Behavioral variant frontotemporal dementia as a model for understanding the cognitive and cerebral determinants of verbal creativity

Abstract

Background: Although creativity is an essential cognitive function to adapt to an ever-changing world, its neurocognitive and cerebral bases still need clarification. Current models highlight the interaction between associative and executive processes underpinned by the default mode (DMN), executive control (ECN) and salience networks (SN). Furthermore, recent neuroimaging studies highlight the key role of the prefrontal cortex (PFC), located at the crossroads of these networks. Hence, behavioral variant frontotemporal dementia (bvFTD), characterized by progressive neurodegeneration principally impacting the prefrontal cortex and the intrinsic connectivity of these three creativity-related networks, represents a unique model to study creativity. In this study involving 14 bvFTD patients and 20 matched controls, we used a simple word-to-word association task (FGAT) to explore the specific cognitive processes involved in remote thinking, i.e., the production of creative semantic associations. Using voxel-based morphometry, we uncovered critical brain regions for each component and then characterized these regions' intrinsic connectivity profiles using resting-state functional connectivity in healthy controls.

Results: We dissociated four key cognitive components underlying remote thinking: spontaneous associative thinking, inhibition of unoriginal responses, intentional remote associative thinking, and verbal initiation; and replicated them in three independent datasets. Spontaneous associative thinking relied on temporal and cerebellar regions involved in low-order and automatic semantic processing, connected with the DMN, ECN and SN. Inhibition of prepotent unoriginal responses depended on key nodes of the SN. The ability to intentionally generate remote semantic associations was underpinned by key regions of the DMN. Finally, initiation of verbal responses relied on the right dorsolateral PFC, connected to the ECN. BvFTD patients were impaired in the last three components. Two components, cognitive inhibition and intentional remote thinking, mediated the link between atrophy in critical regions and an independent measure of creative abilities.

Conclusions: These findings advance our understanding of creative neurocognition, distinguishing components of creative thinking and clarifying their critical cerebral bases, and participate in the characterization of creativity impairment in patients with bvFTD.

Keywords: Creativity; Executive functions; Frontotemporal dementia; Remote thinking; Resting state functional connectivity; Semantic associations; Voxel-based morphometry.

Fig. 1

Schematic representation of the brain imaging analyses pipeline. Whole-brain voxel-based morphometry (VBM) analyses in bvFTD patients and healthy controls (HC) combined were performed to extract the cerebral correlates of each FGAT component. Significant clusters were used as seeds to conduct whole-brain seed-to-voxel resting state functional connectivity (RSFC) analyses in HC, to explore the functional connectivity profile of each region of interest (ROI)

Fig. 2

FGAT components identified by principal component analysis. Results of the PCA computed on all FGAT variables in the whole sample (bvFTD patients and healthy controls, n = 34). Full description of the variables loadings and components' eigenvalues and explained variance are provided in Supplementary Table 1

Fig. 3

Structural brain correlates of the FGAT components for a P_FWE < 0.05 threshold at the voxel level. Significant clusters for whole-brain VBM regressions of GMV on FGAT components’ individual scores on the entire sample (n = 34) are shown. A spatial cluster extent threshold of 100 contiguous voxels was applied to all analyses. The clusters displayed on this figure were used as ROIs for the RSFC analysis of the ‘Verbal Initiation’ and ‘Cognitive Inhibition’ components

Fig. 4

Structural brain correlates of the FGAT Components for a P_uncorrected < 0.001 Threshold at the Voxel Level. Significant clusters for whole-brain VBM regressions of GMV on FGAT components’ individual scores on the entire sample (n = 34) are shown. A spatial cluster extent threshold of 100 contiguous voxels was applied to all analyses. The ‘Remoteness of Spontaneous Semantic Associations’ and ‘Intentional Generation of Remote Semantic Associations’ clusters displayed on this figure were used as ROIs for the RSFC analysis of these two FGAT components

Fig. 5

Significant mediation analyses. Results of the mediation analyses testing mediation effects of each FGAT component’s score on the relationship between GMV within the clusters identified as critical for this component and AUT variables. Results of the significant mediation models are presented in path diagrams (see Supplementary Fig. 6 for non-significant mediation models). Each diagram indicates the standardized estimates of the regression coefficients, with grey matter volume (GMV) in FGAT component’s critical regions as the independent variable (predictor), FGAT component loadings as the mediator, and AUT variable as the dependent variable (outcome). The effect of GMV on the mediator (FGAT component loadings) is indicated by path a, and the effect of the mediator on the AUT variable by path b. The total and direct effects of GMV on the AUT variable are indicated by path c and c′, respectively. The indirect effect of GMV on the AUT variable through the mediation of the FGAT cognitive component is indicated by path d, with the bootstrapped 95% confidence interval (CI) reported. ‘Cognitive Inhibition’ mediated the link between GMV in regions critical for ‘Cognitive Inhibition’ and ‘AUT Creativity Rating’ (A). ‘Intentional Generation of Remote Semantic Associations’ mediated the link between GMV in regions critical for the ‘Intentional Generation of Remote Semantic Associations’ and ‘AUT Minimal Frequency’ (B). P < .05, **P < .01, *** P < .001. Age, gender, education and total intracranial volume were used as covariates of no interest

Fig. 6

ROI-based functional networks of each FGAT component. For each FGAT component, the figure shows the combination of all critical clusters’ functional connectivity profiles, based on whole-brain one-sample t-tests of ROI-to-voxel connectivity in healthy controls. Statistical analyses were thresholded for significance at P_FWE < 0.05 at the voxel level, with an additional spatial cluster extent threshold of 50 contiguous voxels. ROIs are indicated in black. The proportion of overlap of the resulting functional network with Yeo’s 7 intrinsic functional networks [–70] are reported on radar plots for each FGAT component. Results for each ROI individually are shown in Supplementary Figs. 8 to 11