Rostral and caudal prefrontal contribution to creativity: a meta-analysis of functional imaging data

Creativity is of central importance for human civilization, yet its neurocognitive bases are poorly understood. The aim of the present study was to integrate existing functional imaging data by using the meta-analysis approach. We reviewed 34 functional imaging studies that reported activation foci during tasks assumed to engage creative thinking in healthy adults. A coordinate-based meta-analysis using Activation Likelihood Estimation (ALE) first showed a set of predominantly left-hemispheric regions shared by the various creativity tasks examined. These regions included the caudal lateral prefrontal cortex (PFC), the medial and lateral rostral PFC, and the inferior parietal and posterior temporal cortices. Further analyses showed that tasks involving the combination of remote information (combination tasks) activated more anterior areas of the lateral PFC than tasks involving the free generation of unusual responses (unusual generation tasks), although both types of tasks shared caudal prefrontal areas. In addition, verbal and non-verbal tasks involved the same regions in the left caudal prefrontal, temporal, and parietal areas, but also distinct domain-oriented areas. Taken together, these findings suggest that several frontal and parieto-temporal regions may support cognitive processes shared by diverse creativity tasks, and that some regions may be specialized for distinct types of processes. In particular, the lateral PFC appeared to be organized along a rostro-caudal axis, with rostral regions involved in combining ideas creatively and more posterior regions involved in freely generating novel ideas.


INTRODUCTION
Everyone has their own idea of what creativity is. While the realm of artistic creation may be the first that comes to mind, creativity is obviously a cornerstone of many domains of human activity, including science (discovery), technology (invention), and economy (innovation). However, it is not restricted to extraordinary achievements. Finding new solutions to individual problems, achieving something novel, and thinking away from pre-established ideas are all common creative processes that take place in everyday life. According to this point of view, creativity results from a set of mental functions normally found in all humans, and can be studied experimentally. From a neuroscience perspective, creativity is defined as the ability to produce work that is both novel (original) and appropriate or useful (Sternberg and Lubart, 1999;Plucker and Makel, 2010;Runco and Jaeger, 2012). Although this definition may appear reductive or simplistic, it makes experimental testing possible by allowing to form hypotheses about the cognitive processes involved in creativity, and to examine their brain correlates. However, the brain substrates of creativity have been poorly studied. In the existing research, various tasks related to four main theoretical frameworks of creativity have been used: divergent thinking, insight problem solving, combination of remote semantic associations, and artistic creativity. These studies have led to diverse results, with no consensus yet in sight (Fink et al., 2007;Arden et al., 2010;Dietrich and Kanso, 2010;Sawyer, 2011). The present meta-analysis represents an attempt to clarify this small body of literature.
Divergent thinking tasks assess the ability to generate multiple solutions to an open-ended problem that does not have a right or wrong answer (Guilford, 1950). The products of divergent tasks are evaluated according to several criteria, mainly fluency (the quantity of relevant responses), flexibility (the number of different categories of responses), originality (the degree to which responses are uncommon), and elaboration (the degree of enrichment of responses). In a review of six functional imaging studies that used divergent thinking tasks, Dietrich and Kanso (2010) highlighted the importance of the prefrontal cortex (PFC) without pinpointing a specific sub-region. Insight problem-solving tasks usually require one right answer, which allows rating responses as correct or incorrect. In these tasks, the solution often comes to mind with insight (an "eureka" or "aha" moment). Combining words that are remotely semantically related can also evoke an "aha" experience. One popular example of such a combination task is the Remote Associates Task, which consists of finding a word that links three stimulus words, for example, finding the word "cheese" for "rat," "blue," and "cottage" (Mednick et al., 1964b). Functional neuroimaging that uses these tasks has focused on this "aha" aspect rather than on the combinatorial or associative processes that lead to a solution. Dietrich and Kanso (2010) reviewed 11 electrophysiological and nine functional imaging studies on insight, including the Remote Associates Task, and highlighted the frequent involvement of the superior temporal gyrus (STG) and anterior cingulate cortex (ACC). Finally, six studies that examined creativity in the artistic domains of music, dancing, and painting, using ecological tasks were examined in the same review. No region was identified as necessary and sufficient for artistic creativity. Both prefrontal activation and deactivation were reported, possibly suggesting the existence of distinct types of creativity. Overall, these results are in agreement with two recent reviews of neuroimaging and electrophysiological studies of creativity (Arden et al., 2010;Fink and Benedek, 2013) that also highlighted the PFC region without converging to specific prefrontal sub-regions. Studies that used other methods in creativity research, such as voxel-based morphometry, diffusion weighted imaging, or cerebral blood flow (CBF) (Bechtereva et al., 2004;Chavez-Eakle et al., 2007;Jung et al., 2010a,b;Takeuchi et al., 2010a,b) showed a link between creative performance and the lateral frontal and parieto-temporal regions and their connections.
Despite the diversity of tasks and cognitive approaches used to measure creativity, its link with PFC activity is expected. A central role for the PFC during creative behavior is in agreement with cognitive theories according to which several prefrontal functions (such as flexibility, fluency, planning, or working memory) are key cognitive processes of creativity (Carlsson et al., 2000;Zeki, 2001;Dietrich, 2004;Mendez, 2004;Bogousslavsky, 2005;Changeux, 2005;Ward, 2007). However, the precise prefrontal sub-regions involved and their specific roles remain to be clarified. The brain location of functional imaging results was examined qualitatively in two previous reviews (Arden et al., 2010;Dietrich and Kanso, 2010), but not statistically. Consequently, the questions of whether creative thinking is statistically associated with particular sub-regions, and whether different aspects of creativity, measured by different tasks, can be related to distinct sub-regions, remain to be tested.
The aim of this meta-analysis was to identify both shared and unique neural correlates of creative thinking by performing a statistical comparison between multiple studies. We explored brain regions that are most consistently associated with creativity tasks in published functional imaging studies. The results are discussed in light of the data drawn from other methods, including patient studies. The observation of shared regions, despite the diversity of tasks and criteria used to measure creativity, would suggest the existence of a core network for creativity. In addition, in order to determine whether there are process-specific regions, experiments were categorized according to tasks (combination or unusual generation tasks) and stimuli (verbal or non-verbal). The latter distinction aimed at comparing the correlates of creativity in two distinct classical domains of cognition (verbal or non-verbal). The former distinction was based on two separate and influential cognitive theories of creativity mentioned above. The first theory emphasizes the importance of combinatorial processes in creative thinking (new combination of remote associates) and was operationalized in the Remote Associates Task by Mednick (Mednick, 1962;Mednick et al., 1964a,b;Ward and Kolomyts, 2010). Combination tasks involve associating separate and remote elements of information to form a new idea. The second theory, derived from Guilford's work (Guilford, 1950;Runco, 2010), focuses on the level of fluency, flexibility, and originality of generated ideas, and has mainly been operationalized using divergent thinking tasks, such as the Alternate Uses Task. Tasks in the unusual generation category thus consist of producing original or unusual responses to a given stimulus or situation.

