Predicting personality from network-based resting-state functional connectivity

Abstract

Personality is associated with variation in all kinds of mental faculties, including affective, social, executive, and memory functioning. The intrinsic dynamics of neural networks underlying these mental functions are reflected in their functional connectivity at rest (RSFC). We, therefore, aimed to probe whether connectivity in functional networks allows predicting individual scores of the five-factor personality model and potential gender differences thereof. We assessed nine meta-analytically derived functional networks, representing social, affective, executive, and mnemonic systems. RSFC of all networks was computed in a sample of 210 males and 210 well-matched females and in a replication sample of 155 males and 155 females. Personality scores were predicted using relevance vector machine in both samples. Cross-validation prediction accuracy was defined as the correlation between true and predicted scores. RSFC within networks representing social, affective, mnemonic, and executive systems significantly predicted self-reported levels of Extraversion, Neuroticism, Agreeableness, and Openness. RSFC patterns of most networks, however, predicted personality traits only either in males or in females. Personality traits can be predicted by patterns of RSFC in specific functional brain networks, providing new insights into the neurobiology of personality. However, as most associations were gender-specific, RSFC–personality relations should not be considered independently of gender.

Introduction

Inter-individual differences in personality permeate all aspects of life, from affective and cognitive functioning to social relationships. One of the most comprehensive and most widely recognized models of personality is the Five-Factor Model (FFM; Costa and McCrae 1992), consisting of five broad dimensions: Openness to experience/Intellect, Extraversion, Neuroticism, Agreeableness, and Conscientiousness. Openness to experience/Intellect reflects the engagement with aesthetic/sensory and abstract/intellectual information, as well as the degree of appreciation and toleration for the unfamiliar (Nicholson et al. 2002; Fleischhauer et al. 2010; Fayn et al. 2015). Extraversion relates to approach behaviour of driving toward a goal that contains cues for reward, and tendency to experience positive emotions given by the actual attainment of that goal (Depue and Collins 1999; DeYoung 2015). Neuroticism relates to a person’s emotional life and reflects the tendency to heightened emotional reactivity to negative emotions (Goldberg and Rosolack 1994; Rusting and Larsen 1997; Gray and Mcnaughton 2000). Agreeableness relates to interpersonal behaviour and reflects the degree of avoidance of interpersonal conflicts (stability between individuals) (Graziano et al. 2007; Butrus and Witenberg 2013). Conscientiousness reflects the degree to which individuals perform tasks and organize their lives, exhibiting a tendency to show self-discipline, act dutifully, and aim for achievement (stability within individuals) (Ozer and Benet Martínez 2006; Roberts et al. 2009) (cf. for more details McCrae and Costa 2004; DeYoung and Gray 2009).

Since the FFM of personality is based on language descriptors of adjectives applied to human and human behaviour in English lexicon, rather than neurobiological features, many attempts have been made to explore the neural bases of these five factors. At first, each trait has been associated with its most crucial and characterizing psychological functions (e.g., Neuroticism and Extraversion to sensitivity to punishment and reward, respectively, Agreeableness to social processes, Conscientiousness to top–down control of behaviour and Openness cognitive flexibility), and hypotheses have been developed about the associations between brain systems supporting those psychological functions, and the respective trait, paving the way for a biology of personality traits (c.f. DeYoung and Gray 2009). It has, therefore, been suggested that Neuroticism is associated (functionally or structurally) to affective regions that had been linked to respond to threat and punishment like amygdala, hippocampus, cingulate cortex, and medial prefrontal cortex (Kumari 2004; Cremers et al. 2010; DeYoung et al. 2010; Tzschoppe et al. 2014; Madsen et al. 2015; Pang et al. 2016). Extraversion has been linked to regions responding to reward-related stimuli like nucleus accumbens, striatum, amygdala, and orbitofrontal cortex (DeYoung et al. 2010b; Adelstein et al. 2011; Pang et al. 2016, c.f.; Lei et al. 2015). Conscientiousness has been related to the lateral prefrontal cortex (Asahi et al. 2004; Passamonti et al. 2006; DeYoung et al. 2010; Kunisato et al. 2011), deputed to the planning, following complex rule and voluntarily control of behaviour. Similarly, Openness has also been associated with the functions of the lateral PFC (DeYoung et al. 2005; Kunisato et al. 2011), but, in contrast to Conscientiousness, more because of its role in attention, working memory, and cognitive flexibility. Finally, Agreeableness has been associated with regions involved in the processing of social information, such as temporo-parietal junction, superior temporal gyrus and posterior cingulate cortex (Hooker et al. 2008; DeYoung et al. 2010; Adelstein et al. 2011). However, the associations between brain systems underlying specific mental functions and personality traits might be more complex than such one-to-one mapping; instead, it is much more plausible that the mapping between traits and brain systems is rather many-to-many (c.f. Yarkoni 2015; Allen and DeYoung 2016). One example is provided by Neuroticism, which has not only been associated to affective regions, but also to regions exerting cognitive functions, e.g., dlPFC (Kunisato et al. 2011; Pang et al. 2016), or behavioural performances probing attention (MacLean and Arnell 2010), working memory (Studer-Luethi et al. 2012), verbal fluency (Sutin et al. 2011), and explicit memory (Pearman 2009; Denkova et al. 2012). It is, therefore, possible that these systems (affective and executive) both contribute in explaining variance in Neuroticism. The potential contribution of other regions rather than the ones originally suggested also holds for other traits. For example, increasing evidence points to a link between Openness and the functional organization and global efficiency of the default mode network (DeYoung 2014; Sampaio et al. 2014; Beaty et al. 2016). Similarly, even if not directly investigating the trait of Agreeableness, there is evidence (Gazzola et al. 2006; c.f.; Iacoboni 2009) showing a possible association between one of its facet, empathy, with the mirror neuron system.

Furthermore, one of the major challenges of using functional studies for the association between personality traits and brain systems is the fact that the latter can only be based on specific implementations such as behavioural tests or paradigms used in experimental research. Moreover, there is a general consensus that mental functions arise from the coordinated activity within distributed networks rather than any individual brain region (Eickhoff and Grefkes 2011). Therefore, relating a personality trait to a particular function only because a brain region correlates with both is problematic. These considerations have prompted a network-centered perspective of brain organization (c.f. De Vico Fallani et al. 2014), highlighting the importance of functional integration for mental processes and their inter-individual differences. However, this approach, which requires a priori defined seeds, suffers from an important methodological limitation. That is, by choosing pre-defined nodes from a single task-based fMRI study, the findings might be biased toward that particular paradigm operationalization. Furthermore, task-based fMRI literature often suffers from low statistical power and low reproducibility, due to the small sample sizes typically used and considerable heterogeneity in the analysis pipeline (cf. Samartsidis et al. 2017). To solve the problem of a more objective definition of relevant nodes in a given functional network, quantitative meta-analyses of task-based neuroimaging studies aggregate the findings of many individual task-activation studies into a core network representing those locations that are reliably recruited by engaging in a given kind of mental process (cf. Fox et al. 2014). The investigation of RSFC in meta-analytically defined networks representing specific social, affective, executive, or memory functions, therefore, provides a viable approach to capturing the complex intrinsic neural architecture underlying personality (Adelstein et al. 2011; Sampaio et al. 2014).

Given that network connectivity data are almost inevitably high-dimensional, consisting of many correlated features, univariate analyses of associations between connectivity measures and phenotypical traits such as personality may not represent an optimal strategy (Orrù et al. 2012). Moreover, univariate analyses will likely fail to elucidate associations that depend on the pattern of connectivity within a network rather than any specific individual connection. On the other hand, machine learning and multivariate pattern analysis (MVPA), suitable for analysing neuroimaging data (cf. Oktar and Oktar 2015; Gael; Varoquaux and Thirion 2014), provides an approach that overcomes these limitations by searching for patterns in the connectivity matrix that allow the prediction of a continuous target variable (Doyle et al. 2015). In this article, the term “prediction” refers to the out-of-sample evaluation of a statistical model’s ability to predict the personality score for previously unseen individuals based on their RSFC. The potential of such approaches to predict behavioural scores from resting-state connectivity data has already been demonstrated with respect to sustained attention (Rosenberg et al. 2016), autistic traits (Plitt et al. 2015), and impulsivity in economic decision-making (Li et al. 2013). Conversely, personality traits have been predicted from cyber records such as personal websites (Marcus et al. 2006) or social networks (Golbeck 2011; Golbeck et al. 2011; Bachrach et al. 2012) but not yet from neuroimaging data.

Bringing together the different aspects outlined above, the current study explored whether individual levels of five major personality traits can be predicted from RSFC profiles in a priori defined brain networks representing specific cognitive functions. The selection of the networks used a priori knowledge based on the associations reported in the literature between psychological functions (and deputed networks) with personality. Accordingly, we chose functional networks associated with affective (emotion processing, reward, and pain) functions given their main associations with both Extraversion and Neuroticism, social (empathy and face processing) functions in relation to Agreeableness, executive functions as linked to Conscientiousness and Openness (vigilant attention and working memory to represent, respectively, rigid control and flexibility), and memory (autobiographic and semantic) functions as many traits were also found to be associated with them. However, it is important to note that we refrained from having hypotheses about network—predicted traits associations, since we believe that multiple brain systems, among the selected ones, can contribute to explaining inter-individual variance in one trait (e.g., Openness being predicted from networks outside the executive domain). We additionally used a network with whole-brain coverage consisting of 264 nodes (we here refer to it as Connectome; Power et al. 2011) to predict the five personality traits to test if personality can be better predicted by specific functional networks or a rather unspecific whole-brain network. In addition, in light of the previous findings of sexual dimorphism in the relationships between brain structure and personality traits (Nostro et al. 2016) as well as gender differences in RSFC (Allen et al. 2011; Filippi et al. 2013; Hjelmervik et al. 2014; Weis et al. 2017) and personality (Yang et al. 2015), these analyses were performed in a gender-mixed sample as well as separately in male and female subsamples.

Materials and methods

Participants

All data were obtained from the Human Connectome Project (HCP) WU-Minn Consortium as provided in the current “S1200” release (http://www.humanconnectome.org, Van Essen et al. 2013). The HCP was funded by the 16 NIH Institutes and Centers that support the NIH Blueprint for Neuroscience Research; and by the McDonnell Center for Systems Neuroscience at Washington University. Our analyses of the HCP data were approved by the ethics committee of the Heinrich Heine University Düsseldorf.

The HCP sample is composed of monozygotic and dizygotic twins as well as not-twins, the latter including siblings of twins, just siblings, and only-children (including those that have an as-yet not scanned sibling but not twin). Given this structure of related and unrelated subjects, we paid particular attention to select a well-matched sample of males and females that was as large as possible, while, at the same time, controlling for possible effects of heritability by creating a sample of only unrelated subjects. Evidently, we first selected all participants from the HCP sample for whom resting-state fMRI volumes and personality data were available. Out of this sample, we then selected groups of unrelated males and females (i.e., only one representative of a given family), matched for age, years of education, and twin status. This last match (twin or not twin) was preferred over the match for zygosity (not twin, dizygotic or monozygotic) as it enabled us to select a higher number of participants while not introducing dependencies in the sample. In fact, Kolmogorov–Smirnov test showed that zygosity does not lead to any significant difference in the five scores distribution, cf. supplementary Table S1. Importantly, we created a first main sample (Sample 1), where we aimed for the highest number of participants according to the inclusion criteria, but, since a considerable number of individuals were left out from the first selection, we additionally created a “replication” sample, (Sample 2). Sample 2 was thus created by removing the subjects belonging to the Sample 1 from the main release (S1200) and re-applying the selection criteria on the remaining participants.

The final selection procedure of Sample 1 resulted in a total of 420 subjects: 205 males (119 non-twins, 91 twin subjects; aged 22–37 years, mean: 28.3 ± 3.5; years of education: 14.9 ± 1.8) and 205 females (117 non-twins, 93 twin subjects; aged 22–36 years, mean: 28.8 ± 3.5; years of education: 15.0 ± 1.8).

