Introduction

Modern human neuroscience demonstrates that cognitive and behavioral functions emerge from the dynamic interactions of distributed and interconnected brain networks1,2. Functional connectivity (FC), defined as the spontaneous coactivity of distinct brain regions, is typically studied using resting-state functional magnetic resonance imaging (rfMRI), which measures spontaneous BOLD fluctuations in the absence of external task demands3. A seminal shift occurred in the early 2000 s with the discovery of the default mode network (DMN), a set of brain regions more active at rest than during task execution, and strongly involved in self-referential, introspective, and memory-related processes4,5. Since then, research has consistently revealed stable and replicable patterns of intrinsic brain organization, collectively referred to as resting-state networks, which closely resemble the brain’s task-evoked architecture6,7,8. These networks are now considered a core scaffold supporting human cognition9, capable of predicting individual differences in task performance10,11 and evoked brain activity12.

Resting-state networks show remarkable anatomical consistency, with large-scale parcellations identifying up to 98 distinct networks13,14,15,16. However, efforts to standardize these17,18,19,20,21,22 have led to the convergence around a more tractable set of core functional networks (typically 5–8 systems)13,16,21,23,24. These include the DMN, associated with self-reference, memory, and social cognition25; several executive control networks26,27 such as the salience network (SN), involved in filtering salient stimuli and integrating sensory, emotional, and cognitive information28; the frontoparietal network (FPN), supporting flexible cognitive control29,30; the dorsal attention network (DAN), specialized in visuospatial attention31,32; and the cingulo-opercular network (CON), implicated in decision-making and conflict monitoring33. Additional domain-specific systems include the language network (LAN)34,35,36, the somatomotor network (SMN), supporting sensory-motor and auditory functions, and the visual network (VN), involved in visual processing21.

Importantly, beyond reflecting stable neurocognitive traits, resting-state FC also tracks dynamic fluctuations37 in arousal38, autonomic regulation39, and spontaneous mentation40. Whole-brain FC patterns can discriminate between distinct mental contents (including autobiographical thought, inner speech, or music imagery41,42) and relate to fluctuations in self-reported experience43,44,45. This dynamism has prompted a shift from viewing rest as a monolithic state to a more nuanced perspective, in which resting cognition is understood as emerging from the continuous reconfiguration of network coupling. More specifically, it reflects a balance between functional segregation, i.e., strong interaction within networks and minimal interaction across them, and functional integration, characterized by increased interaction and information exchange between distinct networks46.

Recent work has identified distinct recurring patterns of spontaneous thought at rest measured in large samples through self-report and clustering analyses43,47,48. These resting-state thought profiles (RSTPs) reflect variations in richness, modality, and focus of internal experience. Yet, most studies exploring the neural basis of spontaneous thought at rest have relied on small samples (N < 300), and provided only partial insight into how distinct spontaneous thought profiles relate to functional integration and segregation dynamics40,43,44,48,49,50. To address these limitations, we investigated FC in a large sample of 1,717 participants in relation to spontaneous thought across six previously defined RSTPs47: two Evocative profiles (Preoccupied and Reverie), one Fluctuating, one Vigilant, and two modality-predominant profiles (Visual and Verbal), spanning the full diversity of mental resting-state experience. We hypothesized that sparser thought profiles would be associated with greater network functional segregation, while richer, more complex profiles would involve enhanced network functional integration. 

