A total of 42 right-handed healthy young adults were recruited in this study. The inclusion criteria were as follows: (a) non-smoking and without drug addiction; (b) not pregnant; (c) no history of neurological and psychiatric disorder history, and (d) no metal body implants. The entire protocol was approved by the Research Ethics Committee of National Taiwan University (NTU-REC No. 201407EM028). Because two participants could not complete the entire experimental procedure, their datasets were excluded in the analysis. Accordingly, the following results are reported for the remaining 40 participants (19 males, mean age = 24.4 ± 2.8 years, range: 21–33 years).

The stimuli consisted of 56 grayscale images of objects from daily life with background removed. Prior to the actual fMRI experiment, we rated the number of alternative usages for each object stimulus from a separate group of participants (N = 15, 12 males, mean age = 23.4 ± 3 years, range: 20–30 years). During the rating, the 56 objects were presented consecutively to the participants with inter-trial intervals (ITIs) of 26 s. The participants were instructed to think and express alternative usages for each object within 20 s. The mean number of answer for each object was 2.4. We categorized the difficulty of AUT into three levels as follows: normal, 1.99–2.99 answers per picture; difficult, <1.99 answers per picture; and easy, >2.99 answers per picture. Subsequently, we used the task materials to design the AUT in MRI environments. All of the objects were presented against a white background for both rating and fMRI experiment.

After consenting, the participants were instructed to complete a CAQ, containing eight aspects of creativity (Carson et al., 2005), and to undergo the fMRI experiments. The fMRI experiment included seven sessions in total: two resting-state, two 0-back and three AUT sessions. The participants underwent the first rs-fMRI scan (Pre-AUT), performed three consecutive AUTs with two 0-back tasks in between, and went through the second rs-fMRI scan (Post-AUT) at the end. The participants only received the instruction to generate novel ideas after viewing pictures in the AUT sessions without practices to ensure the Pre-AUT resting state was irrelevant to divergent thinking. The multiple AUTs were designed to enhance the divergent thinking process with undemanding tasks (0-back) in between (Baird et al., 2012). Herein, we considered all AUTs and 0-backs as instructed divergent-thinking engagements, and adopted the two sessions of resting-state scans (Pre-AUT and Post-AUT) to discern the alterations of RSFC through the divergent thinking process.

MRI experiments were conducted by a 3T MRI scanner (Prisma, Siemens, Erlangen, Germany) with 20-channel head coil at National Taiwan University. During the experiment, a high-resolution T1-weighted anatomical image was initially scanned using the 3D-MPRAGE sequence with 256 × 256 × 192 matrix size; 0.93 × 0.93 × 0.93 mm³ resolution; inversion time (TI) = 900 ms; repetition time (TR) = 2,000 ms, echo time (TE) = 2.3 ms; flip angle (FA) = 8°; bandwidth = 200 Hz/pixel; NEX = 1. Total scan time was 6 min 14 sec. The fMRI protocol was using a single-shot gradient-echo echo-planar imaging (GE-EPI) sequence with following imaging parameters: TR = 2 s, TE = 34 ms, FA = 84°, bandwidth = 3,005 Hz/pixel, matrix size = 64 × 64 × 33, and FOV = 210 × 210 mm². Stimuli were presented via E-prime 2.0 (Psychology Software Tools, Pittsburgh, PA, USA) with a back-projection projector in 800 × 600 resolution. Participants viewed the stimuli using a mirror mounted on the head-coil and the viewing field was 8.4° (horizontal) by 6.3° (vertical) at a viewing distance of 420 cm. Participants were instructed to respond with a button press using the index and middle fingers of their right hand (Lumina response pad; Cedrus, San Pedro, CA, USA).

During the rs-fMRI scans, the participants were instructed to stay still with their eyes open, to not fall into sleep and to not think of anything in particular. The rs-fMRI scan contained 180 measurements (6 min) per session. During the AUTs, the participants were instructed to view gray-scale pictures of various objects from daily life (through a projector), and to think of as many alternative usages as possible within a fixed duration (20 s). Each AUT session contained 12 trials, including one easy-level picture, one difficult-level picture, and ten normal-level pictures to balance the difficulty level, with fixation and cues of 6 s in between; within each trial one single picture was presented for 20 sec. During the 20 s of the picture presentation, the participants were instructed to press the button immediately when they thought of a special usage; and the button-press numbers and the response time were recorded. Each AUT task contained 156 scans (5 min 12 s) in total. After the cessation of each AUT, the participants were asked to recall the answers retrospectively, and the answers were manually recorded by the experimenter. During the 0-back tasks, one English letter was displayed on screen for 0.5 s, followed by a cross fixation for 1.5 s. The participants were instructed to respond using their right hand to press the right button when the letter appearing on the screen was “X”; otherwise, they were to press the left button. The 0-back fMRI contained 152 scans (5 min 4 s) in each session. The datasets generated for this study are available on request to the corresponding author.