From the remaining subjects not selected for Sample 1, Sample 2 was obtained resulting in a sample of 302 subjects: 151 males (75 non-twins, 76 twins subjects; aged 22–36 years, mean: 28.2 ± 3.4; years of education: 14.8 ± 1.8) and 151 females (76 non-twins, 75 twin subjects; aged 22–35 years, mean: 28.9 ± 3.5; years of education: 15.0 ± 1.8). For an overview on the samples selection, see Fig. 1.

Fig. 1
figure1

Samples selection overview: first Sample 1 (or “main” sample) was created aiming for the largest number of participants. Once 430 subjects were selected for this sample, the same procedure was applied on the remaing subjects of the HCP to generate Sample 2 (or “replication” sample). The two samples result in this was related to each other (as siblings of the subjects in Sample 1 are present in Sample 2), but, within each sample, there are no subjects related to each other

In addition, Sample 1 and Sample 2 were combined to form the largest group of subjects available from the HCP data that are gender-balanced and matched for age and education (Sample 3). This allowed us to investigate the stability of the results discovered in the two unrelated samples (i.e., that did not contain related individuals) and screen for additional relationships. The latter, however, need to be taken with caution, as the pooled sample does systematically contain closely related individuals (siblings and twins). Please refer to the supplementary material for a more detailed overview of the sample and the results of this analysis.

Self-report data

Personality was assessed using the English-language version of the NEO Five-Factor Inventory (NEO-FFI; McCrae and Costa 2004). The NEO-FFI consists of 60 items in the form of statements describing behaviours that are characteristic for a given trait, 12 for each of the five factors (Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism). Each factor is assessed by aggregating individual responses given on five-point Likert-type ratings scales, yielding sum scores between 0 and 60 for each factor. Data were analyzed using SPSS 20 (IBM Corp. Released 2011); scores of males and females were compared via t tests (p < 0.05, Bonferroni-corrected for multiple comparisons) for each personality trait. In case of significant group differences, we estimated effect sizes using Cohen’s d measure (Cohen 1988). Furthermore, correlations among factors were calculated and tested for significance (Bonferroni-corrected) separately for males and females (for details, see supplementary material). Importantly, as reported on the HCP listserv (https://www.mail-archive.com/hcp-users@humanconnectome.org/msg05266.html), the Agreeableness factor score in the HCP database was erroneously calculated due to item 59 not reversed. We addressed this issue by reversing it and using the correct score of Agreeableness.

Meta-analytically derived networks

Selection of networks

We selected nine meta-analytic networks representing regions consistently activated by various social, affective, executive, and memory functions. Specifically, we used two networks related to social cognition: empathy (Emp; Bzdok et al. 2012) and static face perception (Face; Grosbras et al. 2012); three networks related to affective processing: reward (Rew; Liu et al. 2011), physiological stress/pain (Pain; Kogler et al. 2015), and perception of emotional scenes and faces (Emo; Sabatinelli et al. 2011); two networks related to executive functions: working memory (WM; Rottschy et al. 2012) and vigilant attention (VA; Langner and Eickhoff 2013); two networks related to long-term memory: autobiographic memory (AM; Spreng et al. 2008) and semantic processing (SM; Binder et al. 2009).

Selection of coordinates

From each meta-analysis, we selected the reported coordinates of the networks to include in our analyses and modelled a 6-mm sphere around each coordinate. This ensured that all nodes were represented by region of interest of equal size (ROIs) within and across networks. Within each single network, we only selected peaks that either represented different anatomical regions, preventing multiple representations of a single region, or were at least 15 mm apart from each other [according to the SPM anatomy toolbox 2.1; (Eickhoff et al. 2005, 2007)]. In cases of multiple peaks within an anatomical region that were closer to each other, we included the peak showing the highest Z-score. Please note that these criteria were only applied for multiple regions within a single network, while we did not exclude any regions that were found also in another network. That is, even if different networks featured peaks at the same location, these presumably shared nodes were retained. Given that little is yet known about the effect of the networks’ sizes on the outcome predictability, we also had to consider the size of the networks (i.e., number of nodes) to make sure that possible differences in their predictive power were not due to the number of nodes included. As a result, the size of the networks ranged between 16 (VA) and 24 (Emo) nodes. Further details on the meta-analytic networks can be found in Table 1, supplementary Table S3 and supplement Fig S1.

Table 1 Description of the meta-analytic derived networks

Connectome analysis

In addition, we employed a brain-wide network of 264 functional areas from Power and colleagues (Connectome; Power et al. 2011) to compare the predictive power of RSFC from the whole-brain and from meta-analytic networks. For the coordinates of this Connectome, please refer to the supplementary Table S2 of Power et al.

Resting-state fMRI data: acquisition, preprocessing, and functional connectivity analyses

As part of the HCP protocol (Glasser et al. 2013), images were acquired on a Siemens Skyra 3T Human Connectome scanner (http://www.humanconnectome.org/about/project/MR-hardware.html) using a 32-channel head coil. Resting-state (RS)-BOLD data (voxel size = 2 × 2 × 2 mm3, FoV = 208 × 180 mm2, matrix = 104 × 90, 72 slices in a single slab, TR = 720 ms; TE = 33.1 ms, flip angle = 52°) were collected using a novel multi-band echo planar imaging pulse sequence that allows for the simultaneous acquisition of multiple slices (Xu et al. 2013). RS-fMRI data were then cleaned of structured noise through the Multivariate Exploratory Linear Optimized Decomposition into Independent Components (MELODIC) part of FSL toolbox (http://www.fmrib.ox.ac.uk/fsl). This process pairs independent component analysis with a more complex automated component classifier referred to as FIX (FMRIB’s ICA-based X-noisifier) to automatically remove artefactual components (Salimi-Khorshidi et al. 2014).

The FIX-denoised RS-fMRI data were further preprocessed using SPM12 (Statistical Parametric Mapping, Wellcome Department of Imaging Neuroscience, London, UK, http://www.fil.ion.ucl.ac.uk/spm/), running under Matlab R2016a (Mathworks, Natick, MA). For each participant, the first four EPI images were discarded prior to further analyses. Then, EPI images were corrected for head movement by affine registration using a two-pass procedure: in the first step, images were aligned to the first image, and in the second step to the mean of all volumes. Next, the mean EPI image was spatially normalized to the non-linear MNI152 template (Holmes et al. 1998) using the “unified segmentation” approach to account for inter-individual differences in brain morphology (Ashburner and Friston 2005). Finally, images were smoothed with an isotropic Gaussian kernel (full-width at half-maximum = 5 mm).

The activity time series of each voxel was further cleaned by excluding variance that could be explained by mean white-matter and cerebrospinal-fluid signal (Satterthwaite et al. 2013). Data were then band-pass filtered with cut-off frequencies of 0.01 and 0.08 Hz.

To identify participants with aberrant RSFC patterns, we computed each subject’s entire connectome sampled on a 1-cm grid. We then computed the pairwise Euclidean distance between the subjects and identified the nearest neighbour for each subject. We excluded the subjects whose distance to their nearest neighbour was in the highest 2.5% and at least 3 SD away from the average distance. This procedure was done separately for men and women (Sample 1: 5 males, 5 females; Sample 2: 4 males, 4 females). No subjects were excluded due to outlier motion parameters (DVARS and FD both displaying zero-centered values) (Salimi-Khorshidi et al. 2014; Varikuti et al. 2016; Ciric et al. 2017). For RSFC analyses, the subject-specific time series for each node of each network were computed as the first eigenvariate of the activity time courses of all gray-matter voxels within 6 mm of the respective peak coordinate. We then computed pairwise Pearson correlations between the eigenvariates of all nodes in each network, which then were transformed using the Fischer’s Z scores and adjusted (via linear regression) for the effects of age and movement.

RSFC-based prediction of personality traits by relevance vector machine learning

We examined if the RSFC patterns within each network predicted personality scores by means of statistical learning via the Relevance Vector Machine (RVM; Tipping 2001) as implemented in the SparseBayes package (http://www.miketipping.com/index.htm). The RVM is a machine learning technique that can learn to predict a continuous target value given explanatory variables (also called features). In our case, the features were the RSFC values between all nodes of a meta-analytic network, while the score of a specific personality factor scale was the target value.

Briefly, RVM is a multivariate approach that was developed from the Support Vector Machine (SVM) to induce sparseness in the model’s parameters. The RVM, in contrast to SVM, implements a fully probabilistic Bayesian framework: for each possible value of the input vector (e.g., set of FC values), the RVM algorithm provides a probability distribution of the predicted target value (e.g., FFM personality score), unlike a point estimate obtained by the SVM:

$$\hat {y}\left( {x,w} \right)={w_0}\left( {0;\,{\sigma _0}} \right)+\mathop \sum \limits_{{i=1}}^{n} {w_i}\left( {0;\,{\sigma _i}} \right){K_\sigma }\left( {{x_i},\,x} \right).$$

In the RVM formulation above, the kernel K is a multivariate zero-centered Gaussian with standard deviation σ (estimated by the algorithm) and every parameter w i , assigned to each subject x i in the training set, is assumed to follow a Gaussian with mean zero and standard deviation σi. The standard deviations σi that describe the probability distribution of the parameters w i are iteratively estimated from the training data to maximize the likelihood of the model. Sparseness is achieved by discharging parameters w i converged to zero. Once σ0 and σi have been estimated, the trained model can be used to predict the target value (e.g., FFM personality score) from a previously unseen input vector (RSFC data from participants that were not part of the training data) by computing the predictive distribution (for a more detailed description, see Tipping 2001).

In our study, we implemented the RVM algorithm with a 10-fold cross-validation. That is, the sample was randomly split into 10 equally sized groups of which 9 were used for training, while one was held back and used for assessing the performance of the prediction in previously unseen data. Holding out each of the 10 groups in turn then allowed computing the prediction performance across the entire data set. Importantly, this procedure was repeated 250 times using random initial splits of the data to obtain robust estimates of the RVM performance for predicting a given NEO-FFI score from a particular network’s RSFC pattern. For each subject, the predicted values resulting from each cross-validation (i.e., one replication) were averaged over the 250 replications and ultimately correlated with the real score.

As we performed 250 replications of a 10-fold cross-validation, in total 2500 models were computed to predict each trait. We thus quantified the contribution of each connection by the fraction of these 2500 models in which the weight for the respective connection was non-zero. The connections that had a non-zero weight in at least 80% of all models were identified as the connections that were most robustly part of the predictive model. The brain networks were visualized with the BrainNet Viewer (http://www.nitrc.org/projects/bnv/) (Xia et al. 2013).

For both the “main” (Sample 1) and “replication” (Sample 2) samples, predictions were first carried out for all subjects with males and females combined (AllSample1: n = 410 AllSample2: n = 302), and then separately for the male (MenSample1: n = 210; MenSample2: n = 151) and female groups (WomenSample1: n = 210; WomenSample2: n = 151) to assess gender differences in predictability. Predictive power was assessed by computing Pearson correlations between real and predicted NEO-FFI scores and mean absolute error (MAE). Importantly, results were only regarded as significant when they were significant at a threshold of p < 0.05 in both samples (Sample 1 and Sample 2). The p value was computed via permutation testing between real and predicted values with 10,000 runs. For each run, we shuffled the predicted scores across subjects in either the entire sample (for “All”) or in the gender groups (for “Men” and “Women”) without replacement. From here, the definition of the p value as the fraction of runs when the correlation between real and the shuffled predicted score was higher than the one obtained between the real and the original predicted value.

For all significant results in either “All”, “Men” or “Women”, we further tested for significant differences in prediction performance (i.e., correlation between real and predicted value) between males and females in the main sample. Pearson correlation coefficients (r) were transformed into Fisher’s Z and the difference between ZMen and ZWomen calculated and then 95% confidence intervals (CI) were computed based on these difference scores. The difference in correlation coefficients between males and females were regarded as significant if the 95% confidence interval did not contain zero (Lane 2013).

