High transition frequencies of dynamic functional connectivity states in the creative brain

Creativity is thought to require the flexible reconfiguration of multiple brain regions that interact in transient and complex communication patterns. In contrast to prior emphases on searching for specific regions or networks associated with creative performance, we focused on exploring the association between the reconfiguration of dynamic functional connectivity states and creative ability. We hypothesized that a high frequency of dynamic functional connectivity state transitions will be associated with creative ability. To test this hypothesis, we recruited a high-creative group (HCG) and a low-creative group (LCG) of participants and collected resting-state fMRI (R-fMRI) data and Torrance Tests of Creative Thinking (TTCT) scores from each participant. By combining an independent component analysis with a dynamic network analysis approach, we discovered the HCG had more frequent transitions between dynamic functional connectivity (dFC) states than the LCG. Moreover, a confirmatory analysis using multiplication of temporal derivatives also indicated that there were more frequent dFC state transitions in the HCG. Taken together, these results provided empirical evidence for a linkage between the flexible reconfiguration of dynamic functional connectivity states and creative ability. These findings have the potential to provide new insights into the neural basis of creativity.


Torrance Test of Creative Thinking (TTCT)
The creativity performance of each subject was measured using the figural version of Torrance Test of Creative Thinking (TTCT-Figural) (Torrance 1966). The TTCT-Figural measured fluency, originality, flexibility and elaboration, which were based on Guilford's divergent-thinking factors (Guilford, 1959;Torrance, 1996). The TTCT-Figural comprises three parts: picture construction (SI Fig. 1a), picture completion (SI Fig. 1b), and repeated figures of lines (SI Fig. 1c). In picture construction task, the subjects were asked to construct a creative picture that told a story based on a circle. In picture completion part, the subjects needed to complete 10 different lines to novel and interesting pictures. In repeated figures of lines part, subjects were asked to construct novel and meaningful pictures based on 10 pairs of parallel lines. Each part should be finished in ten minus. The creativity scores are determined based on four measures: • Fluency: The number of relevant ideas; shows an ability to produce a number of figural images. • Originality: The number of statistically infrequent ideas; shows an ability to produce uncommon or unique responses. The scoring procedure counts the most common responses as 0 and all other legitimate responses as 1. The originality lists have been prepared for each item on the basis of normative data, which are readily memorized by scorers.
• Elaboration: The number of added ideas; demonstrates the subject's ability to develop and elaborate on ideas.
• Flexibility: scored by the variety of categories of relevant responses

Number of dynamic FC states
To validate the robust of our results, we performed exploratory analyses of k of 3 and 5. We obtained similar results that high creativity group (HCG) has more frequent transition among dynamic functional connectivity (dFC) states than low creativity group (LCG). Specifically, for k = 3, χ 2 (2, N=44) = 8.44, p = 0.02, post hoc analysis revealed that HCG had more frequent transition between dFC state 1 and state 2 than LCG; for k = 5, χ 2 (9, N=44) = 24.59, p < 0.01, post hoc analysis revealed that HCG had more frequent transition between dFC state 4 and state 5 than LCG.

Validation using Multiplication of Temporal Derivatives
To improve the confidence of our results, we performed a confirmatory analysis using 'Multiplication of Temporal Derivatives'. The MTD has been shown to be more sensitive than sliding window correlation methods in detecting dynamic alterations in connectivity structure and less susceptible to spurious connectivity, such as global mean signal fluctuations and head motion. The MTD estimated similar changes over time, specifically a positive value implies time series couple in the same direction (either both increasing or both decreasing), however negative value represents anticoupling (one increasing while the other is decreasing). The value of the MTD can be interpretable as a signed and weighted adjacency matrix with each temporal window.
To avoid the influence of high-frequency noise, we averaged MTD over a temporal window. Briefly, the MTD for the pairwise interaction between region i and j is defined according the following equation: Where dt is the first temporal derivative of the ith or jth time series, σ is the stand deviation od the temporal derivative time series for region i or j, and w is the window length of the simple moving average.
Given that we used a 0.15-Hz low-pass filter, all signals with periods of 15 seconds or shorter would be removed from the data in theory. Therefore, we used a temporal window with window length of 15 time points to calculate the moving average of the MTD. The MTD can provide an estimation of time-varying functional connectivity. We calculated the spatial similarity of the adjacency matrix across all time points and subjects by using spatial Pearson's correlations. We further applied the K-means cluster algorithm (k = 4) to assigned each time point of each subject a cluster index.
Statistical analysis was performed to validate whether the HCG has more frequent dFC states transition.