroids for clustering. Spatial correlations between the k centroid maps and the remaining maps were computed. The maps were assigned to the centroid with which they had the highest spatial correlation, resulting in k clusters of maps. Maps were not assigned to a centroid if the highest correlation was lower than 0.5. The polarity of the maps was ignored by correcting the sign of the spatial correlation.

After all maps were assigned to a cluster, new centroid maps were created by averaging the constituent maps assigned to the given k clusters. Any remaining maps that were not assigned to a cluster in the prior iteration were compared to the newly computed centroids and assigned again based on the correlation criterion. This process continued iteratively until the global explained variance (GEV) between the average centroids and the maps converged to a limit. For each k (1 to 12), this procedure was repeated 100 times, with a new subset of k centroids selected for each iteration. The optimal number of clusters was determined by a meta-criterion defined by 7 independent criteria (Bréchet et al. 2019; Custo et al. 2017).

Second-level Clustering

Once the first-level clustering was completed for each participant, all clusters from the participants were entered into a second-level clustering, and the second-level clustering resulted in the sample-level, global clusters. A set of k (1 to 15) maps was randomly selected to use as initial centroids for clustering. As for the first-level clustering, maps were assigned to a centroid only if the correlation to the centroid was greater than 0.5, and the polarity of the maps was ignored. For each level of k, 200 iterations were run, until the GEV converged to a limit. Then, the k centroid solution was selected based on the same meta criterion used in the first-level clustering (Bréchet et al. 2019, Custo et al. 2017).

Fitting

Finally, the global clusters were fitted to the individual EEG data based on spatial correlations between the EEG topography at each time point and global clusters. Each time point in the individual EEG was labeled with the best matching global cluster. This process resulted in a time series of microstates for the participant. Time points that had lower than 0.5 correlations with any of the global clusters were ignored. Microstate segments that were less than 6 samples (~ 24 msec) were ignored, and the time points were split between the preceding and subsequent microstates in the series.

Microstate Parameters and Transition Probabilities

The following parameters were calculated for each microstate for each participant using the microstate time series: global explained variance (GEV), mean duration, occurrence, and mean global field power (GFP). GEV is the percentage of the observed topographic variance of the individual’s microstate time series explained by a specific microstate. Mean duration is the average of each occurrence duration of the specific microstate. Occurrence refers to the frequency of occurrence of a specific microstate in events per second. Mean GFP is the normalized average of GFP from the microstate time series that corresponds to the average strength of the electric field for a specific microstate.

The first-order Markov-chain transition probabilities were estimated to examine if there was any increase/decrease in transitions from one specific microstate to another. Cartool was used to calculate the observed transition probabilities as well as the chance probabilities which assumed that all transitions are random. The observed probabilities were first compared to the chance probability via permutation tests using the coin package in R with Bonferroni correction. The transition probabilities were further analyzed if the observed probability statistically differed from the chance in at least one of the groups.

The microstates parameters and transition probabilities were analyzed with linear mixed models using the lme4 package in R. Group and microstate type (e.g., microstate topographic configuration A) or transition pairs (e.g., transition from microstates A to B) were entered as fixed effects, and the model parameters were estimated using maximum likelihood. Random subject intercepts were included in all models to represent between-participant variability. Degrees of freedom were estimated using Satterthwaite approximation, and Type-III tests for fixed effects were performed. Additional pairwise posthoc tests via the phia package in R were performed for significant effects to aid interpretations where appropriate. Bonferroni correction was used for the post-hoc tests, and there were 15 comparisons for each linear model with temporal parameters. The post-hoc tests for transition probabilities were corrected for 23 comparisons. Correlations between microstate parameters and clinical/demographic variables were also examined with Bonferroni corrections. Because the clinical correlations were examined only for selected microstates (see below), there were 12 correlations in each clinical/demographic domain.