Results

NEO-FFI scores

Subjects scored in the same range as reported by McCrae and Costa (McCrae and Costa 2004) in both the samples.

Correlations between factors were calculated separately for males and females and in the entire sample (see Supplementary Table S2 for more detailed information). Most of them were significant at p < 0.05 (Bonferroni-corrected) in both males and females and the entire sample. Openness, however, was found to be independent of most of the other factors, except for Agreeableness (in Sample 1 for All, Men, and Women), and Conscientiousness (in All for both Sample 1 and Sample 2). Furthermore, Neuroticism was the only factor correlating negatively with almost all the others (except for Openness in Men of Sample 1 and in All, Men, and Women of Sample 2).

Comparison of the scores for the five personality traits between Men and Women revealed a significant difference for Agreeableness in both samples (Sample 1: t407 = − 4.95; p < 0.05, d = − 0.49; Sample 2: t299 = − 2.2; p < 0.05, d = − 0.27), with females scoring higher than males. For Neuroticism, Women significantly scored higher than Men in Sample 1 (t407 = − 2.8; p < 0.05, d = − 0.28), while in Sample 2, this difference only showed a trend (t299 = − 1.93; p = 0.055, d = − 0.2). For Openness (Sample 1: t407 = 0.1; p = 0.9; Sample 2: t299 = 1.64; p = 0.1) and Extraversion (Sample 1: t407 = 1.1; p = 0.3; Sample 2: t299 = − 0.68; p = 0.5) no significant gender differences were found. For Conscientiousness, Women significantly scored higher than Men in Sample 2 (t299 = − 2.11; p < 0.05, d = − 0.245), while in Sample 1 Women scored higher than Men, but not significantly (t407 = − 0.41; p = 0.15).

RVM: predicting personality traits based on RSFC

Results are only be reported if they were significant both in the main (Sample 1) and in the replication sample (Sample 2).

Predictions in the entire sample (balanced males and females)

In the entire sample, the RSFC pattern of four networks significantly predicted personality factors: Pain and VA predicted Openness, AM predicted Agreeableness and Connectome predicted Neuroticism (see Table 2; Fig. 2 for an overview of the results and Fig. 3 for the correlation plots).

Table 2 Results of the relevance vector machine
Fig. 2
figure2

Emp: empathy; AM: Autobiographic memory; WM: working memory; Emo: emotional processing; Face: face processing; Rew: reward; SM: semantic memory; VA: vigilant attention; Pain: pain processing. Summary of the networks for which FC patterns significantly predicted the five personality traits. For each network-trait combination in either Men or Women, and here, it is reported the conjunction between the correlation coefficients (i.e., minimum r value). Only predictions with r > 0.1 are displayed. While the nine meta-analytic networks are represented as slices (triangles) of the five personality circles, the connectome is represented as well as a circle. Triangles and circles are scaled based on the r values of the predicting networks (r values reported in the axis). Meta-analytic networks are underlined if a significant prediction is detected in either Men or Women. Asterisks mark significant gender differences in Sample 1

Fig. 3
figure3

Scatter plots of the predictions of personality scores significant at p < 0.05 in both samples. Continuous regression lines, dashed lines, representing the standard deviation, and mean absolute errors (MAE) are displayed

Predictions of personality traits in the gender-split groups

In the gender-split groups, we also found a significant prediction of Openness scores based on FC patterns within the Pain network in Women as well as prediction of Neuroticism based on the Connectome FC in Men. In contrast, the VA- and AM-related networks did not significantly predict Openness and Agreeableness in either sub-group. However, in the gender-specific groups, additional significant predictions were observed: in males, Extraversion was predicted by the RSFC patterns of Face and Neuroticism by Emo networks (Table 2; Figs. 2, 3). In females, Openness was predicted by Rew network. Furthermore, in females, Extraversion was predicted by Rew network and the Connectome (Table 2; Figs. 2, 3).

Gender differences in personality predictability

For all the predictions that were significant in at least one group (All/Males/Females), we tested if prediction performance was significantly different between the male and female sub-groups. Significantly better predictability in Men than Women was found for Neuroticism predicted from Emo network (Table 3, supplementary Fig S2). In Women compared with Men, Openness was significantly better predicted from Rew network and Extraversion from the entire Connectome (Table 3, supplementary Fig S2).

Table 3 Gender differences in personality predictability

Notably, not all associations that were only found predictive in one sub-group showed significant differences in predictability between males and females. In particular, no gender differences were found in predicting Openness from Pain, and VA networks, Neuroticism from Connectome, Agreeableness from AM, and Extraversion from Face and Rew networks (Table 3, supplementary Fig S2).

Discussion

Here, we report associations between major dimensions of personality and RSFC in functional brain networks. In particular, individual scores of various personality traits of the Five-Factor Model (McCrae and Costa 2004) could be predicted from patterns of RSFC in specific meta-analytically defined networks as well as from the whole-brain FC pattern. In assessing the generalizability of our findings, we focused on the predictions that replicated in two different samples within the HCP data set.

These results capitalize on the as-yet largely untapped potential (though cf. Schilbach et al. 2016; Varikuti et al. 2016) of neuroimaging meta-analyses to provide robust, functionally specific ROIs to investigate individual task-free data (Lee et al. 2012). These can help to constrain the otherwise vast feature space for statistical learning on resting-state data in a functionally meaningful and anatomically specific manner (Wang et al. 2010). As we demonstrate here, combining meta-analytic network definitions with statistical learning approaches allows, at a moderate level, not only predicting complex individual characteristics such as personality traits, but also the characterization of functional brain networks by their capability to do so. Nonetheless, our results of prediction of personality based on whole-brain FC pattern highlight that, for some traits, it might be crucial to consider the global connectivity as well.

In the overall (gender-mixed) sample, RSFC within networks representing affective and executive brain systems predicted Openness, RSFC within mnemonic network predicted Agreeableness, while RSFC from the whole brain predicted Neuroticism. In the gender-split samples, however, the prediction of Openness from the executive network VA and that of Agreeableness from the mnemonic network AM were not replicated in any of the two sub-groups, an effect likely related to the moderate effect present in the overall sample not specifically driven by a particular sex. In contrast, the prediction from the affective network Pain was also predicted in the female-only subsample, indicating that more information on the respective phenotypes can be gained from RSFC data in one gender. The gender-specific analyses revealed further constellations in which personality traits could be predicted from particular networks (see Fig. 2). In fact, none of the network–trait combination was predictive in both female and male subsamples, but several functional networks were found to differentially predict personality traits in females vs. males. In addition, Connectome successfully predicted Extraversion (in Women) and Neuroticism (in the entire sample, but then also in Men only). This underlines the notion that gender is a fundamental factor with regard to brain–personality relationships.

Methodological considerations and limitations

In our analysis, we combined a priori selection of networks of interest, built upon the existing literature (cf. Kennis et al. 2013; Hu et al. 2011; DeYoung 2010), together with a data-driven approach for learning of the predictive models. The benefits of this approach were twofolds: on one hand, with the a priori selection of networks, we could narrow down the networks of interest, which allowed us for a better functional interpretation of the results as the nodes represent brain regions robustly associated with the respective mental functions; on the other hand, the data-driven predictive models allowed for an explanatory analysis investigating which networks were informative in predicting a single trait, assuming, therefore, that many biological systems could contribute in explaining its inter-individual variance (Yarkoni 2015). Given that if only meta-analytically defined functional networks were employed, less consistently linked yet potentially critical regions might have been left out, we included also a purely explorative analysis employing the whole-brain FC.

In addition, as noted above, using a sparsity inducing method (RVM) which yielded compact regional modes has the advantage of providing regionally specific prediction models. As outlined above, our procedure provided a biologically informed feature reduction, as only the most relevant connections were taken in account in the prediction models. This has the advantage of reducing the complexity of the models avoiding overfitting (Hastie et al. 2009).

With respect to the prediction model, we here employed Relevance Vector Machine (RVM), which, in contrast to support vector regression or ridge regression, yields considerably sparser solutions (Tipping 2001). This allowed for identifying the most used connections and nodes (Fig. 4) that mainly drove the prediction and hence enabled a more specific interpretation of its neurobiological underpinnings. In this context, it is important to note that, for any given model, the entire set of connections with non-zero coefficients provides information about the personality trait (Orrù et al. 2012). For interpretation, however, we focused on the most consistently utilized connections (over 250 replications) as key components of the given prediction.

Fig. 4
figure4

Summary of the most used nodes (i.e., above 80% of the models) between regions from a the reward (Rew), vigilant attention (VA), and pain processing (Pain) networks in the prediction of Openness; b the Rew and face processing (Face) networks in the prediction of Extraversion. Summary of the most used connections between regions from c the autobiographic memory (AM) network in the prediction of Agreeableness, d the Pain and emotional processing (Emo) networks in the prediction of Neuroticism

In accordance with recent recommendations, the current study used 10-fold cross-validation, which has been showed to be less susceptible to overly optimistic estimates as compared with a leave-one-out approach (LOO-CV) (Varoquaux et al. 2016). Moreover, we repeated the cross-validation procedure 250 times, averaging the prediction performance over all replications to obtain robust and generalizable estimates of the capability of different brain networks to predict personality scores in new individuals.

A last important methodological reflection is that, although it might be tempting to make use of the entire HCP sample (which, if requiring an equal number of males and females, and if considered the matching factors of age, education and twin status, would yield about 800 individuals), it systematically consists of related subjects (siblings and twins). In addition, there is considerable evidence for genetic influence on both personality (Jang et al. 1996; Bouchard and McGue 2003; Verweij et al. 2012; Power and Pluess 2015) and brain function (van den Heuvel et al. 2013; Colclough et al. 2017; Ge et al. 2017; Ktena et al. 2017). Consequently, the relationship structure in the HCP data is a critical aspect to this work, as the inclusion of related subjects would potentially hurt the model fitting but even more importantly would introduce an (optimistic) bias into the cross-validation. As a result, we thus performed our analyses primarily in the largest possible set of matched, unrelated subjects, replicate it in the then largest possible independent set of matched, unrelated subjects and only in a supplementary analysis pooled both of these sets for the analysis of around 750 subject.

Our approach, by building upon these methodological considerations, yielded insights into the relationships between brain, behaviour, and personality. However, there are some limitations which are worth consideration in the future studies. First, gender-stratified sub-analyses may reduce statistical power because of the smaller sample sizes. Further studies with a larger sample size, designed to separately analyze men and women, are required, especially monitoring their hormonal levels (Arélin et al. 2015; Weis et al. 2017). Second, even though meta-analytic networks are among the most reliable ways to infer a mental function given a set of brain regions, we acknowledge that some regions of different functional networks can overlap. As a matter of fact, the employment of meta-analytically derived networks does not necessarily ensure a stringent and univocal relationship between the mental function supported by a particular network and a personality trait. Nonetheless, this approach can at least provide some confidence for the implication that a specific trait is related to a particular mental function in terms of the network that subserves them. A third consideration relates to the measurement of personality, i.e., the use of self-reported questionnaires. Self-reported questionnaire might have, indeed, contributed in increasing the noise in the data, as perception and report of own personality traits can be affected by many factors, e.g., men usually scoring low on Neuroticism as socialization effect (Viken et al. 1994).

