Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Resting brain dynamics at different timescales capture distinct aspects of human behavior

Resting brain dynamics at different timescales capture distinct aspects of human behavior ARTICLE https://doi.org/10.1038/s41467-019-10317-7 OPEN Resting brain dynamics at different timescales capture distinct aspects of human behavior 1,2,3 1 1 1 2,3 4,5 Raphaël Liégeois , Jingwei Li , Ru Kong , Csaba Orban , Dimitri Van De Ville , Tian Ge , 6 1,5,7,8 Mert R. Sabuncu & B.T. Thomas Yeo Linking human behavior to resting-state brain function is a central question in systems neuroscience. In particular, the functional timescales at which different types of behavioral factors are encoded remain largely unexplored. The behavioral counterparts of static func- tional connectivity (FC), at the resolution of several minutes, have been studied but beha- vioral correlates of dynamic measures of FC at the resolution of a few seconds remain unclear. Here, using resting-state fMRI and 58 phenotypic measures from the Human Connectome Project, we find that dynamic FC captures task-based phenotypes (e.g., pro- cessing speed or fluid intelligence scores), whereas self-reported measures (e.g., loneliness or life satisfaction) are equally well explained by static and dynamic FC. Furthermore, behaviorally relevant dynamic FC emerges from the interconnections across all resting-state networks, rather than within or between pairs of networks. Our findings shed new light on the timescales of cognitive processes involved in distinct facets of behavior. Department of Electrical and Computer Engineering, Clinical Imaging Research Centre, N.1 Institute for Health and Memory Networks Program, National University of Singapore, Singapore 117583, Singapore. Institute of Bioengineering, Centre for Neuroprosthetics, École Polytechnique Fédérale de Lausanne, 3 4 1015 Lausanne, Switzerland. Department of Radiology and Medical Informatics, University of Geneva, 1205 Geneva, Switzerland. Psychiatric & Neurodevelopmental Genetics Unit, Center for Genomic Medicine, Massachusetts General Hospital, Boston, MA 02114, USA. Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA 02129, USA. School of Electrical and Computer Engineering, Cornell University, 7 8 Ithaca, NY 14853, USA. Centre for Cognitive Neuroscience, Duke-NUS Medical School, Singapore 169857, Singapore. NUS Graduate School for Integrative Sciences and Engineering, National University of Singapore, Singapore 119077, Singapore. Correspondence and requests for materials should be addressed to R.L. (email: Raphael.Liegeois@epfl.ch) or to B.T.T.Y. (email: [email protected]) NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications 1 1234567890():,; ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 rain activity is highly organized in space and in time, even cognitive processes involved in the execution of various 1 30–32 in resting-state conditions . This intrinsic organization, tasks , thereby also supporting the ‘networked-brain’’ para- Bclassically evaluated from resting-state functional con- digm that has emerged in recent years . nectivity (FC), has been shown to encode various behavioral aspects such as integration of cognition and emotions , mon- Results 4 5,6 itoring of external environment , intellectual performance , and Behavioral counterparts of static and dynamic FC. We used emergence of stimulus-independent thoughts . FC has also been data from 419 unrelated HCP subjects to explore the extent to also used as a neuroimaging marker of several pathologies, which behavioral information is encoded in dynamic markers of 8,9 10 including Alzheimer’s disease , major depressive disorders , resting-state functional connectivity (FC), beyond classical static 11 12 13 Parkinson’s disease , schizophrenia , and autism . More measures of FC. We selected 58 behavioral measures from the recently, the advent of large neuroimaging and behavioral data- HCP dataset covering cognitive, social, emotion, and personality sets has allowed the further exploration of the FC behavioral traits (see Supplementary Table 2) from which age, gender, race, counterparts, showing intricate contributions of cognitive, emo- education, and motion (mean FD) were regressed. tional, social, and demographic aspects . FC markers were estimated from the HCP resting-state fMRI Importantly, all these studies use static measures of FC, which dataset. Classical preprocessing was performed, followed by a reflect the average functional organization of entire neuroimaging parcellation into 400 cortical regions of interest (ROIs) and recordings typically running over several minutes. However, there 35 19 subcortical ROIs . Subject-specific static FC markers were is recent converging evidence, suggesting that the resting brain computed by averaging correlation matrices of fMRI time series navigates through different functional connectivity configurations across runs. Dynamic FC markers were defined from an AR-1 15,16 at much faster timescales on the order of seconds . Therefore, model identified from the concatenation of the runs for each new dynamic measures of resting-state FC exploiting these faster subject (Methods). We chose to represent FC dynamics using an changes have been proposed and their behavioral counterparts AR-1 model for several reasons. First, we have shown recently have in turn been explored, showing links to cognitive flex- that AR-1 models, by exploiting the statistical link between 18 19 20 ibility , drug use , and mind-wandering . Yet, the comparison successive time points, capture FC dynamics significantly better of static and dynamic measures of FC has only been proposed in than a hidden Markov model explicitly representing switches specific applications such as temporal lobe epilepsy or the between different states with an equivalent number of para- description of eating behaviors , and a comprehensive analysis meters . Second, the hierarchical organization of brain network exploring which types of behavioral features emerge from func- dynamics was found to be reproduced by an AR-1 model of fMRI tional interactions at different timescales is missing. time series . Finally, lag threads, which also exploit the We explore this question using a discovery dataset (N = 419) sequential ordering information of time series (although they and a replication dataset (N = 328), comprising high-resolution focus on identifying temporal sequences of propagated activity resting-state functional magnetic resonance imaging (fMRI) time rather than connectivity patterns) were shown to provide series and 58 behavioral measures spanning cognitive, emotional, meaningful markers of intrinsic brain function . social, and personality traits from the Human Connectome The link between FC markers and behavioral measures was Project (HCP). First, we compare the extent to which static and 28,38 studied using a variance component model . The model dynamic FC capture behavioral information. We then investigate inputs are (i) a matrix containing the 58 behavioral measures for whether the behavioral relevance of FC markers is preferentially the N = 419 subjects and (ii) at least one N × N matrix, called a encoded in within- or between- network connectivity. Finally, we similarity matrix and denoted by K, whose i,j-th entry encodes test if static and dynamic FC capture complementary behavioral the similarity between (static or dynamic) FC of subjects i and j. information. FC dynamics are evaluated using a first-order Note that static FC matrices are symmetric, whereas dynamic FC 24,25 autoregressive (AR-1) model of resting-state fMRI data . AR-1 matrices are non-symmetric. The model estimates the level of models exploit the temporal ordering of fMRI time series to behavioral variability that is explained by FC variability, both on capture dynamic FC happening at a resolution of a few seconds to average over all behavioral measures, as well as for each which static approaches are blind , without suffering from the behavioral measure (Methods ). 26,27 limitations of classical sliding window methods . FC dynamics are then linked to the 58 behavioral measures using a variance Dynamic FC markers encode more behavioral information.We component model . This model has been extensively used in first compared the level of behavioral variance explained by static genome-wide complex trait analyses and was recently applied to and dynamic FC markers. To this end, we ran the multivariate study the neuroanatomical signatures of traits such as cognitive 29 variance component model twice: once using a similarity matrix or clinical measures . encoding the inter-subject similarity of static FC patterns, and We find that FC dynamics specifically encode behavioral once using similarity of dynamic FC patterns. measures evaluating performance in tasks, whereas self-reported Figure 1a shows that on average over the 58 behavioral measures are explained equally well by static and dynamic FC. measures, dynamic FC markers capture more behavioral variance We argue that this reflects the nature of the functional processes −4 than static FC (p = 8.31 × 10 ; two-tailed t-test), and Fig. 1b involved in the corresponding behavioral experiments. On the presents the results for eight individual phenotypic measures. one hand, task-based metrics engage cognitive processes at Results for the 50 remaining HCP measures are found in timescales on the order of a few seconds that can be captured by Supplementary Fig. 1. FC dynamics. On the other hand, self-reported measures might reflect trait-like properties that are less likely to change over a few seconds, therefore being equally well explained by functional Dynamic FC specifically encodes task-based measures. Even if patterns averaged over longer periods as encoded in static FC. dynamic FC encodes more behavioral information than static FC Furthermore, our results also suggest that task-performance on average, results of Fig. 1b show that some behavioral measures scores are defined by whole-brain FC dynamics involving the are not better explained by dynamic FC (e.g., Meaning of Life, interaction between multiple resting-state networks. Overall, Loneliness or Perceived Stress). In order to explore whether FC more than providing a mere statistical marker of task-perfor- dynamics specifically capture certain types of behavioral mea- mance, these findings offer new insights into the timescales of the sures, we ranked the 58 HCP measures based on the extent to 2 NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 ARTICLE ab Static Dynamic 0.9 0.5 0.8 0.4 0.7 0.6 0.3 0.5 0.4 0.2 0.3 0.2 0.1 0.1 0 0 Fig. 1 Dynamic FC explains more behavioral variance than static FC. a On average over 58 behavioral measures, dynamic FC (blue, 37%) explains more −4 behavioral variance than static FC (red, 19%) (p = 8.31 × 10 ; two-tailed t-test). b Variance explained for eight representative measures. Here, static FC utilizes Pearson’s correlation, while dynamic FC utilizes the coefficient matrix of a first-order autoregressive model. Error bars indicate standard deviation (SD) of the estimates a b 0.3 Static > Dynamic Dynamic > Static 0.2 T 0 0.1 –2 Static Dyn. 0.5 0.4 0.3 0.2 0.1 Self-reported Performance in task Unclassified Static Dyn. Fig. 2 Dynamic FC explains larger behavioral variance than static FC in task-performance measures. a Behavioral measures are ordered based on whether dynamic FC explains more variance than static FC. A positive t-statistic T suggests that dynamic FC explains more variance than static FC. Behavioral measures corresponding to task-performance are marked with a green dot and self-reported measures are marked with an orange dot. b No statistically significant difference (p > 0.10: two-tailed t-test) was found in the mean variance explained by static and dynamic FC in self-reported measures. c −3 Measures of performance in task are on average significantly better explained (p = 1.75 × 10 ; two-tailed t-test) by dynamic FC. Error bars indicate SD of the estimates −3 which dynamic FC better explain their variability, as compared to explained task-performance measures (p = 1.75 × 10 , Fig. 2c), static FC. To this end, we repeated the procedure for the 58 whereas no statistically significant difference could be found in measures and computed 58 t-statistics, denoted by T, of the dif- the capacity of both markers to explain self-reported measures ference between behavioral variance explained by static and (Fig. 2b). We also find that the difference of the differences dynamic FC for each measure (Supplementary Methods). Nega- between static and dynamic explained variances observed in −3 tive values of t-statistics indicate that static FC tends to better Fig. 2b, c is itself different from zero (p = 3.62 × 10 ; two-tailed explain the measure, whereas behavioral measures with positive t-test). This interaction effect confirms that the difference statistics are better explained by dynamic FC, as indicated in observed in Fig. 2c is related to the task condition and not only Fig. 2a. driven by the main effect shown in Fig. 1a. Moreover, the result of This ranking seems to draw a dichotomy between “task- Fig. 2c is reproduced using subcategories of task-based measures performance” and “self-reported” measures. On the one hand, the (Supplementary Fig. 3). Overall, the better average capacity of first category includes metrics that use participant’s performance dynamic FC to explain behavioral measures seems to be driven by in a task to assess a trait (e.g., working memory, spatial its increased capacity to explain task-based measures. orientation) and are marked with green dots in Fig. 2a. On the other hand, “self-reported” measures (orange dots in Fig. 2a) rely Behavior-related FC dynamics arise from network interactions. on subjective appraisal of traits (e.g., loneliness, life satisfaction). Functional interactions between brain networks have been shown No label was attached to the measures with no clear classification 39 to play a key role during the execution of tasks and in the in one of these categories. We find that dynamic FC better description of traits . We tested whether interaction between NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications 3 Static Dynamic Fluid intelligence Sustained attention-spec. Working memory (N-back) Processing speed Emotion face matching Meaning of life Loneliness Perceived stress Openness (NEO) Spatial orientation Perceived rejection Sleep quality Friendship Sustained attention—sens. Social cognition—interaction Meaning of life Loneliness Relational processing Working memory (list sorting) Anger—aggressiveness Grip strength Life satisfaction Emotional support Taste intensity Perceived stress Vocabulary (picture matching) Agreeableness (NEO) Extroversion (NEO) Emotion recog.