FC changes were interpreted in the specific context of rest. Within-network desynchronization (i.e., decreased FC) was taken as a marker of functional segregation, reflecting more differentiated processing within a given network. Conversely, between-network desynchronization was interpreted as a marker of functional integration, reflecting the flexible coordination and interplay between distinct networks. While increased FC is typically associated with greater functional engagement in studies using delayed cognitive assessments51, we argue that, during rest, decreased FC may instead reflect the active recruitment of specific networks involved in spontaneous cognitive processes. This interpretation aligns with empirical findings demonstrating that brain regions exhibiting strong task-evoked activity tend to display low intrinsic connectivity at rest, suggesting a functional dissociation between activation and connectivity52,53. Moreover, task-based dynamic FC studies indicate that decreases in FC can reflect context-specific reorganization or the selective engagement of neural populations, rather than simple network disengagement54. Specifically, such decreases may support functional specialization55 or facilitate selective information routing during cognition56. Finally, theoretical models describing resting-state FC as a dynamic system of transient coupling and decoupling, rather than a static scaffold, further support the notion that desynchronization can accompany active cognitive processing57,58,59.

Our analysis proceeded in two main steps. First, we identified and characterized seven core functional networks using the GINNA atlas, a cognitively annotated functional atlas that enables the identification of functional networks with known psychological correlates18. Second, we assessed both within- and between-network FC differences across the six RSTPs. These profiles were previously derived using a multidimensional clustering strategy applied to subjective reports of thought content at rest47. FC was computed using Fisher’s r-to-z transformed correlation coefficients across all pairwise regions. We analyzed the effects of RSTPs on FC using two mixed-effects models (one for within-network, one for between-network FC), including participant as a random effect.

This approach allowed us to estimate how frequently each core network was implicated in significant FC differences for each RSTP, and which specific connectivity edges defined shared vs. profile-specific patterns of functional integration and segregation. These analyses revealed consistent FC patterns for each spontaneous thought profile, highlighting how network-level reorganization underlies the diversity of spontaneous mental experience at rest.

Results

Core functional brain networks identified from GINNA analysis

To define the brain’s core functional networks, we relied on the 32 resting-state cortical networks from the Groupe d’Imagerie Neurofonctionnelle Network Atlas (GINNA)18. Through hierarchical clustering, we identified seven core functional networks. A key strength of the GINNA atlas lies in its rigorous and cognitively specific annotation of each network, derived through meta-analytic decoding using Neurosynth60. This methodology enabled a valid and precise functional labeling of the core networks, firmly anchored in established task-evoked cognitive domains. Crucially, thanks to this pre-existing annotation, the cognitively informed classification of core networks could be established prior to examining their association with spontaneous thought profiles at rest, thereby providing a robust and interpretable framework. The anatomical layout of these networks is depicted in 3D in Fig. 1 and described in detail in the Methods section.

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

The 3D brain mapping of the seven core functional networks. Social and communication network (SCN, including typically named LAN and social network): spoken language to monitor one’s own speech and interpret the intentions, emotions, and beliefs of others during communication (pink). Self-awareness network (SAN, including typically named DMN and SN): enablers of individuals to use their internal resources - such as memory and self-awareness - to make decisions and behave by promoting reflective and independent thinking (blue). Cognitive flexibility network (CFN, typically named FPN): adaptation to changing circumstances and dynamically adjusting strategies to new information (khaki). Performance monitoring network (PMN, typically named CON): self-regulation and task performance improvement (green). Executive planning network (EPN, comparable to typically named and extended DAN): facilitation of adaptive behavior by enabling environment assessment, informed decisions, and effective goal-directed actions (brown). Somatomotor network (SMN): integration of sensory and motor functions, enabling movement, body awareness, and coordinated interaction with the surrounding environment (violet). Visual network (VN): perception, interpretation, and recognition of visual stimuli (orange).