Beyond the creativity assessment obtained through the CAQ, the two creativity indices of AUT (e.g., originality and fluency) were also evaluated for each participant according to the scoring method of previous studies (Hao et al., 2015). Fluency was estimated on the basis of the total number of ideas given during the AUT trials. Originality was estimated on the basis of statistically infrequent responses using the following means. The generated answers from the AUT were collected into a comprehensive lexicon for comparison. Three trained raters independently assessed the originality of the AUT performance for every participant. If 10% or less of the participants in the sample gave the same response then it was given a score of “1,” and all other responses received scores of “0.” The inter-rater agreement was with intra-class correlation coefficient of 0.97. Subsequently, the originality scores of the AUT performance assigned by the three raters were averaged for every participant.

The data were processed using Analysis of Functional Neuro Images (AFNI) (Cox, 1996). We processed the task data according to the following steps: (a) motion correction; (b) coregistration; (c) smoothing with full width at half maximum (FWHM) = 6 mm, and (d) normalization to the MNI space. During the AUT sessions, participants' responses were recorded and used as the events for divergent thinking, enabling the event-related fMRI analysis. Using the events convolved with the canonical hemodynamic function, we subsequently applied the generalized linear model (GLM) to generate the activation map for each AUT and calculated the average beta map across three AUTs for each individual. The consequent AUT results were equally divided into two groups (n = 20 for each of the high- and low-creative groups) according to their CAQ scores (the cut-off score for the CAQ was 8), and a two-sample t-test was conducted to generate the contrast maps of AUT brain activity.

The rs-fMRI data were processed according to the following steps: (a) field-map correction; (b) motion correction; (c) coregistration; (d) detrending and filtering (0~0.1 Hz); (e) smoothing with FWHM = 5 mm, and (f) normalization; and (g) the physiological noise removal including white matter, cerebrospinal fluid and six motion parameters. Subsequently, the functional connectivity maps before and after AUT were established through the seed-correlation analysis, targeting on two specific brain networks: (1) the bilateral posterior cingulate cortex (PCC) [±3, −53, 26] for assessing DMN (Van Dijk et al., 2010; Yan et al., 2013), and (2) the seeding at the peak of AUT contrasts between the high- and low-creative groups. Subsequently, the group-based RSFC maps underwent the following statistical analyses to estimate the brain reconfigurations.

All group analyses of the fMRI results were conducted on the basis of analysis of variance (ANOVA) and t-test using AFNI. The group difference in brain activity during AUTs were evaluated through a two-sample t-test between the high- and low-creative groups. The two brain networks associated with divergent thinking were compared for RSFC according to the 2 (high- and low-creative) × 2 (pre- and post-AUT) two-way ANOVA (3dMVM), inclusive of the mean framewise displacement to minimize the motion effect on RSFC findings. The multiple comparisons in the group analysis were corrected through the 3dClustSim approach with the auto-correlation function and the significance level was p < 0.05 (uncorrected p < 0.001, cluster size = 80 mm³). To further determine the relationships between creativity scores and the divergent thinking effect of RSFC, the Pearson correlation analysis was used to measure the association between the behavior indices of each individual (originality and fluency) and the RSFC changes (Post-Pre) from selected regions of interested (ROIs). The centers of the spherical ROIs (radius = 5 mm) were placed at the peak value of the significant changes in the contrast maps. The effect of divergent thinking was assessed by calculating the RSFC differences between Pre- and Post-AUT after Fisher's z transformation of the correlation coefficients. The z-transformed RSFC changes from the selected ROIs were correlated with the creativity indices (originality and fluency) with false-discovery rate (FDR) correction across ROI-wise comparisons. The statistical calculations were performed using SPSS 20 (IBM Statistical Package for the Social Sciences).

