Neural, genetic, and cognitive signatures of creativity

The study investigates how divergent thinking (DT)—the capacity to generate multiple ideas for open-ended problems—manifests across neural, cognitive, and genetic levels. Prior task-fMRI studies of DT have reported both consistent activations (e.g., DLPFC, VLPFC, ACC) and varied findings (e.g., angular gyrus, fusiform, middle temporal gyrus), reflecting the complexity of creativity as a high-level cognitive process. Network-level accounts, including the dual-process view, implicate the default mode network (DMN) in spontaneous associative processes and the frontoparietal control network (FPCN) in evaluative control, with their cooperation supporting creativity. A contemporary perspective emphasizes continuous cortical organization via functional gradients, particularly the principal gradient spanning sensory to abstract cognition. The authors aim to build an MVPA-based neural representation of DT, decode its cognitive components, map it onto macroscale gradients, and link it to gene expression and neurotransmitter systems, then test its utility for predicting individual DT performance and generalize to resting-state functional connectivity.

The paper reviews task-based DT neuroimaging showing recurrent engagement of bilateral DLPFC, VLPFC, and right ACC, alongside additional regions such as right angular gyrus, left fusiform gyrus, and left middle temporal gyrus reported across studies. It highlights the dual-process theory of creativity: DMN supports spontaneous idea generation (mind-wandering, free association) while FPCN supports controlled evaluation and selection; their interaction is central to creative thought. Moving beyond discrete regional assignments, recent work on cortical organization via gradients (primary gradient from sensory to transmodal DMN) offers a unifying framework for complex cognition. Cognitive decoding through meta-analytic reverse inference (e.g., NeuroSynth) can map activation patterns to psychological processes. The literature on genetics and neurotransmission indicates cognition, including creativity, is influenced by gene expression and neuromodulatory systems, yet integrated neuroimaging–genetic approaches remain rare in creativity research.

Two independent task-fMRI samples were collected: sample 1 (n=55; after excluding 2 of 57) and sample 2 (n=30; after excluding 1 of 31), all healthy university students. Participants performed an adapted Alternative Uses Task (AUT) with two conditions: novel use (NU; DT) and general use (GU; control). Trials included fixation, cue (NU or GU), object presentation for idea generation (12 s NU; 4 s GU), response indication and brief written response, and rest. After scanning, participants recalled/refined responses and rated originality (1–5). MRI acquisition used Siemens 3T Trio; preprocessing employed fMRIPrep (slice timing, motion correction, CompCor, normalization to MNI, nuisance regressors, motion scrubbing), and GLMs in SPM12 were used to derive condition-level beta images (for MVPA) and single-trial betas (for regression). MVPA: Whole-brain linear SVMs (gray-matter masked; linear kernel C=1) classified NU vs GU separately in each sample with 10×10-fold cross-validation and 5000-label permutation tests. Cross-sample validation applied thresholded weight maps from one sample to the other without refitting. Significant voxel features were identified via 10,000-sample bootstrapping with FDR correction. Given high correspondence between models, thresholded weight maps were averaged to yield the DT brain pattern. Cognitive decoding and network/gradient mapping: Spatial correlations were computed between the DT pattern and NeuroSynth meta-analytic maps to infer cognitive processes (FDR-corrected). Overlaps with seven canonical resting-state networks were quantified for positive and negative predictive weights. Spatial correlations with the first 10 functional connectivity gradients (neuromaps) were assessed using spin-tests for significance. Imaging transcriptomics: Allen Human Brain Atlas microarray data (six donors) were preprocessed with abagen. Gene expression profiles for 15,633 genes were assigned to 574 DT-pattern regions. Pearson correlations between each gene’s expression and DT pattern weights were computed with spatial-autocorrelation-preserving permutation (spin) tests. Ranked genes underwent GO enrichment analysis (GOrilla), with FDR correction. Neurotransmitter mapping: PET-derived atlases for 19 receptors/transporters across nine neurotransmitter systems were obtained via neuromaps. Spatial correlations between each map and the DT pattern used spin tests. Individual-level prediction during task: Relevance Vector Regression (RVR) used single-trial NU beta maps within positively weighted DT-pattern ROIs to predict per-trial originality ratings in each sample with 10×10-fold CV; covariates were gender, age, and head motion. Performance was evaluated by Pearson r and MAE; permutation tests assessed significance. Resting-state generalization: Three datasets (SLIM n=410; GBB n=304; BBP n=600) were preprocessed (SLIM/GBB via fMRIPrep; BBP via SPM/DPABI with nuisance regression, scrubbing, smoothing, band-pass filtering; no GSR). Six ROIs were defined from the DT pattern (left AG; right IFG/bilateral OFC grouped as bilateral OFC in methods figure; left MTG/ITG; left precuneus; bilateral PFC/ACC/thalamus; right cerebellum). For SLIM, 6×6 FC matrices (15 unique edges) fed RVR (10×10-fold CV; covariates as above) to predict AUT creativity scores; bootstrap (10,000) identified significant edges (FDR-corrected). For GBB and BBP, targeted Pearson correlations tested three edges identified in SLIM (left AG–left MTG, left AG–left precuneus, left MTG–right cerebellum) against creativity scores.