Predicting Openness to experience

Our results indicated that self-reported Openness to experience can be linked to RSFC patterns in the networks subserving reward (Rew) and pain (Pain) processing in Women, while, in the overall sample, Openness was significantly predicted by RSFC in the vigilant attention (VA) network and, again, from Pain. Openness to experience has been linked to “need for cognition,” that is, an individual’s tendency to engage in effortful cognitive processing (Fleischhauer et al. 2010): high levels of Openness were found to positively affect work outcomes for highly complex jobs while increasing dissatisfaction when jobs become mechanical and unchallenging (Mohan and Mulla 2013). Such monotonous and intellectually unchallenging tasks were exactly the tasks investigated in the VA meta-analysis of Langner and Eickhoff (2013), which revealed the brain network involved in dealing with sustained attentional demands in boring situations. Thus, the predictability of Openness from FC in the VA network may reflect a neural substrate of the challenge experienced by individuals scoring high on Openness when faced with repetitive tasks and standardized routines. High-Openness participants might, therefore, need to recruit this network differently than low-Openness individuals to keep focused on a tedious, repetitive task over time. Indeed, connections used throughout all prediction models from the VA network of Openness in both samples is between pre-supplementary motor cortex and medial prefrontal cortex (both involved in task-set re-energizing and outcome monitoring), between left inferior occipital gyrus (IOG) and right temporo-parietal junction (crucial for re-orienting the signalling), and left IOG and inferior frontal junction (known for its contribution in the input/output transformation) (see Fig. 4 for the most informative connections and Langner and Eickhoff 2013 for more details on the regions’ functions).

Behaviours associated with the trait of Openness, such as cognitive exploration, have been attributed to high dopamine (DA) functioning (DeYoung et al. 2005). This, indeed, led to the inclusion of Openness in the meta-trait “β” (or plasticity, c.f. DeYoung 2010), a higher order factor representing the shared variance between Openness and Extraversion, which are suggested to be both modulated by the dopaminergic system. DA is the main neurotransmitter modulating the reward network (cf. Berridge and Robinson 1998), and, in line with this, RSFC within the Rew network could predict both Openness and Extraversion (in Women and in Men, respectively), possibly via affecting the reactivity of the dopaminergic system. Interestingly, in predicting Openness, the weights of the nodes (i.e., number of incident edges) most used across the predictive models showed a stronger involvement of the dlPFC, corroborating previous findings that showed an association between Openness and the dopaminergic mesocortical branch, which projects directly onto the dlPFC (DeYoung 2013; Passamonti et al. 2015). On the other hand, regions like amygdala, nucleus accumbens (NAc), and orbitofrontal cortex (OFC), which constitute the other main dopaminergic branch, the mesolimbic pathway, were significantly less recruited. We would thus suggest that DA neurons populating the mesocortical branch, by encoding specifically the saliency of the stimulus (i.e., reward value of information, cf. Bromberg-Martin et al. 2010), can be potentially more informative for high-Open individuals, characterized by the automatic tendency to perceive salient information in everyday experience (DeYoung 2013). Interestingly, we found that Openness could be predicted by FC of the Rew network significantly better in Women, compared to Men (r = 0.17 in Women and r = − 0.06 in Men of Sample 1). This might be explained by the fact that Rew functioning is highly influenced by the ovarian hormones estrogen and progesterone during the menstrual cycle (Dreher et al. 2007). In addition, estrogens have been related to dlPFC functioning, going along with cognitive decline which follows the drop of estrogens in menopause (Shanmugan and Epperson 2014). Despite the lack of studies exploring a direct relationship between females’ hormonal cycling and the trait of Openness, there is evidence for its indirect modulation by estrogen. That is, the catechol-O-methyltransferase gene, which is associated with the trait of Openness (Konishi et al. 2014), is influenced by estrogen (Harrison and Tunbridge 2008). We thus suggest that the influence of ovarian hormones on RSFC in the Rew network as well as on perceived Openness induces joint intra-individual variation (i.e., shared variance), which in turn increases the strength of the neural and phenotypical association across women. This should then result in the observed higher predictability of Openness in female participants.

Across the entire sample, but then also in the female sub-group only, Openness could additionally be predicted in both samples based on FC within the pain network (Pain). Relationships between pain and Openness have been demonstrated in terms of a higher threshold for pain tolerance (Yadollahi et al. 2014) and as protective factor in migraine occurrence (Magyar et al. 2017) in individuals reporting higher levels of Openness. However, very little is known about the association between this trait and the neural correlates of pain. Indirect evidence, however, comes from research in avoidance learning, which suggests that the successful avoiding of an aversive stimulus is experienced as an “intrinsic” reward (Kim et al. 2006). Endogenous opioid peptides, which are highly dense in the pain network (Baumgartner et al. 2006), were, indeed, found to modulate the dopaminergic system in response to aversive stimuli, resulting in the enhancement of a pleasure feeling boosted by DA (Sprouse-Blum et al. 2010). We thus suggest that high- and low-Open individuals differ in their ability to detect possible aversive stimuli (via diverse reactivity of the Pain network) and, by avoiding them, differently experience “intrinsic” reward.

In summary, the predictions from the Rew, VA, and Pain networks of Openness might, therefore, jointly point to the importance of saliency processing of stimuli, which can be rewarding (Rew), monotonous (VA), or aversive (Pain), turning high Open individuals as highly receptive and permeable to relevant information. Ultimately, connections between regions specially targeted by ovarian hormones (e.g., dlPFC) might underlie the significant gender difference in the predictability of Openness from FC in Rew network (Fig. 4).

Predicting Extraversion

Extraversion was predicted by the RSFC patterns within the networks of reward (Rew) in Women and face perception (Face) in Men. Moreover, in Women, this trait was also significantly predicted by the whole-brain (Connectome) RSFC. Extraversion is generally described as behavioural exploration and sensitivity to specific rewards. Importantly, a distinction has been also made between “Agentic Extraversion”, reflected in assertiveness, dominance, and ambition aspects, and a “Affiliative Extraversion” which is more related to sociability and affiliative social bonding (DeYoung et al. 2007; c.f. Allen and DeYoung 2016).

As discussed previously in the paragraph “Predicting Openness to experience”, the traits of Extraversion and Openness exhibit a shared variance, known as “β” factor, and are genetically influenced by the dopaminergic system (c.f. Allen and DeYoung 2016). Notably, while for Openness, Rew’s most used nodes encompassed the mesocortical pathway (see above), for Extraversion, it was regions along the mesolimbic branch that were mostly used (amygdala, NAc and OFC). Thus, we suggest that even though FC of Rew predicts both Openness and Extraversion, the functional connectivity of two different subsystems of the Rew network is informative for the two different traits, namely the mesocortical and mesolimbic pathway, respectively. In favour of this distinction, extraverts were shown to be more sensitive toward the motivational content of the reward stimulus, encoded by DA neurons along the mesolimbic pathway (Bromberg-Martin et al. 2010; DeYoung 2013). We thus believe that the prediction of Extraversion from the FC within Rew might well-capture the “Agentic” dimension of Extraversion, given the motivational value of the rewarding stimuli and drive toward a goal prompted by the dopaminergic mesolimbic system.

While extraversion in Women was found to be associated with FC of Rew, relationships of this trait, in Men, were found with FC in Face network. Faces are arguably the most important social stimuli for humans and it has been shown that extraverts compared to introvert, by spending more time on people, are significantly better at recognizing faces (Li and Liu 2010). Extraversion’s hedonic experience of goal achievement is enclosed in the “Affiliative” component (DeYoung et al. 2007; c.f.; Allen and DeYoung 2016) and its genetic variation has been also pointed to the opiate system, due to its involvement in the hedonic response to the stimulus (Peciña et al. 2006). It is, therefore, possible that the endogenous opioid system via modulation of amygdala and medial prefrontal cortex (Tejeda et al. 2015; Selleck and Baldo 2017), most used regions in the connections of Face, mediate both the perception of faces (Martin et al. 2006) and the social bonding (Pasternak and Pan 2013). We thus suggest that functional connectivity within the Face network in Men is mostly related to the “Affiliative” aspect of Extraversion.

The last prediction of Extraversion is based on whole-brain FC in Women (Sample 1: r = 0.29; Sample 2: r = 0.23, both p < 0.05; for gender comparison in Sample 1, Cohen’s q = 0.323, p < 0.05). However, a major issue using whole-brain connectivity patter might be the lack of anatomical localization for the most informative features, as none of them resulted to be used more than 40% of the predictive models, indicating a heterogeneous mosaic of connections which contribute to the prediction of Extraversion. The only theory in personality neuroscience which relates the functioning of entire cortex to Extraversion (and Neuroticism, see below “Predicting Neuroticism”) is Eysenck’s biological theory of personality (Eysenck 1967). Here, Extraversion is thought to depend on the variability in cortical arousal, with introverted individuals having lower response thresholds consequently more cortical arousal compared to extraverts. In favour of this hypothesis, the topological properties of whole-brain RSFC have shown that brains of more extraverted individuals behave more similarly to a “small-world” compared to a “random” network, with higher clustering coefficient compared to introverts (Gao et al. 2013). A “small-world” clustered configuration, which supports a more modularized information processing and fault tolerance, can, therefore, be associated with higher arousal threshold in extraverts’ cortex. We also observed that this prediction performance was significantly stronger in Women compared to Men (r = 0.29 in Women and r = − 0.03 in Men of Sample 1). Again, a possible cause might be the involvement of ovarian hormones, targeting specifically the most densely interconnected hub structures of the connectome (Alawieh et al. 2015) as well as influencing level of Extraversion (Jokela et al. 2009; Ziomkiewicz et al. 2012). However, more studies are needed to prove this interaction between Extraversion, estrogen, and the topographical properties of whole-brain functional connectivity.

To sum up, connectivity of regions encoding the motivational value and the drive toward a goal (Rew) and the hedonic processing of the goal itself (Face) were informative to predict inter-individual variability in the trait of Extraversion possibly capturing the “Agentic” and “Affiliative” aspects of the trait, respectively (Fig. 4). Importantly, given the modulation of ovarian hormones on both the trait of Extraversion and on the topological properties of the Connectome, we would suggest that sex hormones might be a possible mediator of this trait–network relationship, resulting in better prediction performance in Women.

Predicting Agreeableness

RSFC patterns in the AM network could predict the individual level of perceived Agreeableness while grouping men and women in both samples. This trait reflects a high desire to avoid interpersonal conflicts (Jensen-Campbell and Graziano 2001) and strong affect regulation (Ryan et al. 2011). In line with this, positive correlations have been demonstrated between Agreeableness and regions supporting social functioning (Hooker et al. 2008; DeYoung et al. 2010; Hassabis et al. 2014) and midline regions of the default mode network (DMN), as deputed to self-referential process (Adelstein et al. 2011; Sampaio et al. 2014). Our prediction of Agreeableness from the AM network supports a crucial role of self-reference, strongly linked to autobiographical memory (Molnar-Szakacs and Arzy 2009), in how high agreeable individuals deal with social demands. Self-related cognition has been often discussed at the neural level as the product of interaction between the DMN and the mirror neuron system (MNS), the first responsible for high-level mentalizing function and the second for embodied simulation-based representation (Keysers and Gazzola 2007; Qin and Northoff 2011; c.f.; Molnar-Szakacs and Uddin 2013). As a result, the privileged access to the own physical and mental states would allow a better insight into others’ physical and mental states, and consequent appropriate social responses.

Interestingly, within the AM network, most used connections that informed about the trait in both samples reflected the interaction between the DMN and MNS systems: nodes with highest weights belonged, indeed, to DMN subsystem, such as medial PFC, posterior cingulate cortex, medial temporal lobe (amygdala and hippocampus) and lateral parietal cortex (temporo-parietal junction). The remaining nodes with the highest weights belonged to the MNS, such as inferior frontal gyrus, precentral gyrus, inferior parietal cortex, and superior temporal sulcus. Our result, hence, supports the interplay of these two subsystems in the context of self-processing (here expressed via memory recollection about past experiences, AM) and that this knowledge about the self can significantly predict Agreeableness, the trait most reflecting enhanced social skills.

