Brain Entropy is Associated with Divergent Thinking

Abstract

Creativity is the ability to generate original and useful products, and it is considered central to the progression of human civilization. As a noninherited emerging process, creativity may stem from temporally dynamic brain activity, which, however, has not been well studied. The purpose of this study was to measure brain dynamics using entropy and to examine the associations between brain entropy (BEN) and divergent thinking in a large healthy sample. The results showed that divergent thinking was consistently positively correlated with regional BEN in the left dorsal anterior cingulate cortex/pre-supplementary motor area and left dorsolateral prefrontal cortex, suggesting that creativity is closely related to the functional dynamics of the control networks involved in cognitive flexibility and inhibitory control. Importantly, our main results were cross-validated in two independent cohorts from two different cultures. Additionally, three dimensions of divergent thinking (fluency, flexibility, and originality) were positively correlated with regional BEN in the left inferior frontal gyrus and left middle temporal gyrus, suggesting that more highly creative individuals possess more flexible semantic associative networks. Taken together, our findings provide the first evidence of the associations of regional BEN with individual variations in divergent thinking and show that BEN is sensitive to detecting variations in important cognitive abilities in healthy subjects.

divergent thinking, brain entropy mapping, dorsolateral prefrontal cortex, semantic associative network

Introduction

Creativity is the ability to abstract or generate new knowledge or products and is central to the progression of human civilization, prosperity, and well-being (Runco and Jaeger 2012; Diedrich et al. 2015). Divergent thinking is a key component of creativity that has been found to predict future academic success and social adaptation (Baas et al. 2008). Over the past decade, many studies have adopted functional neuroimaging to search for brain mechanisms that potentially explain creativity. Most have focused on brain activation during the performance of some “creative” task, such as a divergent thinking task (DTT, e.g., alternative uses task [AUT]; Fink et al. 2006, 2009, 2010; Abraham et al. 2012; Kleibeuker et al. 2013b; Cousijn et al. 2014; Sun et al. 2016). However, creativity is a vast and complex construct that requires the involvement of multiple brain regions or networks (Runco and Acar 2012; Jung et al. 2013; Beaty et al. 2016). Previous task fMRI studies that have mainly focused on task-dependent activation may reveal only the local activation of one or several brain regions during a specific creative task. Resting-state brain activity represents a baseline level of activation that is not specific to any overt task and accounts for most of the energy consumed by brain (Raichle et al. 2001). Therefore, a recent research interest is to examine the association between resting-state brain activity and creativity (Beaty et al. 2014; Wei et al. 2014; Chen et al. 2015; Zhu et al. 2017; Beaty et al. 2018).

To date, the existing resting-state literature converges on the hypothesis that creativity is associated with the spontaneous activity of brain regions within and/or between the default mode network (DMN) and cognitive control network (CCN) (Takeuchi et al. 2012; Beaty et al. 2014; Chen et al. 2014; Wei et al. 2014; Li et al. 2016a; Zhu et al. 2017). Most of these studies examined resting-state functional connectivity (FC), including seed-based FC (Takeuchi et al. 2012; Beaty et al. 2014; Wei et al. 2014), large-scale brain network FC (Zhu et al. 2017; Shi et al. 2018), and voxel-wise FC strength (FCS) (Gao et al. 2017; Jiao et al. 2017). Using the medial prefrontal cortex (mPFC), the core node of the DMN, as a seed region, Takeuchi et al. (2012) found that creativity measured by DTTs is positively correlated with the strength of mPFC and posterior cingulate cortex (PCC) connectivity. Similarly, Wei et al. (2014) found that verbal creativity is positively correlated with FC between the mPFC and middle temporal gyrus (MTG). These regions all belong to the DMN, suggesting that FC within the DMN is closely related to creativity. Using the key regions of the CCN, such as the dorsolateral prefrontal cortex (DLPFC) or inferior frontal gyrus (IFG), as the seed regions, previous studies have revealed that figural creativity is positively correlated with FC between the bilateral DLPFC (Li et al. 2016a) and that high-creative individuals exhibit increased FC between the IFG and the entire DMN in comparison with low-creative individuals (Beaty et al. 2014). Based on these findings, creativity is closely related to executive–executive (within-network) coupling and default–executive (between-network) coupling. A large-scale brain network FC analysis further supports this hypothesis. Using an independent component analysis, Zhu et al. (2017) decomposed the brain into large-scale networks and then investigated the relationship between creativity and large-scale brain network FC, and found that both creative domains (visual and verbal creativity) are positively correlated with FC between the DMN and CCN. However, visual creativity was negatively correlated with within-network coupling (both within the posterior DMN and within the right CNN), and verbal creativity was negatively correlated with FC within the anterior DMN. In addition, based on FCS and graph theory analysis, previous studies found that more highly creative groups exhibit higher network efficiency (Jiao et al. 2017) and better information transformation efficiency (Gao et al. 2017) than groups with lower levels of creativity. Furthermore, a study examining the relationship between verbal creativity and the dynamic reconfiguration of brain network FC reported that the temporal variability of FC patterns within the DMN, between the DMN and attention/sensorimotor network, and between the control and sensory networks correlate with verbal creativity, indicating that verbal creativity is associated with high variability of networks involved in spontaneous thought, cognitive control, and attention (Sun et al. 2018).