Synchronizations and desynchronizations of core functional networks linked to spontaneous thought profiles

In a previous study, we identified and characterized six distinct resting-state thought profiles (RSTPs) using the ReSQ 2.0 self-report questionnaire, a method allowing the estimation of spontaneous thoughts emerging during rfMRI acquisition47. In the present study, we used mixed linear regression models to examine variability in between-network and within-network FC across brain core functional networks related to spontaneous thoughts. Fisher’s r-to-z transformed correlation values were computed for 21 between-network edges and for within-network FC, serving as outcome variables. RSTP category (six levels) was entered as a fixed effect, and participant ID was modeled as a random intercept to account for inter-individual variability. Post hoc analyses across the six RSTPs revealed distinct patterns of synchronization (i.e., increased FC) and desynchronization (i.e., reduced FC). Significant differences were observed across 18 of 21 between-network FC edges (F(100,34340) = 2.28, p <.0001; Table 1) and 5 of 7 within-network core functional networks (F(30,12026) = 1.91, p =.002, Table 2), based on a significance threshold of α = 0.05. For each contrast, the unstandardized slope coefficient (b) was reported as an effect size, along with its 95% confidence interval. Tables 1 and 2 present the significant post hoc comparisons, including effect sizes (b), standard errors (SE), degrees of freedom (df), t-values, p-values, and confidence intervals (lower.CL and upper.CL), providing a detailed summary of the connectivity differences across RSTPs.

Table 1 Significant differences of between-network FC according to the RSTPs.
Full size table
Table 2 Significant differences of within-network FC according to the RSTPs.
Full size table

Differential involvement of core functional networks across spontaneous thought profiles

To quantify the involvement of each core functional network in between-group FC differences, we computed a Differential Degree Count (DDC). For each of the seven nodes (networks), we counted the number of times it was involved in an edge that showed a statistically significant difference in between-network FC across the 15 pairwise comparisons between the six RSTP groups (21 possible edges). These differences were further classified as either synchronizations (increased FC in one RSTP relative to another) or desynchronizations (decreased FC). The DDC, inspired by degree centrality, thus reflects the total number of significant between-group differences involving a given node, allowing us to identify which core functional networks are more frequently involved in synchronizations or desynchronizations for each RSTP (Table 3).

Table 3 Differential degree count (DDC) by core functional network and thought profile.
Full size table

We further observed a monotonic positive association between the proportion of between-network FC desynchronizations and the proportion of participants reporting high thought fluctuations, calculated at the RSTP level (Spearman’s ρ = 0.81, p =.049; see Table 4).

Table 4 Correlation between the proportion of between-network FC desynchronizations and the proportion of participants reporting high thought fluctuations across the six RSTPs.
Full size table

The fluctuating profile, marked by the highest proportion of participants reporting high thought fluctuations (94%), was predominantly linked to desynchronizations, accounting for 88% of significant differences. In contrast, the verbal-predominant profile exhibited the lowest proportion of participants reporting high thought fluctuations (64%) and the lowest proportion of desynchronizations, representing only 12% of significant differences.

FC patterns characterizing spontaneous thought profiles

By comparing FC across the six RSTPs, we observed that the two evocative profiles shared a distinctive FC pattern, marked by: (i) SAN-VN integration distinguishing them from the two modality-predominant profiles, (ii) PMN-SMN integration distinguishing them from the vigilant and verbal-predominant profiles, and (iii) CFN-PMN synchronization distinguishing them from the fluctuating and verbal-predominant profiles (Fig. 2). Additionally, we identified unique FC patterns, which reflected specific network configurations consistent with the qualitative content of spontaneous thoughts associated with each profile, as depicted in Fig. 3. The evocative-preoccupied profile was associated with CFN-VN integration, whereas the evocative-reverie profile was linked to CFN-SCN, CFN-SAN, SAN-SCN, and EPN-SAN integrations. The vigilant profile showed integration between the SMN and SAN, as well as between SMN and VN. The fluctuating profile was characterized by CFN-EPN and PMN-VN integrations. In contrast, the visual and verbal-predominant profiles were associated with the segregation of the VN and SCN, respectively.