Predicting Neuroticism

In Men, self-reported Neuroticism was predicted by RSFC within the emotional processing network (Emo). In addition, the RSFC from the whole brain (Connectome) significantly predicted this trait across the entire sample and then specifically in Men only. Neuroticism represents a broad dimension of individual differences in the tendency to experience negative, distressing emotions. High-Neuroticism scores entail the experience of fear, anger, sadness, embarrassment, the incapacity to control cravings and urges, and to cope with stress (Costa and McCrae 1987). Within this trait, it is possible to delineate two major divisions: one related to the experience of anxiety, fear and passive avoidance, and referred in literature as the aspect Withdrawal, and the other related to irritability, anger and active defensive responses, or Volatility (DeYoung et al. 2007). Neuroticism is arguably the most studied personality trait and is an important predictor of many different mental and physical disorders (Lahey 2009). Furthermore, the two aspects of Neuroticism (Withdrawal and Volatility) highly reflect the dimension of Behavioural Inhibition System (BIS) and Fight-Flight-Freeing System (FFFS) from the Gray’s Reinforcement Theory (Gray and Mcnaughton 2000), conceptualized in term of their neurobiology. Interestingly, this distinction between the Volatility/FFFS and Withdrawal/BIS seems to be captured by the two networks showing predictability power for Neuroticism, Emo and Pain. Even though this last prediction (Pain) was found significant in Sample 1 (with r = 0.15, p < 0.05 in Men) but not fully replicated in the Sample 2 (with r = 0.2, p = 0.05 in Men) (Fig. 4), we would still suggest that recruitment of this network in association to Neuroticism might indicate that perception of the aversive stimulus via the Pain network (Iannetti and Mouraux 2010; Hayes and Northoff 2012) could lead high-Neuroticism men to inhibit their behaviours such to avoid potential threats and punishments (Withdrawal). Conversely, Emo network would trigger emotional responses for either escaping or eliminating the threat, but in both cases showing a strong emotional lability (Volatility). Beyond associations with specific networks, Neuroticism could also be predicted from the whole-brain RSFC (Connectome) in Men and across genders. This is nicely in line with graph analysis studies (Gao et al. 2013; Servaas et al. 2015), showing that the neurotic brain displays topological properties of a “random network” and overall weaker FC. Here, cortisol might play a specific role, the hormone that is most closely associated with a biological reaction to stress and found to correlate with Neuroticism. However, the directionality of correlation seems to depend on gender: many studies converged in discovering that Neuroticism was positively correlated with baseline cortisol in men, but the opposite was true in women (Zobel et al. 2004; Oswald et al. 2006; DeSoto and Salinas 2015). Thus, especially in men, the overabundance of cortisol by potentiating neuronal degeneration (Sapolsky 1994) might be responsible for the overall smaller brain volume (Liu et al. 2013), white-matter (Bjørnebekk et al. 2013), and gray-matter (Servaas et al. 2015) functional disconnectivity found in high-Neuroticism individuals compared to the more emotional stable. Given that all the three networks (Emo, Pain, Connectome) showed a stronger predictability in Men compared to Women (statistically significant for the first two, and a strong trend for the third, see Table 3), we suggest that gender may moderate Neuroticism’s relationship to cortisol. However, more (direct) studies are needed to better understand this intricate relationship between RSFC, cortisol, Neuroticism, and gender, and to shed light on the neural mechanisms that make women’s brain more susceptible to Neuroticism-related mental disorders (Jorm 1987).

Implications for the neurobiology of FFM

Contrary to other important theories of personality, such as Cloninger’s Tridimensional Personality Questionnaire (TPQ) or Gray’s Reinforcement Sensitivity Theory (RST), the FFM is not based on biological grounds. However, variability in its personality factors had been associated with the brain, given that personality traits are the product of our actions, emotions and, more generally, cognitive processes. In this way, the cognitive mechanisms work as intermediate bridge between the psychometric constructs of personality and plausible biological substrates. However, the relationships among these factors (brain, behaviour, and personality) can be misleading in the context of personality predictions, which, in fact, were significant only to a moderate level, compared to other findings: contrary to predictions of sustain attention (Rosenberg et al. 2016) or reading comprehension (Cui et al. 2017) which tap predictability of cognitive process itself, personality traits are mostly modulators of these cognitive processes. This may make it more difficult to find brain correlates of personality in specific networks associated with those functions.

In addition, the hierarchy of the FFM model might have contributed in enlarging the gap: in our findings, we highlighted the possibility that the predictions of one trait from different networks could reflect different components within this trait, also known as aspects and facet (cf. DeYoung et al. 2007; Koelsch et al. 2013; Haas et al. 2015). For example, we discussed the prediction of Extraversion from Rew and Face as potentially capturing the “Agentic” and “Affiliative” aspects, respectively, or the prediction of Neuroticism from Pain and Emo as linked to Withdrawal and Volatility. Conversely, when the same network was predicting two different traits (e.g., Rew predicting Openness and Extraversion, discussed in light of the saliency and motivational contribution for the two traits), the prediction might have, indeed, boosted if investigating the meta-trait “β”, which reflects their shared variance within the dopaminergic system and thus more prone to be predicted by the network of reward processing (DeYoung 2013). Therefore, the level of abstraction of the five traits might not mapped well to particular brain systems, and more studies are encouraged for testing both more specific and homogeneous sub-dimensions as well as more heterogeneous higher order factor structure. Finally, many biological mechanisms participate in evoking the same cognitive process, e.g., changes in brain structure, function, or genetic, which are then intrinsically connected with personality. We here used RSFC as “marker” for the individual expression of personality traits, enduring across time and situations. However, a downside of FC in resting conditions might be that it has not so much to do with how personality factors come together to “produce” stable modulations of a whole range of cognitive processes. Therefore, other brain measurements (as structural connectivity, task-based functional activation, or molecular genetics) might be also useful in gaining more knowledge on the biology of personality and its relationship with specific mental functions. Keeping in mind that we cannot expect biological mechanisms to show clear-cut as the respective psychometric dimensions (Yarkoni 2015), but, conversely, many biological mechanisms (function, structure, neurotransmitters) as well as many mental functions can be informative for a given personality trait, we, therefore, support the need for a multi-level approach in future studies as proposed by Yarkoni to achieve a unified description of the biological bases of personality traits.

However, even though all these aspects might affect the relationship between brain function (and structure) and personality, we here do provide insights on the relation between brain and personality: when analysing the entire sample while adjusting for gender effects, only two predictions (VA predicting Openness and AM predicting Agreeableness) can be found not specifically driven by one gender-group. However, when looking at men and women separately, we observed much more and larger effects, evidence which highly remarks the importance of gender while investigating the neural correlates of personality. Specifically, the current findings propose a link between Openness and executive and affective domain. Agreeableness with memory domain. Extraversion with social and affective networks and lastly Neuroticism with the affective system. Interestingly, these last two traits could be predicted as well from the entire Connectome. An interesting consideration is that Openness could be significantly predicted by three different, barely overlapping networks (Pain, Rew, VA), but could not be predicted from the whole-brain, which was covering the nodes of all the three at the same time. We thus argue for a better predictability of Openness from specific and separate functional networks. Contrarily, Extraversion and Neuroticism could be significantly predicted by both meta-analytic networks and the whole brain, pointing to the importance of also global effects, besides specific functions. This is particularly true for Extraversion, which showed significantly higher prediction performance from global RSFC (Connectome) with a very vast nodes contribution, rather than from the specific networks of Rew and Face, thus favouring the global effects over the specific functions for this trait.

Conclusions

Using multivariate machine learning, we showed that personality traits can be predicted from RSFC patterns in affective, social, executive, and memory networks of the brain, as well as from the whole-brain. Our observation that for most of these networks predictive power was gender-specific complements previous morphometric findings (Nostro et al. 2016) in highlighting the crucial role of gender when trying to understand the neurobiology of personality. In addition, the many-to-many associations between mental functions and personality traits indicate the complexity of the biological substrates of personality, as many functional systems may contribute to the observable differences in each trait (for a critical review see Yarkoni 2015). Maybe, even more fundamental are the implications for the concept of personality, given that even a trait as complex and broad as, for instance, Openness, seems to have a neurobiological underpinning in pre-defined functional networks that enables estimation of the individual level of that trait in a new subject.

References

  1. Adelstein JS, Shehzad Z, Mennes M et al (2011) Personality is reflected in the brain’s intrinsic functional architecture. PLoS One 6:e27633. https://doi.org/10.1371/journal.pone.0027633

    PubMed  PubMed Central  CAS  Article  Google Scholar 

  2. Alawieh A, Sabra Z, Sabra M et al (2015) A rich-club organization in brain ischemia protein interaction network. Sci Rep. https://doi.org/10.1038/srep13513

    PubMed  PubMed Central  Article  Google Scholar 

  3. Allen TA, DeYoung CG (2016) Personality neuroscience and the five factor model

  4. Allen EA, Erhardt EB, Damaraju E et al (2011) A baseline for the multivariate comparison of resting-state networks. Front Syst Neurosci 5:2. https://doi.org/10.3389/fnsys.2011.00002

    PubMed  PubMed Central  Article  Google Scholar 

  5. Arélin K, Mueller K, Barth C et al (2015) Progesterone mediates brain functional connectivity changes during the menstrual cycle—a pilot resting state MRI study. Front Neurosci 9:1–11. https://doi.org/10.3389/fnins.2015.00044

    Article  Google Scholar 

  6. Asahi S, Okamoto Y, Okada G et al (2004) Negative correlation between right prefrontal activity during response inhibition and impulsiveness: a fMRI study. Eur Arch Psychiatry Clin Neurosci 254:245–251. https://doi.org/10.1007/s00406-004-0488-z

    PubMed  Article  Google Scholar 

  7. Ashburner J, Friston KJ (2005) Unified segmentation. Neuroimage 26:839–851. https://doi.org/10.1016/j.neuroimage.2005.02.018

    PubMed  Article  Google Scholar 

  8. Bachrach Y, Kosinski M, Graepel T et al (2012) Personality and patterns of Facebook usage. Proc 3rd Annu ACM Web Sci Conf-WebSci’. 12 24–32. https://doi.org/10.1145/2380718.2380722

  9. Baumgartner U, Buchholz HG, Bellosevich A et al (2006) High opiate receptor binding potential in the human lateral pain system. Neuroimage 30:692–699. https://doi.org/10.1016/j.neuroimage.2005.10.033

    PubMed  Article  Google Scholar 

  10. Beaty RE, Kaufman SB, Benedek M et al (2016) Personality and complex brain networks: the role of openness to experience in default network efficiency. Hum Brain Mapp 37:773–779. https://doi.org/10.1002/hbm.23065

    PubMed  Article  Google Scholar 

  11. Berridge KC, Robinson TE (1998) What is the role of dopamine in reward: Hedonic impact, reward learning, or incentive salience? Brain Res Rev 28:309–369. https://doi.org/10.1016/S0165-0173(98)00019-8

    PubMed  CAS  Article  Google Scholar 

  12. Binder JR, Desai RH, Graves WW, Conant LL (2009) Where is the semantic system? A critical review and meta-analysis of 120 functional neuroimaging studies. Cereb Cortex 19:2767–2796. https://doi.org/10.1093/cercor/bhp055

    PubMed  PubMed Central  Article  Google Scholar 

  13. Bjørnebekk A, Fjell AM, Walhovd KB et al (2013) Neuronal correlates of the five factor model (FFM) of human personality: multimodal imaging in a large healthy sample. Neuroimage 65:194–208. https://doi.org/10.1016/j.neuroimage.2012.10.009

    PubMed  Article  Google Scholar 

  14. Bouchard TJ, McGue M (2003) Genetic and environmental influences on human psychological differences. J Neurobiol 54:4–45

    PubMed  Article  Google Scholar 

  15. Bromberg-Martin ES, Matsumoto M, Hikosaka O (2010) Dopamine in motivational control: rewarding, aversive, and alerting. Neuron 68:815–834

    PubMed  PubMed Central  CAS  Article  Google Scholar 

  16. Butrus N, Witenberg RT (2013) Some personality predictors of tolerance to human diversity: the roles of openness, agreeableness, and empathy. Aust Psychol 48:290–298. https://doi.org/10.1111/j.1742-9544.2012.00081.x

    Article  Google Scholar 

  17. Bzdok D, Schilbach L, Vogeley K et al (2012) Parsing the neural correlates of moral cognition: ALE meta-analysis on morality, theory of mind, and empathy. Brain Struct Funct 217:783–796. https://doi.org/10.1007/s00429-012-0380-y