Two other resting-state brain activity metrics, regional homogeneity (ReHo) and fractional amplitude of low frequency fluctuations (fALFF), have also been used in imaging creativity association studies. Chen et al. (2015) found that people with a high verbal creative ability exhibit low ReHo in the right precuneus. Additionally, based on a large sample (1277 subjects), Takeuchi et al. (2017) identified associations between verbal creativity and ReHo in the left anterior temporal lobe and fALFF in the precuneus, middle cingulate gyrus, right middle frontal gyrus, and cerebellum. ReHo measures the temporal homogeneity of neural activity and reflects the inter-voxel coherence within the neighboring voxels during rest (Zang et al. 2004), while fALFF measures the amplitude of neural activity and is thought to be a direct index of the spontaneous signal fluctuations during rest (Zang et al. 2007). However, these measures do not actually assess brain activity dynamics at the local voxel level. fALFF is the only measure that can partially examine the temporal dynamics but it has a purely empirical cutoff threshold. Assessment of full-band temporal dynamics may reveal important brain–behavior relationships.

In contrast to other resting-state brain measures such as ALFF and ReHo, brain entropy (BEN) is a new method that has recently been developed to quantify the state of the temporal brain dynamics by measuring the signal regularity of a time series (Yang et al. 2013; Smith et al. 2014; Wang et al. 2014). Entropy is a well-defined physical and statistical concept that measures the complexity, randomness, and predictability of a dynamic process (Pincus 1991; Sandler 2017). The entropy value reflects the extent to which a signal is temporally ordered (low entropy), complex (medium entropy), or uncorrelated (high entropy). Recently, BEN was defined as the number of neural states a given brain can access (Saxe et al. 2018). The human brain is a dynamic functional system that exhibits ongoing fluctuations in activity and its dynamic range relates to the information processing capacity (Xue et al. 2019). A brain capable of large variability in neural states will more easily understand and predict variable external events. Therefore, a higher BEN value indicates a large information processing capacity and irregularities in brain activity. A previous study that examined the relationship between BEN and intelligence identified a positive correlation between BEN at rest and individuals’ intellectual ability, suggesting that access to variable neural states predicts complex behavioral performance (Saxe et al. 2018). Thus, BEN is an emerging method that characterizes the resting-state temporal dynamics of the brain and represents a physiologically meaningful approach to assess the dynamic status of intrinsic activity in temporal patterns (Bassett et al. 2012; Wang et al. 2014).

BEN mapping is still new in the literature, although it has successfully been applied in several neuroimaging studies. Regional BEN has been reliably mapped in the normal brain (Wang et al. 2014). However, the normal BEN distribution may be altered during aging (Yao et al. 2013), and in patients with attention deficit hyperactivity disorder (Sokunbi et al. 2013), schizophrenia (Sokunbi et al. 2014), smoking habits (Li et al. 2016b), and cocaine addiction (Wang et al. 2017b). For example, patients with Alzheimer’s disease exhibit lower entropy values in the frontal, temporal, and occipital lobes than healthy controls (Wang et al. 2017a) while patients with relapsing-remitting multiple sclerosis patients show increased BEN in the motor areas, spatial coordination areas, memory areas, and executive control areas compared with controls (Zhou et al. 2016). Additionally, a study applying BEN to a healthy sample found that compared with controls, chronic smokers exhibited globally higher BEN (Li et al. 2016b). However, most of these studies focused on the comparison between two groups (e.g., patients vs. healthy controls), and few studies have directly examined the relationship between regional BEN and individual differences in behavioral variables, such as divergent thinking.

