Prediction of stimulus-independent and task-unrelated thought from functional brain networks

Mind wandering, characterized by self-generated thoughts decoupled from immediate environmental inputs and current tasks, occupies a significant portion of our waking lives. Its prevalence varies considerably among individuals, and this variability is strongly linked to mental health consequences in conditions like attention-deficit/hyperactivity disorder (ADHD) and Alzheimer's disease. Understanding the neural mechanisms underlying mind wandering is therefore a crucial goal in cognitive and clinical neuroscience. Recent research has effectively employed neuroimaging combined with online experience sampling to link brain dynamics with mind-wandering episodes. Initial studies using fMRI revealed an association between mind wandering and default mode network (DMN) activation, a finding corroborated by subsequent fMRI and intracranial electrophysiology research. However, these studies also highlighted the involvement of distributed, dynamic network-level interactions beyond the DMN, including the frontoparietal control network (FPCN) and sensory/motor regions. Lesion and neurostimulation studies further implicated specific DMN and FPCN regions. Despite this accumulating evidence, inconsistencies in the definition and experimental contexts of mind wandering have limited the generalizability of findings. The critical question remains: can a common functional network pattern serve as a specific, generalizable marker of a particular aspect of mind wandering across individuals, clinical groups, and diverse contexts? Predictive modeling of neuroimaging data offers a promising avenue to address this question, allowing for the development of generalizable functional network markers of cognitive and affective functions. Prior research hinted at the predictive potential of multivariate features based on DMN connectivity for task-unrelated thought, but broader generalizability remained to be established. This study focused on identifying and validating a brain network marker for stimulus-independent, task-unrelated thought (SITUT), a core component of mind wandering. The research employed a multi-stage approach: first establishing a predictive model in healthy adults, then testing its validity and specificity; second, verifying its generalizability to adults with ADHD; and finally, assessing its ability to predict individual differences in SITUT across three independent resting-state fMRI (rs-fMRI) datasets using various mind wandering measures.

The existing literature on the neural correlates of mind wandering is extensive but fragmented. Early fMRI studies primarily focused on the default mode network (DMN), showing increased activation during mind wandering. However, this view has been refined by subsequent studies highlighting the crucial role of dynamic interactions between multiple brain networks. The frontoparietal control network (FPCN), responsible for executive functions, has emerged as a key player, with its interplay with the DMN influencing the degree of mind wandering. Furthermore, other brain regions, such as sensory and motor areas, have been implicated, suggesting a more complex, distributed network involvement. The challenge has been the lack of a consistent definition of mind wandering across studies. This variability in operationalization has hampered the development of generalized and robust neural markers. Some studies focus solely on task-unrelated thoughts, while others incorporate the element of stimulus-independence. This lack of consensus has made it difficult to compare results across different research groups and establish reliable neurobiological signatures. The use of predictive modeling has been proposed as a solution to address this issue, offering a data-driven approach to identify robust markers that can generalize across different conditions and contexts.

This study employed a multi-pronged methodology involving fMRI data from multiple sources and diverse analytical techniques. The researchers first defined a connectome-based predictive model (CPM) of stimulus-independent, task-unrelated thought (SITUT) using data from a group of healthy adults performing the Gradual-Onset Continuous Performance Task (gradCPT) while undergoing fMRI scans with interspersed thought probes. The gradCPT allowed for the assessment of SITUT while controlling for performance fluctuations. Participants in the main fMRI experience sampling study provided subjective ratings of task-related focus and retrospective reports on the type of off-task thoughts (SITUT, external distractions, task-related interferences). Only participants whose off-task thoughts were primarily attributed to SITUT were included in the primary analysis. The CPM analysis employed a leave-one-participant-out cross-validation approach. Functional connectivity matrices were computed for 30-second windows preceding thought probes, and connectome-wide correlations between edge weights and SITUT ratings were determined. A linear model predicting SITUT from these functional connectivity patterns was constructed. This model was then used to predict SITUT in held-out participants, and the correlation between predicted and actual SITUT ratings was compared to a null distribution derived from permuted data to assess the model's predictive validity. Several control analyses were conducted to assess the specificity and robustness of the SITUT-CPM. These included controlling for head motion, reaction time variability, and comparing the SITUT-CPM with previously published CPMs for sustained attention and creativity. The researchers also examined the functional neuroanatomical basis of the SITUT-CPM by investigating the contributions of different brain networks (using the Schaefer300 and Yeo-Krienen 7 and 17-network atlases). The generalizability of the SITUT-CPM was then tested on three independent datasets: 1. An independent sample of adults with ADHD undergoing the same gradCPT experience sampling paradigm; 2. A large resting-state fMRI dataset from healthy adults (Superstruct dataset), using the Daydreaming Frequency Scale (DDFS) as a trait measure of SITUT; 3. Another resting-state fMRI dataset from healthy adults (Leipzig dataset), using the Mind Wandering Deliberate and Spontaneous scales (MW-D, MW-S) as trait measures; 4. A resting-state fMRI dataset from adults with ADHD (MIT dataset), using the Mind Wandering Questionnaire (MWQ) to assess trait SITUT. Finally, the researchers analyzed the dynamics of the SITUT-CPM across multiple resting-state fMRI scans within the Leipzig dataset, correlating it with self-reported experience sampling after each scan to assess the relationship between state and trait SITUT.