    PubMed  PubMed Central  Article  Google Scholar 

  18. Ciric R, Wolf DH, Power JD et al (2017) Benchmarking of participant-level confound regression strategies for the control of motion artifact in studies of functional connectivity. Neuroimage 154:174–187. https://doi.org/10.1016/j.neuroimage.2017.03.020

    PubMed  Article  PubMed Central  Google Scholar 

  19. Cohen J (1988) Statistical power analysis for the behavioral sciences. Stat Power Anal Behav Sci 2nd:567

  20. Colclough GL, Smith SM, Nichols TE et al (2017) The heritability of multi-modal connectivity in human brain activity. Elife 6:e20178. https://doi.org/10.7554/ELIFE.20178

    PubMed  PubMed Central  Article  Google Scholar 

  21. Costa PT, McCrae RR (1987) Neuroticism, somatic complaints, and disease: is the bark worse than the bite? J Pers 55:299–316. https://doi.org/10.1111/j.1467-6494.1987.tb00438.x

    PubMed  Article  Google Scholar 

  22. Costa PT, McCrae RR (1992) Professional manual: revised NEO personality inventory (NEO-PI-R) and NEO five-factor inventory (NEO-FFI)

  23. Cremers HR, Demenescu LR, Aleman A et al (2010) Neuroticism modulates amygdala-prefrontal connectivity in response to negative emotional facial expressions. Neuroimage 49:963–970. https://doi.org/10.1016/j.neuroimage.2009.08.023

    PubMed  Article  Google Scholar 

  24. Cui Z, Su M, Li L et al (2017) Individualized prediction of reading comprehension ability using gray matter volume individualized prediction of reading comprehension ability using gray matter volume. https://doi.org/10.1093/cercor/bhx061

  25. De Vico Fallani F, Richiardi J, Chavez M, Achard S (2014) Graph analysis of functional brain networks: practical issues in translational neuroscience. Philos Trans R Soc B Biol Sci 369:20130521–20130521. https://doi.org/10.1098/rstb.2013.0521

    Article  Google Scholar 

  26. Denkova E, Dolcos S, Dolcos F (2012) Reliving emotional personal memories: affective biases linked to personality and sex-related differences. Emotion 12:515–528. https://doi.org/10.1037/a0026809.supp

    PubMed  Article  Google Scholar 

  27. Depue RA, Collins PF (1999) Neurobiology of the structure of personality: dopamine, facilitation of incentive motivation, and extraversion. Behav Brain Sci 22:491–517

    PubMed  CAS  Google Scholar 

  28. DeSoto MC, Salinas M (2015) Neuroticism and cortisol: the importance of checking for sex differences. Psychoneuroendocrinology 62:174–179. https://doi.org/10.1016/j.psyneuen.2015.07.608

    PubMed  CAS  Article  Google Scholar 

  29. DeYoung CG (2010) Personality neuroscience and the biology of traits. Soc Personal Psychol Compass 4:1165–1180. https://doi.org/10.1111/j.1751-9004.2010.00327.x

    Article  Google Scholar 

  30. DeYoung CG (2013) The neuromodulator of exploration: a unifying theory of the role of dopamine in personality. Front Hum Neurosci. https://doi.org/10.3389/fnhum.2013.00762

    PubMed  PubMed Central  Article  Google Scholar 

  31. DeYoung C (2014) Openness/intellect: a dimension of personality reflecting cognitive exploration. APA Handb Personal Soc Psychol Personal Process Individ Differ 4:369–399. https://doi.org/10.1037/14343-017

    Article  Google Scholar 

  32. DeYoung CG (2015) Cybernetic big five theory. J Res Pers 56:33–58. https://doi.org/10.1016/j.jrp.2014.07.004

    Article  Google Scholar 

  33. DeYoung CG, Gray JR (2009) Personality neuroscience: explaining individual differences in affect, behaviour and cognition

  34. DeYoung CG, Peterson JB, Higgins DM (2005) Sources of openness/intellect: cognitive and neuropsychological correlates of the fifth factor of personality. J Pers 73:825–858. https://doi.org/10.1111/j.1467-6494.2005.00330.x

    PubMed  Article  Google Scholar 

  35. DeYoung CG, Quilty LC, Peterson JB (2007) Between facets and domains: 10 aspects of the big five. J Pers Soc Psychol 93:880–896. https://doi.org/10.1037/0022-3514.93.5.880

    PubMed  Article  Google Scholar 

  36. DeYoung CG, Hirsh JB, Shane MS, Papademetris X (2010) Testing predictions from personality neuroscience: brain structure and the big five. 21:820–828. https://doi.org/10.1177/0956797610370159.Testing

  37. Doyle OM, Mehta MA, Brammer MJ (2015) The role of machine learning in neuroimaging for drug discovery and development. Psychopharmacology 232:4179–4189. https://doi.org/10.1007/s00213-015-3968-0

    PubMed  CAS  Article  Google Scholar 

  38. Dreher J-C, Schmidt PJ, Kohn P et al (2007) Menstrual cycle phase modulates reward-related neural function in women. Proc Natl Acad Sci USA 104:2465–2470. https://doi.org/10.1073/pnas.0605569104

    PubMed  PubMed Central  CAS  Article  Google Scholar 

  39. Eickhoff SB, Grefkes C (2011) Approaches for the integrated analysis of structure, function and connectivity of the human brain. Clin EEG Neurosci 42:107–121. https://doi.org/10.1177/155005941104200211

    PubMed  Article  Google Scholar 

  40. Eickhoff SB, Stephan KE, Mohlberg H et al (2005) A new SPM toolbox for combining probabilistic cytoarchitectonic maps and functional imaging data. Neuroimage 25:1325–1335. https://doi.org/10.1016/j.neuroimage.2004.12.034

    PubMed  Article  Google Scholar 

  41. Eickhoff SB, Paus T, Caspers S et al (2007) Assignment of functional activations to probabilistic cytoarchitectonic areas revisited. Neuroimage 36:511–521. https://doi.org/10.1016/j.neuroimage.2007.03.060

    PubMed  Article  Google Scholar 

  42. Eysenck HJ (1967) Biological basis of personality. Nature 199:1031–1034. https://doi.org/10.1038/1991031a0

    Article  Google Scholar 

  43. Filippi M, Valsasina P, Misci P et al (2013) The organization of intrinsic brain activity differs between genders: a resting-state fMRI study in a large cohort of young healthy subjects. Hum Brain Mapp 34:1330–1343. https://doi.org/10.1002/hbm.21514

    PubMed  Article  Google Scholar 

  44. Fleischhauer M, Enge S, Brocke B et al (2010) Same or different? Clarifying the relationship of need for cognition to personality and intelligence. Personal Soc Psychol Bull 36:82–96. https://doi.org/10.1177/0146167209351886

    Article  Google Scholar 

  45. Fox PT, Lancaster JL, Laird AR, Eickhoff SB (2014) Meta-analysis in human neuroimaging: computational modeling of large-scale databases peter. Annu Rev Neurosci 37:409–434. https://doi.org/10.1002/aur.1474.Replication

    PubMed  PubMed Central  CAS  Article  Google Scholar 

  46. Gao Q, Xu Q, Duan X et al (2013) Extraversion and neuroticism relate to topological properties of resting-state brain networks. Front Hum Neurosci 7:257. https://doi.org/10.3389/fnhum.2013.00257

    PubMed  PubMed Central  Article  Google Scholar 

  47. Gazzola V, Aziz-Zadeh L, Keysers C (2006) Empathy and the somatotopic auditory mirror system in humans. Curr Biol 16:1824–1829. https://doi.org/10.1016/j.cub.2006.07.072

    PubMed  CAS  Article  Google Scholar 

  48. Ge T, Holmes AJ, Buckner RL et al (2017) Heritability analysis with repeat measurements and its application to resting-state functional connectivity. Proc Natl Acad Sci 114:5521–5526. https://doi.org/10.1073/pnas.1700765114

    PubMed  PubMed Central  CAS  Article  Google Scholar 

  49. Glasser MF, Sotiropoulos SN, Wilson JA et al (2013) The minimal preprocessing pipelines for the Human Connectome Project. Neuroimage 80:105–124. https://doi.org/10.1016/j.neuroimage.2013.04.127

    PubMed  PubMed Central  Article  Google Scholar 

  50. Golbeck J (2011) Predicting personality with social media. Proc 2011 Annu Conf Ext Abstr Hum factors Comput Syst CHI EA. 11 253–262. https://doi.org/10.1145/1979742.1979614

  51. Golbeck J, Robles C, Edmondson M, Turner K (2011) Predicting personality from twitter. Proc—2011 IEEE Int Conf Privacy., Secur Risk Trust IEEE Int Conf Soc Comput PASSAT/SocialCom 149–156. https://doi.org/10.1109/PASSAT/SocialCom.2011.33

  52. Goldberg LR, Rosolack TK (1994) The big five factor structure as an integrative framework: an empirical comparison with Eysenck’s P-E-N model. In: Halverson Jr. CF, Kohnstamm GA, Martin RP (eds) The developing structure of temperament and personality from infancy to adulthood. Lawrence Erlbaum, New York, pp 7–35

    Google Scholar 

  53. Gray J, Mcnaughton N (2000) The neuropsychology of anxiety: an enquiry into the functions of the septo-hippocampal system, second. Oxford Psychol Ser Second Edi:433. https://doi.org/10.1017/S0140525X00013066

    Article  Google Scholar 

  54. Graziano WG, Habashi MM, Sheese BE, Tobin RM (2007) Agreeableness, empathy, and helping: A person × situation perspective. J Pers Soc Psychol 93:583–599. https://doi.org/10.1037/0022-3514.93.4.583

    PubMed  Article  Google Scholar 

  55. Grosbras MH, Beaton S, Eickhoff SB (2012) Brain regions involved in human movement perception: a quantitative voxel-based meta-analysis. Hum Brain Mapp 33:431–454. https://doi.org/10.1002/hbm.21222

    PubMed  Article  Google Scholar 

  56. Haas BW, Brook M, Remillard L et al (2015) I know how you feel: The warm-altruistic personality profile and the empathic brain. PLoS One 10:1–15. https://doi.org/10.1371/journal.pone.0120639

    CAS  Article  Google Scholar 

  57. Harrison PJ, Tunbridge EM (2008) Catechol-O-methyltransferase (COMT): a gene contributing to sex differences in brain function, and to sexual dimorphism in the predisposition to psychiatric disorders. Neuropsychopharmacology 33:3037–3045. https://doi.org/10.1038/sj.npp.1301543

    PubMed  CAS  Article  Google Scholar 

  58. Hassabis D, Spreng RN, Rusu AA et al (2014) Imagine all the people: How the brain creates and uses personality models to predict behavior. Cereb Cortex 24:1979–1987. https://doi.org/10.1093/cercor/bht042

    PubMed  Article  Google Scholar 

  59. Hastie T, Tibshirani R, Friedman J (2009) The elements of statistical learning. Elements 1:337–387. https://doi.org/10.1007/b94608

    Article  Google Scholar 

  60. Hayes DJ, Northoff G (2012) Common brain activations for painful and non-painful aversive stimuli. BMC Neurosci 13:60. https://doi.org/10.1186/1471-2202-13-60

    PubMed  PubMed Central  Article  Google Scholar 

  61. Hjelmervik H, Hausmann M, Osnes B et al (2014) Resting states are resting traits—an fMRI study of sex differences and menstrual cycle effects in resting state cognitive control networks. PLoS One 9:32–36. https://doi.org/10.1371/journal.pone.0103492

    CAS  Article  Google Scholar 

  62. Holmes CJ, Hoge R, Collins L et al (1998) Enhancement of MR images using registration for signal averaging. J Comput Assist Tomogr 22:324–333. https://doi.org/10.1097/00004728-199803000-00032

    PubMed  CAS  Article  Google Scholar 

  63. Hooker CI, Verosky SC, Miyakawa A et al (2008) The influence of personality on neural mechanisms of observational fear and reward learning. Neuropsychologia 46:2709–2724. https://doi.org/10.1016/j.neuropsychologia.2008.05.005