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

FC Patterns associated with several RSTPs. Chord diagrams show FC variations associated with at least two RSTPs. Links indicate a significant difference for the RSTP compared to at least two others. Dark grey links indicate significant desynchronization (i.e., decreased FC). Light grey links indicate significant synchronization (i.e., increased FC). Radar plots show RSTP characterization by the ReSQ 2.0 categories (N = 1779). The background gray profile represents the most frequent categories in the study sample. 0 = never, - = rarely, + + = often, + + + = very often, except for voice vividness: - = rather vague, + + = rather precise and image vividness: - = few details, + + = moderate details, + + + = many details. Detailed characterization was displayed in47.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

FC Patterns specifically associated with each RSTP. Chord diagrams show FC variations associated with each RSTP. Links indicate a significant difference for the RSTP compared to at least two others. Dark grey links indicate significant desynchronization (i.e., decreased FC). Light grey links indicate significant synchronization (i.e., increased FC). Radar plots show RSTP characterization by the ReSQ 2.0 categories (N = 1779). The background gray profile represents the most frequent categories in the study sample. 0 = never, - = rarely, + + = often, + + + = very often, except for voice vividness: - = rather vague, + + = rather precise and image vividness: - = few details, + + = moderate details, + + + = many details. Detailed characterization was displayed in47.

Discussion

Using the GINNA atlas, we identified seven core functional networks whose degree of desynchronization was positively correlated with the richness of spontaneous thought dynamics across participants. FC comparisons across the six RSTPs revealed that the two evocative profiles shared a common FC pattern that reliably distinguished them from the others. In addition, each RSTP displayed distinct FC signatures that aligned with their respective thought content, highlighting differential engagement of large-scale brain networks in shaping spontaneous mental activity.

We observed a consistent association between reduced FC and spontaneous thought profiles, suggesting that greater network involvement in mental content is accompanied by lower synchronization, both within and between core functional networks. These results align with studies emphasizing a dissociation between neural activation and functional connectivity53,61,62,63, supporting the notion that cognitive engagement can reduce FC rather than increase it. Furthermore, our results are consistent with the observation that high neural activity often coincides with lower FC in the resting brain52. Our findings reinforce theoretical models of FC dynamics, which propose that high connectivity may correspond to a resting or standby state, while desynchronization may indicate active, task-like processing57,58,59. In this view, decreased FC reflect network dynamic recruitment in spontaneous cognitive operations.

The two evocative spontaneous thought profiles represent rich and diverse mental states characterized by internally oriented, self-referential, and empathic thoughts, accompanied by vivid imagery and emotionally salient recollections or imaginative scenes. These profiles were associated with common patterns of FC, which contrasted with those observed in other profiles along three main dimensions (Fig. 2).

First, SAN-VN desynchronization was observed in the two evocative profiles, contrasting with synchronization in the two modality-predominant profiles (visual-predominant and verbal-predominant). This synchronization in the latter may reflect the relative paucity of their thought content: the visual-predominant profile was marked by sparse mental content, while the verbal-predominant profile primarily centers on surroundings and cardiorespiratory awareness. The involvement of the SAN in the evocative profiles is consistent with prior evidence linking the DMN to self-referential processing, social cognition, and autobiographical memory25 (e.g., vivid internal experiences during working memory maintenance64, and perspective-taking or simulation of others’ experiences65). Furthermore, SAN-VN integration has been associated with mental imagery66,67, creativity68, and heightened vividness linked to early visual cortex activation69. In this context, SAN-VN coupling may contribute to the vividness and creativity in the evocative profiles.

Second, CFN-PMN synchronization in the evocative profiles contrasted with desynchronization in the fluctuating and verbal-predominant profiles. These latter two are defined by alternating internal and external attention, and their CFN-PMN integration likely supports this cognitive flexibility and attentional reorientation, consistent with their known roles in task switching and attentional control25,30,70.