The study yielded several key findings supporting the existence and generalizability of a neural signature for SITUT: 1. **Internal Validation:** A connectome-based predictive model (SITUT-CPM) was developed in healthy adults that accurately predicted trial-by-trial fluctuations in SITUT ratings (M±SD within-participant r = 0.11 ± 0.16, P = 0.019). This demonstrated the model's capacity for internal validation. 2. **Specificity:** Control analyses confirmed the SITUT-CPM's specificity to SITUT, showing that its predictions remained significant even after controlling for head motion, reaction time variability, and the influence of previously defined CPMs for sustained attention and creativity. The SITUT-CPM and SA-CPM masks showed little overlap, and their network strengths were negatively correlated, suggesting distinct neural substrates. 3. **Neuroanatomical Basis:** The SITUT-CPM was characterized by increased coupling (or reduced antagonism) between the DMN and FPCN, specifically between DMNA and FPCNA subnetworks, consistent with prior research on the neural underpinnings of mind wandering. It was also associated with increased anticorrelation between the DMN and primary sensory/motor regions (SMNA). 4. **External Validation in ADHD:** The SITUT-CPM successfully predicted SITUT fluctuations in an independent sample of adults with ADHD (r = 0.045 ± 0.086, P = 0.028), demonstrating external validation. Furthermore, adults with ADHD exhibited higher SITUT ratings and greater SITUT-CPM network strength than healthy controls. 5. **External Validation with rs-fMRI:** Across three independent resting-state fMRI datasets (total n=1115), the SITUT-CPM showed a consistent, albeit modest, association with trait-level SITUT measures, as assessed by multiple questionnaires. This finding held true in both healthy adults and individuals with ADHD. 6. **State-Dependent Dynamics:** Analysis of the Leipzig dataset showed that across multiple resting-state fMRI runs, the SITUT-CPM network strength increased over time, coinciding with a decrease in self-reported thoughts involving the surroundings, indicating a shift toward more stimulus-independent thought content.

This study provides robust evidence for a common neural signature underlying SITUT across multiple contexts and individual differences. The SITUT-CPM, based on a distributed pattern of brain network interactions, successfully predicted both state and trait SITUT across healthy and ADHD populations, highlighting its generalizability. The model's sensitivity to both state and trait SITUT suggests that it captures both transient and enduring aspects of the underlying neural mechanisms. The strong association between the SITUT-CPM and DMN-FPCN interactions aligns with theoretical models positing a crucial role for these networks in internally focused cognition and top-down control of thought processes. The observed differences in SITUT-CPM predictions between healthy individuals and those with ADHD highlight the potential clinical relevance of this model for understanding the neurobiological basis of ADHD symptoms. This research advances the neuroscience of mind wandering by providing a more nuanced and comprehensive view of its neural correlates. By focusing on a well-defined aspect of mind wandering (SITUT), the study successfully bridged the gap between previous research employing different operationalizations and experimental contexts. The findings also underscore the potential of predictive modeling in neuroimaging research for creating generalizable biomarkers of cognitive and clinical phenomena.

This study successfully developed and validated a functional brain network model (SITUT-CPM) that accurately predicts both state and trait stimulus-independent task-unrelated thought (SITUT), a key component of mind wandering. The model’s generalizability across healthy adults and those with ADHD, as well as its sensitivity to various SITUT measures in independent resting-state fMRI datasets, highlights its potential as a robust neuroimaging biomarker. Future research should explore the SITUT-CPM's applicability to other clinical populations and refine the model to further enhance its predictive accuracy and specificity. Investigating the temporal dynamics of SITUT-CPM network activity could provide further insights into the underlying mechanisms of mind wandering.

Several limitations should be considered when interpreting the study's results. Firstly, the SITUT-CPM was initially defined within a specific task context (gradCPT), which might limit the model's generalizability to other task conditions. The 30-second window used for functional connectivity estimation might not fully capture the rapid dynamics of cognitive processes during SITUT. The retrospective nature of some SITUT assessments could also introduce recall bias. Finally, while the modest effect sizes observed in the resting-state fMRI analyses are comparable to those seen in similar studies, they still emphasize the need for larger samples to confirm and enhance these predictions.