The purpose of the present study was to investigate the relationship between regional BEN and divergent thinking using resting-state fMRI (rs-fMRI) from a large cohort of healthy subjects. Given that previous studies have shown that creativity is associated with the dynamic interactions among the DMN, Salience Network (SN), and CCN (Beaty et al. 2016; Zhu et al. 2017; Beaty et al. 2018; Sun et al. 2018), we hypothesized that divergent thinking is associated with the entropy value of brain regions within the DMN (e.g., the PCC and MTG), SN (e.g., the dorsal anterior cingulate cortex [dACC]), and CCN (e.g., the DLPFC and IFG). First, individuals’ divergent thinking scores were assessed using the verbal form of Torrance Tests of Creative Thinking (TTCT; (Torrance 1974). Then, the whole-brain BEN map was computed using the Brain Entropy Mapping toolbox (BENtbx; Wang et al. 2014). Finally, a correlation analysis was performed to examine the relationship between regional BEN and individual differences in divergent thinking. In addition, we attempted to replicate the results in two independent samples to improve the reliability and strength of our work.

Materials and Methods

Participants

The main sample was drawn from the time point 1 (T1) data of the Southwest University Longitudinal Imaging Multimodal Dataset (International Neuroimaging Data-Sharing Initiative (INDI), http://fcon_1000.projects.nitrc.org/) (Liu et al. 2017). This sample consists of 401 college students recruited from Southwest University. All participants were healthy and right-handed, with no history of neurological or psychological disorders and provided written informed consent. All participants were required to complete behavior tests (including a creativity test and intelligence tests) and brain imaging data acquisition. Fifteen subjects were excluded due to the absence of imaging data, leaving a final sample of 386 subjects (182 males), aged 17–27 years (mean ± standard deviation [SD] = 20.0 ± 1.16). The study was approved by the Southwest University Brain Imaging Center Institutional Review Board.

Assessment of Divergent Thinking

The verbal form of the Chinese version TTCT (Torrance 1974; Ye et al. 1988) has often been used to assess an individual’s divergent thinking ability, which is a core aspect of creativity (Runco and Acar 2012). It consists of seven subtasks (Torrance 1974). Three subtasks require subjects to respond to a scenario presented pictorially by generating questions, causes, and consequences for 5 min. The fourth subtask requires subjects to improve a product (toy elephant) by proposing creative ideas for 10 min. The fifth subtask requires subjects to generate as many unusual uses of a cardboard box as possible within 10 min. The sixth subtask requires subjects to generate unusual questions related to a cardboard box for 5 min. The seventh subtask requires subjects to imagine the consequences of an impossible situation for 5 min. For each subtask, the scores comprise three different creative dimensions: fluency (the number of relevant and meaningful responses), flexibility (the number of different categories of responses), and originality (the degree of originality of the responses). Three trained raters evaluated the responses recorded in all tasks, and the inter-rater correlation coefficient was high (0.9). The total score was the sum of the fluency, flexibility, and originality scores of all subtasks.

Assessment of General Intelligence

The Combined Raven’s Test (CRT; Chinese revised edition) was used to examine subjects’ intellectual ability (Wang et al. 2006). This test displays good reliability and validity (Ming 1989; Tang et al. 2012). The CRT contains three Raven’s standard progressive matrices (C, D, and E sets) and three Raven’s color progressive matrices (A, B, and AB sets), with a total of 72 nonverbal items. Participants were asked to choose the correct answer from six or eight alternatives by completing the matrix with a missing piece. The intelligence score is equal to the number of correct answers.

Image Acquisition

All structural and functional MRI scans were performed on a 3T Trio scanner (Siemens Medical Systems, Erlangen, Germany) at the Brain Imaging Center, Southwest University. Rs-fMRI images were obtained using a gradient echo-planar imaging sequence: repetition time (TR) = 2000 ms, echo time (TE) = 30 ms, slices = 32, thickness = 3 mm, resolution matrix = 64 × 64, flip angle = 90°, field of view (FOV) = 220 × 220 mm², and voxel size = 3.4 × 3.4 × 4 mm³. High-resolution T1-weighted structural images were acquired using a magnetization prepared rapid acquisition gradient-echo sequence: TR/TE = 1900 ms/2.52 ms, inversion time = 900 ms, flip angle = 9°, FOV = 256 × 256 mm², slices = 176, thickness = 1.0 mm, and voxel size = 1 × 1 × 1 mm³.