Topographical Analysis

Topographic dissimilarities were compared between groups (FXS vs. TDC) using multivariate distance matrix regression (MDMR) (McArdle et al. 2001, McArtor et al. 2017) with the MDMR package in R (McArtor 2018). MDMR is a person-centered regression method for estimating statistical associations between multivariate outcomes and categorical or continuous predictors based on dissimilarities among sets of data (McArtor et al. 2017). MDMR has been extended to the mixed-effects modeling framework to account for dependent or hierarchically nested data (McArtor et al. 2017; McArtor 2018).

Topographic dissimilarities between the map centroids from individual-level clustering were quantified by Euclidean distance. The resulting Euclidean distances were compared between groups and microstate types using MDMR. Map topographies were included in the model as nested within individuals when participants had more than one map for a particular microstate type. Analytic p-values are reported for MDMR significance tests (McArtor et al. 2017). A measure of variance among topographies was calculated as the map discrepancy (Studer et al. 2011). Map discrepancy is conceptually akin to a measure of the spatial standard deviation and reflects the average dissimilarity of all topographies within a grouping to the multivariate center of that grouping. A pseudo-R2 was also estimated based on partitions of sums of squares of dissimilarities based on the fixed effects MDMR model and represents the proportion of total discrepancy among sequences accounted for by covariates.

Relationship to Individual Alpha Power

Previous studies have shown that microstate parameters are strongly correlated with alpha power (Croce et al. 2020), and previous studies of FXS have reported alterations in alpha activity in FXS (Pedapati et al. 2022; Smith et al. 2021; Wang et al. 2017). Thus, our analysis below on spectral characteristics focuses on the alpha band. We first estimated the individual alpha frequency (IAF) by finding the center of gravity in 7–14 Hz to derive the IAF peak, then we set the lower and upper bounds by respectively multiplying the peak by 0.6 and 1.2 (Klimesch 1999; Saggar et al. 2012). Spectral decomposition was performed using spectopo function in EEGLAB (Delorme et al. 2004), and IAF power was extracted using the individually set window. IAF peak and power were compared between groups using a t-test with Welch correction for degrees of freedom in R. Correlations between microstate temporal parameters and IAF variables were Bonferroni corrected, with 36 correlations each for the IAF peak and power related analyses.

Results

Microstate Parameters

The k-means clustering analysis resulted in 6 global clusters, which accounted for 82.60% of the topographic variance. The global clusters are shown in Fig. 1. Our 6 global clusters showed a close resemblance to previous work from our group and others that used the same method and data-driven criteria but used different independent samples (Custo et al. 2017; Takarae et al. 2022; Zanesco et al. 2020b). Following the naming convention in the field, microstates A through D were labeled based on similarities to the classic microstate topographies that have been replicated in previous studies (Koenig et al. 2002, Michel et al. 2018). Microstates E and F were named based on the labels used for these configurations in our prior work (Takarae et al. 2022; Zanesco et al. 2020b). After fitting the global clusters to individual EEG data, the percentages of samples that were assigned the global microstates were similar between groups, t = 1.41, n.s., means(sd): M = 79.71%(10.36) for FXS and M = 82.06%(8.76) for TDC.

Fig. 1
figure 1

Voltage scalp maps of the global clusters obtained in the current study. Each map shows the centroid of each cluster. Voltage maps are 2D isometric projections with the nasion oriented towards the top of the map

Full size image

In the primary analysis of microstate parameters presented below, the effect of microstate type (A through F) was significant in all the analyses. However, because the primary focus of this study is the differences relating to the FXS diagnosis, we focus on the main effects of group and the interaction between group and microstate type. The effects of microstate type are not discussed in the subsequent text. Means, standard deviations, and standardized effect sizes of group differences on microstate parameters are presented in Tables 2 and 3, respectively.