• SVM classification of NU vs GU achieved robust accuracy: sample 1 accuracy 80% ± 3.8%, AUC 0.90, sensitivity 80% (95% CI 70–90%), specificity 80% (95% CI 69–89%); sample 2 accuracy 85% ± 4.6%, AUC 0.94, sensitivity 96% (95% CI 89–100%), specificity 73% (95% CI 56–89%). Cross-sample validation: model1 on sample 2 accuracy 70% ± 5.9% (p=0.0027); model2 on sample 1 accuracy 78% ± 3.9% (p<0.00001).
• The two classifiers’ thresholded weight maps were highly correlated (r=0.838, p<0.001); the averaged thresholded map defined the DT brain pattern (FDR q<0.05).
• Positive predictive weights: bilateral DLPFC, bilateral DMPFC, left VLPFC, bilateral ACC, bilateral OFC, left angular gyrus, left middle temporal gyrus, bilateral thalamus, right cerebellum. Negative predictive weights: right superior parietal lobule, right precuneus, right inferior lateral occipital cortex.
• Cognitive decoding: positive weights associated with higher-order processes (Emotion, Cognitive, Memories, Judgments, Retrieval, Reasoning). Negative weights associated with perceptual/visual processes (Visual, Selective, Location, Perception, Sensory) (FDR p<0.05).
• Network overlap: positive weights concentrated in Default Mode, Frontoparietal Control, and Limbic networks; negative weights in dorsal Attention, Visual, and Somatomotor networks.
• Functional gradients: strong positive correlation with principal gradient G1 (r=0.66, PFDR=0.0005) and G9 (r=0.47, PFDR=0.0005); negative with G8 (r=−0.40, PFDR=0.0007), indicating DT aligns with the sensory-to-transmodal hierarchy.
• Gene expression associations: significant GO enrichment among genes positively correlated with the DT pattern included negative regulation of multicellular organismal processes, detoxification of copper ion, and detoxification of inorganic compounds (all FDR q<0.05). No significant enrichment for the reverse-ranked gene set.
• Neurotransmitter maps: strongest positive correlations with μ-opioid receptor (MOR; r=0.67, pspin<0.001), CB1 (OMAR; r=0.45, pspin<0.001), H3 (r=0.41, pspin<0.001), and mGluR5 (r=0.39, pspin<0.001), consistent with systems influencing dopamine release and reward/motivation.
• Task-based individual prediction: RVR predicted originality ratings—sample 1 average r=0.35 (Ppt<0.001), MAE=0.82; sample 2 average r=0.14 (Ppt<0.001), MAE=1.82.
• Resting-state generalization: In SLIM, FC among six DT ROIs predicted AUT scores (average r=0.15, Ppt=0.002, MAE=0.76); significant edges included left AG–left MTG, left AG–left precuneus, and left MTG–right cerebellum. Validation: left AG–left MTG FC correlated with creativity in GBB (r=0.14, p=0.018) and BBP (r=0.10, p=0.017); other edges were non-significant in validation cohorts.

Findings demonstrate that DT engages a distributed neural pattern emphasizing DMN and FPCN regions, with relative suppression in primary sensory/visual networks. Cognitive decoding aligns this pattern with higher-order memory, judgment, and reasoning processes, while downweighting perceptual processing, consistent with internal mentation during idea generation. Mapping onto the principal functional connectivity gradient shows DT organization follows the cortex’s sensory-to-transmodal hierarchy, suggesting creative thinking relies on integrating concrete sensory information with abstract, internally oriented cognition. The biogenetic analyses link the DT pattern to gene expression related to neurotransmitter synthesis/release and to neurotransmitter systems (MOR, CB1, H3, mGluR5) that modulate dopamine, reward, and motivation—states known to facilitate creativity. Importantly, the DT brain pattern not only discriminates DT from control but also predicts individual originality ratings during task and generalizes to resting-state connectivity, indicating it captures meaningful individual differences in creative cognition. These results situate creativity within a macroscale organizational framework and connect it to specific neuromodulatory substrates, advancing a unified multiscale account of DT.

The study establishes a multivariate neural signature of divergent thinking that generalizes across samples and imaging modalities, aligns with the brain’s principal functional gradient, and is interpretable via cognitive decoding and biogenetic associations. It contributes a bridge between task-evoked patterns, intrinsic functional organization, gene expression, and neurotransmitter systems. Future research should extend this integrated framework to other DT modalities (e.g., visual), convergent thinking, and diverse creative domains (music, art, science), and employ more balanced samples to enhance generalizability.

The study used the Alternative Uses Task (AUT), which, despite its prevalence in DT research, has known reliability issues. The authors mitigated some concerns by constraining response procedures and expanding/validating stimulus items, but task-specific variability remains. The sample was predominantly female, limiting generalizability; future work should balance sex distribution. Not all resting-state FC edges derived from the task-based pattern replicated across validation datasets, indicating partial generalizability of connectivity features.