Image Preprocessing and BEN Mapping

Data preprocessing was performed using batch scripts provided with ASLtbx (Wang et al. 2008) based on SPM (http://www.fil.ion.ucl.ac.uk/spm/). Briefly, the first 10 volumes of rs-fMRI images were discarded to allow the signal to reach a steady state. The remaining rs-fMRI images were corrected for slice timing and head motion. The high-resolution structural images were first reoriented to the original image and then segmented into gray matter, white matter (WM), and cerebrospinal fluid (CSF) using the segmentation tool in SPM8. Then, rs-fMRI images were registered to the Montreal Neurological Institute (MNI) standard space via the T1-weighted anatomical images. Subsequently, temporal nuisance filtering was performed (including bandpass filtering [0.01–0.08 Hz] and nuisance correction [head motion parameters, WM and CSF signals]) and then smoothed with a 6 mm full-width half-maximum (FWHM) Gaussian kernel.

After preprocessing, each subjects’s BEN map was computed using the BENtbx (https://cfn.upenn.edu/~zewang/BENtbx.php; Wang et al. 2014). BEN was calculated at each voxel using Sample Entropy (SampEn; Lake et al. 2002). SampEn is one of the approximate entropy measures that quantifies the temporal coherence of a time series by computing the “logarithmic likelihood” that a small section (within a window of length “m”) of the data that matches with other sections will still match the others if the section length increases by 1. “Match” is defined by a cutoff threshold “r”. Based on previous studies (Wang et al. 2014), the optimal window length “m” was set to 3 and the optimal cutoff threshold ‘r’ was set to 0.6. The entropy values for all voxels that formed the BEN map were calculated using the “batch_calc_BEN” code in BENtbx. Subsequently, the BEN map was normalized to the MNI standard space using SPM8 and resampled with a resolution of 2x2x2 mm³ and then smoothed with an isotropic Gaussian kernel (FWHM = 6 mm³).

Table 1

Descriptive statistics of the demographic and behavioral assessments

	Discovery dataset (N = 386)	Independent sample 1 (N = 431)	Independent sample 2 (N = 132)
Age	20.00 ± 1.16	19.56 ± 1.47	22.67 ± 5.99
Sex (M/F)	182/204	119/312	36/96
IQ	65.60 ± 3.96	127.91 ± 8.53	8.74 ± 1.89
Originality	45.30 ± 16.40	24.57 ± 10.33	\
Flexibility	26.70 ± 6.30	6.46 ± 2.19	\
Fluency	57.30 ± 18.60	10.62 ± 4.22	\
Total score	129.00 ± 39.30	41.66 ± 16.25	1.70 ± 0.27

Figure 1

Top: regions of positive correlations between regional BEN and total creativity scores. Bottom: scatter plots depicting significant correlations between regional entropy values and total creativity scores. dACC/pre-SMA: dorsal anterior cingulate cortex/pre-supplementary motor area; DLPFC: dorsolateral prefrontal cortex; IFG_Tri: the triangular region of the inferior frontal gyrus; IFG_Oper: the opercular region of the inferior frontal gyrus.

BEN-Behavior Correlation Analysis

A multiple regression analysis was performed using SPM8 to explore the brain regions in which the regional BEN value was correlated with the creativity score. Age, sex, and the intelligence score were included as nuisance covariates. The small-volume correction (SVC) method was performed across the regions of interest (ROIs) defined by previous studies (Fink et al. 2009; Abraham et al. 2012; Jung et al. 2013; Chen et al. 2014; Wu et al. 2015; Beaty et al. 2016). The statistical significance level was set at P < 0.05, with SVC for multiple comparisons (family-wise error, FWE corrected) in ROIs. These ROIs are all key nodes of the DMN, CCN, and SN, including the PCC, MTG, DLPFC, IFG, and dACC, and were created using the Wake Forest University Pick Atlas (Maldjian et al. 2003).

Validation Analysis

To improve the strength and reliability of our work, we validated our main results in two independent samples (see Supplementary Materials and Methods). First, we produced binary masks for those regions that showed a significant correlation with divergent thinking in the main findings. Then, each region’s average entropy value was extracted from the BEN map in two independent samples by applying the above binary masks described above. Finally, a partial correlation analysis was performed to examine whether the entropy values of these regions also correlated with individuals’ divergent thinking in the two independent samples after regressing out the age, sex, and intelligence score. Multiple comparisons were performed using the false discovery rate.

Results

Behavioral Assessments

Table 1 shows the means and SDs of the demographic and behavioral assessments of all subjects involved in the present study.