—happiness Instrumental support Odor identification Positive affect Walking speed Contrast sensitivity Processing speed Sadness Perceived hostility Neuroticism (NEO) Visual episodic memory Emotion recog.—total Self-efficacy Conscientiousness (NEO) Emotion recog.—sadness Pain interference survey Verbal episodic memory Anger—affect Vocabulary (pronunciation) Story comprehension Arithmetic Emotion recog.—anger Emotion recog.—neutral Fear-affect Inhibition (Flanker task) Emotion recog.—fear Sustained attention—spec. Fluid intelligence Walking endurance Anger—hostility Dexterity Cognitive status (MMSE) Cognitive flexibility Fear—somatic arousal Emotion face matching Working memory (N-back) Social cognition—random Delay discounting Mean variance explained Variance explained Mean variance Mean variance in self-report in task ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 a b Within networks 0.1 0.10 Vis 0.08 0.1 SM 0.06 0.1 0.04 D-Att 0.02 0.1 Sal 0.1 Between networks Lim 0.10 0.1 FP 0.08 0.1 0.06 DMN 0.04 0.1 Sub 0.02 Vis SM D-Att Sal Lim FP DMN Sub Static Dynamic Fig. 3 Dynamic FC does not explain more behavioral variance than static FC within (pairs of) networks. a Behavioral variance explained by within-network (shaded diagrams) and between-network (unshaded diagrams), network static and dynamic FC. Seven cortical networks were used: visual (VIS), somatomotor (SM), dorsal attention (D-Att), salience (Sal), limbic (Lim), frontoparietal (FP), default mode network (DMN) and we also gathered the 19 subcortical areas (Sub). b There is no statistically significant difference in behavioral variance explained by within-network static and dynamic FC. c −3 Between-network static FC explains more behavioral variance than between-network dynamic FC (p = 8.31 × 10 ; two-tailed t-test). Error bars indicate SD of the estimates resting-state networks were also critical for extracting behavioral The average behavioral variance explained by combining static information from FC. To this end, the same model as described and dynamic FC is shown in dark blue in Fig. 4a. Results for eight above was used but similarity matrices were not computed from representative measures are shown in Fig. 4b and results for the the whole static or dynamic FC matrices. Instead, only sub-blocks 50 remaining traits are found in Supplementary Fig. 2. In Fig. 4a, of the FC matrices corresponding to (pairs of) well-known rest- the combined value is significantly higher than static FC (p = −4 ing-state networks were used. In other words, we tested how 4.73 × 10 ; two-tailed t-test), confirming the fact that FC behavioral variability is encoded in the variability of (pairs of) dynamics contains information above and beyond static FC. resting-state networks connectivity patterns. We used a common However, no statistical difference was found between average partition in seven cortical resting-state networks and included combined results and dynamic FC (p > 0.10, see Supplementary subcortical areas (Methods), as shown in Fig. 3. Table 1 for details), which suggests that the information encoded The average behavioral variance explained by static and by static FC is largely encoded in dynamic FC. dynamic FC restricted to within or between networks is under 10% for almost all the pairs of networks. Not surprisingly this is Dynamic FC interactions driving task-performance. We now lower than the behavioral variance explained from the whole- explore which dynamic FC interactions contribute to the overall brain connectivity patterns (19% for static FC and 37% for association with task-performance (Fig. 2c). We used a refor- dynamic FC), as anticipated from previous findings showing that mulation of the variance component model defined in Eq. (2) that individual FC fingerprinting is distributed throughout the revealed the relative contribution of the interaction between each brain . More unexpected is the fact that FC dynamics do not pair of (sub)networks to the overall explained variance (Supple- seem to carry more behavioral information than static FC. On the mentary Eq. (10)). The results are shown in Fig. 5. It can be seen contrary, on average over all inter-network connections (Fig. 3, that default C and frontoparietal C, together with the subcortical unshaded diagrams), static FC explained more behavioral regions, are contributing the most to the association between −3 variance than dynamic FC (p = 8.31 × 10 ; two-tailed t-test), dynamic FC and task-performance. whereas no statistically significant difference was found for within-networks connections (Fig. 3, shaded diagrams). Replication dataset. The findings shown in Figs. 1–4 were replicated in a second group of 328 unrelated HCP subjects. More precisely, all significant differences found in Figs. 1–4 were also Testing complementarity between static and dynamic FC.We found to be significant in the replication dataset (more details are have shown that on average FC dynamics encode more behavioral found in Supplementary Figs. 5–9 and Supplementary Table 1). information than static FC (Fig. 1), especially for task- The replication dataset was composed of the second subject of performance measures (Fig. 2). However, this does not mean each HCP family containing more than one person. We note that that static FC is not capturing any additional behavioral infor- it is therefore not completely independent from the discovery mation not encoded by dynamic FC. To test this, we used a dataset. generalized version of the multivariate variance component model that takes multiple similarity matrices -in our case two: the Additional control analyses. We performed a series of control ones computed from static and dynamic FC- as inputs and esti- analyses to evaluate the impact of various processing steps in our mates the level of behavioral variance explained by the combi- baseline analysis. More specifically, we tested the impact of (i) nation of these similarity matrices (Supplementary Methods). including the variance of the mean cortical grayordinate signal as 4 NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications Static Dynamic Mean var. expl. Mean var. expl. c NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 ARTICLE Static Dynamic Combined ab 0.5 1 0.9 0.4 0.8 0.7 0.6 0.3 0.5 0.4 0.2 0.3 0.2 0.1 0.1 Fig. 4 Combined static and dynamic FC does not capture more behavioral variance than dynamic FC alone. a Average variance explained across 58 behavioral measures using static FC (red), dynamic FC (light blue), and the combination of these two (dark blue). b Variance explained for eight representative measures. Error bars indicate SD of the estimates functional connectivity (FC), and recent evidence has shown that exploiting the dynamical properties of FC instead of the classical B A static FC metrics could open new avenues to interpret brain A B functioning at different timescales. In this study, we aim at refining our understanding of the behavioral information carried by FC dynamics. To this end, we explored the extent to which resting-state static and dynamic FC measures relate to a large repertoire of measures covering cognitive, social, emotion, and personality traits. We first show that, on average over 58 selected behavioral measures, FC dynamics encode significantly more behavioral information than a common static FC metric. This confirms current findings that have highlighted the advantage of resting-state dynamic FC measures over their static counterparts 43 21 22 A in describing mindfulness , disease , and eating behaviors . Interestingly, FC dynamics within well-known resting-state networks, or between pairs of networks, did not capture more A C behavioral information than static FC in the same networks (Fig. 3). For example, static inter-network FC was shown to explain more behavioral variance than dynamic inter-network FC Fig. 5 Dynamic FC interactions contributing the most to the association on average. These results might seem counter-intuitive at first with task-performance. Networks and corresponding colors are the same as sight and suggest that similarity measures derived from local in Fig. 3, and subnetworks are defined following the 17-network parcellation 35 patterns of FC do not complement each other in the same way in of Schaefer et al. , as reported in Supplementary Fig. 4. The colors of the the static and dynamic cases. In other words, the advantage of edges are defined by their destination and only connections surviving an dynamic FC in explaining the behavioral information observed in FDR correction at the level q= 0.05 are shown Fig. 1 is encoded in the global dynamic FC interaction patterns. From a methodological point of view, this also indicates that even a covariate in the variance component model, (ii) evaluating the if dynamic FC uses richer statistical information than static FC by static and dynamic FC matrices from fMRI time series from relaxing the static assumption associated to this metric, dynamic which the mean cortical grayordinate signal was not regressed, FC should not a priori and automatically be considered as a better (iii) including head motion metrics as covariates in the variance neuroimaging marker than static FC. component model, (iv) evaluating the static and dynamic FC In Fig. 2, we show that FC dynamics specifically encode matrices from full (i.e., uncensored) fMRI time series, (v) the measures of performance in a task, such as working memory number of behavioral measures considered in the variance tasks, whereas static and dynamic FC explain self-reported component model, and (vi) the relative contributions of static and measures, such as the perception of loneliness, equally well. This dynamic FC to the overall variance explained within the com- additional information is found to be encoded in the global bined variance component model. The variance component dynamic FC patterns, and not confined to single areas or net- model appeared to be robust to these changes and in each case, works (Fig. 3). More precisely, Fig. 5 suggests that the default our main findings were reproduced (Supplementary Figs. 10–12). mode and frontoparietal networks drive the integration of the dynamic FC coming from other networks. This is in line with previous findings identifying these areas as hubs of the dynamic Discussion functional connectome , and further supports the importance of Exploring how resting-state functional organization is linked to coupled default network and frontoparietal activities during task- various behavioral traits is a central neuroimaging research performance . question. This organization is classically evaluated from NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications 5 Static Dynamic Combined Fluid intelligence Sustained attention—spec. Working memory (N-back) Processing speed Emotion face matching Meaning of life Loneliness Perceived stress Mean variance explained Variance explained S ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 Taken together, these results support the line of current find- Table 2). However, we note that classifying these measures in one ings indicating that different phenotypic measures such as pain , or the other category did not significantly change the results in 46 47 perception , and vigilance are encoded in dynamic interactions Fig. 2. That said, considering only two categories of behavioral between multiple areas and not within single networks. Then, measures disregards the multifactorial nature of behavior and the even if they concern interactions between resting-state networks, repertoire of behavioral measures could be approached using our results interestingly echo the nature of network interactions other classification criteria such as trait vs. state or intrinsic vs. involved in task-based conditions. Indeed, the execution of tasks extrinsic . relies on a coordinated activation of different networks at faster We referred to the AR model of BOLD time series as ‘dynamic’’ 31,32 63 timescales that can be captured by dynamic FC , and to which following the systems theory literature . This nomenclature is static measures of FC are blind. On the contrary, self-reported motivated by the fact that such models, by accounting for the measures could be considered as trait-like properties that might memory present in the time series (i.e., x depends on x ), are t t−1 therefore be explained equally well by average FC patterns able to reproduce empirical fluctuations in the multivariate time encoded in static FC. Converging results also suggest that simple series of interest much better than memoryless (or ‘static’’, fol- tasks exhibit segregated activation