The average CAQ score for the 40 participants was 10.8 ± 10.4, ranging from 1 to 62. Regarding the overall AUT performances, the evaluated originality and fluency were 0.33 ± 0.11 and 3.1 ± 1.0, respectively. The CAQ scores were adjusted by the log-transform (Form and Kaernbach, 2018), and the log(CAQ) scores exhibited a positive trend toward the AUT fluency (r = 0.30, p = 0.06), without prominent relationships with originality (r = 0.22, p = 0.16) for all participants. To test the difference in brain activity between high and low creativity individuals, the following results were divided into two groups with equal sample size (n = 20 for each group); where their CAQ scores were 16.6 ± 12.2 for the high-creative group and 5.0 ± 1.9 for the low-creative group. In addition, the two groups showed no significant differences in age (p = 0.06), gender (p = 0.53) and motion during the two rs-fMRI scans (framewise displacement: p = 0.39).

The AUT-associated brain activations were located at the left superior temporal gyrus, anterior cingulate gyrus, declive of the cerebellum, caudate and occipital visual cortex (Figure 1, corrected p < 0.01). Figures 1A,B illustrate the average activation maps of AUT for the high-creative and low-creative groups, respectively, where the high-creative group exhibited relatively broad spatial extents in AUT engagements relatively. Negative activations were found lateralized to the right hemisphere, specifically located at the right PCC and right inferior parietal lobule. Figure 1C depicts the contrast between the two groups by independent t-test (corrected p < 0.05). A difference between the groups was observed only in the left IFG (BA47, [−32, 24, −14]), which was taken as the seed localization for the subsequent RSFC analyses. Supplementary Figure 1 shows the overall AUT activation map across all participants and Supplementary Table 1 lists the detailed description of AUT-related brain activities.

To further elucidate the alterations of brain integrity over the divergent thinking process, we applied a seed-correlation approach to investigate RSFC. Two brain networks were targeted with the seed locations in bilateral PCC for DMN and left IFG for IFG-associated network (IFGN) to assess the group (high- vs. low-creative) and divergent thinking (Pre- vs. Post-AUT) effects. The upper panel of Figure 2 illustrates the DMN connectivity maps exhibiting strong connections to the bilateral medial and middle frontal, PCC, and bilateral angular gyrus (corrected p < 0.01). The lower panel of Figure 2 illustrates the IFGN connectivity maps (corrected p < 0.01). The IFGN was associated with the bilateral superior frontal, medial frontal, anterior cingulate, insula, superior, and middle temporal gyri. The detailed information of the group × divergent thinking comparisons (high-creative, low-creative, Pre-AUT and Post-AUT) for IFGN and DMN is listed in Table 1. Prior to the AUT engagements (Pre-AUT), the only significant difference in IFGN between the groups was in the right angular gyrus (AG) and right inferior parietal lobule (IPL), but no group difference was observed in DMN at the baseline level. However, the RSFC of both networks changed after performing the AUT. Table 1 demonstrates that the group difference of RSFC_DMN became prominent after AUT engagements, and the RSFC_IFGN showed different network reconfigurations between the two groups.

An ROI analysis was conducted to determine the relationship between RSFC changes and creativity scores. The spherical ROIs were placed according to the results in Table 1. Figure 3 presents the RSFC changes between Pre-AUT and Post-AUT and their relationships with AUT originality and fluency in both groups. The RSFC changes in IFGN at the right AG and right IPL were positively correlated with the AUT performances: originality and fluency (originality-ΔFC_IFG−AG: r = 0.51, p < 0.02; fluency-ΔFC_IFG−AG: r = 0.49, p < 0.03; originality-ΔFC_IFG−IPL: r = 0.63, p < 0.001) in the high-creative group, but the correlations in the low-creative group were non-significant. The Steiger's Z-test presented significant group differences in the correlation coefficients of originality-ΔFC_IFG−AG (p < 0.001), fluency-ΔFC_IFG−AG (p < 0.004) and originality-ΔFC_IFG−IPL (p < 0.001). Supplementary Table 2 summarizes all correlations between RSFC and creativity assessments [log(CAQ), AUT fluency and originality].