BEN-Behavior Correlation Analysis

First, no significant result was obtained after applying the voxel-level FWE correction (P < 0.05, cluster size > 10) at the whole-brain level. Then, based on our strong hypothesis, the SVC method (P < 0.05, FWE corrected) was used for multiple comparisons. The results are presented in Figure 1 and Table 2. Significant positive correlations between the total creativity score and regional BEN were observed in the left dACC extending into the pre-supplementary motor area (dACC/pre-SMA, cluster size = 44 voxels; peak coordinates in MNI: x, y, z = −8, 18, 54, t_peak = 4.99, r = 0.222, P_(svc) < 0.05), left DLPFC (cluster size = 154 voxels; peak coordinates in MNI: x, y, z = −16, 40, 50, t_peak = 3.82, r = 0.220, P_(svc) < 0.05), left opercular region of the IFG (IFG_Oper: cluster size = 53 voxels; peak coordinates in MNI: x, y, z = −36, 16, 34, t_peak = 4.48, r = 0.223, P_(svc) < 0.05), and left triangular region of the IFG (IFG_Tri, cluster size = 66 voxels; peak coordinates in MNI: x, y, z = −56, 28, 26, t_peak = 4.13, r = 0.219, P_(svc) < 0.05). No significant negative correlation between the total creativity score and regional BEN in the brain was found.

Table 2

Brain regions in which the regional BEN correlate with verbal creativity

Creativity	Region	Side	Peak coordination (MNI)			Cluster size
			X	Y	Z	k (voxels)	T-score
Total creativity score
	dACC/pre-SMA	L	−8	18	54	44	4.99*
	DLPFC	L	−16	40	50	154	3.82*
	IFG_Oper	L	−36	16	34	53	4.48*
	IFG_Tri	L	−56	28	26	66	4.13*
Flexibility score
	dACC/pre-SMA	L	−8	18	54	42	5.12*
	DLPFC	L	−16	40	50	197	3.97*
	IFG_Oper	L	−36	16	34	53	4.62*
	IFG_Tri	L	−56	28	26	76	4.51*
Fluency score
	dACC/pre-SMA	L	−8	18	54	51	4.61*
	DLPFC	L	−16	48	34	118	3.72*
	IFG_Oper	L	−34	14	36	65	4.75*
	MTG	L	−56	−16	−12	38	4.31*
Originality score
	dACC/pre-SMA	L	−8	18	54	28	4.38*
	IFG_Oper	L	−38	18	34	27	3.97*

Note: MNI = Montreal Neurological Institute; dACC/pre-SMA = dorsal anterior cingulate cortex/pre-supplementary motor area; DLPFC = dorsolateral prefrontal cortex; IFG_Tri = the triangular region of the inferior frontal gyrus, IFG_Oper = the opercular region of the inferior frontal gyrus; MTG = middle temporal gyrus. ^*Significant levels for correction were set at P < 0.05, small-volume-corrected.

Figure 2

Top: regions of positive correlations between regional BEN and the three dimensions of verbal creativity (originality/fluency/flexibility scores). Bottom: scatter plots depicting significant correlations between regional entropy values and the three dimensions of verbal creativity (originality/fluency/flexibility scores). dACC/pre-SMA: dorsal anterior cingulate cortex\pre-supplementary motor area; DLPFC: dorsolateral prefrontal cortex; IFG_Tri: the triangular region of the inferior frontal gyrus; IFG_Oper: the opercular region of the inferior frontal gyrus; MTG: middle temporal gyrus.

Figure 3

Scatter plots depicting significant correlations between regional entropy values of the left dACC/pre-SMA and/or DLPFC and creativity scores in independent sample 1 (top: total score; middle: originality score; bottom: flexibility score). dACC/pre-SMA: dorsal anterior cingulate cortex/pre-supplementary motor area; DLPFC: dorsolateral prefrontal cortex.