Table 2 Means and standard deviations for temporal parameters of each microstate by group. Standard deviations are presented in parentheses
Full size table
Table 3 Standardized effect size d for FXS vs. TDC comparisons for each microstate. Positive values indicate higher values in FXS, and negative values indicate higher values in TDC.
Full size table

Global Explained Variance (GEV)

There was no significant overall group difference in GEV, but there was a significant interaction of group and microstate type (Fig. 2), F (5, 798) = 6.71, p < .0001. The interaction effect was driven by greater GEV for microstates C and D in the FXS group compared to the TDC group, whereas GEV was lower for other microstates in the FXS group. The group difference for microstate C was statistically different than microstates A, E, F (p < .01), and B (p < .05) after Bonferroni correction. The group difference for microstate D also significantly differed from microstates A, E (p’s < 0.01), and F(p < .05) after Bonferroni correction.

Fig. 2
figure 2

Global explained variance (GEV) for each microstate by group. Group means and standard errors of the mean are presented. Asterisks indicate significant differences from microstates C (black) or D (light blue)

Full size image

Mean Duration

There was a significant group x microstate type interaction, F(5, 665) = 3.46, p < .01, but no significant main effect of group. The group difference for microstate D was significantly different from those for microstates A and E (p’s < 0.05, Bonferroni corrected) (Fig. 3).

Fig. 3
figure 3

Mean duration of each microstate by group. Group means and standard errors of the mean are presented. Asterisks indicate significant differences from microstate D (light blue)

Full size image

Occurrence Per Second

The microstate occurrence did not significantly differ between groups, but there was a significant group by microstate type interaction (Fig. 4), F (5, 665) = 4.62, p < .001. Microstates C and D occurred more frequently in the FXS group than the TDC group, while other microstates were less frequent in the FXS group. The post-hoc comparisons with Bonferroni correction suggest that the group difference in the occurrence of microstate C was statistically different from microstate A (p < .05), and the group difference for microstate D was different from the effects for microstates A (p < .01), E, and F (p’s < 0.05).

Fig. 4
figure 4

Occurrence per second for each microstate by group. Group means and standard errors of the mean are presented. Asterisks indicate significant differences from microstates C (black) or D (light blue)

Full size image

Mean Global Field Power (GFP)

There was no significant effect of group, but there was a significant group by microstate type interaction (Fig. 5), F (5, 665) = 8.40, p < .0001. Group differences for the mean GFP were greater for microstates C and D with higher mean GFP for the FXS group, while group differences for other microstates were much smaller. The post-hoc analysis suggested that the group difference for microstate C was significantly different from microstates A (p < .01) and E and F (p’s < 0.05). The group difference for microstate D significantly differed from the effects for microstates A, E, and F (p’s < 0.001) and B (p < .01).

Fig. 5
figure 5

Mean global field power (GFP) for each microstate by group. Group means and standard errors of the mean are presented. Asterisks indicate significant differences from microstates C (black) or D (light blue)

Full size image

Transition Probabilities

Although the overall group difference was not significant, the group by transition pairs interaction was significant, F(22, 2926) = 3.12, p < .0001. The post-hoc analysis revealed that the group difference in the B to D transition probability was significantly different compared to transitions from A to F (p < .05) after Bonferroni correction. While the B to D transition occurred more frequently in the FXS group (means(sd): M = 0.13(0.07) for FXS and M = 0.10(0.06) for TDC), the A to F transition occurred less frequently in the FXS group than in the TDC group (M = 0.06(0.04) for FXS and M = 0.08(0.04) for TDC).

Topographical Dissimilarity Analysis

MDMR indicated no main effect of group, but a significant interaction between group and microstate (p < .001) was observed. We followed up on this finding by comparing groups for each separate microstate. This process resulted in a significant effect of group for microstate B (p < .05), and microstate D (p < .05) (Fig. 6), and the analytic p values were confirmed using permutation tests for both microstates with 1,000 permutations each. Group explained 3.01% (pseudo-R2 = 0.0301) of the variance in dissimilarities for topographies in the cluster for microstate B and 3.08% (pseudo-R2 = 0.0308) for microstate D. The relatively low R2 values suggest that, although statistically significant, there were only minor topographic group differences for both microstates. There was less discrepancy among maps for microstate B for FXS participants (discrepancy = 0.157) than controls (discrepancy = 0.186). A similar pattern was observed for microstate D with a slightly lower discrepancy in the FXS group (discrepancy = 0.172) than in the TDC group (discrepancy = 0.187). The topographic discrepancy suggests that the microstate configurations were slightly more homogeneous for these microstates among FXS participants than the TDC participants.