patterns whereas complex tasks lowing the same nomenclature) models . The AR model is also involving multiple cognitive processes (e.g., working memory or used to compute the dominant dynamic modes shaping resting- visuospatial attention) require an integrated activation of multiple state brain function . Overall, this model can be seen as a 39,48,49 intrinsic networks and activate flexible brain regions such compact way to summarize the temporal fluctuations of BOLD 50,51 as connector hubs . It does not seem unreasonable to assume and FC time series that are directly exploited by time-varying 17,65 that task-performance measures, in which subjects are incited to models . reach a high score in a test that often involves multiple or In summary, static measures of FC provide a measure of brain coordinated actions, capture more complex behavioral traits -as function averaged over several minutes. This is an over- defined above- than self-reported measures. More generally, these simplification and new dynamic measures capturing the temporal results constitute empirical evidence supporting the ‘networked- changes of brain function on the order of a few seconds have been brain’’ paradigm that has emerged in recent years. This paradigm proposed. While these new measures were shown to capture more essentially views the brain as a multiscale network producing statistical properties of fMRI data, their behavioral relevance complex spatio-temporal activity patterns rather than an above and beyond static FC remains unclear. Here, we have ensemble of neuronal populations with localized shown using 747 HCP subjects and 58 behavioral measures that 33,52,53 specification . FC dynamics specifically capture measures of performance in The fact that FC dynamics seem to capture more complex tasks by leveraging the dynamic information encoded in multiple- network interactions than static FC is also supported by the links network interactions. On the contrary, self-reported measures are that have been drawn between FC and the underlying brain equally well explained by static and dynamic measures of FC. anatomy, or structural connectivity (SC). As static FC is on Overall, we believe our work opens up future possibilities to a average closer to SC, the dynamic FC repertoire also captures better characterization of the cognitive processes shaping the excursions from SC that are characterized by higher efficiency various facets of human behavior. and lower modularity . These fluctuations of modular organi- zation, encoded in FC dynamics, were shown to operate at dif- Methods ferent timescales and to support the periodic (de)coupling of Data and preprocessing. We used data of the HCP 1200-subjects release com- 55 prising structural MRI, resting-state functional MRI, and behavioral measures of resting-state networks , possibly constituting a signature young (ages 22–35) and healthy participants drawn from a population of siblings . of consciousness and allowing for a more efficient transfer of All imaging data were acquired on a 3-T Siemens Skyra scanner using a multi-band neuronal information . sequence. Functional images have a temporal resolution of 0.72 s and a 2-mm Other converging findings suggest that the amplitude of FC isotropic spatial resolution whereas structural images are 0.7-mm isotropic. For each subject, four 14.4 min runs (1200 frames) of functional time series were dynamics variability decreases during task as compared to acquired . Resting-state fMRI data was projected to the fs_LR surface space using 58,59 resting-state . Altogether with the results of Figs. 2 and 3 23,66 the multimodal surface matching method (MSM-All . Both cortical and sub- indicating that task-based behavioral measures are specifically 67,68 cortical data were cleaned using the ICA-FIX method and saved in CIFTI encoded by resting-state FC dynamics, we might hypothesize that grayordinate format. This cleaning procedure included the regression of 24 motion-related parameters (six classical motion parameters, their derivatives, and there exists a resting-state ‘dynamic reservoir’’ that is recruited the squares of these 12 parameters). Motion censoring was then applied by when leaving rest and, which defines task-performance. This removing frames with FD > 0.2 mm or DVARS > 75, as well as one frame before dynamic reservoir, encoded by FC dynamics and not by simpler 69,70 and two frames after these frames . Remaining segments containing less than static FC markers as shown in Fig. 4, emerges from highly inte- five frames were also removed and runs with >50% of censored frames were discarded. Linear trends and mean cortical grayordinate signal were regressed and grated connections involving multiple intrinsic networks. As censored frames were ignored to compute regression coefficients. Mean cortical such, we could also interpret this dynamic connectivity structure grayordinate signal regression was performed because this step was shown recently as a signature of the human connectome evolution that tends to to strengthen the association between FC metrics and behavioral measures . attain efficient organizations to perform complex tasks . However, we note that not regressing mean cortical grayordinate signal yielded The dichotomy of behavioral measures proposed in Fig. 2 is similar conclusions (Supplementary Fig. 10). Finally, fMRI time series were par- cellated into 419 regions of interest (ROIs) comprising 400 cortical areas and motivated by the statistical difference between static and dynamic 19 subcortical areas defined in Freesurfer. Static functional connectivity was FC that seems to capture distinct behavioral properties. This obtained for each subject based on the pearson’s correlation matrices, computed dichotomy presents some limitations. First, the classification of from uncensored frames for each run, which were Fisher z-transformed, averaged behavioral measures in one of the two proposed categories, ‘task- over runs and transformed back to r-space. Dynamic functional connectivity measures were estimated from the model parameter of a first-order autoregressive performance’’ or ‘self-reported’’, was not always straightforward. representation of fMRI time series: For example, the Delay Discounting task measure was left unclassified as one could argue that it is not capturing a perfor- x ¼ A  x þ ϵ ð1Þ t t1 t mance, associated with an underlying truth or optimal score, in N ´ 1 where x 2R represents the fMRI time series in the N = 419 ROIs at time t, the same way a classical task such as the Working Memory task N ´ N R R A 2R is the model parameter that encodes the linear relationship between is. For the same reason, other measures (grip strength, odor N ´ 1 successive time points, and ε 2R are the residuals of the model . The model identification, walking speed, contrast sensitivity, taste intensity, parameter A was identified from the concatenation of the uncensored sections of and walking endurance) were not classified (Supplementary the different runs, while ignoring transitions between uncensored sections and 6 NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 ARTICLE transitions between runs. The proportion of variance explained by this model, mean difference of explained variance is: jjCovðε Þjj t F 2 defined as R ¼ 1  analogous to the definition of R for univariate AR jjCovðx Þjj X t F N  1 ^ ^ ^ V ¼ M  M : ð5Þ models, and where ||⋅|| denotes the Frobenius norm , is 69.3% ± 11.2% (com- jack i jack i¼1 puted over all subjects, including the replication dataset). qffiffiffiffiffiffiffiffiffi We selected 58 behavioral measures, including cognitive, social, emotion and ^ ^ With a large sample size (N = 419), the estimator M  M = V is jack jack personality traits (Supplementary Table 2). These measures consist of metrics from assumed to follow a standard normal distribution under the null hypothesis (M = the NIH Toolbox and some well-known non-NIH measures (e.g., NEO-FFI). 0) and a two-tailed p-value can be computed. All significant results survived FDR Details about behavioral measures can be found in HCP S1200 Data Dictionary correction at q < 0.05 (see Supplementary Table 1 for details). and Barch et al. . These measures were classified either as ‘task-performance’’, ‘self-reported’’,or ‘unclassified’’ if no clear belonging to one of the first two categories applied (Supplementary Table 2). Reporting summary. Further information on research design is available in Among the HCP 1200-subjects release, 1029 subjects had at least one run that the Nature Research Reporting Summary linked to this article. was not discarded after applying the preprocessing rules. Excluding the subjects with missing or problematic entries for some behavioral measures further reduced Data availability the dataset to 953 subjects belonging to 419 families. To avoid the influence of The HCP data is publicly available at http://www.humanconnectomeproject.org/data/; shared genetic and environmental factors, we kept the first subject from each informed consent was obtained from all HCP participants . family leading to a final set of N = 419 unrelated subjects. Among the 419 families used for the initial set, 91 ‘families’’ contained only one subject and hence these families were discarded in the construction of the replication dataset that contained Code availability 328 subjects. All code is publicly available at https://github.com/RaphaelLiegeois/FC-Behavior/. Variance component model. We regressed age, gender, race, education, and Received: 14 November 2018 Accepted: 3 May 2019 motion (mean FD) from the 58 phenotypic measures, which were then quantile normalized. We used the multivariate variance component model developed by Ge et al. to link FC markers and behavioral measures: Y ¼ C þ E ð2Þ where Y, C, and E are 419 × 58 matrices. Y contains the 58 processed behavioral References measures for all 419 subjects. VecðCÞ NðÞ 0; Σ  F and VecðEÞ NðÞ 0; Σ  I , c e 1. Damoiseaux, J. et al. Consistent resting-state networks across healthy subjects. where Vec(.) is the matrix vectorization operator, ⊗ is the Kronecker product of Proc. Natl Acad. Sci. USA 103, 13848–13853 (2006). matrices, and I is the identity matrix. F is a similarity matrix such that F(i, j) 2. Friston, K. J. Functional and effective connectivity: a review. Brain Connect. 1, encodes the (static or dynamic) FC similarity between subjects i and j, and is 13–36 (2011). defined as the correlation between the static FC (or dynamic FC) matrices of the 3. Greicius, M. D., Krasnow, B., Reiss, A. L. & Menon, V. Functional connectivity two subjects. Σ and Σ are unknown 58 × 58 matrices to be estimated from F and c e in the resting brain: a network analysis of the default mode hypothesis. Proc. Y. The variance explained by (static or dynamic) FC markers, denoted by M,is Natl Acad. Sci. USA 100, 253–258 (2003). computed as: 4. Gusnard, D. A. & Raichle, M. E. Searching for a baseline: functional imaging and the resting human brain. Nat. Rev. Neurosci. 2, 685–694 (2001). TrðΣ Þ M ¼ ; ð3Þ 5. Song, M. et al. Brain spontaneous functional connectivity and intelligence. TrðΣ Þþ TrðΣ Þ c e Neuroimage 41, 1168–1176 (2008). 6. van den Heuvel, M. P., Stam, C. J., Kahn, R. S. & Hulshoff Pol, H. E. Efficiency where Tr(.) is the trace operator. M measures how much inter-subject behavioral of functional brain networks and intellectual performance. J. Neurosci. 29, variability is explained by inter-subject FC variability . Equation (3) is computed 7619–7624 (2009). twice, once using static FC to build the FC similarity matrix in (2) and once using dynamic FC to build this matrix. M for a single behavioral measure is given by 7. Mason, M. F. et al. Wandering minds: the default network and stimulus- independent thought. Science 315, 393–395 (2007). M = Σ (i, i)/(Σ (i, i) + Σ (i, i)). We used the variance component model because it i c c e provides an easy way to summarize high-dimensional FC patterns and produces 8. Wang, K. et al. Altered functional connectivity in early alzheimer’s disease: one scalar (variance explained) for each behavioral measure that is normalized a resting-state fmri study. Hum. Brain. Mapp. 28, 967–978 (2007). (between 0 and 1) and directly comparable across measures and processing 9. Meskaldji, D.-E. et al. Prediction of long-term memory scores in mci based on methods. resting-state fmri. Neuroimage Clin. 12, 785–795 (2016). 10. Drysdale, A. T. et al. Resting-state connectivity biomarkers define neurophysiological subtypes of depression. Nat. Med. 23,28–38 (2017). Complexity of static and dynamic variance component models. Static FC 11. Gorges, M. et al. To rise and to fall: functional connectivity in cognitively similarity matrices are built from the correlation matrix of resting-state fMRI time normal and cognitively impaired patients with parkinson’s disease. Neurobiol. series. These correlation matrices are symmetric and of size N × N , where ROI ROI Aging 36, 1727–1735 (2015). N is the number of regions of interest, and hence their degrees of freedom ROI 12. Bassett, D. S., Nelson, B. G., Mueller, B. A., Camchong, J. & Lim, K. O. Altered (d.o.f.) is N ×(N − 1)/2. On the other hand, dynamic FC is built from the ROI ROI resting state complexity in schizophrenia. Neuroimage 59, 2196–2207 (2012). non-symmetric coefficient matrix of a first-order AR model of size N × N , ROI ROI 13. Anderson, J. S. et al. Functional connectivity magnetic resonance imaging and hence their d.o.f. is N . It should be noted, however, that this difference in d. ROI classification of autism. Brain 134, 3742–3754 (2011). o.f. is not present in the static and dynamic variance component models as the 14. Smith, S. M. et al. A positive-negative mode of population covariation links static and dynamic FC matrices are only used in intermediate steps of the com- brain connectivity, demographics and behavior. Nat. Neurosci. 