Significant positive correlations between the three dimensions of divergent thinking and regional BEN were also observed in the resting brain (Figure 2 and Table 2). Positive correlations between flexibility and regional BEN were found in the left dACC/pre-SMA (cluster size = 42 voxels; peak coordinates in MNI space: x, y, z = −8, 18, 54, t_peak = 5.12, r = 0.214, P_(svc) < 0.05), left DLPFC (cluster size = 197 voxels; peak coordinates in MNI space: x, y, z = −16, 40,50, t_peak = 3.97, r = 0.184, P_(svc) < 0.05), left IFG_Oper (cluster size = 53 voxels; peak coordinates in MNI space: x, y, z = −36, 16, 34, t_peak = 4.62, r = 0.223, P_(svc) < 0.05), and left IFG_Tri (left: cluster size = 76 voxels; peak coordinates in MNI space: x, y, z = −56, 28, 26, t_peak = 4.51, r = 0.170, P_(svc) < 0.05). Positive correlations between fluency and regional BEN were observed in the left dACC/pre-SMA (cluster size = 51 voxels; peak coordinates in MNI space: x, y, z = −8, 18, 54, t_peak = 4.61, r = 0.220, P_(svc) < 0.05), left DLPFC (cluster size = 118 voxels; peak coordinates in MNI space: x, y, z = −16, 48, 34, t_peak = 3.72, r = 0.217, P_(svc) < 0.05), and left MTG (cluster size = 38 voxels; peak coordinates in MNI space: x, y, z = −56, −16, −12, t_peak = 4.31, r = 0.160, P_(svc) < 0.05). Positive correlations between originality and regional BEN were observed in the left dACC/pre-SMA (cluster size = 28 voxels; peak coordinates in MNI space: x, y, z = −8, 18, 54, t_peak = 4.38, r = 0.202, P_(svc) < 0.05) and left IFG_Oper (cluster size = 27 voxels; peak coordinates in MNI space: x, y, z = −38, 18, 34, t_peak = 3.97, r = 0.200, P_(svc) < 0.05).

Validation Analysis

The associations between BEN and divergent thinking identified in the main cohort were cross-validated in two independent cohorts.

Independent sample 1: This sample consisted of 431 Chinese subjects from the Gene-Brain- Behavior project of Southwest University. The behavioral and functional imaging tests were the same as those used in our main discovery dataset (see Supplementary Materials and Methods). The validation results showed a positive correlation between the entropy value of the left dACC/pre-SMA and the total score (r = 0.134, P = 0.005), and the entropy value of the left DLPFC was marginally positively correlated with the total score (r = 0.094, P = 0.051; Figure 3). Similarly, positive correlations between the entropy value of the left dACC/pre-SMA with originality (r = 0.146, P = 0.002) and flexibility (r = 0.141, P = 0.003), and between the entropy value of the left DLPFC with originality (r = 0.103, P = 0.032) and flexibility (r = 0.108, P = 0.025) were also observed (Figure 3).

Independent sample 2: This sample included 132 subjects from the University of North Carolina at Greensboro (Beaty et al. 2018). The subjects completed a different verbal creativity test, namely, the AUT (see SI Materials and Methods). However, consistent results were obtained. The regional entropy value of the left dACC/pre-SMA was positively correlated with the creativity score (r = 0.232, P = 0.007). Similarly, the entropy value of the left DLPFC was also positively correlated with the creativity score (r = 0.195, P = 0.025; Figure 4).

Figure 4

Scatter plots depicting significant correlations between regional entropy values of the left dACC/pre-SMA and/or DLPFC and creativity score in independent sample 2. dACC/pre-SMA: dorsal anterior cingulate cortex/pre-supplementary motor area; DLPFC: dorsolateral prefrontal cortex.

Given that the discovery sample and independent sample 1 were both Chinese populations, we also combined the two Chinese samples to examine the relationship between creativity and resting BEN within a larger sample. A multiple regression analysis was performed on the combined sample, controlling for age, gender, mean FD, and scanning session of two samples. The SVC method (P < 0.05 corrected) was also applied here for multiple comparisons. Similar results were observed whatever using the combined Chinese sample (see Supplementary Table 2) or two separate Chinese samples.

Discussion

The present study aimed to examine the associations between regional BEN and divergent thinking using a large cohort of healthy subjects. BEN is an emerging brain activity measure that has been increasingly used to assess brain states but has rarely been used to assess normal cognitive abilities. Based on our results, divergent thinking is consistently positively correlated with the regional BEN of the left dACC/pre-SMA and left DLPFC in both the discovery dataset and two validation datasets. In addition, significant correlations were observed between the three dimensions of divergent thinking (flexibility, fluency, and originality) and regional BEN in the left IFG and left MTG. Together, these findings provide the first evidence of the associations of regional BEN with individual variations in divergent thinking.