Third, PMN-SMN desynchronization in the evocative profiles contrasted with PMN-SMN synchronization in the vigilant and modality-predominant profiles. The evocative profiles were also associated with reduced movement during scanning, unlike the verbal-predominant and vigilant profiles, which showed greater in-scanner motion. Additionally, within-network SMN desynchronization was observed in the vigilant and verbal-predominant profiles, possibly reflecting increased sensorimotor engagement due to higher mobility. Conversely, PMN-SMN integration in the evocative profiles may reflect efficient motor suppression or stabilized motor control, facilitating the internally immersive experience. This interpretation aligns with the PMN’s proposed role in maintaining stable implementation of task sets in downstream sensorimotor processors across trials30 and its FC with the SMN26,71.

In addition to the shared features, distinct FC patterns were also observed for specific thought profiles, indicating that each profile engages partially dissociable network dynamics (Fig. 3).

The evocative-reverie profile was specifically linked to an extensive integration of functional core networks, including the CFN, EPN, SAN, and SCN. Empathic, past-oriented, and imaginative thoughts of the evocative-reverie profile align with previous research linking CFN-SAN interaction to creative thinking72,73,74,75 and temporally extended evaluation of self-generated thoughts76. The specific involvement of the SCN, encompassing regions such as the anterior temporoparietal junction, inferior frontal gyrus, and posterior middle temporal gyrus, known for their roles in processing language and depictions of human behavior65, suggests that the evocative-reverie profile engaged in mentalizing social scenes through an allocentric perspective. Consistently, several authors emphasize that allocentric representations require broad inter-network interactions to support their complexity77,78. The integration between the SAN and EPN (an extended version of the DAN) might have seemed intriguing, given the frequently documented anticorrelation between the SAN and EPN. However, Dixon et al.79 have shown that SAN-EPN FC fluctuates over time and contexts, coordinated by broader network dynamics involving the CFN, a tripartite interaction also observed in the evocative-reverie profile. Notably, they observed that the frontal eye fields (part of the EPN) and the dorsomedial prefrontal cortex (part of the SAN) exhibited stronger negative FC during a movie-watching condition compared to rest. The similarity between their findings and ours supports that the evocative-reverie profile favored an outward-focused perspective. Within SAN-EPN integration, it is plausible that EPN involvement facilitated flexibility in supporting allocentric spatial representations80.

In contrast to the evocative-reverie profile, the high co-occurrence of self-referential and empathic thoughts in the evocative-preoccupied profile strongly aligns it with a first-person perspective81,82. Accordingly, the evocative-preoccupied profile exhibited limited specific functional integration, involving only the CFN-VN. We hypothesize that the evocative-preoccupied profile adopted an egocentric perspective in which individuals “see” themselves within the scene, engaging visual simulation processes analogous to those activated during real-world perception83 while requiring significantly less functional integration than the allocentric perspective of the evocative-reverie profile. In support of this hypothesis, CFN-VN integration is associated with mental imagery process67, with the CFN contributing to subjective vividness and the VN enhancing its precision84,85.

The fluctuating profile is a diverse profile mixing internally and externally oriented thoughts and was characterized by the highest level of thought fluctuations (Table 4). The fluctuating profile was associated with a specific FC pattern integrating CFN-EPN and PMN-VN. This finding is consistent with previous literature showing that executive function networks likely work as an ensemble to produce accurate and timely individual switching behaviors86.

The vigilant profile is focused on externally and bodily-oriented thoughts and was linked to the integration of SMN-VN and SAN-SMN. SAN-SMN integration was previously linked to low-level frequency daydreaming87 and observed in task-focused trials compared to stimulus-independent, task-unrelated thought88. The prominent involvement and integration of the SAN and SMN in the vigilant profile (Table 3) likely underpin the heightened experience of painful and unpleasant sensations previously associated with this profile47. These findings suggest that the SAN (comprising the DMN and SN) may play a pivotal role in modulating integrated self-consciousness by linking external stimuli to interoception and nociception.