18, 1565–1567 putation of the similarity matrices of the variance component model. These (2015). similarity matrices are in both the static and dynamic case symmetric and of size N with N ×(N − 1)/2 degrees of freedom, where N is the number of subjects. 15. Abrol, A. et al. Replicability of time-varying connectivity patterns in large resting state fmri samples. Neuroimage 163, 160–176 (2017). Therefore, there is no complexity bias giving an advantage to one or the other approach. 16. Liégeois, R., Laumann, T. O., Snyder, A. Z., Zhou, J. & Yeo, B. T. T. Interpreting temporal fluctuations in resting-state functional connectivity MRI. Neuroimage 163, 437–455 (2017). Statistic inference. We use the delete-1 Jackknife approach to evaluate the 17. Preti, M. G., Bolton, T. A. & Van De Ville, D. The dynamic functional statistical difference between the variance explained by two different methods (e.g., connectome: State-of-the-art and perspectives. Neuroimage 160,41–54 A B ^ ^ when M is computed using only static FC vs. only dynamic FC). Let M and M be (2017). the estimates of behavioral variance explained using the methods A and B to be 18. Chen, T., Cai, W., Ryali, S., Supekar, K. & Menon, V. Distinct global brain compared. Denoting by M the estimates computed from the whole dataset dynamics and spatiotemporal organization of the salience network. PLoS Biol. without subject i, the delete-1 Jackknife estimate of the mean difference between 14, e1002469 (2016). methods A and B is: 19. Vergara, V. M., Weiland, B. J., Hutchison, K. E. & Calhoun, V. D. The impact of combinations of alcohol, nicotine, and cannabis on dynamic brain N N X X 1 1 A B ^ ^ ^ ^ M ¼ M  M ¼ M ; ð4Þ connectivity. Neuropsychopharmacology 43, 877–890 (2018). jack i i i N N i¼1 i¼1 20. Christoff, K., Irving, Z. C., Fox, K. C., Spreng, R. N. & Andrews-Hanna, J. R. Mind-wandering as spontaneous thought: a dynamic framework. Nat. Rev. A B ^ ^ ^ where N = 419 is the number of subjects, and M ¼ M  M is the difference i i i Neurosci. 17, 718–731 (2016). of explained variance for the i-th Jackknife sample. The variance estimate of the NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 21. Chiang, S. et al. Temporal and spectral characteristics of dynamic functional 53. Wang, P. et al. Inversion of a large-scale circuit model reveals a cortical connectivity between resting-state networks reveal information beyond static hierarchy in the dynamic resting human brain. Sci. Adv. 5, eaat7854 connectivity. PLoS ONE 13, e0190220 (2018). (2019). 22. Park, B.-Y., Moon, T. & Park, H. Dynamic functional connectivity analysis 54. Liégeois, R. et al. Cerebral functional connectivity periodically (de) reveals improved association between brain networks and eating behaviors synchronizes with anatomical constraints. Brain Struct. Funct. 221, 2985–2997 compared to static analysis. Behav. Brain. Res. 337, 114–121 (2018). (2016). 23. Van Essen, D. C. et al. The WU-Minn human connectome project: an 55. Betzel, R. F., Fukushima, M., He, Y., Zuo, X. -N. & Sporns, O. Dynamic overview. Neuroimage 80,62–79 (2013). fluctuations coincide with periods of high and low modularity in resting-state 24. Valdés-Sosa, P. A. et al. Estimating brain functional connectivity with sparse functional brain networks. Neuroimage 127, 287–297 (2016). multivariate autoregression. Philos. Trans. R. Soc. Lond. B. Biol. Sci. 360, 56. Barttfeld, P. et al. Signature of consciousness in the dynamics of resting-state 969–981 (2005). brain activity. Proc. Natl Acad. Sci. USA 112, 887–892 (2015). 25. Rogers, B. P., Katwal, S. B., Morgan, V. L., Asplund, C. L. & Gore, J. C. 57. Zalesky, A., Fornito, A., Cocchi, L., Gollo, L. L. & Breakspear, M. Time- Functional mri and multivariate autoregressive models. Magn. Reson. Imaging resolved resting-state brain networks. Proc. Natl Acad. Sci. USA 111, 28, 1058–1065 (2010). 10341–10346 (2014). 26. Leonardi, N. & Van De Ville, D. On spurious and real fluctuations of dynamic 58. Elton, A. & Gao, W. Task-related modulation of functional connectivity functional connectivity during rest. Neuroimage 104, 430–436 (2015). variability and its behavioral correlations. Hum. Brain. Mapp. 36, 3260–3272 27. Hindriks, R. et al. Can sliding-window correlations reveal dynamic functional (2015). connectivity in resting-state fmri? Neuroimage 127, 242–256 (2016). 59. Hutchison, R. M. & Morton, J. B. Tracking the brain’s functional coupling 28. Yang, J., Lee, S. H., Goddard, M. E. & Visscher, P. M. Gcta: a tool for genome- dynamics over development. J. Neurosci. 35, 6849–6859 (2015). wide complex trait analysis. Am. J. Hum. Genet. 88,76–82 (2011). 60. van den Heuvel, M. P., Bullmore, E. T. & Sporns, O. Comparative 29. Sabuncu, M. R. et al. Morphometricity as a measure of the neuroanatomical connectomics. Trends Cogn. Sci. 20, 345–361 (2016). signature of a trait. Proc. Natl Acad. Sci. USA 113, E5749–E5756 (2016). 61. Grös, D. F., Antony, M. M., Simms, L. J. & McCabe, R. E. Psychometric 30. Bassett, D. S. et al. Dynamic reconfiguration of human brain networks during properties of the state-trait inventory for cognitive and somatic anxiety learning. Proc. Natl Acad. Sci. USA 108, 7641–7646 (2011). (sticsa): comparison to the state-trait anxiety inventory (stai). Psychol. Assess. 31. Cocchi, L., Zalesky, A., Fornito, A. & Mattingley, J. B. Dynamic cooperation 19, 369–381 (2007). and competition between brain systems during cognitive control. Trends 62. DeLeeuw, K. E. & Mayer, R. E. A comparison of three measures of cognitive Cogn. Sci. 17, 493–501 (2013). load: Evidence for separable measures of intrinsic, extraneous, and germane 32. Cohen, J. R. The behavioral and cognitive relevance of time-varying, dynamic load. J. Educ. Psychol. 100, 223–234 (2008). changes in functional connectivity. Neuroimage 180, 515–525 (2018). 63. Oppenheim, A. & Willsky, A. S. Signals and Systems (Prentice Hall, 1997). 33. Betzel, R. F. & Bassett, D. S. Multi-scale brain networks. Neuroimage 160, 64. Casorso, J. et al. Dynamic mode decomposition of resting-state and task fmri. 73–83 (2017). Neuroimage 194,42–54 (2019). 34. Smith, S. M. et al. Resting-state fmri in the human connectome project. 65. Hutchison, R. M. et al. Dynamic functional connectivity: promise, issues, and Neuroimage 80, 144–168 (2013). interpretations. Neuroimage 80, 360–378 (2013). 35. Schaefer, A. et al. Local-global parcellation of the human cerebral cortex from 66. Robinson, E. C. et al. Multimodal surface matching: fast and generalisable intrinsic functional connectivity mri. Cereb. Cortex 28, 3095–3114 (2018). cortical registration using discrete optimisation. Inf. Process. Med. Imaging 23, 36. Vidaurre, D., Smith, S. M. & Woolrich, M. W. Brain network dynamics are 475–486 (2013). hierarchically organized in time. Proc. Natl Acad. Sci. USA 114, 12827–12832 67. Salimi-Khorshidi, G. et al. Automatic denoising of functional mri data: (2017). combining independent component analysis and hierarchical fusion of 37. Mitra, A., Snyder, A. Z., Blazey, T. & Raichle, M. E. Lag threads organize the classifiers. Neuroimage 90, 449–468 (2014). brain’s intrinsic activity. Proc. Natl Acad. Sci. USA 112, E2235–E2244 (2015). 68. Griffanti, L. et al. Ica-based artefact removal and accelerated fmri acquisition 38. Ge, T. et al. Multidimensional heritability analysis of neuroanatomical shape. for improved resting state network imaging. Neuroimage 95, 232–247 (2014). Nat. Commun. 7, 13291 (2016). 69. Jenkinson, M., Bannister, P., Brady, M. & Smith, S. Improved optimization for 39. Cohen, J. R. & D’Esposito, M. The segregation and integration of distinct the robust and accurate linear registration and motion correction of brain brain networks and their relationship to cognition. J. Neurosci. 36, images. Neuroimage 17, 825–841 (2002). 12083–12094 (2016). 70. Power, J. D., Barnes, K. A., Snyder, A. Z., Schlaggar, B. L. & Petersen, S. E. 40. Cheng, J. C. et al. Multivariate machine learning distinguishes cross-network Spurious but systematic correlations in functional connectivity mri networks dynamic functional connectivity patterns in state and trait neuropathic pain. arise from subject motion. Neuroimage 59, 2142–2154 (2012). Pain 159, 1764–1776 (2018). 71. Li, J. et al. Global signal regression strengthens association between resting- 41. Yeo, B. T. T. et al. The organization of the human cerebral cortex estimated by state functional connectivity and behavior. bioRxiv (2019). intrinsic functional connectivity. J. Neurophysiol. 106, 1125–1165 (2011). 72. Hamilton, J. Time Series Analysis. (Princeton University Press, Princeton, NJ, 42. Finn, E. S. et al. Functional connectome fingerprinting: identifying individuals 1994). using patterns of brain connectivity. Nat. Neurosci. 18, 1664–1671 (2015). 73. Barch, D. M. et al. Function in the human connectome: task-fmri and 43. Lim, J., Teng, J., Patanaik, A., Tandi, J. & Massar, S. A. A. Dynamic functional individual differences in behavior. Neuroimage 80, 169–189 (2013). connectivity markers of objective trait mindfulness. Neuroimage 176, 193–202 74. Efron, B. The Jackknife, the Bootstrap and Other Resampling Plans Vol. 38 (2018). (Society for Industrial and Applied Mathematics, 1982). 44. Schaefer, A. et al. Dynamic network participation of functional connectivity hubs assessed by resting-state fmri. Front. Hum. Neurosci. 8, 195 (2014). 45. Spreng, R. N., Stevens, W. D., Chamberlain, J. P., Gilmore, A. W. & Schacter, Acknowledgements D. L. Default network activity, coupled with the frontoparietal control We would like to thank Russ Poldrack, Timothy Laumann, and Jean-Baptiste Poline for network, supports goal-directed cognition. Neuroimage 53, 303–317 (2010). their insightful feedback. This research was supported in parts by the CHIST-ERA IVAN 46. Sadaghiani, S., Poline, J. -B., Kleinschmidt, A. & D’Esposito, M. Ongoing project (20CH21 174081), the Center for Biomedical Imaging (CIBM) of the Geneva- dynamics in large-scale functional connectivity predict perception. Proc. Natl Lausanne Universities and the EPFL, the Singapore MOE Tier 2 (MOE2014-T2-2-016), Acad. Sci. USA 112, 8463–8468 (2015). NUS Strategic Research (DPRT/944/09/14), NUS SOM Aspiration Fund 47. Wang, C., Ong, J. L., Patanaik, A., Zhou, J. & Chee, M. W. Spontaneous eyelid (R185000271720), Singapore NMRC (CBRG/ 0088/2015), NUS YIA, Singapore NRF closures link vigilance fluctuation with fmri dynamic connectivity states. Proc. fellowship (NRF-NRFF2017-06), NIH grant K99AG054573, and the Neuroimaging Natl Acad. Sci. 113, 9653–9658 (2016). Informatics and Analysis Center (1P30NS098577). The computational work was partially 48. Fornito, A., Harrison, B. J., Zalesky, A. & Simons, J. S. Competitive and performed on resources of the National Supercomputing Center, Singapore https://www. cooperative dynamics of large-scale brain functional networks supporting nscc.sg. Data were provided by the Human Connectome Project, WU-Minn Consortium recollection. Proc. Natl Acad. Sci. USA 109, 12788–12793 (2012). (Principal Investigators: David Van Essen and Kamil Ugurbil; 1U54MH091- 657) funded 49. Davison, E. N. et al. Brain network adaptability across task states. PLoS. by the 16 NIH Institutes and Centers that support the NIH Blueprint for Neuroscience Comput. Biol. 11, e1004029 (2015). Research; and by the McDonnell Center for Systems Neuroscience at Washington 50. Bertolero, M. A., Yeo, B. T. T. & D’Esposito, M. The modular and integrative University. functional architecture of the human brain. Proc. Natl Acad. Sci. USA 112, E6798–E6807 (2015). Author contributions 51. Yeo, B. T. T. et al. Functional specialization and flexibility in human R.L., C.O., T.G., M.R.S., and B.T.T.Y. designed research; R.L., J.L, and R.K. performed association cortex. Cereb. Cortex 25, 3654–3672 (2015). research; R.L., J.L., R.K., and T.G. contributed new reagents/analytic tools; R.L., J.L., T.G., 52. Demirtaş, M. et al. Hierarchical heterogeneity across human cortex shapes and B.T.T.Y. analyzed data; and R.L., D.V.D.V., T.G., M.R.S., and B.T.T.Y. wrote the paper. large-scale neural dynamics. Neuron 101, 1181–1194.e13 (2019). 8 NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 ARTICLE Additional information Open Access This article is licensed under a Creative Commons Supplementary Information accompanies this paper at https://doi.org/10.1038/s41467- Attribution 4.0 International License, which permits use, sharing, 019-10317-7. adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Competing interests: The authors declare no competing interests. Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless Reprints and permission information is available online at http://npg.nature.com/ indicated otherwise in a credit line to the material. If material is not included in the reprintsandpermissions/ article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from Journal peer review information: Nature Communications thanks Michael Breakspear, the copyright holder. To view a copy of this license, visit http://creativecommons.org/ Pablo Barttfeld, and the anonymous reviewer(s) for their contribution to the peer review licenses/by/4.0/. of this work. Peer reviewer reports are available. Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in © The Author(s) 2019 published maps and institutional affiliations. NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications 9 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Nature Communications Springer Journals

Resting brain dynamics at different timescales capture distinct aspects of human behavior

Download PDF
9 pages

Loading next page...