SELECTION OF THE STUDIES
Studies were all peer-reviewed and published in English before June 2012. The PubMed and Scopus Medline databases were searched using the following keywords in text and/or abstract/title and Boolean operators: "creativity, creative thinking, creative process, unusualness, hypothesis generation, idea generation, aha, eureka, novel ideas, original ideas, originality, insight problemsolving, insight solution, artistic" AND "brain imaging, cerebral imaging, MRI, fMRI, PET, neural correlates, cerebral correlates, brain activation, functional magnetic resonance." In order to ensure that inclusion criteria was as unbiased as possible, we did not systematically search for studies on domains that may be relevant to creativity (such as imagination, metaphors, music improvisation or expression, mental imagery, counterfactual thinking), or studies that explore various processes presumably involved in creativity (such as cognitive flexibility, inhibition of prepotent responses, working memory, planning, and so on), but such studies were included if the authors related explicitly to creativity in their work.
In addition, in order to be included in the meta-analysis, studies had to meet the following inclusion criteria: (1) using functional imaging in healthy adults, (2) reporting whole-brain results of signal changes in stereotactic space in 3D coordinates (x, y, z) in the Montreal Neurological Institute space (MNI; Evans et al., 1993) or the Talairach space (Talairach and Tournoux, 1988), and (3) reporting the peak coordinates in these spaces. We reviewed activation contrasts between tasks performed during the scanning of one or several groups of participants. In each study, only independent contrasts were included. If several contrasts in the same study were dependent, only results from the contrast reporting the most significant maxima were included. Betweengroup comparisons based on level of expertise were not included because their interpretation is difficult in terms of neurocognitive processes.
We analyzed a total of 443 activation foci reported in 44 independent contrasts from 34 experiments carried out in 622 healthy participants, from studies listed in Table 1. CONTRAST CATEGORIES (TABLE 1) Each study was categorized in order to look for dissociations between networks associated with distinct creativity domains or operations. As it was difficult to group heterogeneous tasks in categories based on task used, they were classified into larger categories based on type of process involved (for example, combination of information vs. self-generation of unusual responses) and domain of information processing (verbal or not) used. First, we categorized each experiment according to the type of creativity processes: combination or free unusual generation tasks. Tasks that involved an explicit request to freely generate an unusual response were gathered in the "unusual generation" category, while those that required the combination of separate and remote elements were categorized as "combination." Tasks that did not fall into one of these categories were not included, leaving 29 experiments in this sub-analysis. The second classification was based on the verbal or nonverbal nature of the stimuli used. The non-verbal category included visual and musical domains. While these domains are different, they were gathered into the non-verbal category because the number of experiments in each domain separately was insufficient for statistical testing.
All contrasts were classified according to these categories by two double-blind authors (GGY, EV). The few disagreements that occurred were all solved by discussion between the co-authors.

