Dynamic Functional Connectivity ofNeurocognitive Networks in Children

Hilary A. Marusak,1* Vince D. Calhoun,2,3 Suzanne Brown,4

Laura M. Crespo,5 Kelsey Sala-Hamrick,5 Ian H. Gotlib,6 andMoriah E. Thomason7,8,9

1Department of Pharmacy Practice, Wayne State University, Detroit, Michigan2The Mind Research Network, Albuquerque, North Mexico

3Department of Electrical and Computer Engineering, The University of New Mexico,Albuquerque, North Mexico

4School of Social Work, Wayne State University, Detroit, Michigan5Department of Psychology, Wayne State University, Detroit, Michigan

6Department of Psychology, Stanford University, Stanford, California7Perinatology Research Branch, NICHD/NIH/DHHS, Bethesda, Maryland

8Department of Pediatrics, Wayne State University School of Medicine, Detroit, Michigan9Merrill Palmer Skillman Institute for Child and Family Development, Wayne State

University, Detroit, Michigan

r r

Abstract: The human brain is highly dynamic, supporting a remarkable range of cognitive abilities thatemerge over the course of development. While flexible and dynamic coordination between neural sys-tems is firmly established for children, our understanding of brain functional organization in early lifehas been built largely on the implicit assumption that functional connectivity (FC) is static. Under-standing the nature of dynamic neural interactions during development is a critical issue for cognitiveneuroscience, with implications for neurodevelopmental pathologies that involve anomalies in brainconnectivity. In this work, FC dynamics of neurocognitive networks in a sample of 146 youth from var-ied sociodemographic backgrounds were delineated. Independent component analysis, sliding timewindow correlation, and k-means clustering were applied to resting-state fMRI data. Results revealedsix dynamic FC states that re-occur over time and that complement, but significantly extend, measuresof static FC. Moreover, the occurrence and amount of time spent in specific FC states are related to thecontent of self-generated thought during the scan. Additionally, some connections are more variableover time than are others, including those between inferior parietal lobe and precuneus. These regionscontribute to multiple networks and likely play a role in adaptive processes in childhood. Age-relatedincreases in temporal variability of FC among neurocognitive networks were also found. Taken togeth-er, these findings lay the groundwork for understanding how variation in the developing

Additional Supporting Information may be found in the onlineversion of this article.

The authors declare no conflict of interest.Contract grant sponsors: The Merrill Palmer Skillman Instituteand the Department of Pediatrics, Wayne State University Schoolof Medicine, a NARSAD Young Investigator Award (MET),NIEHS awards P30ES020957 and R21ES026022 (MET), NIGMSawards P20GM103472 and R01EB020407 (VDC), NSF grant#1539067 (VDC), American Cancer Society award 129368-PF-16-057-01-PCSM (HAM), and NIMH R21 MH074849 (IHG)

*Correspondence to: Hilary A. Marusak, Ph.D., Department ofPharmacy Practice, Wayne State University, 259 Mack Ave, Suite2190, Detroit, MI 48202. E-mail: hmarusak@med.wayne.edu

Received for publication 6 June 2016; Accepted 1 August 2016.

DOI: 10.1002/hbm.23346Published online 18 August 2016 in Wiley Online Library(wileyonlinelibrary.com).

r Human Brain Mapping 38:97–108 (2017) r

VC 2016 Wiley Periodicals, Inc.

chronnectome is related to risk for neurodevelopmental disorders. Understanding how brain systemsreconfigure with development should provide insight into the ontogeny of complex, flexible cognitiveprocesses. Hum Brain Mapp 38:97–108, 2017. VC 2016 Wiley Periodicals, Inc.

Key words: resting-state; brain development; intrinsic connectivity; fMRI; intrinsic activity; indepen-dent component analysis; state variability; default mode network; salience network; central executivenetwork

r r

INTRODUCTION

The human brain is intrinsically organized into a complexset of functional networks. These networks engender andconstrain a broad range of complex cognitive functions,including language, reasoning, and cognitive control. Thestructure of cognition is undergirded by ability of brainfunctional networks to dynamically express various func-tional configurations. Until recently, however, our under-standing of intrinsic connectivity networks (ICNs) has reliedon static measures of functional neural connectivity (FC) thatassume spatial temporal stationarity throughout the mea-surement period. Observations that FC within and betweenICNs is dynamic, or fluctuating over time [Allen et al., 2014;Chang and Glover, 2010; Hutchison et al., 2013; Sakogluet al., 2010], underscores the need for detailed examinationof FC across discrete time windows. These discoveries of thehuman chronnectome, or time-varying functional couplingin ICNs, are increasing our understanding of fundamentalproperties of functional brain networks, and of how neuralsystems flexibly coordinate to support cognitive and behav-ioral dynamics [Calhoun et al., 2014].

High-level attentional and cognitive control processesdevelop over the course of childhood and adolescence[Bunge and Wright, 2007; Luna et al., 2010]. It is increas-ingly clear that the emergence of these complex processesis supported by reconfiguration and maturation of func-tional brain networks. Studies of conventional static FCdemonstrate that although ICNs are established in earlylife, even during the fetal period [Thomason et al., 2015],they undergo continued refinement over the first two dec-ades of life by a protracted trajectory of brain developmentand experiential learning [Fair et al., 2007; Fransson et al.,2011]. Research shows that ICNs become more segregatedwith age, with age-related increases in within-network FC[Dosenbach et al., 2010; Fair et al., 2007; Sherman et al.,2014; Thomason et al., 2008; Uddin et al., 2011]. There arealso age-related changes in between-network integration,with children displaying weaker between-network FC com-pared with adults [Uddin et al., 2011].