The integrity of brain network can be modulated by the previous cognitive engagements within minutes, and the functional re-organization may be important to the cognitive performances. Previous research investigated the correlation between the IFG connectivity and the subsequent memory (Stevens et al., 2010), disclosing the connectivity facilitates the recognition accuracy. In the current study, we focused on the dynamic RSFC changes of the two networks, DMN and IFGN, corresponded to the networks related with high-creative thinking ability reported in literature (Beaty et al., 2018). Interestingly, the same study also demonstrated the evidence of dynamic RSFC alterations following divergent thinking. In their supporting information, Beaty et al. presented fairly similar correlations between network strengths and creativity scores based on a resting-state dataset (r = 0.13 for high-creative networks and r = 0.11 for low-creative networks). However, the RSFC–creativity correlation became elevated for high-creative networks when performing the AUT (r = 0.35 and 0.28 for two separate datasets), but the correlation turned out non-significant for low-creative networks during AUT engagements (r = 0.02 to −0.04 for two separate datasets). The evidence implies that dynamic network reorganizations in AUT can be associated with the creativity performances. In this study, assuming that the short-term network reorganizations induced by AUT could sustain after the task cessation (the second resting state in Post-AUT), we demonstrated that dynamic RSFC alterations were associated with the AUT performance (originality and fluency). Specifically, we prescribed the ROIs from both the creativity trait effect (High-creative > Low-creative defined by CAQ in Table 1) and the creativity state effect (Post-AUT > Pre-AUT), because the CAQ (trait) and AUT performances (state) were highly correlated within each group (Supplementary Table 2). Our results showed that DMN demonstrated the trait difference between groups; however, no prominent state effect was found in DMN. Relatively, IFGN contained both trait and state effect (Table 1), indicating that the IFGN indeed presented the network reconfiguration after AUT engagements. The concept of network reconfiguration was supported by Wei et al. (2014), in which they demonstrated that the RSFC strengths was modulated in general after the performance of a cognitive stimulation task. The dynamic alterations of RSFC following cognitive tasks can be regarded as a short-term functional reconfiguration of brain circuits to facilitate associated cognitive tasks, which fits the neurophysiological perspective of the dynome framework (Kopell et al., 2014). For example, Wang et al. presented the dynamic reorganizations of DMN during and after a word-picture matching task, so as the corresponding whole-brain small-world topology (Wang et al., 2012). However, previous studies did not specifically present the relationship between the network reconfiguration and the cognitive performances. In the current work, Figure 3 illustrates that dynamic IFGN reconfigurations were in accordance with the creativity performances, supporting the short-term reconfiguration and sustenance of network integrity underlying the mental process of divergent thinking.

Three brain regions in the left hemisphere were reported to be involved in divergent thinking—the IFG, pre-/post-central gyri, and the MTG (Boccia et al., 2015), and these are similar to our results of brain activity in AUT. Generally, the lateralization of brain function in the left hemisphere in AUT is considered to be the verbal processing and semantic memory (Thompson-Schill, 2003). In addition, the left hemisphere executes the functions of interpreting and reasoning about the sentences and the causal relationships (Marinsek et al., 2014). By presenting the CAQ-based group difference of brain activation in AUT (Figure 1), the left IFG was highlighted in the performance of creativity, regarded as the functional localizer for the following RSFC analysis.

Figure 3 illustrates the relationship between originality and the corresponding RSFC changes after AUTs; the RSFC changes are prominent between the left IFG, right AG, and right IPL. Previously, the SPL/IPL was the core for top-down attention or abstract thinking (Shomstein, 2012), and the IPL was determined to usually play roles involving the voluntary orienting of attention to a location (Corbetta et al., 2000). In addition, the left IFG and right IPL involved the verbal working memory (LaBar et al., 1999), and the strength of RSFC between the left IFG and right IPL decreased in high working memory loading (Liu et al., 2017). Furthermore, a previous study reported that the low working memory loading task could help the divergent thinking and (Baird et al., 2012). For the multifunction of AG, the right side was determined to play the role of spatial cognition for perceptual learning and shifting attention to relevant information (Seghier, 2013). Therefore, our result demonstrated that the high-creative group leveraged from the frontoparietal reorganizations for elevating the abstract thinking, verbal working memory or spatial cognition involved in AUT. In contrast, the changes of RSFC in the low-creative group showed no relationship associated with the creative indexes. Following the finding, our results revealed the possibility that the low-creative group might adopt distinct strategies with respect to the semantic or sensory processing in AUT.