General principles
We performed a meta-analysis of functional neuroimaging data on creativity using Activation Likelihood Estimation (GingerALE) software (http://brainmap.org/ale/cli.html; Laird et al., 2005;Eickhoff et al., 2009Eickhoff et al., , 2012Turkeltaub et al., 2012). ALE is a coordinate-based meta-analysis method that uses published activation peaks reported in functional imaging studies in a normalized coordinate referential. ALE delineates the regions in the brain where convergence across all included studies is higher than would be expected by chance (null distribution of randomly generated activation likelihoods) (Eickhoff et al., 2009). In other words, ALE evaluates the "inter-experiment" reliability of the involvement of brain regions in given processes-in this case in creativity tasks.

Global analysis
The ALE analyses were conducted using the GingerALE software v2.2 (www.brainmap.org; Eickhoff et al., 2009Eickhoff et al., , 2012Turkeltaub et al., 2012). Coordinates collected from studies that were reported in Talairach space were converted into the MNI space using the tal2mni algorithm implemented in Matlab (http:// imaging.mrc-cbu.cam.ac.uk/imaging/MniTalairach). In the first step, activation foci from each included study were modeled as Gaussian distributions and merged into a single 3D volume. To address the problem of the independence of observation within the same study, we used the modified ALE algorithm (Turkeltaub et al., 2012) and organized datasets according to subject groups. The algorithm also modeled spatial uncertainty (Eickhoff et al., 2009-and thus probability distributionof each focus, using an estimation of the inter-subject and inter-laboratory variability typically observed in neuroimaging experiments, rather than a pre-specified full-width half maximum (FWHM). Thus, the number of participants in a given study influenced the spatial extent of the Gaussian function used. GingerALE first modeled the probability of activation over all studies at each spatial point in the brain, returning localized "activation likelihood estimates" or ALE values. In a second step, ALE values were compared to a null distribution created from simulated datasets with randomly placed foci in order to identify significantly activated clusters. ALE maps were calculated using 10,000 permutations. We used a cluster correction for multiple comparisons  with a false discovery rate (FDR) corrected threshold at p < 0.05 for cluster-formation and then a p < 0.05 for cluster thresholding. Only clusters with a size exceeding the cluster size recommended by ALE were reported. We used an extentthreshold because cluster-level inference may represent a compromise between uncorrected thresholding with additional arbitrary extent correction and voxel-level corrected inference. Moreover, cluster-level thresholding seems to provide a better balance between sensitivity and specificity than the highly conservative voxel-level family-wise error (FWE) correction .

Focused sub-analyses
In order to analyse specific task categories, an ALE analysis was first performed separately for each task category (combination and unusual generation tasks, verbal and non-verbal). We proceeded as described for the global analysis (with a cluster thresholding), but entered only selected corresponding foci.

Task comparisons
Differences between task categories were tested by first performing an ALE analysis separately for each condition (thresholded at p < 0.05 uncorrected) and then computing the voxel-wise difference between the resulting ALE maps (Laird et al., 2005). The difference in ALE value between two ALE maps was computed at each voxel, and statistical significance was tested using permutations. An FDR correction at p < 0.05 was used with a minimum cluster size of 200 mm 3 in order to address the problem of multiple comparisons.

ALE results
Anatomical labels of final cluster locations were provided by the Talairach Daemon (http://www.talairach.org/daemon.html) and available as a GingerALE output. Each ALE map was visualized using Mango (http://ric.uthscsa.edu/mango) and Anatomist (http://brainvisa.info/), and was overlaid on the anatomical Colin27 Template for visual inspection and representation purposes using Anatomist.

ALL STUDIES (SEE TABLE 2, FIGURE 1)
The global ALE map revealed a network consistently associated with various creativity tasks, including the bilateral inferior and left superior PFC (BA 44,47,46,9,10), the medial PFC (BA 6, 9), the bilateral ACC and insula, the left anterior (BA 37) and posterior lateral temporal gyri (BA 22, 37), the right fusiform gyrus, the left supramarginal gyrus (BA 40) and precuneus (BA 7), the bilateral occipital cortex, the bilateral anterior and posterior cerebellum, and the left thalamus.
Combination and generation tasks overlapped in several focused regions, including the inferior frontal junction (IFJ), the inferior frontal gyrus (IFG), the posterior middle frontal gyrus, the parieto-occipital cortex, and the medial wall (Figure 2A).
When comparing these two task categories statistically ( Figure 2B, Table 5), ALE showed regions more consistently associated with combination than generation tasks. These regions were located in the left rostrolateral PFC (BA 10) and the left inferior and middle frontal gyri (BA 45,46), in the right IFG (BA 45, 46) and insula, in the left posterior middle temporal gyrus (BA 21/37), and in the left posterior parietal region.
Conversely, ALE showed regions that were more strongly associated with generation than combination tasks within the bilateral cerebellum (Lobules IV, V, VI, and VIIb), the bilateral thalamus, the left inferior parietal lobule (BA 40), the right posterior cingulate (BA 29), and the left middle frontal gyrus (BA 9).