Fig. 6
figure 6

Symmetrical matrices showing the pairwise Euclidean distances between microstate topographies for microstate B (top left) and D (top right). Pairwise distances between topographies vary from 0 (most similar – blue) to 1 (most dissimilar – red) and are organized by group. The average topographic map for each microstate configuration is shown below each distance matrix as a 2D topographic representation

Full size image

Sex Differences

Linear mixed models to examine the effects of group and microstate type were re-run after adding sex as an additional factor. Because of overlaps with the previously presented 2-factor group x microstate type models, we report only the 3-way group by sex by microstate type interaction effect. The 3-way interaction was significant for mean GFP, F(5, 665) = 2.24, p < .05 (Fig. 7A), and there was also a trend toward significance for GEV, F(5, 798) = 2.00, p = .076. Because the trend for GEV is likely to reflect the same functionally related neural system impairments as reflected by differences in mean GFP, only the mean GFP model was further examined. When only male participants were entered into the group x microstate type model, the group x microstate type interaction effect was significant, F(5, 380) = 8.00, p < .0001 (Fig. 7B). The post-hoc analysis identified significant group differences for microstates C and D (microstate C vs. A (p < .001), B (p < .05), E (p < .001), F (p < .001), and microstate D vs. A (p < .001). E (p < .05), F (p < .05)). The female only model also showed a significant group by microstate type interaction, F(5, 285) = 5.35, p < .001 (Fig. 7B). The post-hoc analysis, however, only indicated significant differences for microstate D compared to other microstates (D vs. A (p < .001), B (p < .01), C (p < .05), E (p < .001), and F (p < .01)). When participants were separated by diagnostic group, the FXS only model showed a significant sex by microstate type interaction (F(5, 325) = 6.84, p < .0001) with the post-hoc tests indicating differences involving microstates C and D (microstate C vs. A (p < .01), B (p < .05), E (p < .01), and F (p < .05), microstate D vs. A (p < .01), B (p < .01), E (p < .01), F (p < .05)). In contrast, the TDC only model showed a significant sex by microstate type interaction (F(5, 340) = 10.96, p < .0001) with post-hoc significant differences involving only microstate D (p’s < 0.0001 for differences from all other microstates).

Fig. 7
figure 7

(A) Mean global field power (GFP) of each microstate by group and sex. Means and standard errors of the mean are presented. B) Mean global field power (GFP) of each microstate by group for Female/Male or FXS/TDC participants only. Means and standard errors of the mean are presented. Asterisks indicate significant differences from microstates C (black) or D (light blue)

Full size image

Relationship to Alpha Band Activity

The power spectrum and detailed characteristics of alpha band activities including source estimations have previously been reported using the same sample (Pedapati et al. 2022; Smith et al. 2021). Thus, we only report mean characteristics and focus on relationships to microstate parameters in the current study. Individual alpha frequency (IAF) peaks were lower in the FXS group than in the TDC group, t(129) = 7.24, p < .0001, (means(sd): M = 9.38 Hz (0.44) for FXS and M = 9.91 Hz (0.41) for TDC). The IAF power across all electrodes was significantly higher in the FXS than TDC group, t(130) = 2.78, p < .01 (means(sd): M = 0.65 dB (3.22) for FXS and M = -0.94 dB (3.35) for TDC). We calculated correlations for the mean duration, occurrence, and mean GFP of microstates with individual alpha peak frequency and power (Fig. 8). Correlations involving GEV were not calculated because of the GEV’s dependency on other microstate parameters.