 
/lp/springer-journals/resting-brain-dynamics-at-different-timescales-capture-distinct-PYpvJAopgs

References (83)

Publisher
Springer Journals
Copyright
Copyright © 2019 by The Author(s)
Subject
Science, Humanities and Social Sciences, multidisciplinary; Science, Humanities and Social Sciences, multidisciplinary; Science, multidisciplinary
eISSN
2041-1723
DOI
10.1038/s41467-019-10317-7
Publisher site
See Article on Publisher Site

Abstract

ARTICLE https://doi.org/10.1038/s41467-019-10317-7 OPEN Resting brain dynamics at different timescales capture distinct aspects of human behavior 1,2,3 1 1 1 2,3 4,5 Raphaël Liégeois , Jingwei Li , Ru Kong , Csaba Orban , Dimitri Van De Ville , Tian Ge , 6 1,5,7,8 Mert R. Sabuncu & B.T. Thomas Yeo Linking human behavior to resting-state brain function is a central question in systems neuroscience. In particular, the functional timescales at which different types of behavioral factors are encoded remain largely unexplored. The behavioral counterparts of static func- tional connectivity (FC), at the resolution of several minutes, have been studied but beha- vioral correlates of dynamic measures of FC at the resolution of a few seconds remain unclear. Here, using resting-state fMRI and 58 phenotypic measures from the Human Connectome Project, we find that dynamic FC captures task-based phenotypes (e.g., pro- cessing speed or fluid intelligence scores), whereas self-reported measures (e.g., loneliness or life satisfaction) are equally well explained by static and dynamic FC. Furthermore, behaviorally relevant dynamic FC emerges from the interconnections across all resting-state networks, rather than within or between pairs of networks. Our findings shed new light on the timescales of cognitive processes involved in distinct facets of behavior. Department of Electrical and Computer Engineering, Clinical Imaging Research Centre, N.1 Institute for Health and Memory Networks Program, National University of Singapore, Singapore 117583, Singapore. Institute of Bioengineering, Centre for Neuroprosthetics, École Polytechnique Fédérale de Lausanne, 3 4 1015 Lausanne, Switzerland. Department of Radiology and Medical Informatics, University of Geneva, 1205 Geneva, Switzerland. Psychiatric & Neurodevelopmental Genetics Unit, Center for Genomic Medicine, Massachusetts General Hospital, Boston, MA 02114, USA. Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA 02129, USA. School of Electrical and Computer Engineering, Cornell University, 7 8 Ithaca, NY 14853, USA. Centre for Cognitive Neuroscience, Duke-NUS Medical School, Singapore 169857, Singapore. NUS Graduate School for Integrative Sciences and Engineering, National University of Singapore, Singapore 119077, Singapore. Correspondence and requests for materials should be addressed to R.L. (email: Raphael.Liegeois@epfl.ch) or to B.T.T.Y. (email: [email protected]) NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications 1 1234567890():,; ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 rain activity is highly organized in space and in time, even cognitive processes involved in the execution of various 1 30–32 in resting-state conditions . This intrinsic organization, tasks , thereby also supporting the ‘networked-brain’’ para- Bclassically evaluated from resting-state functional con- digm that has emerged in recent years . nectivity (FC), has been shown to encode various behavioral aspects such as integration of cognition and emotions , mon- Results 4 5,6 itoring of external environment , intellectual performance , and Behavioral counterparts of static and dynamic FC. We used emergence of stimulus-independent thoughts . FC has also been data from 419 unrelated HCP subjects to explore the extent to also used as a neuroimaging marker of several pathologies, which behavioral information is encoded in dynamic markers of 8,9 10 including Alzheimer’s disease , major depressive disorders , resting-state functional connectivity (FC), beyond classical static 11 12 13 Parkinson’s disease , schizophrenia , and autism . More measures of FC. We selected 58 behavioral measures from the recently, the advent of large neuroimaging and behavioral data- HCP dataset covering cognitive, social, emotion, and personality sets has allowed the further exploration of the FC behavioral traits (see Supplementary Table 2) from which age, gender, race, counterparts, showing intricate contributions of cognitive, emo- education, and motion (mean FD) were regressed. tional, social, and demographic aspects . FC markers were estimated from the HCP resting-state fMRI Importantly, all these studies use static measures of FC, which dataset. Classical preprocessing was performed, followed by a reflect the average functional organization of entire neuroimaging parcellation into 400 cortical regions of interest (ROIs) and recordings typically running over several minutes. However, there 35 19 subcortical ROIs . Subject-specific static FC markers were is recent converging evidence, suggesting that the resting brain computed by averaging correlation matrices of fMRI time series navigates through different functional connectivity configurations across runs. Dynamic FC markers were defined from an AR-1 15,16 at much faster timescales on the order of seconds . Therefore, model identified from the concatenation of the runs for each new dynamic measures of resting-state FC exploiting these faster subject (Methods). We chose to represent FC dynamics using an changes have been proposed and their behavioral counterparts AR-1 model for several reasons. First, we have shown recently have in turn been explored, showing links to cognitive flex- that AR-1 models, by exploiting the statistical link between 18 19 20 ibility , drug use , and mind-wandering . Yet, the comparison successive time points, capture FC dynamics significantly better of static and dynamic measures of FC has only been proposed in than a hidden Markov model explicitly representing switches specific applications such as temporal lobe epilepsy or the between different states with an equivalent number of para- description of eating behaviors , and a comprehensive analysis meters . Second, the hierarchical organization of brain network exploring which types of behavioral features emerge from func- dynamics was found to be reproduced by an AR-1 model of fMRI tional interactions at different timescales is missing. time series . Finally, lag threads, which also exploit the We explore this question using a discovery dataset (N = 419) sequential ordering information of time series (although they and a replication dataset (N = 328), comprising high-resolution focus on identifying temporal sequences of propagated activity resting-state functional magnetic resonance imaging (fMRI) time rather than connectivity patterns) were shown to provide series and 58 behavioral measures spanning cognitive, emotional, meaningful markers of intrinsic brain function . social, and personality traits from the Human Connectome The link between FC markers and behavioral measures was Project (HCP). First, we compare the extent to which static and 28,38 studied using a variance component model . The model dynamic FC capture behavioral information. We then investigate inputs are (i) a matrix containing the 58 behavioral measures for whether the behavioral relevance of FC markers is preferentially the N = 419 subjects and (ii) at least one N × N matrix, called a encoded in within- or between- network connectivity. Finally, we similarity matrix and denoted by K, whose i,j-th entry encodes test if static and dynamic FC capture complementary behavioral the similarity between (static or dynamic) FC of subjects i and j. information. FC dynamics are evaluated using a first-order Note that static FC matrices are symmetric, whereas dynamic FC 24,25 autoregressive (AR-1) model of resting-state fMRI data . AR-1 matrices are non-symmetric. The model estimates the level of models exploit the temporal ordering of fMRI time series to behavioral variability that is explained by FC variability, both on capture dynamic FC happening at a resolution of a few seconds to average over all behavioral measures, as well as for each which static approaches are blind , without suffering from the behavioral measure (Methods ). 26,27 limitations of classical sliding window methods . FC dynamics are then linked to the 58 behavioral measures using a variance Dynamic FC markers encode more behavioral information.We component model . This model has been extensively used in first compared the level of behavioral variance explained by static genome-wide complex trait analyses and was recently applied to and dynamic FC markers. To this end, we ran the multivariate study the neuroanatomical signatures of traits such as cognitive 29 variance component model twice: once using a similarity matrix or clinical measures . encoding the inter-subject similarity of static FC patterns, and We find that FC dynamics specifically encode behavioral once using similarity of dynamic FC patterns. measures evaluating performance in tasks, whereas self-reported Figure 1a shows that on average over the 58 behavioral measures are explained equally well by static and dynamic FC. measures, dynamic FC markers capture more behavioral variance We argue that this reflects the nature of the functional processes −4 than static FC (p = 8.31 × 10 ; two-tailed t-test), and Fig. 1b involved in the corresponding behavioral experiments. On the presents the results for eight individual phenotypic measures. one hand, task-based metrics engage cognitive processes at Results for the 50 remaining HCP measures are found in timescales on the order of a few seconds that can be captured by Supplementary Fig. 1. FC dynamics. On the other hand, self-reported measures might reflect trait-like properties that are less likely to change over a few seconds, therefore being equally well explained by functional Dynamic FC specifically encodes task-based measures. Even if patterns averaged over longer periods as encoded in static FC. dynamic FC encodes more behavioral information than static FC Furthermore, our results also suggest that task-performance on average, results of Fig. 1b show that some behavioral measures scores are defined by whole-brain FC dynamics involving the are not better explained by dynamic FC (e.g., Meaning of Life, interaction between multiple resting-state networks. Overall, Loneliness or Perceived Stress). In order to explore whether FC more than providing a mere statistical marker of task-perfor- dynamics specifically capture certain types of behavioral mea- mance, these findings offer new insights into the timescales of the sures, we ranked the 58 HCP measures based on the extent to 2 NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 ARTICLE ab Static Dynamic 0.9 0.5 0.8 0.4 0.7 0.6 0.3 0.5 0.4 0.2 0.3 0.2 0.1 0.1 0 0 Fig. 1 Dynamic FC explains more behavioral variance than static FC. a On average over 58 behavioral measures, dynamic FC (blue, 37%) explains more −4 behavioral variance than static FC (red, 19%) (p = 8.31 × 10 ; two-tailed t-test). b Variance explained for eight representative measures. Here, static FC utilizes Pearson’s correlation, while dynamic FC utilizes the coefficient matrix of a first-order autoregressive model. Error bars indicate standard deviation (SD) of the estimates a b 0.3 Static > Dynamic Dynamic > Static 0.2 T 0 0.1 –2 Static Dyn. 0.5 0.4 0.3 0.2 0.1 Self-reported Performance in task Unclassified Static Dyn. Fig. 2 Dynamic FC explains larger behavioral variance than static FC in task-performance measures. a Behavioral measures are ordered based on whether dynamic FC explains more variance than static FC. A positive t-statistic T suggests that dynamic FC explains more variance than static FC. Behavioral measures corresponding to task-performance are marked with a green dot and self-reported measures are marked with an orange dot. b No statistically significant difference (p > 0.10: two-tailed t-test) was found in the mean variance explained by static and dynamic FC in self-reported measures. c −3 Measures of performance in task are on average significantly better explained (p = 1.75 × 10 ; two-tailed t-test) by dynamic FC. Error bars indicate SD of the estimates −3 which dynamic FC better explain their variability, as compared to explained task-performance measures (p = 1.75 × 10 , Fig. 2c), static FC. To this end, we repeated the procedure for the 58 whereas no statistically significant difference could be found in measures and computed 58 t-statistics, denoted by T, of the dif- the capacity of both markers to explain self-reported measures ference between behavioral variance explained by static and (Fig. 2b). We also find that the difference of the differences dynamic FC for each measure (Supplementary Methods). Nega- between static and dynamic explained variances observed in −3 tive values of t-statistics indicate that static FC tends to better Fig. 2b, c is itself different from zero (p = 3.62 × 10 ; two-tailed explain the measure, whereas behavioral measures with positive t-test). This interaction effect confirms that the difference statistics are better explained by dynamic FC, as indicated in observed in Fig. 2c is related to the task condition and not only Fig. 2a. driven by the main effect shown in Fig. 1a. Moreover, the result of This ranking seems to draw a dichotomy between “task- Fig. 2c is reproduced using subcategories of task-based measures performance” and “self-reported” measures. On the one hand, the (Supplementary Fig. 3). Overall, the better average capacity of first category includes metrics that use participant’s performance dynamic FC to explain behavioral measures seems to be driven by in a task to assess a trait (e.g., working memory, spatial its increased capacity to explain task-based measures. orientation) and are marked with green dots in Fig. 2a. On the other hand, “self-reported” measures (orange dots in Fig. 2a) rely Behavior-related FC dynamics arise from network interactions. on subjective appraisal of traits (e.g., loneliness, life satisfaction). Functional interactions between brain networks have been shown No label was attached to the measures with no clear classification 39 to play a key role during the execution of tasks and in the in one of these categories. We find that dynamic FC better description of traits . We tested whether interaction between NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications 3 Static Dynamic Fluid intelligence Sustained attention-spec. Working memory (N-back) Processing speed Emotion face matching Meaning of life Loneliness Perceived stress Openness (NEO) Spatial orientation Perceived rejection Sleep quality Friendship Sustained attention—sens. Social cognition—interaction Meaning of life Loneliness Relational processing Working memory (list sorting) Anger—aggressiveness Grip strength Life satisfaction Emotional support Taste intensity Perceived stress Vocabulary (picture matching) Agreeableness (NEO) Extroversion (NEO) Emotion recog.