VERBAL vs. NON-VERBAL TASKS (TABLES 6, 7, 8 AND FIGURES 3A,B)
For verbal tasks only ( Figure 3A, Table 6), significant activation was found bilaterally with a left dominance within lateral prefrontal regions, including the IFG (BA 44, 47), the middle frontal gyrus (BA 8,9,46), and extending into the rostral PFC (medial and lateral BA 10), and the superior frontal gyrus (BA 6, 8). Additional regions were observed in the left anterior temporal fusiform gyrus and in the posterior part of the lateral temporal region extending into the inferior parietal lobule (BA 39/40), in the middle temporal gyrus caudally, in the left fusiform gyrus, and the anterior STG. The bilateral insula, superior parietal lobule, cerebellum, and subcortical structures were also involved.
Verbal and non-verbal tasks overlapped in several focused left regions, including the IFJ, the posterior IFG, the parieto-occipital  cortex, the posterior middle frontal gyrus, the medial wall, and the cerebellum ( Figure 3A). When comparing these two task domains statistically ( Figure 3B, Table 8), ALE revealed some regions to be more consistently associated with verbal tasks: mainly the left and right lateral PFC (BA 8,9,44,46,47,10). Additional regions in the left ACC, the left posterior STG (BA 22/37), the right lingual gyrus, and the left thalamus were also observed.
The reverse contrast showed a few regions more associated with non-verbal than verbal tasks, within the right and left premotor regions (medial and lateral BA 6), the left middle frontal gyrus (BA 9), and the left occipital cortex.

GENERAL FEATURES OF THE SHARED CREATIVITY NETWORK
To our knowledge, the present study is the first quantitative metaanalysis to focus on creativity tasks. It reveals, despite the variety of tasks employed (Table 1), a statistical convergence across experiments in a set of brain regions (Figure 1): the caudal part of the lateral PFC, both ventrally and dorsally, the medial and lateral portion of the left rostral PFC, the inferior parietal lobule, and the lateral temporal gyrus. In this set of brain regions, the PFC is of central importance. This finding is in agreement with previous reviews (Fink et al., 2007;Dietrich and Kanso, 2010) as well as with the small number of lesion studies that has examined the cerebral bases of creativity in neurological patients (Miller and Tippett, 1996;Reverberi et al., 2005;de Souza et al., 2010;Shamay-Tsoory et al., 2011;Abraham et al., 2012a). In particular, Shamay-Tsoory et al. (2011) and Abraham et al. (2012a) demonstrated that damage to the rostral PFC impaired performance on divergent creativity tests such as the Alternate Uses test and the Torrance Test of Creative Thinking (TTCT, Torrance, 1979). The present meta-analysis similarly pointed to the rostral PFC (BA 10) as an important region for creativity tasks.
This set of brain regions shared by functional imaging studies is also consistent with those observed using other methods, in both the prefrontal and posterior regions (Chavez-Eakle et al., 2007;de Souza et al., 2010). For instance, de Souza et al. (2010) used SPECT to examine 17 patients with behavioral variant of fronto-temporal lobar degeneration (bvFTD), and showed brain regions in which perfusion correlated with creativity performance on the TTCT. Several of the reported regions overlapped or were very close to the ones shown in the present meta-analysis, in particular in the left IFG (BA 47), the left posterior inferior and middle temporal gyri (BA 37), the left inferior parietal lobule (BA39/40), and the left precuneus (BA 23).
FIGURE 1 | ALE map of all foci, thresholded at a whole-brain cluster-level corrected p < 0.05. Activations are displayed on the surface rendering of the Colin27 template (Holmes et al., 1998) in the MNI space.
FIGURE 2 | (A) ALE maps of combination task foci (in cyan) and generation task foci (in red). Overlaps between maps are shown in black. All maps were thresholded at a whole-brain cluster-level corrected p < 0.05. (B) ALE maps resulting from contrast studies of combination vs. generation tasks foci (in cyan) and generation vs. combination task foci (in red). These contrast maps were thresholded at a whole-brain FDR corrected p < 0.05. In (A,B), ALE maps are displayed on a surface rendering of the Colin27 template (Holmes et al., 1998)   The shared creativity network evidenced here includes regions usually associated with cognitive rather than affective processing. This finding suggests that this set of brain regions supports cognitive processes rather than affective, conative, or motivational processes (Lubart, 2003). This does not imply that the latter processes are not involved in creative thinking. One should keep in mind that this review was specifically designed to investigate the cognitive aspects of creative thinking rather than affective factors. The brain regions shared by creativity tasks appear to be predominantly distributed in the left hemisphere (Figure 1).
When comparing the number of left and right foci reported in the reviewed studies, the number of left foci (n = 266) was significantly greater than the right (n = 173) ones [paired t-test: t (33) = 3.43, p = 0.002]. This predominance of left colocalizations, which was also observed in previous studies (Arden et al., 2010;de Souza et al., 2010;Dietrich and Kanso, 2010), does not support the hypothesis of right dominance for creativity (Bowden and Jung-Beeman, 2003;Jung-Beeman et al., 2004;Friedman and Forster, 2005;Howard-Jones et al., 2005;Arden et al., 2010;Dietrich and Kanso, 2010  left-but not right-dorsolateral PFC and left anterior temporal lobe have been shown to be critical for creativity tasks in brain stimulation studies (Cerruti and Schlaug, 2009;Snyder, 2011, 2012;Metuki et al., 2012). The leftward asymmetry observed in the present study is unlikely due to a domain effect, since both verbal and non-verbal stimuli were associated with a left-dominant network (60% of the foci were left-sided in both verbal and non-verbal experiments). Overall, available evidence shows that both hemispheres are involved in creative thinking, and it is possible that right regions are specialized for specific processes (see further discussion below in relation to the combination vs. generation comparison).