In this study, divergent thinking was found to be consistently positively correlated with the regional BEN of the left dACC/pre-SMA and left DLPFC in both the discovery dataset and two validation datasets, suggesting that highly creative individuals may exhibit more irregular and variable activity in these brain regions. This finding is consistent with results from a recent study reporting a correlation between verbal creativity and the temporal variability of the FC patterns of the control networks (Sun et al. 2018). Creativity is closely related to the ability to switch between remote semantic concepts and organize them into creative associations in a flexible manner (Bossomaier et al. 2009; Fink et al. 2009), which may be reflected by brain variability or flexibility during the resting state. Larger variability of the left dACC/pre-SMA and left DLPFC in our results indicated more variable and flexible resting brain activity, which may facilitate executive functions that are subserved by these regions. Supporting evidence from previous studies revealed that the dACC/pre-SMA and DLPFC are implicated in various processes related to creativity. For example, the dACC/pre-SMA is often activated during different creative tasks, such as insight problem solving (Kounios et al. 2008; Qiu et al. 2010), word association (Bechtereva et al. 2004), divergent thinking (Carlsson et al. 2000), and musical improvisation (Bengtsson et al. 2007). The function of the dACC/pre-SMA is involved in the response selection and inhibition linked to response flexibility (Nee et al. 2007; Dosenbach et al. 2008), supporting the executive control aspect of creative cognition (de Manzano and Ullén 2012). Similarly, the main functions of the DLPFC are inhibitory control and cognitive flexibility (Miller and Cohen 2001; Koechlin et al. 2003; Alvarez and Emory 2006). The inhibitory control process points to the role of executive processes in suppressing ordinary responses (Benedek et al. 2014), while cognitive flexibility is involved in the executive processes required for shifting fixed mindsets and applying novel criteria to combine remote concepts (Howard-Jones et al. 2005; Sawyer 2011), all of which are crucial for creative information processing (Dietrich 2004; Dietrich and Kanso 2010). For example, behavioral studies have reported a close association between higher creativity and higher cognitive flexibility (Zabelina and Robinson 2010; Benedek et al. 2014; Chen et al. 2014). As noted above, entropy is a measure of the variety of change patterns in a time-series signal and depicts irregularities in brain activity (Shannon and Weaver 1950). A higher entropy value indicates large, irregular, and variable brain fluctuations. Entropy is also related to the ability to shift among different states, which is defined as flexibility (Shuiabi and Thomsonab 2005). Supporting evidence from a rs-fMRI study showed that BEN measured at rest reveals the overall flexibility or readiness of the brain to respond to unpredictable stimuli. This study even identified a close association between entropy in a variety of widespread brain regions and the performance on the Matrix Reasoning task, which requires flexible reasoning about novel stimuli (Saxe et al. 2018). Therefore, together with the previous finding that people who can easily shift between different concepts or states during creative problem solving and are capable of cognitive flexibility may obtain a high creativity score (Fink et al. 2009), the large irregularity and variable brain fluctuations that are manifested as high entropy values may be closely related to cognitive flexibility during creative problem solving. Taken together, the high entropy of the left DLPFC in highly creative individuals suggests that creativity is closely related to the functional dynamics of the control networks involved in cognitive flexibility and inhibitory control.

In addition, a higher entropy value indicates higher frequency fluctuations. Technically, the presence of more high frequency fluctuations in correlated signals may reduce their correlation (Wang et al. 2014; Zhou et al. 2016). Thus, a higher entropy value of the control network may reflect a weaker FC within the control network. Creativity is negatively associated with executive–executive (within-network) coupling during both resting-state and task-based fMRI (Zhu et al. 2017; Shi et al. 2018), which may partially support our findings of the positive correlations between verbal creativity and the regional entropy value of the control network.

In the present study, the three dimensions of divergent thinking (flexibility, fluency, and originality) positively correlated with regional BEN in the left IFG and left MTG. Based on accumulating evidence, creativity is related to the involvement of the IFG and MTG. For example, structural MRI studies revealed associations between increased cortical thickness and/or volume of the bilateral IFG with higher verbal creativity (Takeuchi et al. 2012; Zhu et al. 2013). Resting-state studies also observed greater FC of the IFG and other networks (e.g., DMN) in more highly creative individuals (Beaty et al. 2014a, 2014b). Consistent with these findings, previous studies have observed increased activation of the MTG during DTTs (Kleibeuker et al. 2013b, 2017). The IFG and left MTG belong to the semantic system (Foster et al. 2005; Binder et al. 2009). The function of the IFG is the retrieval and selection of relevant remote associations (Benedek et al. 2014a, 2014b), while the MTG is associated with the activation of long-term memory (Martin 2001), which is closely implicated in the creative idea generation process (Fink et al. 2009; Abraham 2014). Thus, highly creative individuals have a more flexible semantic associative network, as reflected by the large entropy values of the IFG and MTG.