—happiness Instrumental support Odor identification Positive affect Walking speed Contrast sensitivity Processing speed Sadness Perceived hostility Neuroticism (NEO) Visual episodic memory Emotion recog.—total Self-efficacy Conscientiousness (NEO) Emotion recog.—sadness Pain interference survey Verbal episodic memory Anger—affect Vocabulary (pronunciation) Story comprehension Arithmetic Emotion recog.—anger Emotion recog.—neutral Fear-affect Inhibition (Flanker task) Emotion recog.—fear Sustained attention—spec. Fluid intelligence Walking endurance Anger—hostility Dexterity Cognitive status (MMSE) Cognitive flexibility Fear—somatic arousal Emotion face matching Working memory (N-back) Social cognition—random Delay discounting Mean variance explained Variance explained Mean variance Mean variance in self-report in task ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 a b Within networks 0.1 0.10 Vis 0.08 0.1 SM 0.06 0.1 0.04 D-Att 0.02 0.1 Sal 0.1 Between networks Lim 0.10 0.1 FP 0.08 0.1 0.06 DMN 0.04 0.1 Sub 0.02 Vis SM D-Att Sal Lim FP DMN Sub Static Dynamic Fig. 3 Dynamic FC does not explain more behavioral variance than static FC within (pairs of) networks. a Behavioral variance explained by within-network (shaded diagrams) and between-network (unshaded diagrams), network static and dynamic FC. Seven cortical networks were used: visual (VIS), somatomotor (SM), dorsal attention (D-Att), salience (Sal), limbic (Lim), frontoparietal (FP), default mode network (DMN) and we also gathered the 19 subcortical areas (Sub). b There is no statistically significant difference in behavioral variance explained by within-network static and dynamic FC. c −3 Between-network static FC explains more behavioral variance than between-network dynamic FC (p = 8.31 × 10 ; two-tailed t-test). Error bars indicate SD of the estimates resting-state networks were also critical for extracting behavioral The average behavioral variance explained by combining static information from FC. To this end, the same model as described and dynamic FC is shown in dark blue in Fig. 4a. Results for eight above was used but similarity matrices were not computed from representative measures are shown in Fig. 4b and results for the the whole static or dynamic FC matrices. Instead, only sub-blocks 50 remaining traits are found in Supplementary Fig. 2. In Fig. 4a, of the FC matrices corresponding to (pairs of) well-known rest- the combined value is significantly higher than static FC (p = −4 ing-state networks were used. In other words, we tested how 4.73 × 10 ; two-tailed t-test), confirming the fact that FC behavioral variability is encoded in the variability of (pairs of) dynamics contains information above and beyond static FC. resting-state networks connectivity patterns. We used a common However, no statistical difference was found between average partition in seven cortical resting-state networks and included combined results and dynamic FC (p > 0.10, see Supplementary subcortical areas (Methods), as shown in Fig. 3. Table 1 for details), which suggests that the information encoded The average behavioral variance explained by static and by static FC is largely encoded in dynamic FC. dynamic FC restricted to within or between networks is under 10% for almost all the pairs of networks. Not surprisingly this is Dynamic FC interactions driving task-performance. We now lower than the behavioral variance explained from the whole- explore which dynamic FC interactions contribute to the overall brain connectivity patterns (19% for static FC and 37% for association with task-performance (Fig. 2c). We used a refor- dynamic FC), as anticipated from previous findings showing that mulation of the variance component model defined in Eq. (2) that individual FC fingerprinting is distributed throughout the revealed the relative contribution of the interaction between each brain . More unexpected is the fact that FC dynamics do not pair of (sub)networks to the overall explained variance (Supple- seem to carry more behavioral information than static FC. On the mentary Eq. (10)). The results are shown in Fig. 5. It can be seen contrary, on average over all inter-network connections (Fig. 3, that default C and frontoparietal C, together with the subcortical unshaded diagrams), static FC explained more behavioral regions, are contributing the most to the association between −3 variance than dynamic FC (p = 8.31 × 10 ; two-tailed t-test), dynamic FC and task-performance. whereas no statistically significant difference was found for within-networks connections (Fig. 3, shaded diagrams). Replication dataset. The findings shown in Figs. 1–4 were replicated in a second group of 328 unrelated HCP subjects. More precisely, all significant differences found in Figs. 1–4 were also Testing complementarity between static and dynamic FC.We found to be significant in the replication dataset (more details are have shown that on average FC dynamics encode more behavioral found in Supplementary Figs. 5–9 and Supplementary Table 1). information than static FC (Fig. 1), especially for task- The replication dataset was composed of the second subject of performance measures (Fig. 2). However, this does not mean each HCP family containing more than one person. We note that that static FC is not capturing any additional behavioral infor- it is therefore not completely independent from the discovery mation not encoded by dynamic FC. To test this, we used a dataset. generalized version of the multivariate variance component model that takes multiple similarity matrices -in our case two: the Additional control analyses. We performed a series of control ones computed from static and dynamic FC- as inputs and esti- analyses to evaluate the impact of various processing steps in our mates the level of behavioral variance explained by the combi- baseline analysis. More specifically, we tested the impact of (i) nation of these similarity matrices (Supplementary Methods). including the variance of the mean cortical grayordinate signal as 4 NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications Static Dynamic Mean var. expl. Mean var. expl. c NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 ARTICLE Static Dynamic Combined ab 0.5 1 0.9 0.4 0.8 0.7 0.6 0.3 0.5 0.4 0.2 0.3 0.2 0.1 0.1 Fig. 4 Combined static and dynamic FC does not capture more behavioral variance than dynamic FC alone. a Average variance explained across 58 behavioral measures using static FC (red), dynamic FC (light blue), and the combination of these two (dark blue). b Variance explained for eight representative measures. Error bars indicate SD of the estimates functional connectivity (FC), and recent evidence has shown that exploiting the dynamical properties of FC instead of the classical B A static FC metrics could open new avenues to interpret brain A B functioning at different timescales. In this study, we aim at refining our understanding of the behavioral information carried by FC dynamics. To this end, we explored the extent to which resting-state static and dynamic FC measures relate to a large repertoire of measures covering cognitive, social, emotion, and personality traits. We first show that, on average over 58 selected behavioral measures, FC dynamics encode significantly more behavioral information than a common static FC metric. This confirms current findings that have highlighted the advantage of resting-state dynamic FC measures over their static counterparts 43 21 22 A in describing mindfulness , disease , and eating behaviors . Interestingly, FC dynamics within well-known resting-state networks, or between pairs of networks, did not capture more A C behavioral information than static FC in the same networks (Fig. 3). For example, static inter-network FC was shown to explain more behavioral variance than dynamic inter-network FC Fig. 5 Dynamic FC interactions contributing the most to the association on average. These results might seem counter-intuitive at first with task-performance. Networks and corresponding colors are the same as sight and suggest that similarity measures derived from local in Fig. 3, and subnetworks are defined following the 17-network parcellation 35 patterns of FC do not complement each other in the same way in of Schaefer et al. , as reported in Supplementary Fig. 4. The colors of the the static and dynamic cases. In other words, the advantage of edges are defined by their destination and only connections surviving an dynamic FC in explaining the behavioral information observed in FDR correction at the level q= 0.05 are shown Fig. 1 is encoded in the global dynamic FC interaction patterns. From a methodological point of view, this also indicates that even a covariate in the variance component model, (ii) evaluating the if dynamic FC uses richer statistical information than static FC by static and dynamic FC matrices from fMRI time series from relaxing the static assumption associated to this metric, dynamic which the mean cortical grayordinate signal was not regressed, FC should not a priori and automatically be considered as a better (iii) including head motion metrics as covariates in the variance neuroimaging marker than static FC. component model, (iv) evaluating the static and dynamic FC In Fig. 2, we show that FC dynamics specifically encode matrices from full (i.e., uncensored) fMRI time series, (v) the measures of performance in a task, such as working memory number of behavioral measures considered in the variance tasks, whereas static and dynamic FC explain self-reported component model, and (vi) the relative contributions of static and measures, such as the perception of loneliness, equally well. This dynamic FC to the overall variance explained within the com- additional information is found to be encoded in the global bined variance component model. The variance component dynamic FC patterns, and not confined to single areas or net- model appeared to be robust to these changes and in each case, works (Fig. 3). More precisely, Fig. 5 suggests that the default our main findings were reproduced (Supplementary Figs. 10–12). mode and frontoparietal networks drive the integration of the dynamic FC coming from other networks. This is in line with previous findings identifying these areas as hubs of the dynamic Discussion functional connectome , and further supports the importance of Exploring how resting-state functional organization is linked to coupled default network and frontoparietal activities during task- various behavioral traits is a central neuroimaging research performance . question. This organization is classically evaluated from NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications 5 Static Dynamic Combined Fluid intelligence Sustained attention—spec. Working memory (N-back) Processing speed Emotion face matching Meaning of life Loneliness Perceived stress Mean variance explained Variance explained S ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 Taken together, these results support the line of current find- Table 2). However, we note that classifying these measures in one ings indicating that different phenotypic measures such as pain , or the other category did not significantly change the results in 46 47 perception , and vigilance are encoded in dynamic interactions Fig. 2. That said, considering only two categories of behavioral between multiple areas and not within single networks. Then, measures disregards the multifactorial nature of behavior and the even if they concern interactions between resting-state networks, repertoire of behavioral measures could be approached using our results interestingly echo the nature of network interactions other classification criteria such as trait vs. state or intrinsic vs. involved in task-based conditions. Indeed, the execution of tasks extrinsic . relies on a coordinated activation of different networks at faster We referred to the AR model of BOLD time series as ‘dynamic’’ 31,32 63 timescales that can be captured by dynamic FC , and to which following the systems theory literature . This nomenclature is static measures of FC are blind. On the contrary, self-reported motivated by the fact that such models, by accounting for the measures could be considered as trait-like properties that might memory present in the time series (i.e., x depends on x ), are t t−1 therefore be explained equally well by average FC patterns able to reproduce empirical fluctuations in the multivariate time encoded in static FC. Converging results