SEMANTIC AND EXECUTIVE ROLES OF THE SHARED CREATIVITY NETWORK
The shared creativity regions evidenced by this meta-analysis include areas involved in semantic processing (Buckner et al., 2008;Binder et al., 2009;Price, 2010;Seghier et al., 2010;Vigneau et al., 2010): the IFG (BA47), the left posterior parietal lobule, and the left posterior part of the lateral middle temporal region. Some of these regions, namely the left IFG and posterior part of the left lateral temporal cortex, were more associated with verbal than non-verbal tasks in the subsequent analysis, reinforcing the hypothesis that these regions fulfil the semantic requirements of creativity tasks. The left IFG is indeed thought to play a crucial role in the controlled retrieval of information in semantic memory and/or in the selection of semantic associates in competition during retrieval (Thompson-Schill et al., 1997;Wagner et al., 2001;Thompson-Schill, 2003;Badre andWagner, 2004, 2007;Kan and Thompson-Schill, 2004;Martin and Cheng, 2006;Thompson-Schill and Botvinick, 2006). According to its multimodal integrative functions and its role in semantic memory (Binder et al., 2009), the posterior parietal lobule (BA 39) may be essential to the integration of different types of semantic  information. The lateral temporal cortex has been associated with the activation of semantic concepts and the integration of their meaning (Price, 2010). The shared creativity network includes several prefrontalparietal sub-regions. Parieto-prefrontal networks have also been found associated with fluid reasoning (e.g., the P-FIT theory from Jung and Haier, 2007) and executive functions, though their exact location is difficult to compare to the current results. Among the present prefrontal regions, overlaps were found between taskdependent maps in several discrete regions (overlap between combination and generation maps in Figure 2A, overlap between verbal and non-verbal tasks maps in Figure 3A). Both overlaps included a frontal region located in the caudal part of the IFG and the IFJ, extending to the adjacent middle frontal gyrus (BA 44,45/47). This region has been associated with several executive processes, including cognitive control (Koechlin et al., 2003;Derrfuss et al., 2005;Azuar et al., 2010), inhibition, and flexibility (Miller and Tippett, 1996;Aron et al., 2003;Rieger et al., 2003;Derrfuss et al., 2005;Picton et al., 2007;Volle et al., 2012), fluency (Perret, 1974;Bates et al., 2003;Krainik et al., 2003;Hillis et al., 2004;Kinkingnehun et al., 2007), and working memory (Goldman-Rakic, 1987;Duncan and Owen, 2000;Mottaghy et al., 2002;Curtis and D'Esposito, 2003;Sakai and Passingham, 2003;Courtney, 2004;Volle et al., 2005Volle et al., , 2008Mohr et al., 2006;Mottaghy, 2006;Postle, 2006;Sala and Courtney, 2007;Tsuchida and Fellows, 2009).