The visual-predominant thoughts are distinguished by their exclusive use of images, whereas the verbal-predominant thoughts rely solely on inner speech. Accordingly, the visual-predominant profile was associated with VN segregation, and the verbal-predominant profile with SCN segregation, both without any connection to a specific integration pattern. These sparse and modality-predominant profiles were the only ones linked to FC segregation, underscoring the unique relationship between cognitive modes and functional network organization.

Conclusion

We present a comprehensive and integrated framework for understanding the dynamics of core functional network integration and segregation in relation to spontaneous thought profiles during rfMRI scans. Our findings reveal that the sparsest thought profiles are characterized by pronounced segregation of core networks, aligning with specific cognitive modes. In contrast, rich and diverse thought profiles exhibit distinct patterns of functional integration consistent with the established roles of the brain regions forming these networks. These results support a broader conceptualization of the ‘default mode,’ extending beyond the narrow confines of the DMN to encompass dynamic and distributed brain activity. We propose that the resting-state reflects a personalized integration of core functional networks, producing distinct FC patterns. By modeling inter-individual variability across six well-characterized thought profiles, we demonstrate that this process underpins unique cognitive and introspective functions.

Methods

Participants

We leveraged data from the MRi-Share project, a large-scale prospective cohort of French university students, to explore neuroanatomical maturation and variability in healthy young adults89. From the 1,870 MRi-Share participants, native French speakers with usable MRI data were included, alongside those who completed the ReSQ 2.0, a tool for evaluating the quality and content of thoughts during resting-state periods, resulting in a final sample of 1,717 individuals (mean age 22 ± 2.25 years, 72% women). All procedures were performed in compliance with the declaration of Helsinki and were approved by the Comité de Protection des Personnes Sud-Ouest et Outre-Mer (local ethics committee CPP SOOMIII) with agreement 2015-A00850-49. All participants signed an informed written consent form.

Resting-state spontaneous thoughts assessment

The ReSQ 2.0 (24-item self-report) evaluates spontaneous mental activity at rest47. To classify individuals into homogeneous thought subgroups, we applied agglomerative hierarchical clustering with Ward’s method and Euclidean distances based on multiple correspondence analysis components, followed by K-means clustering for refinement. This analysis identified six distinct resting-state thought profiles (RSTPs), summarized below:

  • An evocative-preoccupied profile: A profile focused on internally oriented thoughts, self-relevant preoccupations, thoughts about others with empathy, past events, and future thinking, with precise images and voices and negative emotional valence (RSTP 1, n = 152).

  • A fluctuating profile: The most fluctuating profile, with diverse thoughts alternating between externally and internally oriented thoughts, is close to the average profile of all participants (RSTP 2, n = 488).

  • An evocative-reverie profile: A profile exclusively involving internally oriented thoughts about others with empathy, past events, and imaginative scenes, with precise images and voices and positive emotional valence (RSTP 3, n = 355).

  • A vigilant profile: A profile centered on externally and bodily-oriented thoughts, future thinking, and lectures, with a negative emotional valence (RSTP 4, n = 282).

  • A visual-predominant profile: The least defined profile, with exclusive use of images (RSTP 5, n = 352).

  • A verbal-predominant profile: The least fluctuating profile, concentrating on surroundings and cardiorespiratory awareness, exclusively using inner speech (RSTP 6, n = 88).

Additionally, analysis of the quality control metric for head motion (mean absolute translation67) revealed that the two internally-oriented evocative profiles were associated with significantly less movement compared to the vigilant and verbal-predominant profiles (RSTP 1 vs.4: b = − 0.0126, t = − 2.29, p =.022; RSTP 1 vs. 6: b = − 0.0172, t = −2.36, p =.018; RSTP 3 vs.4: b = − 0.0113, t = −2.6, p =.009; RSTP 3 vs.6: b = 0.016, t = 2.46, p =.014).