Mean microstate duration was longer in individuals with higher IAF peak frequency for most microstates in both groups: A (rho = 0.53, p < .001), B (rho = 0.46, p < .01), C (rho = 0.38, p = .07), E (rho = 0.46, p < .01) and F (rho = 0.50, p < .001) in the FXS group, and B (rho = 0.48, p < .01), D (rho = 0.41, p < .05), E (rho = 0.39, p < .05), and F(rho = 38, p = .06) in the TDC group, after Bonferroni correction. In addition, the occurrence and mean GFP for microstate D were significantly correlated with IAF peak frequency in the FXS group (rho = -0.47, p < .01 for occurrence and rho = -0.46, p < .01 for mean GFP), but not in the TDC group. Lower IAF peaks were associated with higher mean GFP and more frequent occurrence for microstate D in the FXS group (Fig. 8). No other correlations involving IAF peaks were significant.

Fig. 8
figure 8

Scatter plot for the IAF peak frequency with occurrence (left) and mean GFP (right) of microstate D

Full size image

Longer mean durations were generally associated with lower amplitude alpha activity: microstates A (rho = -0.58, p < .0001), B (rho = -0.49, p < .01), C (rho = -0.51), p < .001), E (rho = -0.46, p < .01), and F (rho = -0.53, p < .001) in the FXS group, and A (-0.43, p < .05), B (rho = -0.42, p < .05),C (rho = -0.42, p < .05), D (rho = -0.42, p < .05), E (rho = -0.46, p < .01), and F (rhp = -0.41, p < .05) in the TDC group. Additionally, there was a trend toward significance for the occurrence of microstates B (rho = 0.38, p = .053) and F (rho = 0.38, p = .053), only in the TDC group. No other correlations were significant.

Relationship to Other Clinical and Demographic Variables

Because most of the significant group differences in the primary analysis were related to microstates C and D, we correlated their mean duration, occurrence, and mean GFP with NVIQ and age to aid the interpretations of the findings.

NVIQ-related Changes

There was a significant negative correlation between NVIQ scores and the mean GFP of Microstate C, suggesting that lower values of GFP for microstate C are associated with higher NVIQ scores in the FXS group (Spearman’s rho = -0.42, p < .05 after Bonferroni correction) (Fig. 9). Other correlations were not significant.

Fig. 9
figure 9

Scatterplot for the relationship between non-verbal IQ and mean GFP for microstate C

Full size image

Age-related Changes

Chronological age was positively correlated with the mean duration of microstates C and D in both groups, though the correlations were higher in the TDC group than the FXS group (microstate C: rho = 0.435, p < .01 for FXS group and rho = 0.672, p < .0001 for TDC group, microstate D: rho = 0.540, p < .0001 for FXS and rho = 0.789, p < .0001 for TDC). Age was also positively correlated with the mean GFP of microstate D in both groups (rho = 0.378, p < .05 for FXS and rho = 0.379, p < .05), but not with microstate C.

Discussion

Despite substantial group differences in functional outcomes (e.g., IQ) between participants with Fragile X syndrome (FXS) and typically developing participants, we did not find large, extensive disruption in the dynamics of all microstates. Rather, group differences were specific to the activity of particular brain circuits each reflecting distinct microstates. We observed significant group by microstate type interactions for GEV, mean duration, occurrence, and mean GFP in a series of linear mixed models. The interaction was driven by group differences in microstate C and D temporal parameters. In addition, transition pattern analysis suggests that the probability of microstates B to D transitions increased in the FXS group compared to the TDC group. Likewise, the topographic configuration of microstates B and D differed between groups, and the topography of microstates B and D was less heterogeneous in the FXS group.