Frontiers in Human Neuroscience
www.frontiersin.org August 2013 | Volume 7 | Article 465 | 11 FIGURE 3 | (A) ALE maps of verbal task foci (in purple) and non-verbal task foci (in orange). Overlaps between maps are shown in red. All maps were thresholded at a whole-brain cluster-level corrected p < 0.05. (B) ALE maps resulting from contrast studies of verbal vs. non-verbal task foci (in purple) and non-verbal vs. verbal task foci (in orange). These contrast maps were thresholded at a whole-brain FDR corrected p < 0.05. In (A,B), ALE maps are displayed on a surface rendering of the Colin27 template (Holmes et al., 1998) in the MNI space.
Although this meta-analgfysis was not designed to determine the specific executive processes supported by these regions, it is nevertheless interesting to consider their link with creativity tasks, as several theoretical frameworks rely on the involvement of the executive processes in creative thinking (Carlsson et al., 2000;Dietrich, 2004;Bogousslavsky, 2005;Changeux, 2005). Among these processes, fluency is critical for divergent thinking tasks, such as the TTCT. Chavez-Eakle et al. (2007) showed a region within the left IFG (BA 47,11) in which CBF correlated with fluency performance on the TTCT in healthy subjects. More specifically, among the criteria measured by the TTCT, the inferior frontal region was related to the fluency (as well as flexibility) aspects of the task, whereas CBF in a more anterior region in the rostral PFC (BA 10) co-varied with the originality of responses. Cognitive flexibility (set shifting and task switching tasks) has been consistently associated with the IFJ, together with the posterior parietal cortex Kim et al., 2011). In contrast to classical set shifting or task switching paradigms, shifts in creativity tasks are not specified by an instruction or by feedback, but are initiated spontaneously by the individual. In relation to spontaneous flexibility, previous patient studies (Miller and Tippett, 1996;Goel and Grafman, 2000) have suggested a role for the right inferior frontal region in hypothesis generation with set-shift transformation-processes that may be necessary in most creativity tasks. The lateral PFC, and especially the IFJ and/or right IFG have also been associated with inhibition of prepotent but inappropriate responses and switching to an alternative response (Miller and Tippett, 1996;Garavan et al., 1999;Konishi et al., 1999;Liddle et al., 2001;Menon et al., 2001;Aron et al., 2003;Rieger et al., 2003;Brass et al., 2005;Derrfuss et al., 2005;Picton et al., 2007;Xue et al., 2008;Kenner et al., 2010;Walther et al., 2010;Volle et al., 2012).
Cognitive flexibility and inhibition of prepotent responses could be related to processes that enable thinking away from conventional or constrained ideas (Munakata et al., 2011), a fundamental principle of most creativity tasks, including divergent thinking and problem-solving tasks. In divergent thinking tasks, originality depends on the ability to provide unusual answers and may require the suppression of more obvious responses. In problem-solving tasks, problems are typically biased by constraints that are implicitly induced by the problem and that prevent participants from considering and evaluating the correct solutions (Knoblich et al., 1999;Frith, 2000; see also Reverberi et al., 2005;Chi and Snyder, 2011). Relaxing constraints in the semantic domain, for instance in a sentence completion task (Burgess and Shallice, 1997), also relies on the lateral PFC (Nathaniel-James and Frith, 2002). Further exploration is needed to determine whether thinking away from constraints and more classical executive functions, namely cognitive flexibility and inhibition, rely on similar lateral prefrontal regions (the IFG or middle frontal gyrus).
Overall, several regions-especially in the lateral PFCmay support the semantic and executive processes involved in various creativity tasks. These processes may participate in knowledge activation and its control, enabling ideation ( Table 9).

Regions showing greater activity for combination than unusual generation tasks (Table 9)
This meta-analysis also suggests that specific brain regions may support specific creative tasks, with combination and generation tasks activating partly non-overlapping brain regions (Figure 2). The rostral portion of the PFC (BA 10) was particularly sensitive to this distinction. Statistical comparison between task categories ( Figure 2B) showed that combination tasks, more than the other task types, recruited the lateral rostral PFC together with the posterior lateral temporal and temporo-parietal regions. The lateral rostral PFC is generally activated by tasks that require integration of multiple relations (Christoff et al., 2001;Kroger et al., Frontiers in Human Neuroscience www.frontiersin.org August 2013 | Volume 7 | Article 465 | 12    2002; Smith et al., 2007), analogical reasoning, and similarity judgment (Wharton et al., 2000;Bunge et al., 2005;Geake and Hansen, 2005;Green et al., 2006;Wendelken et al., 2008;Crone et al., 2009;Garcin et al., 2012;Vartanian, 2012), abstract thinking (Badre, 2008;Christoff et al., 2009aChristoff et al., , 2001 as well as coordinating goals and sub-goals (Koechlin et al., 1999;Koechlin and Hyafil, 2007). All of these functions may be involved in combination tasks. Therefore, the lateral rostral PFC could play a role in enabling subjects to find combinatorial solutions based on remote semantic associations or on relational similarity. This hypothesis is consistent with recent models that place the lateral rostral PFC at the top of a hierarchical organization of prefrontal functions according to progressively higher levels of abstraction (Christoff et al., 2001;Hampshire et al., 2010;Krawczyk et al., 2010), or to a greater distance from external stimuli when building internal thoughts (Christoff et al., 2003;Burgess et al., 2007). More posterior areas of the PFC are thought to be involved in the systematic control of representations necessary for these higher-level processes (Kroger et al., 2002;Brass et al., 2005;Cho et al., 2010;Wendelken and Bunge, 2010), and may be sufficient in some creativity tasks, such as free generation tasks. The rostral PFC is likely to operate as part of a network, together with other regions, such as the posterior STG and the temporo-parietal junction (BA 39), as suggested by the combination map and its contrast to the

