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

Abstract

Neural substrates of "mind wandering" have been widely reported, yet experiments have varied in their contexts and their definitions of this psychological phenomenon, limiting generalizability. We aimed to develop and test the generalizability, specificity, and clinical relevance of a functional brain network-based marker for a well-defined feature of mind wandering-stimulus-independent, task-unrelated thought (SITUT). Combining functional MRI (fMRI) with online experience sampling in healthy adults, we defined a connectome-wide model of inter-regional coupling-dominated by default-frontoparietal control subnetwork interactions-that predicted trial-by-trial SITUT fluctuations within novel individuals. Model predictions generalized in an independent sample of adults with attention-deficit/hyperactivity disorder (ADHD). In three additional resting-state fMRI studies (total n = 1115), including healthy individuals and individuals with ADHD, we demonstrated further prediction of SITUT (at modest effect sizes) defined using multiple trait-level and in-scanner measures. Our findings suggest that SITUT is represented within a common pattern of brain network interactions across time scales and contexts.

Fig. 1. Functional connectivity-based predictive modeling of trial-wise SITUT fluctuations within healthy adults.

a A schematic of the analysis pipeline. Within training data, SITUT ratings and pre-rating functional connectivity matrices (based on a 268-node whole-brain atlas) were extracted for each trial. Red and blue, respectively, indicate positive and negative functional connectivity. Edges that were correlated with SITUT rating were identified at a threshold of P < 0.01 (uncorrected), and a summary score was obtained for each trial based on a subtraction of positive and negative edge sums. b A linear model, based on summary scores, was used to correlate 36 predicted versus observed SITUT ratings in a held-out participant (top). Subsequently, that correlation value (indicated with orange line) was compared with a null distribution of r values derived from 1000 permutations of shuffled SITUT ratings (bottom). c Predicted versus observed SITUT correlations, and mean null correlations, within each held-out participant. The gray bar indicates the mean across individuals (i.e., mean across white bars; n = 17). At the group level, predicted versus observed correlations were significantly greater than mean null correlations (P = 0.019, two-sided, Wilcoxon signed-rank test). Source data are provided as a Source data file. d Edges strongly contributing positively (blue) and negatively (red) to the predictive model. A degree threshold of 6 was applied; i.e., nodes involved in at least 6 contributing edges are displayed. Error bars indicate standard error of the mean. SITUT stimulus-independent, task-unrelated thought. *P < 0.05.

Fig. 2. Functional neuroanatomical basis of the SITUT-CPM network.

a The number of edges, among those within the SITUT-CPM positive mask, assigned to each within- or between-network pair based on the Schaefer300 and Yeo-Krienen 7-network atlases. b The mean correlation between CPM network-pair regions as a function of SITUT rating, shown for the top five positive network-pair features. “SITUT” and “Task-Focused” trial values were derived from a median split of trials, based on SITUT ratings, within each participant (n = 17 participants). Error bars indicate standard error of the mean. c Same as a, except for the SITUT-CPM negative mask. d Same as b, except for edges negatively correlated with SITUT (n = 17 participants). e Same as a, except for the Yeo-Krienen 17-network atlas. f Same as b, except for the Yeo-Krienen 17-network atlas. g Summary schematic of major inter-network pairs contributing to the CPM-based SITUT network. DAN dorsal attention network, DMN default mode network, FCPN frontoparietal control network; LIM limbic network, SAL salience network, SMN sensorimotor network, TP temporal-parietal network, VIS visual network, SITUT stimulus-independent, task-unrelated thought, CPM connectome-based predictive model. Source data are provided as a Source data file.