    PubMed  PubMed Central  Article  Google Scholar 

  64. Hu X, Erb M, Ackermann H et al (2011) Voxel-based morphometry studies of personality: issue of statistical model specification—effect of nuisance covariates. Neuroimage 54:1994–2005. https://doi.org/10.1016/j.neuroimage.2010.10.024

    PubMed  Article  Google Scholar 

  65. Iacoboni M (2009) Imitation, empathy, and mirror neurons. Annu Rev Psychol 60:653–670. https://doi.org/10.1146/annurev.psych.60.110707.163604

    PubMed  Article  Google Scholar 

  66. Iannetti GD, Mouraux A (2010) From the neuromatrix to the pain matrix (and back). Exp Brain Res 205:1–12

    PubMed  CAS  Article  Google Scholar 

  67. IBM Corp. Released (2011) IBM SPSS statistics for windows, Version 20.0

  68. Jang KL, Livesley WJ, Vernon P (1996) Heritability of the big five personality dimensions and their facets: a twin study. J Pers 64:577–591. https://doi.org/10.1111/j.1467-6494.1996.tb00522.x

    PubMed  CAS  Article  Google Scholar 

  69. Jensen-Campbell LA, Graziano WG (2001) Agreeableness as a moderator of interpersonal conflict. J Pers 69:323–361. https://doi.org/10.1111/1467-6494.00148

    PubMed  CAS  Article  Google Scholar 

  70. Jokela M, Kivimäki M, Elovainio M, Keltikangas-Järvinen L (2009) Personality and having children: a two-way relationship. J Pers Soc Psychol 96:218–230. https://doi.org/10.1037/a0014058

    PubMed  Article  Google Scholar 

  71. Jorm AF (1987) Sex differences in neuroticism: a quantitative synthesis of published research. Aust N Z J Psychiatry 21:501–506. https://doi.org/10.3109/00048678709158917

    PubMed  CAS  Article  Google Scholar 

  72. Kennis M, Rademaker AR, Geuze E (2013) Neural correlates of personality: an integrative review. Neurosci Biobehav Rev 37:73–95. https://doi.org/10.1016/j.neubiorev.2012.10.012

    PubMed  Article  Google Scholar 

  73. Keysers C, Gazzola V (2007) Integrating simulation and theory of mind: from self to social cognition. Trends Cogn Sci 11:194–196

    PubMed  Article  Google Scholar 

  74. Kim H, Shimojo S, O’Doherty JP (2006) Is avoiding an aversive outcome rewarding? Neural substrates of avoidance learning in the human brain. PLoS Biol 4:1453–1461. https://doi.org/10.1371/journal.pbio.0040233

    CAS  Article  Google Scholar 

  75. Koelsch S, Skouras S, Jentschke S (2013) Neural correlates of emotional personality: a structural and functional magnetic resonance imaging study. PLoS One. https://doi.org/10.1371/journal.pone.0077196

    Article  PubMed  PubMed Central  Google Scholar 

  76. Kogler L, Müller VI, Chang A et al (2015) Psychosocial versus physiological stress—meta-analyses on deactivations and activations of the neural correlates of stress reactions. Neuroimage 119:235–251. https://doi.org/10.1016/j.neuroimage.2015.06.059

    PubMed  PubMed Central  Article  Google Scholar 

  77. Konishi Y, Tanii H, Otowa T et al (2014) Gender-specific association between the COMT Val158Met polymorphism and openness to experience in panic disorder patients. Neuropsychobiology 69:165–174. https://doi.org/10.1159/000360737

    PubMed  CAS  Article  Google Scholar 

  78. Ktena SI, Arslan S, Parisot S, Rueckert D (2017) Exploring heritability of functional brain networks with inexact graph matching. Proc Int Symp Biomed Imaging 354–357

  79. Kumari V (2004) Personality predicts brain responses to cognitive demands. J Neurosci 24:10636–10641. https://doi.org/10.1523/JNEUROSCI.3206-04.2004

    PubMed  CAS  Article  Google Scholar 

  80. Kunisato Y, Okamoto Y, Okada G et al (2011) Personality traits and the amplitude of spontaneous low-frequency oscillations during resting state. Neurosci Lett 492:109–113. https://doi.org/10.1016/j.neulet.2011.01.067

    PubMed  CAS  Article  Google Scholar 

  81. Lahey BB (2009) Public health significance of neuroticism. Am Psychol 64:241–256. https://doi.org/10.1037/a0015309

    PubMed  PubMed Central  Article  Google Scholar 

  82. Lane DM (2013) Introduction to statistics. Introd to Stat 454–458. https://doi.org/10.1016/B978-0-12-370483-2.00006-0

  83. Langner R, Eickhoff SB (2013) Sustaining attention to simple tasks: a meta-analytic review of the neural mechanisms of vigilant attention. Psychol Bull 139:870–900. https://doi.org/10.1037/a0030694

    PubMed  Article  Google Scholar 

  84. Lee M, Smyser C, Shimony J (2012) {Resting-State} {fMRI:} a review of methods and clinical applications. Am J Neuroradiol. https://doi.org/10.3174/ajnr.A3263

    Article  PubMed  Google Scholar 

  85. Lei X, Yang T, Wu T (2015) Functional neuroimaging of extraversion-introversion. Neurosci Bull 31:663–675. https://doi.org/10.1007/s12264-015-1565-1

    PubMed  PubMed Central  Article  Google Scholar 

  86. Li J, Liu J (2010) Extraversion predicts individual differences in face recognition. pp 295–298

  87. Li N, Ma N, Liu Y et al (2013) Resting-state functional connectivity predicts impulsivity in economic decision-making. J Neurosci 33:4886–4895. https://doi.org/10.1523/JNEUROSCI.1342-12.2013

    PubMed  CAS  Article  Google Scholar 

  88. Liu X, Hairston J, Schrier M, Fan J (2011) Common and distinct networks underlying reward valence and processing stages: a meta-analysis of functional neuroimaging studies. Neurosci Biobehav Rev 35:1219–1236. https://doi.org/10.1016/j.neubiorev.2010.12.012

    PubMed  Article  Google Scholar 

  89. Liu W-Y, Weber B, Reuter M et al (2013) The big five of personality and structural imaging revisited: a VBM–DARTEL study. Neuroreport 24:375–380. https://doi.org/10.1097/WNR.0b013e328360dad7

    PubMed  Article  Google Scholar 

  90. MacLean MH, Arnell KM (2010) Personality predicts temporal attention costs in the attentional blink paradigm. Psychon Bull Rev 17:556–562. https://doi.org/10.3758/PBR.17.4.556

    PubMed  Article  Google Scholar 

  91. Madsen MK, Mc Mahon B, Andersen SB et al (2015) Threat-related amygdala functional connectivity is associated with 5-HTTLPR genotype and neuroticism. Soc Cogn Affect Neurosci 11:140–149. https://doi.org/10.1093/scan/nsv098

    PubMed  PubMed Central  Article  Google Scholar 

  92. Magyar M, Gonda X, Pap D et al (2017) Decreased openness to experience is associated with migraine-type headaches in subjects with lifetime depression. Front Neurol. https://doi.org/10.3389/fneur.2017.00270

    PubMed  PubMed Central  Article  Google Scholar 

  93. Marcus B, Machilek F, Schütz A (2006) Personality in cyberspace: personal Web sites as media for personality expressions and impressions. J Pers Soc Psychol 90:1014–1031. https://doi.org/10.1037/0022-3514.90.6.1014

    PubMed  Article  Google Scholar 

  94. Martin L, Clair J, Davis P et al (2006) Enhanced recognition of facial expressions of disgust in opiate users receiving maintenance treatment. Addiction 101:1598–1605. https://doi.org/10.1111/j.1360-0443.2006.01574.x

    PubMed  Article  Google Scholar 

  95. McCrae RR, Costa PT (2004) A contemplated revision of the NEO Five-Factor Inventory. Pers Individ Dif 36:587–596. https://doi.org/10.1016/S0191-8869(03)00118-1

    Article  Google Scholar 

  96. Mohan G, Mulla ZR (2013) Openness to experience and work outcomes: exploring the moderating effects of conscientiousness and job complexity

  97. Molnar-Szakacs I, Arzy S (2009) Searching for an integrated self-representation. Commun Integr Biol 2:365–367

    PubMed  PubMed Central  Article  Google Scholar 

  98. Molnar-Szakacs I, Uddin LQ (2013) Self-processing and the default mode network: interactions with the mirror neuron system. Front Hum Neurosci. https://doi.org/10.3389/fnhum.2013.00571

    PubMed  PubMed Central  Article  Google Scholar 

  99. Nicholson N, Fenton-O’Creevy M, Soane E, Willman P (2002) Risk propensity and personality. London Edu/Docs/Risk 1–33. https://doi.org/10.1080/1366987032000123856

    Article  Google Scholar 

  100. Nostro AD, Müller VI, Reid AT, Eickhoff SB (2016) Correlations between personality and brain structure: a crucial role of gender. Cereb Cortex 1–15. https://doi.org/10.1093/cercor/bhw191

  101. Oktar N, Oktar Y (2015) Machine learning and neuroimaging. J Neurol Sci [Turkish] 32:1–4

    Google Scholar 

  102. Orrù G, Pettersson-Yeo W, Marquand AF et al (2012) Using Support Vector Machine to identify imaging biomarkers of neurological and psychiatric disease: a critical review. Neurosci Biobehav Rev 36:1140–1152. https://doi.org/10.1016/j.neubiorev.2012.01.004

    PubMed  Article  Google Scholar 

  103. Oswald LM, Zandi P, Nestadt G et al (2006) Relationship between cortisol responses to stress and personality. 1583–1591. https://doi.org/10.1038/sj.npp.1301012

  104. Ozer DJ, Benet-Martínez V (2006) Personality and the prediction of consequential outcomes. Annu Rev Psychol 57:401–421. https://doi.org/10.1146/annurev.psych.57.102904.190127

    PubMed  Article  Google Scholar 

  105. Pang Y, Cui Q, Wang Y et al (2016) Extraversion and neuroticism related to the resting-state effective connectivity of amygdala. Sci Rep. https://doi.org/10.1038/srep35484

    Article  PubMed  PubMed Central  Google Scholar 

  106. Passamonti L, Fera F, Magariello A et al (2006) Monoamine oxidase—a genetic variations influence brain activity associated with inhibitory control: new insight into the neural correlates of impulsivity. Biol Psychiatry 59:334–340. https://doi.org/10.1016/j.biopsych.2005.07.027

    PubMed  CAS  Article  Google Scholar 

  107. Passamonti L, Terracciano A, Riccelli R et al (2015) Increased functional connectivity within mesocortical networks in open people. Neuroimage 104:301–309. https://doi.org/10.1016/j.neuroimage.2014.09.017

    PubMed  CAS  Article  Google Scholar 

  108. Pasternak GW, Pan Y-X (2013) Mu opioids and their receptors: evolution of a concept. Pharmacol Rev 65:1257–1317. https://doi.org/10.1124/pr.112.007138

    PubMed  PubMed Central  CAS  Article  Google Scholar 

  109. Pearman A (2009) Predictors of subjective memory in young adults. 101–107. https://doi.org/10.1007/s10804-009-9063-1

  110. Peciña S, Smith KS, Berridge KC (2006) Hedonic hot spots in the brain. Neurosci 12:500–511. https://doi.org/10.1177/1073858406293154

    Article  Google Scholar 

  111. Plitt M, Barnes KA, Wallace GL et al (2015) Resting-state functional connectivity predicts longitudinal change in autistic traits and adaptive functioning in autism. Proc Natl Acad Sci 112:E6699-6706. https://doi.org/10.1073/pnas.1510098112

    CAS  Article  Google Scholar 

  112. Power R, Pluess M (2015) Heritability estimates of the Big Five personality traits based on common genetic variants. Transl Psychiatry 5:e604. https://doi.org/10.1038/tp.2015.96