Regions Hypothetical roles SHARED REGIONS: MAY REFLECT COGNITIVE CONTROL AND SEMANTIC MEMORY REQUIRED FOR IDEATION
left IFJ (BA44/6) extension to DLPFC Flexible cognitive control on information retrieved from memory and activation of task representations  In interaction with dorsal ACC (BA 32) (Beckmann et al., 2009) Left IFG (BA45/ 47) Controlled retrieval and/or selection of remote information from semantic memory (Martin and Cheng, 2006;Thompson-Schill and Botvinick, 2006) May control retrieval in connected parietal systems (BA 39) and allow higher levels of abstraction (Binder et al., 2009) Left GA (BA 39) Concept retrieval from episodic and semantic memory, integration of different types of semantic information (Binder et al., 2009;Bonner et al., 2013)

engaged in combination
Left posterior MTG (BA 37/21) Storage, activation or retrieval of perceptual information about objects and their attributes, rules and concepts, integration of their meaning (Binder et al., 2009;Price, 2010) Right IFG (BA 44/45) Suppression of inappropriate mindsets or responses (Aron et al., 2003;Volle et al., 2012) and switching to alternatives Lateral transformation of the problem

UNUSUAL GENERATION ORIENTED: MAY BE RELATED TO INCREASED WORKING MEMORY REQUIREMENTS
Left DLPFC (BA 45/46) Updating and manipulation of mindsets in working memory (verbal/semantic content) Free selection of responses in working memory (Rowe et al., 2008) Left SMG (BA 40) Maintenance on mindsets in working memory (Smith and Jonides, 1999) Medial rostral PFC (BA10) * Evocation of unusual semantic associates (Shamay-Tsoory et al., 2011;Green et al., 2012) in an associative mode of activation of mental representations * This region was found to be associated with unusual generation tasks, but was not significant when contrasting unusual generation with combination tasks.
generation map. Further clarification is needed to determine the respective role of each region in detecting similarities and combining different elements. Anatomically, these co-activations may be supported by direct connections between the PFC and superior lateral temporal areas, as shown in monkeys by Petrides and Pandya (2007). A right-lateralized IFG activation was more prominent for combination than for unusual generation. This finding may be related to the fact that tasks classified in the combination category included insight problem-solving tasks, which have been shown to involve the right IFG in relation to shifts in hypothesis generation . The right IFG is also associated with suppression of inappropriate mindsets or responses (Aron et al., 2003;Volle et al., 2012), which may be more important in combination than in unusual generation tasks, in order to suppress unsuitable self-generated responses. Insight in problem solving has also been associated with the right temporal pole, a region closely connected with IFG through the uncinate fasciculus (Jung-Beeman et al., 2004;Bowden et al., 2005). That this result rather reflects a stronger interaction between the two hemispheres in order to combine ideas cannot be ruled out (Takeuchi et al., 2010b).