also suggest that simple series of interest much better than memoryless (or ‘static’’, fol- tasks exhibit segregated activation patterns whereas complex tasks lowing the same nomenclature) models . The AR model is also involving multiple cognitive processes (e.g., working memory or used to compute the dominant dynamic modes shaping resting- visuospatial attention) require an integrated activation of multiple state brain function . Overall, this model can be seen as a 39,48,49 intrinsic networks and activate flexible brain regions such compact way to summarize the temporal fluctuations of BOLD 50,51 as connector hubs . It does not seem unreasonable to assume and FC time series that are directly exploited by time-varying 17,65 that task-performance measures, in which subjects are incited to models . reach a high score in a test that often involves multiple or In summary, static measures of FC provide a measure of brain coordinated actions, capture more complex behavioral traits -as function averaged over several minutes. This is an over- defined above- than self-reported measures. More generally, these simplification and new dynamic measures capturing the temporal results constitute empirical evidence supporting the ‘networked- changes of brain function on the order of a few seconds have been brain’’ paradigm that has emerged in recent years. This paradigm proposed. While these new measures were shown to capture more essentially views the brain as a multiscale network producing statistical properties of fMRI data, their behavioral relevance complex spatio-temporal activity patterns rather than an above and beyond static FC remains unclear. Here, we have ensemble of neuronal populations with localized shown using 747 HCP subjects and 58 behavioral measures that 33,52,53 specification . FC dynamics specifically capture measures of performance in The fact that FC dynamics seem to capture more complex tasks by leveraging the dynamic information encoded in multiple- network interactions than static FC is also supported by the links network interactions. On the contrary, self-reported measures are that have been drawn between FC and the underlying brain equally well explained by static and dynamic measures of FC. anatomy, or structural connectivity (SC). As static FC is on Overall, we believe our work opens up future possibilities to a average closer to SC, the dynamic FC repertoire also captures better characterization of the cognitive processes shaping the excursions from SC that are characterized by higher efficiency various facets of human behavior. and lower modularity . These fluctuations of modular organi- zation, encoded in FC dynamics, were shown to operate at dif- Methods ferent timescales and to support the periodic (de)coupling of Data and preprocessing. We used data of the HCP 1200-subjects release com- 55 prising structural MRI, resting-state functional MRI, and behavioral measures of resting-state networks , possibly constituting a signature young (ages 22–35) and healthy participants drawn from a population of siblings . of consciousness and allowing for a more efficient transfer of All imaging data were acquired on a 3-T Siemens Skyra scanner using a multi-band neuronal information . sequence. Functional images have a temporal resolution of 0.72 s and a 2-mm Other converging findings suggest that the amplitude of FC isotropic spatial resolution whereas structural images are 0.7-mm isotropic. For each subject, four 14.4 min runs (1200 frames) of functional time series were dynamics variability decreases during task as compared to acquired . Resting-state fMRI data was projected to the fs_LR surface space using 58,59 resting-state . Altogether with the results of Figs. 2 and 3 23,66 the multimodal surface matching method (MSM-All . Both cortical and sub- indicating that task-based behavioral measures are specifically 67,68 cortical data were cleaned using the ICA-FIX method and saved in CIFTI encoded by resting-state FC dynamics, we might hypothesize that grayordinate format. This cleaning procedure included the regression of 24 motion-related parameters (six classical motion parameters, their derivatives, and there exists a resting-state ‘dynamic reservoir’’ that is recruited the squares of these 12 parameters). Motion censoring was then applied by when leaving rest and, which defines task-performance. This removing frames with FD > 0.2 mm or DVARS > 75, as well as one frame before dynamic reservoir, encoded by FC dynamics and not by simpler 69,70 and two frames after these frames . Remaining segments containing less than static FC markers as shown in Fig. 4, emerges from highly inte- five frames were also removed and runs with >50% of censored frames were discarded. Linear trends and mean cortical grayordinate signal were regressed and grated connections involving multiple intrinsic networks. As censored frames were ignored to compute regression coefficients. Mean cortical such, we could also interpret this dynamic connectivity structure grayordinate signal regression was performed because this step was shown recently as a signature of the human connectome evolution that tends to to strengthen the association between FC metrics and behavioral measures . attain efficient organizations to perform complex tasks . However, we note that not regressing mean cortical grayordinate signal yielded The dichotomy of behavioral measures proposed in Fig. 2 is similar conclusions (Supplementary Fig. 10). Finally, fMRI time series were par- cellated into 419 regions of interest (ROIs) comprising 400 cortical areas and motivated by the statistical difference between static and dynamic 19 subcortical areas defined in Freesurfer. Static functional connectivity was FC that seems to capture distinct behavioral properties. This obtained for each subject based on the pearson’s correlation matrices, computed dichotomy presents some limitations. First, the classification of from uncensored frames for each run, which were Fisher z-transformed, averaged behavioral measures in one of the two proposed categories, ‘task- over runs and transformed back to r-space. Dynamic functional connectivity measures were estimated from the model parameter of a first-order autoregressive performance’’ or ‘self-reported’’, was not always straightforward. representation of fMRI time series: For example, the Delay Discounting task measure was left unclassified as one could argue that it is not capturing a perfor- x ¼ A  x þ ϵ ð1Þ t t1 t mance, associated with an underlying truth or optimal score, in N ´ 1 where x 2R represents the fMRI time series in the N = 419 ROIs at time t, the same way a classical task such as the Working Memory task N ´ N R R A 2R is the model parameter that encodes the linear relationship between is. For the same reason, other measures (grip strength, odor N ´ 1 successive time points, and ε 2R are the residuals of the model . The model identification, walking speed, contrast sensitivity, taste intensity, parameter A was identified from the concatenation of the uncensored sections of and walking endurance) were not classified (Supplementary the different runs, while ignoring transitions between uncensored sections and 6 NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 ARTICLE transitions between runs. The proportion of variance explained by this model, mean difference of explained variance is: jjCovðε Þjj t F 2 defined as R ¼ 1  analogous to the definition of R for univariate AR jjCovðx Þjj X t F N  1 ^ ^ ^ V ¼ M  M : ð5Þ models, and where ||⋅|| denotes the Frobenius norm , is 69.3% ± 11.2% (com- jack i jack i¼1 puted over all subjects, including the replication dataset). qffiffiffiffiffiffiffiffiffi We selected 58 behavioral measures, including cognitive, social, emotion and ^ ^ With a large sample size (N = 419), the estimator M  M = V is jack jack personality traits (Supplementary Table 2). These measures consist of metrics from assumed to follow a standard normal distribution under the null hypothesis (M = the NIH Toolbox and some well-known non-NIH measures (e.g., NEO-FFI). 0) and a two-tailed p-value can be computed. All significant results survived FDR Details about behavioral measures can be found in HCP S1200 Data Dictionary correction at q < 0.05 (see Supplementary Table 1 for details). and Barch et al. . These measures were classified either as ‘task-performance’’, ‘self-reported’’,or ‘unclassified’’ if no clear belonging to one of the first two categories applied (Supplementary Table 2). Reporting summary. Further information on research design is available in Among the HCP 1200-subjects release, 1029 subjects had at least one run that the Nature Research Reporting Summary linked to this article. was not discarded after applying the preprocessing rules. Excluding the subjects with missing or problematic entries for some behavioral measures further reduced Data availability the dataset to 953 subjects belonging to 419 families. To avoid the influence of The HCP data is publicly available at http://www.humanconnectomeproject.org/data/; shared genetic and environmental factors, we kept the first subject from each informed consent was obtained from all HCP participants . family leading to a final set of N = 419 unrelated subjects. Among the 419 families used for the initial set, 91 ‘families’’ contained only one subject and hence these families were discarded in the construction of the replication dataset that contained Code availability 328 subjects. All code is publicly available at https://github.com/RaphaelLiegeois/FC-Behavior/. Variance component model. We regressed age, gender, race, education, and Received: 14 November 2018 Accepted: 3 May 2019 motion (mean FD) from the 58 phenotypic measures, which were then quantile normalized. We used the multivariate variance component model developed by Ge et al. to link FC markers and behavioral measures: Y ¼ C þ E ð2Þ where Y, C, and E are 419 × 58 matrices. Y contains the 58 processed behavioral References measures for all 419 subjects. VecðCÞ NðÞ 0; Σ  F and VecðEÞ NðÞ 0; Σ  I , c e 1. Damoiseaux, J. et al. Consistent resting-state networks across healthy subjects. where Vec(.) is the matrix vectorization operator, ⊗ is the Kronecker product of Proc. Natl Acad. Sci. USA 103, 13848–13853 (2006). matrices, and I is the identity matrix. F is a similarity matrix such that F(i, j) 2. Friston, K. J. Functional and effective connectivity: a review. Brain Connect. 1, encodes the (static or dynamic) FC similarity between subjects i and j, and is 13–36 (2011). defined as the correlation between the static FC (or dynamic FC) matrices of the 3. Greicius, M. D., Krasnow, B., Reiss, A. L. & Menon, V. Functional connectivity two subjects. Σ and Σ are unknown 58 × 58 matrices to be estimated from F and c e in the resting brain: a network analysis of the default mode hypothesis. Proc. Y. The variance explained by (static or dynamic) FC markers, denoted by M,is Natl Acad. Sci. USA 100, 253–258 (2003). computed as: 4. Gusnard, D. A. & Raichle, M. E. Searching for a baseline: functional imaging and the resting human brain. Nat. Rev. Neurosci. 2, 685–694 (2001). TrðΣ Þ M ¼ ; ð3Þ 5. Song, M. et al. Brain spontaneous functional connectivity and intelligence. TrðΣ Þþ TrðΣ Þ c e Neuroimage 41, 1168–1176 (2008). 6. van den Heuvel, M. P., Stam, C. J., Kahn, R. S. & Hulshoff Pol, H. E. Efficiency where Tr(.) is the trace operator. M measures how much inter-subject behavioral of functional brain networks and intellectual performance. J. Neurosci. 29, variability is explained by inter-subject FC variability . Equation (3) is computed 7619–7624 (2009). twice, once using static FC to build the FC similarity matrix in (2) and once using dynamic FC to build this matrix. M for a single behavioral measure is given by 7. Mason, M. F. et al. Wandering minds: the default network and stimulus- independent thought. Science 315, 393–395 (2007). M = Σ (i, i)/(Σ (i, i) + Σ (i, i)). We used the variance component model because it i c c e provides an easy way to summarize high-dimensional FC patterns and produces 8. Wang, K. et al. Altered functional connectivity in early alzheimer’s disease: one scalar (variance explained) for each behavioral measure that is normalized a resting-state fmri study. Hum. Brain. Mapp. 28, 967–978 (2007). (between 0 and 1) and directly comparable across measures and processing 9. Meskaldji, D.-E. et al. Prediction of long-term memory scores in mci based on methods. resting-state fmri. Neuroimage Clin. 12, 785–795 (2016). 10. Drysdale, A. T. et al. Resting-state connectivity biomarkers define neurophysiological subtypes of depression. Nat. Med. 23,28–38 (2017). Complexity of static and dynamic variance component models. Static FC 11. Gorges, M. et al. To rise and to fall: functional connectivity in cognitively similarity matrices are built from the correlation matrix of resting-state fMRI time normal and cognitively impaired patients with parkinson’s disease. Neurobiol. series. These correlation matrices are symmetric and of size N × N , where ROI ROI Aging 36, 1727–1735 (2015). N is the number of regions of interest, and hence their degrees of freedom ROI 12. Bassett, D. S., Nelson, B. G., Mueller, B. A., Camchong, J. & Lim, K. O. Altered (d.o.f.) is N ×(N − 1)/2. On the other hand, dynamic FC is built from the ROI ROI resting state complexity in schizophrenia. Neuroimage 59, 2196–2207 (2012). non-symmetric coefficient matrix of a first-order AR model of size N × N , ROI ROI 13. Anderson, J. S. et al. Functional connectivity magnetic resonance imaging and hence their d.o.f. is N . It should be noted, however, that this difference in d. ROI classification of autism. Brain 134, 3742–3754 (2011). o.f. is not present in the static and dynamic variance component models as the 14. Smith, S. M. et al. A positive-negative mode of population covariation links static and dynamic FC matrices are only used in intermediate steps of the com- brain connectivity, demographics and behavior. Nat. Neurosci. 