FC analyses and core functional networks delineation

One thousand seven hundred seventeen participants completed a 15-minute rfMRI scan with closed eyes. Data acquisition and preprocessing details can be found elsewhere89. FC was computed using the GINNA atlas18 and is available for download at https://github.com/Achillegillig/ginna. The non-overlapping version of this atlas includes 439 brain regions, each assigned to one of 33 functional networks and classified into one of three tissue types: Cortex, Basal Ganglia, or Cerebellum. For the present analysis, only cortical regions were included, resulting in a decomposition into 32 networks (one exclusively covers basal ganglia regions) with a minimum of 2 and a maximum of 17 regions per network.

Core functional networks were identified through a hierarchical ascending classification of the 32 networks, following the method outlined in Doucet et al.13. The average time course of voxels within each GINNA atlas region was computed. The time course of each functional network was calculated as the average of the time courses of its constituent regions. FC values between the 32 networks were quantified as the Fisher z-transformed Pearson correlation coefficients between the time courses of any two functional networks. Then, FC values were averaged across participants, and hierarchical clustering was implemented using the average-linkage method. Correlation values were converted into dissimilarity distances and aggregated based on minimal dissimilarity. Clustering reliability was assessed using the pvclust R package, employing a bootstrap resampling approach on 100,000 datasets, ranging from 50 to 140% of the sample size90. This analysis identified seven core functional networks, including 33 to 69 regions each.

Within-network FC of core functional networks was calculated as the average Pearson correlation coefficient (Fisher z-transformed) between the time courses of all region pairs within a single core functional network.

Between-network FC of core functional networks was defined in the same manner as for the GINNA atlas (see above).

The average Pearson correlation coefficient (Fisher z-transformed) across all region pairs (APCC) was calculated to estimate the drowsiness effects38.

Anatomical and cognitive profiling of seven core functional networks

The anatomical organization of seven core functional networks is visualized in 3D in Fig. 1. Each network was assigned a synthesized cognitive label based on the corresponding cognitive functions described in the GINNA atlas18, independent of its anatomical description (Table 5).

Table 5 The anatomical and cognitive profiling of seven core functional networks.
Full size table

The Social and Communication Network (SCN) clusters frontoparietal regions previously associated with language91 and dynamic representations of social interactions65. The SCN includes social-related areas like the dorsomedial prefrontal cortex92,93; semantic-associated areas, such as the superior frontal gyrus94; and regions associated with both functions, including the left inferior frontal gyrus (pars triangularis and pars orbitalis left), anterior temporal lobe, supplementary motor area, superior temporal gyrus, and temporoparietal junction94,95,96,97,98,99. The association between language and social networks is sometimes observed at rest94,100 and emerges from Kong’s meta-analysis of eight FC atlases20.

The Self-Awareness Network (SAN) outlines a medial frontoparietal network. It includes regions commonly identified in the literature as the DMN, specifically the posterior cingulate cortex, ventral precuneus, medial prefrontal cortex, medial temporal lobe (i.e., hippocampus and parahippocampal cortex), medial orbitofrontal cortex, caudal angular gyrus, posterior inferior parietal lobule, and parieto-occipital sulcus21,101,102. The SAN also encompasses the SN, including the ventral anterior insula and rostral anterior cingulate cortex14. The DMN is an integrated system for autobiographical and semantic memory, self-monitoring, and empathic processing25,65. The SN is a hub for consciously integrating autonomic feedback and responses with internal goals and external demands28. The ventral anterior insula plays a central role in interoceptive awareness103, body movement awareness, and self-recognition104. It supports homeostatic demands28 and emotional awareness104,105 in conjunction with the anterior cingulate. Thus, the SAN may support a context-sensitive self, integrating the introspective capacities of the DMN with the situational awareness and emotional relevance provided by the SN. Note that this SN is distinct from the one described by Menon et al.25 which is more closely associated with the PMN. By uniting essential components of self-related processing and emotional experience, the SAN presents a system of integrated self-consciousness in line with the ‘sense of self’ concept attributed to the DMN25.