    PubMed  PubMed Central  CAS  Article  Google Scholar 

  113. Power JD, Cohen AL, Nelson SM et al (2011) Functional network organization of the human brain. Neuron 72:665–678. https://doi.org/10.1016/j.neuron.2011.09.006.Functional

    PubMed  PubMed Central  CAS  Article  Google Scholar 

  114. Qin P, Northoff G (2011) How is our self related to midline regions and the default-mode network? Neuroimage 57:1221–1233

    PubMed  Article  Google Scholar 

  115. Quercia D, Kosinski M, Stillwell D, Crowcroft J. Our twitter profiles, our selves: predicting personality with twitter

  116. Roberts BW, Jackson JJ, Fayard JV et al (2009) Conscientiousness. Handb Individ Differ Soc Behav 369–381

  117. Rosenberg MD, Finn ES, Scheinost D et al (2016) A neuromarker of sustained attention from whole-brain functional connectivity. Nat Neurosci 19:165–171. https://doi.org/10.1038/nn.4179

    PubMed  CAS  Article  Google Scholar 

  118. Rottschy C, Langner R, Dogan I et al (2012) Modelling neural correlates of working memory: a coordinate-based meta-analysis. Neuroimage 60:830–846. https://doi.org/10.1016/j.neuroimage.2011.11.050

    PubMed  CAS  Article  Google Scholar 

  119. Rusting CL, Larsen RJ (1997) Extraversion, neuroticism, and susceptibility to positive and negative affect: A test of two theoretical models. Pers Individ Dif 22:607–612. https://doi.org/10.1016/S0191-8869(96)00246-2

    Article  Google Scholar 

  120. Ryan JP, Sheu LK, Gianaros PJ (2011) Resting state functional connectivity within the cingulate cortex jointly predicts agreeableness and stressor-evoked cardiovascular reactivity. Neuroimage 55:363–370. https://doi.org/10.1016/j.neuroimage.2010.11.064

    PubMed  Article  Google Scholar 

  121. Sabatinelli D, Fortune EE, Li Q et al (2011) Emotional perception: Meta-analyses of face and natural scene processing. Neuroimage 54:2524–2533. https://doi.org/10.1016/j.neuroimage.2010.10.011

    PubMed  Article  Google Scholar 

  122. Salimi-Khorshidi G, Douaud G, Beckmann CF et al (2014) Automatic denoising of functional MRI data: combining independent component analysis and hierarchical fusion of classifiers. Neuroimage 90:449–468. https://doi.org/10.1016/j.neuroimage.2013.11.046

    PubMed  PubMed Central  Article  Google Scholar 

  123. Samartsidis P, Montagna S, Nichols TE, Johnson TD (2017) The coordinate-based meta-analysis of neuroimaging data. Stat Sci Volume 32:580–599

    Article  Google Scholar 

  124. Sampaio A, Soares JM, Coutinho J et al (2014) The big five default brain: functional evidence. Brain Struct Funct 219:1913–1922. https://doi.org/10.1007/s00429-013-0610-y

    PubMed  Article  Google Scholar 

  125. Sapolsky RM (1994) Glucocorticoids, stress and exacerbation of excitotoxic neuron death. Semin Neurosci 6:323–331

    CAS  Article  Google Scholar 

  126. Satterthwaite TD, Elliott MA, Gerraty RT et al (2013) An improved framework for confound regression and filtering for control of motion artifact in the preprocessing of resting-state functional connectivity data. Neuroimage 64:240–256. https://doi.org/10.1016/j.neuroimage.2012.08.052

    PubMed  Article  Google Scholar 

  127. Schilbach L, Derntl B, Aleman A et al (2016) Differential patterns of dysconnectivity in mirror neuron and mentalizing networks in schizophrenia. Schizophr Bull sbw015. https://doi.org/10.1093/schbul/sbw015

    Article  Google Scholar 

  128. Selleck RA, Baldo BA (2017) Feeding-modulatory effects of mu-opioids in the medial prefrontal cortex: a review of recent findings and comparison to opioid actions in the nucleus accumbens. Psychopharmacology 234:1439–1449

    PubMed  PubMed Central  CAS  Article  Google Scholar 

  129. Servaas MN, Geerligs L, Renken RJ et al (2015) Connectomics and neuroticism: an altered functional network organization. Neuropsychopharmacology 40:296–304. https://doi.org/10.1038/npp.2014.169

    PubMed  Article  Google Scholar 

  130. Shanmugan S, Epperson CN (2014) Estrogen and the prefrontal cortex: Towards a new understanding of estrogen’s effects on executive functions in the menopause transition. Hum Brain Mapp 35:847–865

    PubMed  Article  Google Scholar 

  131. Spreng RN, Mar RA, Kim ASN (2008) The common neural basis of autobiographical memory, prospection, navigation, theory of mind, and the default mode: a quantitative meta-analysis. J Cogn Neurosci 21:489–510. https://doi.org/10.1162/jocn.2008.21029

    Article  Google Scholar 

  132. Sprouse-Blum AS, Smith G, Sugai D, Parsa FD (2010) Understanding endorphins and their importance in pain management. Hawaii Med J 69:70–71

    PubMed  PubMed Central  Google Scholar 

  133. Studer-Luethi B, Jaeggi SM, Buschkuehl M, Perrig WJ (2012) Influence of neuroticism and conscientiousness on working memory training outcome. Pers Individ Dif 53:44–49. https://doi.org/10.1016/j.paid.2012.02.012

    Article  Google Scholar 

  134. Sutin AR, Terracciano A, Kitner-Triolo MH et al (2011) Personality traits prospectively predict verbal fluency in a lifespan sample. Psychol Aging 26:994–999. https://doi.org/10.1037/a0024276

    PubMed  PubMed Central  Article  Google Scholar 

  135. Tejeda HA, Hanks AN, Scott L et al (2015) Prefrontal cortical kappa opioid receptors attenuate responses to amygdala inputs. Neuropsychopharmacology 40:2856–2864. https://doi.org/10.1038/npp.2015.138

    PubMed  PubMed Central  CAS  Article  Google Scholar 

  136. Tipping M (2001) Sparse Bayesian learning and the relevance vector mach. J Mach Learn Res 1:211–244. https://doi.org/10.1162/15324430152748236

    Article  Google Scholar 

  137. Tzschoppe J, Nees F, Banaschewski T et al (2014) Aversive learning in adolescents: modulation by amygdala-prefrontal and amygdala-hippocampal connectivity and neuroticism. Neuropsychopharmacology 39:875–884. https://doi.org/10.1038/npp.2013.287

    PubMed  Article  Google Scholar 

  138. Van Essen DC, Smith SM, Barch DM et al (2013) The WU-Minn human connectome project: an overview. Neuroimage 80:62–79. https://doi.org/10.1016/j.neuroimage.2013.05.041

    PubMed  PubMed Central  Article  Google Scholar 

  139. van den Heuvel MP, van Soelen ILC, Stam CJ et al (2013) Genetic control of functional brain network efficiency in children. Eur Neuropsychopharmacol 23:19–23. https://doi.org/10.1016/j.euroneuro.2012.06.007

    PubMed  CAS  Article  Google Scholar 

  140. Varikuti DP, Hoffstaedter F, Genon S et al (2016) Resting-state test retest reliability of a priori defined canonical networks over different preprocessing steps. Brain Struct Funct 1–22. https://doi.org/10.1007/s00429-016-1286-x

  141. Varoquaux G, Thirion B (2014) How machine learning is shaping cognitive neuroimaging. Gigascience 3:28. https://doi.org/10.1186/2047-217X-3-28

    PubMed  PubMed Central  Article  Google Scholar 

  142. Varoquaux G, Raamana P, Engemann D et al (2016) Assessing and tuning brain decoders: cross-validation, caveats, and guidelines. arXiv:160605201 [statML] 1–14. https://doi.org/10.1016/j.neuroimage.2016.10.038

  143. Verweij KJH, Yang J, Lahti J et al (2012) Maintenance of genetic variation in human personality: testing evolutionary models by estimating heritability due to common causal variants and investigating the effect of distant inbreeding. Evolution 66:3238–3251. https://doi.org/10.1111/j.1558-5646.2012.01679.x

    PubMed  PubMed Central  Article  Google Scholar 

  144. Viken RJ, Rose RJ, Kaprio J, Koskenvuo M (1994) A developmental genetic analysis of adult personality: extraversion and neuroticism from 18 to 59 years of age. J Pers Soc Psychol 66:722–730. https://doi.org/10.1037//0022-3514.66.4.722

    PubMed  CAS  Article  Google Scholar 

  145. Wang Y, Fan Y, Bhatt P, Davatzikos C (2010) High-dimensional pattern regression using machine learning: from medical images to continuous clinical variables. Neuroimage 50:1519–1535. https://doi.org/10.1016/j.neuroimage.2009.12.092

    PubMed  PubMed Central  Article  Google Scholar 

  146. Weis S, Hodgetts S, Hausmann M (2017) Sex differences and menstrual cycle effects in cognitive and sensory resting state networks. Brain Cogn

  147. Xia M, Wang J, He Y (2013) BrainNet viewer: a network visualization tool for human brain connectomics. PLoS One. https://doi.org/10.1371/journal.pone.0068910

    Article  PubMed  PubMed Central  Google Scholar 

  148. Xu J, Moeller S, Auerbach EJ et al (2013) Evaluation of slice accelerations using multiband echo planar imaging at 3T. Neuroimage 83:991–1001. https://doi.org/10.1016/j.neuroimage.2013.07.055

    PubMed  Article  Google Scholar 

  149. Yadollahi P, Khalaginia Z, Vedadhir A et al (2014) The study of predicting role of personality traits in the perception of labor pain. Iran J Nurs Midwifery Res 19:S97–S102

    PubMed  PubMed Central  Google Scholar 

  150. Yang W, Cun L, Du X et al (2015) Gender differences in brain structure and resting-state functional connectivity related to narcissistic personality. Sci Rep 5:10924. https://doi.org/10.1038/srep10924

    PubMed  PubMed Central  CAS  Article  Google Scholar 

  151. Yarkoni T (2015) Neurobiological substrates of personality: a critical overview. APA Handb Personal Soc Psychol 4:61–83. https://doi.org/10.1017/CBO9781107415324.004

  152. Ziomkiewicz A, Wichary S, Bochenek D et al (2012) Temperament and ovarian reproductive hormones in women: evidence from a study during the entire menstrual cycle. Horm Behav 61:535–540. https://doi.org/10.1016/j.yhbeh.2012.01.017

    PubMed  CAS  Article  Google Scholar 

  153. Zobel A, Barkow K, Schulze-Rauschenbach S et al (2004) High neuroticism and depressive temperament are associated with dysfunctional regulation of the hypothalamic-pituitary-adrenocortical system in healthy volunteers. Acta Psychiatr Scand 109:392–399. https://doi.org/10.1111/j.1600-0447.2004.00313.x

    PubMed  CAS  Article  Google Scholar 

Download references

Acknowledgements

This study was supported by the Deutsche Forschungsgemeinschaft (DFG, EI 816/4-1, LA 3071/3-1), the National Institute of Mental Health (R01-MH074457), the Helmholtz Portfolio Theme “Supercomputing and Modelling for the Human Brain”, and the European Union’s Horizon 2020 Research and Innovation Programme under Grant Agreement no. 7202070 (HBP SGA1).

Author information

Affiliations

Authors

Corresponding author

Correspondence to Alessandra D. Nostro.

Ethics declarations

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary material 1 (DOCX 6044 KB)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Nostro, A.D., Müller, V.I., Varikuti, D.P. et al. Predicting personality from network-based resting-state functional connectivity. Brain Struct Funct 223, 2699–2719 (2018). https://doi.org/10.1007/s00429-018-1651-z

Download citation

Keywords

  • Functional networks
  • Gender differences
  • Hormonal influence
  • Machine learning
  • NEO-FFI
  • Resting-state functional connectivity