Regions showing greater activity for unusual generation than combination tasks (Table 9)
While combination tasks engaged the lateral rostral PFC, unusual generation maps showed the involvement of its medial part (Figure 2A). Although this rostromedial PFC region was not significant when contrasting unusual generation to combination maps (Figure 2B), this result is in agreement with a lesion study that showed that the medial rostral PFC region is critical for unusual generation performance (Alternate Uses tasks and TTCT) and, more specifically, that it is associated with the unusualness (originality) of the responses (Shamay-Tsoory et al., 2011). The role of the medial rostral PFC (BA 10) may not be limited to the evocation of unusual responses in generation tasks. Green et al. (2012) found more activation in this region when the domains compared in analogical reasoning were remote. Thus, semantic distance or information dissimilarity might be coded in this region. This may explain that activation in the rostromedial PFC was not statistically significant when comparing directly unusual generation to combination tasks (Figure 2B), since the semantic distance/dissimilarity factor may have an effect on both. It is noteworthy that the link between the medial rostral PFC and novelty/unusualness has been made outside the scope of creativity studies. For example, Krueger et al. (2007) showed scripts of real life events to subjects participating in an fMRI experiment and found that the medial rostral PFC was involved in coping with unusual situations more than with frequent ones. This region has also been associated with counterfactual thinking (Gomez Beldarrain et al., 2005;Van Hoeck et al., 2013), prospective memory and future thinking (Hassabis et al., 2007;Schacter et al., 2007;Abraham et al., 2008;Addis et al., 2009;Szpunar et al., 2009;Burgess et al., 2011;Volle et al., 2011), mentalizing (Frith and Frith, 2006;Andrews-Hanna et al., 2010;Gilbert et al., 2010), and daydreaming (Christoff et al., 2009b;Mason et al., 2007). These cognitive functions may be involved in the search for alternative responses in generation tasks. The relative role of medial vs. lateral rostral PFC in creative thinking will be an interesting question to address in future studies, in terms of representations or processes, and in the light of existing theories (Burgess et al., 2007;Buckner et al., 2008;Christoff et al., 2009b). Unusual generation tasks, when compared statistically to combination tasks (Figure 2B), were associated with the dorsolateral prefrontal area (BA 9), the anterior inferior parietal region (left inferior parietal lobule and dorsal supramarginal gyrus, BA 40), and the cerebellar lobes ( Figure 2B). This result is consistent with a previous finding from Chavez-Eakle et al. (2007) based on correlations between CBF and performance on TTCT. The supramarginal gyrus and middle frontal gyrus are anatomically connected (Catani and Thiebaut de Schotten, 2012). The involvement of this fronto-parietal network in generation tasks could be related to its role in working memory, monitoring, and/or attention-a set of functions more involved in generation than in combination tasks. The fronto-parietal regions have also been associated with the free and/or random generation of actions (Frith et al., 1991;Rowe et al., 2008) that requires the freedom of choice of one's responses as well as their selection and monitoring. Both are likely to participate in creative generation tasks that engage spontaneous willed actions. The role of the cerebellum in human cognition is poorly understood. A recent metaanalysis of functional imaging studies showed the involvement of the cerebellar lobes in six high-level functions: emotion, working memory, executive functions, music, timing, and language (E et al., 2012). All these domains may be involved in creativity tasks, and it is difficult to draw precise conclusions from the present study. Nevertheless, cerebellar lobes appeared to be more associated with verbal than non-verbal tasks in the present metaanalysis, which is in agreement with the particular involvement of lobule VI (Declive) in language reported by E et al. (2012).

Verbal vs. non-verbal experiments
The separated ALE analyses for verbal and non-verbal experiments ( Figure 3A) and the statistical comparison between them ( Figure 3B) showed that tasks using verbal material engage more inferior regions than those using non-verbal ones in the lateral prefrontal, occipital, and medial frontal regions. A ventral/dorsal dissociation according to verbal/non-verbal domain of information in the caudal part of the lateral PFC has often been reported (Goldman-Rakic, 1987;Mottaghy et al., 2002;Curtis and D'Esposito, 2003;Sakai and Passingham, 2003;Courtney, 2004;Mottaghy, 2006;Mohr et al., 2006;Postle, 2006;Sala and Courtney, 2007;Volle et al., 2008). The analysis of non-verbal tasks showed bilateral activation in dorsal prefrontal areas (superior frontal gyrus), in regions implicated in attention, visuospatial processing, and working memory. Compared to verbal tasks, non-verbal tasks were also associated with more activation foci within the left rostral PFC, in the left occipital cortex, and in bilateral dorso-caudal prefrontal regions. However, because the non-verbal experiments gathered heterogeneous stimuli (visual, spatial, music), it is difficult to interpret these results and their apparent left dominance.
Finally, this second set of results supports the view that distinct creativity tasks could make different demands on cognitive processes that are subserved by different brain regions, in particular in rostral vs. caudal prefrontal areas. These regions, when damaged, may affect some aspects of creativity but each in a different manner.

CONCLUSION
The present findings highlighted the importance of caudal and rostral prefrontal regions, together with inferior parietal and posterior temporal areas, for the cognitive aspects of creativity. We further showed that some of these regions (mainly prefrontal ones) were shared by all task categories investigated, whereas other regions were more specifically associated with particular tasks. The core creativity network outlined by this meta-analysis is consistent with previous findings from different approaches in both healthy subjects and patients. Within this network, the lateral PFC (and especially the left IFJ) has been associated with various executive processes, such as fluency, flexibility, inhibition of prepotent responses, and cognitive control. These processes may represent components of creative thinking. In addition, this core network includes semantic regions, i.e., the left angular gyrus, STG and IFG, which have been related to the retrieval or connection of semantic associates. Retrieving and activating distant mental representations may constitute some of the mechanisms that allow creativity to emerge in both combination and free generation tasks. Subsequent task-dependent analyses revealed more specific regions in rostral PFC and in parietotemporal regions. Among them, the lateral rostral PFC and posterior temporal regions, associated with combination tasks, may more specifically support the ability to combine information in new ways, bridging semantic distances and/or superficial dissimilarities between them. A more caudal dorsolateral PFC region together with the inferior parietal lobule, associated with generation tasks, might rather support the free production of unusual or alternative responses. However, the cognitive processes involved in creativity are not yet understood, and their identification was outside the scope of this study. This meta-analysis does not enable us to determine whether or not the observed regions support processes specific to creative thinking. Further studies should explore whether and how original ideas emerge automatically from remote activation in semantic networks or whether they result from an effortful cognitive set of processes.