Three networks represent the executive functions. First, the Cognitive Flexibility Network (CFN) includes portions of the lateral prefrontal cortex and posterior parietal cortex, e.g., the middle and inferior frontal lobe, pre-supplementary motor area, rostral angular gyrus, left inferior parietal lobule, and a portion of the inferior temporal lobe. The CFN is, therefore, a lateral frontoparietal network aligned with the well-described FPN23,30,34. This similarity further reinforces the role of the CPN as a flexible hub for cognitive control30. Second, the Performance Monitoring Network (PMN) comprises the dorsal anterior insula and anterior midcingulate cortex (i.e., the dorsal anterior cingulate cortex), along with regions of the middle and superior frontal lobe and the dorsal posterior supramarginal gyrus37. The PMN is a midcingulo-insular network closely resembling the CON, which Dosenbach et al. renamed the action-mode network106, aligning with its role in postural107 and motor control37. Third, the Executive Planning Network (EPN) incorporates core regions of the DAN, including the frontal eye fields, superior posterior parietal lobule, and intraparietal sulcus108. The EPN encompasses additional regions, most of which had been previously associated with the DAN, such as the dorsal anterior precuneus109, dorsal anterior supramarginal gyrus98, posterior inferotemporal area108,110,111, left lateral fusiform gyrus112, right middle frontal lobe108,113, and right lateral orbitofrontal cortex. The EPN broadens the DAN’s framework (i.e., visuospatial attention) by integrating regions tied to critical aspects of executive planning, encompassing high-level visual processing, like categorical recognition, and priority mapping, alongside evaluating potential outcomes and the sequencing of movements112,114,115,116.

Classically, the SMN is a pericentral network encompassing the superior temporal gyrus, mid-posterior insula, superior anterior parietal lobe, and central regions such as Rolandic operculum, precentral, postcentral, and central sulcus. Also typical, the VN is located in the occipital lobe21.

Statistical analysis

Associations of spontaneous thoughts with variations in FC

To investigate how spontaneous thoughts relate to the integration and segregation of core functional networks, we analyzed between-network and within-network FC in relation to the RSTPs using two separate mixed linear regression models. Between-network FC (21 edges, calculated as 7 × 6 ÷ 2) and within-network FC were determined using Fisher’s r-to-z transformed correlation values and served as the outcomes. The RSTP category (6 levels) was included as the predictor, with participant ID specified as a random effect.

Estimation and hypothesis testing of coefficients

A resampling bootstrap method was used with 1,000-fold replication to evaluate the uncertainty about fixed effect coefficient estimates and validate the internal reliability of the effects117. This method showed good consistency among fixed effect coefficient estimates. Then, the significance of fixed effects was assessed using ANOVA components and residuals were visually inspected for normality and homoscedasticity.

Addressing the influence of confounding factors

Fixed-effect covariates, including total intracranial volume, age, gender/sex, and drowsiness (i.e., APCC), were incorporated into the mixed linear regression models to account for their potentially confounding effects38,118.

Comparison of FC across the six RSTPs and interpretation threshold

To compare the FC of the six RSTPs, post hoc analyses were conducted using an alpha level of 0.05 for the significance threshold of hypothesis tests. The unstandardized slope coefficient (b) with its confidence interval was reported as effect size119. Detailed post hoc results, including effect sizes (b), standard errors (SE), degrees of freedom (df), t-values, p-values, and confidence intervals (lower.CL and upper.CL), are provided in Tables 1 and 2. Aiming to ensure the consistency of the observed differences, we focused exclusively on those involving one RSTP relative to at least two others to visualize and interpret the results.

Statistical analyses comparing RSTPs were conducted using RStudio (R 4.0.0)120,121. Mixed linear regression models were fitted with the lmerTest package (version 3.1-3)122 and post hoc analyses were performed with the emmeans package (version 1.4.7)123.