Sex differences between male and female individuals with FXS were also observed for microstates C and D. While past studies have shown sex differences in mean duration and occurrence in certain microstates, including C and D, in typically developing individuals (Bagdasarov et al. 2022; Tomescu et al. 2018; Zanesco et al. 2020b), we did not observe group by sex by microstate type interactions for duration and occurrence, suggesting that sex differences in neurophysiological processes that are related to these parameters may not be strongly influenced by FXS. We, however, observed a significant group by sex by microstate type interaction for mean GFP. While the prior studies investigating sex differences did not include mean GFP measures, the mean GFP is considered to reflect the degree of neural synchronization among the neural generators of the network, contributing to the overall strength of the scalp electric field (Khanna et al. 2015). Potential excitation to inhibition imbalances in FXS (Nomura 2021; Paluszkiewicz et al. 2011; Pedapati et al. 2022) could affect mean GFP because of the proposed role of inhibition in neural synchronization (Adesnik 2018; Chauhan et al. 2018). Thus, the mean GFP could be sensitive to the severity and degree of neurophysiological alterations within FXS; Certain aspects of the alterations could be associated with sex because of the male-female difference in FMRP production in FXS.

Correlational patterns involving clinical and demographic variables suggest that alterations for microstates C and D may relate to different aspects of cognition in FXS. Microstate C was more prominent in the FXS group than the TDC group, as observed as elevated GEV, occurrence, and mean GFP. The greater mean GFP of microstate C was associated with lower NVIQ scores in FXS. The increase in mean GFP appeared specific to male participants who tend to have more profound intellectual impairments. Intellectual impairments in FXS are highly correlated with FMRP levels (Kim et al. 2019), which are generally lower in males. Thus, the unusually high neural synchronization associated with microstate C may be particularly relevant to the intellectual impairments among FXS. Microstate C has been associated with the default mode network through source localization studies, and default network maturation has been associated with intellectual development in developmental disabilities, including FXS (Hall et al. 2013; Lee et al. 2022).

The unusually high neural synchronization is also reminiscent of the previously reported greater sensory-related ERP amplitude in FXS (Ethridge et al. 2019, Knoth et al. 2012). ERPs are generated via synchronization of neuronal activities in response to sensory stimulation and have strong representations of lower frequency signals (Makeig et al. 2002). The current data add that neural synchronization in lower frequencies may also be increased at rest in FXS within a specific network. In contrast, previous studies have suggested that evoked responses in higher frequency ranges may be reduced in FXS (Ethridge et al. 2017, 2019). Further investigations are needed to clarify frequency specific alterations of neural synchronization in FXS.

Temporal parameters for microstate D showed different patterns of group differences and correlations compared to microstate C. The GEV, mean duration, occurrence, and mean GFP were elevated in the FXS group compared to the TDC group with respect to this microstate. The mean GFP was greater in both male and female FXS participants compared to their TDC counterparts, and the group difference was similar between male and female participants. There was no significant correlation between NVIQ and microstate D parameters, suggesting that it may not relate to general intellectual disabilities.

Source localization studies have associated microstate D with the frontoparietal attention network (Britz et al. 2010b; Custo et al. 2017), and its parameters change as a function of attention demands in the task and arousal levels (Seitzman et al. 2017). Attention impairments are frequently observed and considered prognostic of FXS (Scerif et al. 2012). While the occurrence of microstate D was overall higher in the FXS group compared to the TDC group, the transition analysis suggests that specific transitions from microstate B to D may contribute to the increased occurrence of microstate D in the FXS group. It appears that engagement of attention control (microstate D) following sensory processing (microstate B) is more critical to perform the resting state task (i.e., passively viewing a silent movie) in the FXS group.