Furthermore, in previous studies, patients with schizophrenia exhibited high entropy values at the mean whole-brain level (particularly in the frontal lobe), which also partially support our results. A close relationship between psychiatric disorders (particularly schizophrenia) and creativity has been identified. For example, a behavioral study showed that individuals who exhibit greater performance in artistic and creative fields have higher schizotypy scores (Burch et al. 2011). Genetic research has also revealed an association between a creative profession or an artistic society membership and higher polygenic risk scores for schizophrenia, suggesting that creativity and psychosis may share genetic roots (Power et al. 2015). Together with the finding that patients with schizophrenia have more complex fMRI signals that are reflected by higher entropy values than healthy controls (Sokunbi et al. 2014), our results showing a positive correlation between creativity and regional entropy in the resting brain further support the hypothesis that people with schizophrenia and highly creative individuals share common psychological attributes from the perspective of brain function dynamics.

More importantly, our main results were well replicated in both independent samples from two different cultures. Positive correlations between the total verbal creativity score and regional BEN in the left dACC/pre-SMA and left DLPFC were found in both independent samples, although the correlation between the total score and regional BEN in the left DLPFC was marginally significant in independent sample 1. However, the results were well replicated in independent sample 2, even using a different creativity test. Regarding the three dimensions of verbal creativity, regional BEN in the left DLPFC was positively correlated with originality and flexibility scores in independent sample 1, which is slightly different from our main results that regional BEN in the left DLPFC positively correlated with fluency and flexibility scores. However, the flexibility score consistently correlated with the regional BEN in the left DLPFC in both the discovery and validation datasets. As noted above, the entropy value depicts irregular and variable fluctuations in resting-state brain activity, and the entropy value of the DLPFC may be related to cognitive flexibility, which is associated with creativity (Zabelina and Robinson 2010; Chen et al. 2014). Thus, these findings may explain the robust relationship between the flexibility score and regional entropy value of the left DLPFC.

Finally, the present study has several limitations. First, creativity is a multidimensional process, and divergent thinking is a critical indicator of creativity (Runco and Acar 2012). The present study mainly focused on the association between BEN and divergent thinking. However, convergent thinking is also considered an aspect of creativity (Cropley 2006; Jung et al. 2013). Therefore, the current findings should be explained within the context of divergent thinking and researchers designing future studies should considering examining the association between BEN and convergent thinking. Second, BEN is a data-driven voxel-based SampEn approach. SampEn is largely independent of data length and displays relative consistency over a broader range of possible parameters (Richman and Moorman 2000). It is also fairly unaffected by low-level noise and is robust to large or small artifacts (Zhang and Roy 2001). However, the calculation of SampEn is based on a prespecified distance threshold (r = 0.6 in this study), which may have affected the estimated entropy but should not have affected the group-level entropy analysis. Future studies should explore the effects of different thresholds and attempt to identify the optimal threshold for your particular datasets.

Conclusion

In summary, our results first identified a robust correlation between verbal creativity and regional BEN in the resting brain and showed that a higher BEN and higher information processing capacity predict higher potential creativity. Specifically, verbal creativity is consistently positively correlated with regional BEN in the left dACC/pre-SMA and left DLPFC, which is involved in cognitive flexibility and inhibitory control. In addition, the three dimensions of verbal creativity (fluency, flexibility, and originality scores) positively correlate with regional BEN in the left IFG and left MTG, which indicates a flexible semantic associative network in highly creative individuals. More importantly, the findings were successfully replicated in two different cultural populations, confirming the strength and reliability of our results. In general, these findings provide new insights that improve our understanding of the relationship between the temporal dynamics of the resting brain and verbal creativity.

Funding

National Natural Science Foundation of China (31470981, 31571137, 31500885, 31600878, 31771231); Project of the National Defense Science and Technology Innovation Special Zone; Chang Jiang Young Scholar; National Program for Special Support of Eminent Professionals (National Program for Support of Top-notch Young Professionals); Program for the Top-notch Young Professionals by Chongqing; Fundamental Research Funds for the Central Universities (SWU1609177); Natural Science Foundation of Chongqing (cstc2015jcyjA10106); Fok Ying Tung Education Foundation (151023); Research Program Funds of the Collaborative Innovation Center of Assessment toward Basic Education Quality at Beijing Normal University.

Conflict of Interest: None declared.

Author notes

^†Liang Shi and Roger E. Beaty contributed equally to this work