18, 1565–1567 putation of the similarity matrices of the variance component model. These (2015). similarity matrices are in both the static and dynamic case symmetric and of size N with N ×(N − 1)/2 degrees of freedom, where N is the number of subjects. 15. Abrol, A. et al. Replicability of time-varying connectivity patterns in large resting state fmri samples. Neuroimage 163, 160–176 (2017). Therefore, there is no complexity bias giving an advantage to one or the other approach. 16. Liégeois, R., Laumann, T. O., Snyder, A. Z., Zhou, J. & Yeo, B. T. T. Interpreting temporal fluctuations in resting-state functional connectivity MRI. Neuroimage 163, 437–455 (2017). Statistic inference. We use the delete-1 Jackknife approach to evaluate the 17. Preti, M. G., Bolton, T. A. & Van De Ville, D. The dynamic functional statistical difference between the variance explained by two different methods (e.g., connectome: State-of-the-art and perspectives. Neuroimage 160,41–54 A B ^ ^ when M is computed using only static FC vs. only dynamic FC). Let M and M be (2017). the estimates of behavioral variance explained using the methods A and B to be 18. Chen, T., Cai, W., Ryali, S., Supekar, K. & Menon, V. Distinct global brain compared. Denoting by M the estimates computed from the whole dataset dynamics and spatiotemporal organization of the salience network. PLoS Biol. without subject i, the delete-1 Jackknife estimate of the mean difference between 14, e1002469 (2016). methods A and B is: 19. Vergara, V. M., Weiland, B. J., Hutchison, K. E. & Calhoun, V. D. The impact of combinations of alcohol, nicotine, and cannabis on dynamic brain N N X X 1 1 A B ^ ^ ^ ^ M ¼ M  M ¼ M ; ð4Þ connectivity. Neuropsychopharmacology 43, 877–890 (2018). jack i i i N N i¼1 i¼1 20. Christoff, K., Irving, Z. C., Fox, K. C., Spreng, R. N. & Andrews-Hanna, J. R. Mind-wandering as spontaneous thought: a dynamic framework. Nat. Rev. A B ^ ^ ^ where N = 419 is the number of subjects, and M ¼ M  M is the difference i i i Neurosci. 17, 718–731 (2016). of explained variance for the i-th Jackknife sample. The variance estimate of the NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 21. Chiang, S. et al. Temporal and spectral characteristics of dynamic functional 53. Wang, P. et al. Inversion of a large-scale circuit model reveals a cortical connectivity between resting-state networks reveal information beyond static hierarchy in the dynamic resting human brain. Sci. Adv. 5, eaat7854 connectivity. PLoS ONE 13, e0190220 (2018). (2019). 22. Park, B.-Y., Moon, T. & Park, H. Dynamic functional connectivity analysis 54. Liégeois, R. et al. Cerebral functional connectivity periodically (de) reveals improved association between brain networks and eating behaviors synchronizes with anatomical constraints. Brain Struct. Funct. 221, 2985–2997 compared to static analysis. Behav. Brain. Res. 337, 114–121 (2018). (2016). 23. Van Essen, D. C. et al. The WU-Minn human connectome project: an 55. Betzel, R. F., Fukushima, M., He, Y., Zuo, X. -N. & Sporns, O. Dynamic overview. Neuroimage 80,62–79 (2013). fluctuations coincide with periods of high and low modularity in resting-state 24. Valdés-Sosa, P. A. et al. Estimating brain functional connectivity with sparse functional brain networks. Neuroimage 127, 287–297 (2016). multivariate autoregression. Philos. Trans. R. Soc. Lond. B. Biol. Sci. 360, 56. Barttfeld, P. et al. Signature of consciousness in the dynamics of resting-state 969–981 (2005). brain activity. Proc. Natl Acad. Sci. USA 112, 887–892 (2015). 25. Rogers, B. P., Katwal, S. B., Morgan, V. L., Asplund, C. L. & Gore, J. C. 57. Zalesky, A., Fornito, A., Cocchi, L., Gollo, L. L. & Breakspear, M. Time- Functional mri and multivariate autoregressive models. Magn. Reson. Imaging resolved resting-state brain networks. Proc. Natl Acad. Sci. USA 111, 28, 1058–1065 (2010). 10341–10346 (2014). 26. Leonardi, N. & Van De Ville, D. On spurious and real fluctuations of dynamic 58. Elton, A. & Gao, W. Task-related modulation of functional connectivity functional connectivity during rest. Neuroimage 104, 430–436 (2015). variability and its behavioral correlations. Hum. Brain. Mapp. 36, 3260–3272 27. Hindriks, R. et al. Can sliding-window correlations reveal dynamic functional (2015). connectivity in resting-state fmri? Neuroimage 127, 242–256 (2016). 59. Hutchison, R. M. & Morton, J. B. Tracking the brain’s functional coupling 28. Yang, J., Lee, S. H., Goddard, M. E. & Visscher, P. M. Gcta: a tool for genome- dynamics over development. J. Neurosci. 35, 6849–6859 (2015). wide complex trait analysis. Am. J. Hum. Genet. 88,76–82 (2011). 60. van den Heuvel, M. P., Bullmore, E. T. & Sporns, O. Comparative 29. Sabuncu, M. R. et al. Morphometricity as a measure of the neuroanatomical connectomics. Trends Cogn. Sci. 20, 345–361 (2016). signature of a trait. Proc. Natl Acad. Sci. USA 113, E5749–E5756 (2016). 61. Grös, D. F., Antony, M. M., Simms, L. J. & McCabe, R. E. Psychometric 30. Bassett, D. S. et al. Dynamic reconfiguration of human brain networks during properties of the state-trait inventory for cognitive and somatic anxiety learning. Proc. Natl Acad. Sci. USA 108, 7641–7646 (2011). (sticsa): comparison to the state-trait anxiety inventory (stai). Psychol. Assess. 31. Cocchi, L., Zalesky, A., Fornito, A. & Mattingley, J. B. Dynamic cooperation 19, 369–381 (2007). and competition between brain systems during cognitive control. Trends 62. DeLeeuw, K. E. & Mayer, R. E. A comparison of three measures of cognitive Cogn. Sci. 17, 493–501 (2013). load: Evidence for separable measures of intrinsic, extraneous, and germane 32. Cohen, J. R. The behavioral and cognitive relevance of time-varying, dynamic load. J. Educ. Psychol. 100, 223–234 (2008). changes in functional connectivity. Neuroimage 180, 515–525 (2018). 63. Oppenheim, A. & Willsky, A. S. Signals and Systems (Prentice Hall, 1997). 33. Betzel, R. F. & Bassett, D. S. Multi-scale brain networks. Neuroimage 160, 64. Casorso, J. et al. Dynamic mode decomposition of resting-state and task fmri. 73–83 (2017). Neuroimage 194,42–54 (2019). 34. Smith, S. M. et al. Resting-state fmri in the human connectome project. 65. Hutchison, R. M. et al. Dynamic functional connectivity: promise, issues, and Neuroimage 80, 144–168 (2013). interpretations. Neuroimage 80, 360–378 (2013). 35. Schaefer, A. et al. Local-global parcellation of the human cerebral cortex from 66. Robinson, E. C. et al. Multimodal surface matching: fast and generalisable intrinsic functional connectivity mri. Cereb. Cortex 28, 3095–3114 (2018). cortical registration using discrete optimisation. Inf. Process. Med. Imaging 23, 36. Vidaurre, D., Smith, S. M. & Woolrich, M. W. Brain network dynamics are 475–486 (2013). hierarchically organized in time. Proc. Natl Acad. Sci. USA 114, 12827–12832 67. Salimi-Khorshidi, G. et al. Automatic denoising of functional mri data: (2017). combining independent component analysis and hierarchical fusion of 37. Mitra, A., Snyder, A. Z., Blazey, T. & Raichle, M. E. Lag threads organize the classifiers. Neuroimage 90, 449–468 (2014). brain’s intrinsic activity. Proc. Natl Acad. Sci. USA 112, E2235–E2244 (2015). 68. Griffanti, L. et al. Ica-based artefact removal and accelerated fmri acquisition 38. Ge, T. et al. Multidimensional heritability analysis of neuroanatomical shape. for improved resting state network imaging. Neuroimage 95, 232–247 (2014). Nat. Commun. 7, 13291 (2016). 69. Jenkinson, M., Bannister, P., Brady, M. & Smith, S. Improved optimization for 39. Cohen, J. R. & D’Esposito, M. The segregation and integration of distinct the robust and accurate linear registration and motion correction of brain brain networks and their relationship to cognition. J. Neurosci. 36, images. Neuroimage 17, 825–841 (2002). 12083–12094 (2016). 70. Power, J. D., Barnes, K. A., Snyder, A. Z., Schlaggar, B. L. & Petersen, S. E. 40. Cheng, J. C. et al. Multivariate machine learning distinguishes cross-network Spurious but systematic correlations in functional connectivity mri networks dynamic functional connectivity patterns in state and trait neuropathic pain. arise from subject motion. Neuroimage 59, 2142–2154 (2012). Pain 159, 1764–1776 (2018). 71. Li, J. et al. Global signal regression strengthens association between resting- 41. Yeo, B. T. T. et al. The organization of the human cerebral cortex estimated by state functional connectivity and behavior. bioRxiv (2019). intrinsic functional connectivity. J. Neurophysiol. 106, 1125–1165 (2011). 72. Hamilton, J. Time Series Analysis. (Princeton University Press, Princeton, NJ, 42. Finn, E. S. et al. Functional connectome fingerprinting: identifying individuals 1994). using patterns of brain connectivity. Nat. Neurosci. 18, 1664–1671 (2015). 73. Barch, D. M. et al. Function in the human connectome: task-fmri and 43. Lim, J., Teng, J., Patanaik, A., Tandi, J. & Massar, S. A. A. Dynamic functional individual differences in behavior. Neuroimage 80, 169–189 (2013). connectivity markers of objective trait mindfulness. Neuroimage 176, 193–202 74. Efron, B. The Jackknife, the Bootstrap and Other Resampling Plans Vol. 38 (2018). (Society for Industrial and Applied Mathematics, 1982). 44. Schaefer, A. et al. Dynamic network participation of functional connectivity hubs assessed by resting-state fmri. Front. Hum. Neurosci. 8, 195 (2014). 45. Spreng, R. N., Stevens, W. D., Chamberlain, J. P., Gilmore, A. W. & Schacter, Acknowledgements D. L. Default network activity, coupled with the frontoparietal control We would like to thank Russ Poldrack, Timothy Laumann, and Jean-Baptiste Poline for network, supports goal-directed cognition. Neuroimage 53, 303–317 (2010). their insightful feedback. This research was supported in parts by the CHIST-ERA IVAN 46. Sadaghiani, S., Poline, J. -B., Kleinschmidt, A. & D’Esposito, M. Ongoing project (20CH21 174081), the Center for Biomedical Imaging (CIBM) of the Geneva- dynamics in large-scale functional connectivity predict perception. Proc. Natl Lausanne Universities and the EPFL, the Singapore MOE Tier 2 (MOE2014-T2-2-016), Acad. Sci. USA 112, 8463–8468 (2015). NUS Strategic Research (DPRT/944/09/14), NUS SOM Aspiration Fund 47. Wang, C., Ong, J. L., Patanaik, A., Zhou, J. & Chee, M. W. Spontaneous eyelid (R185000271720), Singapore NMRC (CBRG/ 0088/2015), NUS YIA, Singapore NRF closures link vigilance fluctuation with fmri dynamic connectivity states. Proc. fellowship (NRF-NRFF2017-06), NIH grant K99AG054573, and the Neuroimaging Natl Acad. Sci. 113, 9653–9658 (2016). Informatics and Analysis Center (1P30NS098577). The computational work was partially 48. Fornito, A., Harrison, B. J., Zalesky, A. & Simons, J. S. Competitive and performed on resources of the National Supercomputing Center, Singapore https://www. cooperative dynamics of large-scale brain functional networks supporting nscc.sg. Data were provided by the Human Connectome Project, WU-Minn Consortium recollection. Proc. Natl Acad. Sci. USA 109, 12788–12793 (2012). (Principal Investigators: David Van Essen and Kamil Ugurbil; 1U54MH091- 657) funded 49. Davison, E. N. et al. Brain network adaptability across task states. PLoS. by the 16 NIH Institutes and Centers that support the NIH Blueprint for Neuroscience Comput. Biol. 11, e1004029 (2015). Research; and by the McDonnell Center for Systems Neuroscience at Washington 50. Bertolero, M. A., Yeo, B. T. T. & D’Esposito, M. The modular and integrative University. functional architecture of the human brain. Proc. Natl Acad. Sci. USA 112, E6798–E6807 (2015). Author contributions 51. Yeo, B. T. T. et al. Functional specialization and flexibility in human R.L., C.O., T.G., M.R.S., and B.T.T.Y. designed research; R.L., J.L, and R.K. performed association cortex. Cereb. Cortex 25, 3654–3672 (2015). research; R.L., J.L., R.K., and T.G. contributed new reagents/analytic tools; R.L., J.L., T.G., 52. Demirtaş, M. et al. Hierarchical heterogeneity across human cortex shapes and B.T.T.Y. analyzed data; and R.L., D.V.D.V., T.G., M.R.S., and B.T.T.Y. wrote the paper. large-scale neural dynamics. Neuron 101, 1181–1194.e13 (2019). 8 NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10317-7 ARTICLE Additional information Open Access This article is licensed under a Creative Commons Supplementary Information accompanies this paper at https://doi.org/10.1038/s41467- Attribution 4.0 International License, which permits use, sharing, 019-10317-7. adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Competing interests: The authors declare no competing interests. Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless Reprints and permission information is available online at http://npg.nature.com/ indicated otherwise in a credit line to the material. If material is not included in the reprintsandpermissions/ article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from Journal peer review information: Nature Communications thanks Michael Breakspear, the copyright holder. To view a copy of this license, visit http://creativecommons.org/ Pablo Barttfeld, and the anonymous reviewer(s) for their contribution to the peer review licenses/by/4.0/. of this work. Peer reviewer reports are available. Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in © The Author(s) 2019 published maps and institutional affiliations. NATURE COMMUNICATIONS | (2019) 10:2317 | https://doi.org/10.1038/s41467-019-10317-7 | www.nature.com/naturecommunications 9

Journal

Nature Communications – Springer Journals

Published: May 24, 2019

There are no references for this article.