As evident in Supplementary Table 1, the majority of AUT-induced negative activations were located in the right hemisphere (the superior temporal gyrus, MTG, medial frontal gyrus, precuneus, precentral gyrus, and superior frontal gyrus) in both groups. Previously, the mechanism of the negative activation may have originated from the neural inhibition hypothesis, the affected neurotransmitter caused the reduction of local cerebral blood flow or the elevation of the cerebral metabolic rate of oxygen consumption in the cerebrovasculature (Sten et al., 2017). Interestingly, Benedek et al. also presented similar lateralized negative activation during idea generation (Benedek et al., 2014). To further verify the role of negative AUT activations in creativity, the negative brain activations were associated with the log-transform CAQ, originality, and fluency in both the high- and low-creative groups. We noted significantly positive correlations between the log(CAQ) and the right MTG [52, −10, −16] activity (r = 0.52, p < 0.02) in the high-creative group, as well as a significantly positive relationship between originality and the right MTG activity (r = 0.51, p < 0.02). However, the association between the negative activation and creativity performances is beyond the scope of the current work. Future studies are warranted to discern the underlying mechanism of divergent thinking.

About grouping of CAQ scores, we chose the median split because of a positively-skewed distribution of CAQ in Carson's report. Additionally, the average CAQ score of Carson's report was as high as 14.4, the cut-off line would be around 8 to 9 when we adopted the median split (Carson et al., 2005). Therefore, the cut-off threshold of 8 points was in agreement with Carson's report. Although we used the CAQ (the creative traits) for splitting participants into two groups, we confirmed the functional brain distinction between the high- and low-creative groups when performing AUT (Figure 1), revealing the active role of the left IFG in AUT engagements. By definition, the participants with high CAQ scores were believed to possess the capability of openness to experience (Carson et al., 2005), in part explaining that the high-creative group uses the brain systems of abstract thinking to achieve superior originality. Moreover, the between-group comparison disclosed that the DMN connectivity did not exhibit significant differentiation at the baseline level before AUT engagements (Table 1). In IFGN, the baseline connectivity strength between the left IFG and right IPL in the high-creative group was higher than that in the low-creative group at the Pre-AUT condition. This finding was contradictory to the previous report (Beaty et al., 2014). The disparity might be originated from the cultural difference, because previous study showed that frontal-parietal attentional control network involves in the visual and attention task and these functions exhibits cultural disparity (Han and Northoff, 2008; Hedden et al., 2008). Then, the connection from IFG to IPL involved the lexical judgment in Chinese studies (Deng et al., 2012). Future studies are warranted to prove such conjecture. In addition, when comparing the RSFC_IFGN changes between the two groups in Table 1, the low-creative group demonstrated more Pre-Post connectivity differences in the posterior brain, whereas the high-creative group mostly remained unchanged. Wei et al. observed similar between-group differences in RSFC between MTG and MPFC (Wei et al., 2014). They suggested that the group with higher TTCT scores exhibited fewer RSFC differences after creativity training, though the RSFC changes were not associated with the creativity performances. Collectively, it is speculated that each individual may possess specific brain-network plasticity to facilitate the performances of divergent thinking, where such brain reconfigurations are distinct between the high- and low-creative groups.

This study has several limitations. First, the sample size of the current study was above the request of 36 samples for sufficient fMRI replicability (Turner et al., 2018); however, the statistical power might be reduced after splitting the samples into two equal-size groups. Compared with literature, the AUT-induced brain activities were in well-agreement with previous findings, but the RSFC might be inconsistent with other studies. The sample size higher than 20 in each group is recommended for future neuroimaging studies with AUT engagements. Second, we observed only the two most addressed functional networks in AUT, namely DMN and IFGN, by which we might miss additional AUT-related brain reconfigurations in other brain networks, such as the reported global architectures involved in divergent thinking (Beaty et al., 2018). Future studies are warranted to test the whole-brain functional changes other than DMN and IFGN. Third, the group separation (based on CAQ) and their creative performances (i.e., originality and fluency) were all defined by the subjective preferences from the participants or the raters. However, creativity judgment is a relatively challenging task due to the lack of objective definitions, as aforementioned. Therefore, we used the AUT activity to verify the effectiveness of group separation and employed three raters to minimize the subjectivity involved in the creativity scoring. Fourth, we did not include any other cognitive tasks to test whether such network reconfiguration is specific to the divergent thinking. It was unanswered because the assumption that the network plasticity is subject to distinct cognitive tasks has yet been studied systematically. The post-cognition network plasticity is pending for further investigations in the near future. At last, the causal relation between the network reorganization and creativity performance remains limited, because the findings in this study was built upon the assumption that the AUT-associated network reconfigurations can sustain in a short period of time after the cessation of tasks. The short-term sustenance of functional organizations in brain circuits shall be further tested before confirming the causal relationship between network plasticity and creativity.