Neurocognitive patterns dissociating semantic processing from executive control are linked to more detailed off-task mental time travel

Abstract

Features of ongoing experience are common across individuals and cultures. However, certain people express specific patterns of thought to a greater extent than others. Contemporary psychological theory assumes that individual differences in thought patterns occur because different types of experience depend on the expression of different neurocognitive processes. Consequently, individual variation in the underlying neurocognitive architecture is hypothesised to determine the ease with which certain thought patterns are generated or maintained. Our study (N = 178) tested this hypothesis using multivariate pattern analysis to infer shared variance among measures of cognitive function and neural organisation and examined whether these latent variables explained reports of the patterns of on-going thoughts people experienced in the lab. We found that relatively better performance on tasks relying primarily on semantic knowledge, rather than executive control, was linked to a neural functional organisation associated, via meta-analysis, with task labels related to semantic associations (sentence processing, reading and verbal semantics). Variability of this functional mode predicted significant individual variation in the types of thoughts that individuals experienced in the laboratory: neurocognitive patterns linked to better performance at tasks that required guidance from semantic representation, rather than those dependent on executive control, were associated with patterns of thought characterised by greater subjective detail and a focus on time periods other than the here and now. These relationships were consistent across different days and did not vary with level of task demands, indicating they are relatively stable features of an individual's cognitive profile. Together these data confirm that individual variation in aspects of ongoing experience can be inferred from hidden neurocognitive architecture and demonstrate that performance trade-offs between executive control and long-term semantic knowledge are linked to a person's tendency to imagine situations that transcend the here and now.

Figure 1

Two reliable neurocognitive modes uncovered through CCA. In both sets of word clouds, the size of the word indicates the strength of the relationship and the colour the direction of the association. The left-hand panel shows the task loadings for each mode in the form of a word cloud (see supplemental material Figure S1 for canonical vector loadings as heat maps). The central panel presents the brain regions that contributed the most to each mode created by calculating the sum of the canonical vectors derived from the functional connectivity. The orange-yellow maps present the overall positively connected nodes and the blue maps show the overall negatively connected modes (see https://github.com/htwangtw/Wang2020_taskSCCA/tree/master/reports/manuscript for raw canonical vector loadings of the functional connectivity data). The right-hand panel shows word clouds that describe the results of a quantitative meta-analysis of this functional data performed using Neurosynth (see “Methods”).

Figure 2

The patterns in the experience sampling questions associated with mode 2. A The neurocognitive associations that constitute mode 2. B The multivariate relationship between mode 2 and patterns of ongoing thought displayed in the form of a word cloud in which the size and colour of the items reflect the strength and direction of the relationship. Individuals who tended to be better at tasks dependent on semantic knowledge to generate task responses tended to endorse patterns of thoughts high on subjective detail, and that were unrelated to the task, and focused on other periods (past and future). C Standardised task performance scores (y-axis) plotted alone the percentile of mode 2 score (x-axis). The task performance scores are the original measures used in SCCA (see “Cognitive tasks”). The mode 2 score is the addition of the SCCA component score of cognitive tasks and functional connectivity (see “Group level regression analysis”). The ribbon plot shows that mode 2 captures patterns of individual differences in the trade-offs between executive function and semantic knowledge. The shaded bars describe the 95% confidence intervals of the mean.

Figure 3

The stability of mode 2 in predicting thought patterns across task context and day. A Similar multivariate associations between mode 2 and the patterns of ongoing thought emerged in both by tasks conditions; B correlations between the patterns of ongoing thought observed in general and across both tasks; all p-value < 0.001; C similar patterns of ongoing thought were observed across each day of the study; D correlations between the patterns of ongoing thought observed in general and across each day of the study; all p-value < 0.001.

Figure 4

Exploratory cortical thickness analysis. We used a general linear model to explore whether individual variation in the canonical modes was also reflected in the structural organisation of the cortex. We found that the task component in mode 2 is related to the variation in a region of subgenual anterior cingulate among. Thickness in this region was higher for individuals better at semantic than executive tasks. The functional connectivity component of mode 2 was associated with cortical thickness in a region of motor cortex immediately posterior to the central sulcus. Thickness in this region was smaller towards the semantic end of this dimension.

Figure 5

Analysis flowchart. The flowchart of the analysis pipeline. We first conducted SCCA to uncover the hidden structure that combines the task measures and the functional connectivity data. The SCCA model selection is detailed in “Model selection” and Fig. 6. The theoretical validity of latent variables was later examined by predicting the ongoing thought report. We also explore the associated cortical thickness change associated with the functional neurocognitive modes. For the cortical thickness analysis, please refer to “Cortical thickness analysis”. For the details of multiple multivariate regression, please see “Group level regression analysis”.

Figure 6

Multivariate pattern analysis pipeline. Top: parameter tuning combined with k-fold cross-validation to search for the best sparsity constraint with maximal out-of-sample the rank-1 canonical correlation. The selected set of parameters was then used as a basis to recompute CCA on the full dataset. Bottom: Permutation test for mode selection.