The occurrence and mean GFP of microstate D were significantly correlated with the individual alpha peak frequency. Alpha activities play an essential role in generation of microstates, and early depictions of microstates were conducted using alpha band EEG (Lehmann 1971). Microstate parameters are most strongly correlated with alpha power while the correlations with other frequency bands are much weaker (Croce et al. 2020). It has been shown that the periodicity of microstates occurs twice the alpha frequency (Lehmann 1971; von Wegner et al. 2017) and is associated with the phase of resting alpha oscillations (von Wegner et al. 2021). The thalamus plays a critical role in generating cortical alpha (Halgren et al. 2019), and structural alterations in the thalamus have been reported in FXS. For instance, reduced gray matter density in the thalamus (Hall et al. 2013) and lower fractional anisotropy in regions connecting the thalamus and neocortical regions (Swanson et al. 2018) have been documented. These abnormalities may yield changes in both alpha rhythm and temporal dynamics of microstates in FXS.

While the alpha band is typically defined as 8-12 Hz, its peak frequency is known to vary across individuals. Alpha peak frequencies generally increase from childhood to adolescence (Freschl et al. 2022) then decrease through adulthood (Klimesch 1999). Further, notable shifts in the alpha peak frequency (thalamocortical dysrhythmia) have been reported in several clinical disorders, including epilepsy, tinnitus, and depression (Ridder et al. 2015; Llinás et al. 1999, 2005). In some of these conditions, such as tinnitus, it has been suggested that deafferentation of sensory input reduces the information flow to the thalamus, which results in the slowing of the alpha rhythm (Vanneste et al. 2018). It has been shown that the alpha peak shifts towards the theta frequency in FXS (Van der Molen et al. 2013, Smith et al. 2021) as well as showing more anterior localization (Pedapati et al. 2022). The slowing of the alpha rhythm may alter optimal temporal windows for inhibitory control over sensory input, which could result in less efficient top-down modulation or increased efforts to achieve such modulation. Though speculative, the aforementioned increased occurrence of B to D transitions may reflect compensation for such inefficiency.

Limitations

One important limitation of the current study is that EEG was only collected during an eyes-open resting condition. Temporal parameters of microstates during eyes-open and -closed could differ (Zanesco et al. 2020b), and previous studies of developmental disabilities have found differences between eyes conditions in EEG microstates (Nagabhushan Kalburgi et al. 2020) and fMRI-based connectivity (Nair et al. 2018). Thus, exploring differences by the eye condition in resting state EEG is an important topic in future research of FXS.

EEG in the current study was recorded while the participant passively viewed a silent movie. Movies are often used to obtain data from young participants or those with developmental disabilities when participants’ cooperation is otherwise difficult to obtain. While the microstate topographies observed in the current study closely resemble previously published work, the use of the movie may have affected the relative occurrence, duration, and strength of the signal for each microstate. Likewise, the sensory stimulation of the movie may have interacted with the attentional and intellectual capacities of the participants, influencing patterns of microstate dynamics that are functionally associated with cognition. Thus, the generalization from and to other studies may be limited.

While our sample was relatively large for a study of developmental disabilities, the sample size was not sufficient to perform a detailed characterization of subgroups. Previous studies have reported sex and age differences in microstate parameters, as well as interactions of these demographic variables (Bagdasarov et al. 2022; Koenig et al. 2002; Tomescu et al. 2018; Zanesco et al. 2020b). Investigating these demographic variables is important in the context of X chromosome-linked developmental disabilities. Our sample was, however, not large enough to investigate age-related changes beyond a preliminary check to potentially detect gross changes in developmental trends. The sample was also not sufficient to investigate the potentially complex age-sex relationships. Limited data were also available to examine behavioral and other cognitive correlates to allow deeper insights into the functional relevance of microstates for understanding the pathophysiology of FXS.

Conclusion

Our data showed specific alterations of microstate C and D parameters in FXS participants relative to age-matched TDC participants. Specific alterations in these microstates are potentially related to two of the most prominent cognitive characteristics of FXS: intellectual disabilities and attention impairments. These findings suggest that microstate parameters could serve as clinically relevant markers to investigate cognitive phenotypes and evaluate treatment-related changes. Slowing of the peak IAF and its correlation to microstate D parameters may also suggest changes in the thalamocortical loop in FXS, which may be related to attention control.