Fig. 3. The SITUT-CPM is sensitive to intra-individual mind wandering in adults with ADHD.

a Predicted versus observed SITUT correlations, and mean null correlations, within each held-out participant with ADHD, based on the SITUT-CPM generated in healthy control (HC) participants. At the group level (n = 20 individuals with ADHD), predicted versus observed correlations were significantly greater than mean null correlations (*P = 0.028, two-sided, Wilcoxon signed-rank test). The gray bar indicates the mean across individuals (i.e., mean across white bars). b Same as a but for held-out HC participants (n = 17) based on a CPM generated to predict SITUT within ADHD participants (*P = 0.015, two-sided, Wilcoxon signed-rank test). c Mean SITUT ratings were significantly greater in  the ADHD compared to HC group (*P = 9.6 × 10⁻⁵, two-sided, Wilcoxon rank-sum test). d SITUT-CPM network strength, averaged across gradCPT runs within each participant, was significantly greater in the ADHD compared to HC group (*P = 0.039, two-sided, Wilcoxon rank-sum test). Error bars indicate standard error of the mean. Solid black lines in c and d indicate median. SITUT stimulus-independent, task-unrelated thought, CPM connectome-based predictive model. *P < 0.05. Source data are provided as a Source data file.

Fig. 4. SITUT-CPM predictions from resting-state fMRI correlate with individual differences in daydreaming frequency scale scores (Superstruct dataset).

a Significant positive correlation between SITUT-CPM-predicted versus observed scores on the daydreaming frequency scale within the Superstruct rs-fMRI dataset. The trend line (for visualization purposes) is based on locally weighted regression fitting with a second-order polynomial. Statistical testing was performed based on Spearman’s rank correlation (coefficient and two-sided p value shown). Source data are provided as a Source data file. b Correlations (ordered by rank) between SITUT-CPM prediction and 67 behavioral and self-report outcomes in the Superstruct dataset. Among all outcomes, daydreaming frequency (denoted as “MindWandering_Freq”) showed the highest correlation with model prediction. Light to dark blue scale indicates lower to higher absolute correlation value. See ref. for a phenotype legend of labels shown. SITUT stimulus-independent, task-unrelated thought, CPM connectome-based predictive model.

Fig. 5. SITUT-CPM predictions from resting-state fMRI data correlate with individual differences in deliberate and spontaneous mind wandering questionnaire scores (Leipzig dataset).

a Significant positive correlation between SITUT-CPM-predicted versus observed scores on the Mind Wandering Deliberate scale within a dataset collected at the University of Leipzig. b Same as a but for the scores on the Mind Wandering Spontaneous scale. Trend lines (for visualization purposes) are based on locally weighted regression fitting with a second-order polynomial. For both a and b, statistical testing was performed based on Spearman’s rank correlation (coefficients and two-sided p value shown without correction for multiple comparisons). SITUT stimulus-independent, task-unrelated thought, CPM connectome-based predictive model, MW mind wandering. Source data are provided as a Source data file.

Fig. 6. SITUT-CPM network strength during rs-fMRI is greater in adults with ADHD with high compared to low Mind Wandering Questionnaire scores.

Individuals were classified as people showing high levels of mind wandering and people showing low levels of mind wandering based on the Mind Wandering Questionnaire. SITUT-CPM network strength was computed based on the dot product of SITUT-CPM suprathreshold edges (see “Methods” and Fig. 1) and connectivity values in a given resting-state scan (P = 0.043, two-sided, Wilcoxon rank-sum test). Solid black horizontal line indicates median. SITUT stimulus-independent, task-unrelated thought, CPM connectome-based predictive model, MW mind wandering. *P < 0.05. Source data are provided as a Source data file.

Fig. 7. Across rs-fMRI scans, thoughts about surroundings decrease while SITUT-CPM network strength increases over time.

In the University of Leipzig dataset, experience sampling following each ~15 min rs-fMRI run (within one session) revealed that self-reported thoughts involving surroundings decreased (i.e., stimulus-independence of thoughts increased) after the first run and then remained relative stable (black; participants included in analyses: n = 164 for run 1; n = 164 for run 2; n = 164 for run 3; n = 162 for run 4). Conversely, SITUT-CPM strength increased after the first run (blue; participants included in analyses: n = 140 for run 1; n = 165 for run 2; n = 140 for run 3; n = 146 for run 4). Solid dots indicate mean, and error bars indicate standard error of the mean. Statistical tests were corrected for multiple comparisons (*P_FDR < 0.05, two-sided, Wilcoxon rank-sum test). Significant increases were found for run 3 compared to 1, 3 compared to 2, and 4 compared to 1 (P_FDR = 0.0027, 0.0039, and 0.035, respectively). SITUT stimulus-independent, task-unrelated thought, CPM connectome-based predictive model. Source data are provided as a Source data file.