While static FC approaches have provided us with criti-cal information about the topological organization of func-tional brain networks during development, they say littleabout the dynamic interplay within and between function-al brain networks at the time scale in which cognitive pro-cesses occur. Recent advances in the field of dynamic FC

have made this all the more accessible [see reviews by Cal-houn et al., 2014; Hutchison et al., 2013]. Dynamic FCstudies have demonstrated reoccurring patterns of brainFC, or FC “states,” that are reproducible over time andacross individuals [Allen et al., 2014; Chang and Glover,2010]. FC dynamics have shown promise for predictingmental and vigilance states [Shirer et al., 2012] and forcharacterizing disease [Damaraju et al., 2014a; Rashidet al., 2014]. Because dynamic FC approaches more closelyapproximate the unfolding of cognitive processes in realtime, they may also help us to better understand develop-mental gains in cognitive and behavioral flexibility. Staticand dynamic FC approaches are complementary, and bothare needed to understand functional neural organizationduring early life, and how cognitive functions emerge andevolve over the course of maturation.

Emerging data demonstrate the utility of dynamic FCapproaches for understanding functional neurodevelop-ment. In a recent study of 51 individuals (ages 9–32 years),Hutchison and Morton [2015] found that although thestructure and topology of FC states is stable across age,there are age-related changes in the frequency and dwelltime (how long an individual spends in a given state) ofcertain FC states. Additionally, older individuals showed ahigher frequency of state transitions during a cognitivecontrol task, but not during rest. Other work demonstratesthat connections among ICNs are more variable withincreasing age [Allen et al., 2011; Qin et al., 2015]. Thesefindings are in agreement with seminal research by McIn-tosh et al. [2008] in children (ages 8–15) and young adults(ages 20–33) showing that electrophysiological variabilityincreases with age, which corresponds with reduced behav-ioral variability and more accurate performance. Together,these data have led to prevailing theory that (1) higherbrain variability reflects greater complexity and a greatercapacity for information processing, and (2) developmentalgains in cognitive functioning are undergirded byincreased neural temporal dynamics [Hutchison and Mor-ton, 2016; McIntosh et al., 2008]. Importantly, knowledgeof FC dynamics and their possible associations with men-tal activity during developmental years, when higher-order cognitive functions are emerging, is limited.

In the present study we aimed to characterize dynamicFC, or variation in FC over time, in a sample of 146 youth,ages 7–16 years. To increase the generalizability of ourfindings, we combined two youth cohorts from varied

r Marusak et al. r

r 98 r

socioeconomic and demographic backgrounds. We focusedon interactions within and between subsystems of coreneurocognitive networks thought to be critical for develop-mental gains in cognitive functioning [see reviews byMenon, 2011, 2013]. These include (i) the frontoparietalcentral executive network (CEN) anchored in dorsolateralprefrontal cortex and posterior parietal cortex; (ii) thedefault mode network (DMN) anchored in posterior cingu-late cortex, medial prefrontal cortex, and medial temporallobe, and (iii) the salience network (SN) anchored in ante-rior insula and anterior cingulate cortex. For completeness,secondary analyses examining whole brain ICNs were alsoconducted. ICNs were identified using group independentcomponent analysis (ICA), and time-varying FC withinand between ICNs was estimated using a series of slidingwindows. To identify connectivity states that are quasi-stable, that is, that reoccur over time and are reproducibleacross subjects, we applied k-means clustering to win-dowed FC correlation matrices. We leveraged the uncon-strained nature of the resting-state, as dynamics arethought to be more prominent during rest periods whenindividuals freely engage in several forms of mental activi-ty [Delamillieure et al., 2010] that are posited to differen-tially affect brain functional organization. We also aimedto examine the functional significance of observed FCstates by evaluating their correspondence with measuresof thought content during the scan, and tested for age-related change in variability in FC between networks, andin frequency of FC states.

The present study extends prior work in at least threeways. First, the prior study by Hutchison and Morton[2015] in 51 individuals (ages 9–32 years) provided criticalinsight into the functional relevance of particular FC statesby evaluating changes in their frequency during a cognitivecontrol task versus rest. The present study complementsthis work by evaluating, for the first time, the relevance ofdynamic FC states for self-generated mental activity in chil-dren during the resting-state, which is particularly condu-cive to mind-wandering [Filler and Giambra, 1973]. We alsoextend this work by testing a large sample of children fromvaried demographic backgrounds, consistent with calls forlarger, more representative neuroimaging studies [Falket al., 2013]. Finally, we apply nearly identical analytic strat-egies as our prior study in adults [Allen et al., 2014], facili-tating comparison between studies.

MATERIALS AND METHODS

Participants

Resting-state scans from 146 children and adolescents(youth) were used in this study (mean age: 12.28 years,range 7–16, 57% female). An additional 11 participantswere scanned but excluded from the present studybecause their movement exceeded 2 mm, as describedbelow. Half of the participants (n 5 73) were scanned at

Stanford University’s (SU) Richard M. Lucas Center forImaging (Stanford, CA), and the other half (n 5 73) atWayne State University’s (WSU) Magnetic ResonanceResearch Facility (Detroit, MI). These samples overlappedin age, t(144) 5 1.01, P 5 0.311, and data were collectedusing similar imaging parameters. Importantly, however,the two samples differed in their ethnic and socioeconomicdistribution. As shown in Table I, the SU sample was pre-dominantly Caucasian with annual household incomes of$100,000 or more. In contrast, the majority of WSU partici-pants were African American with household incomes lessthan $40,000 per year. There were more females in theWSU than in the SU sample, v2(1) 5 6.28, P 5 0.009. Thesesamples were combined to achieve a larger, more diverseyouth cohort (see Supporting Information Fig. S1 for ageand gender distribution, by site).

All participants were fluent in English and had noreported history of brain injury. Pubertal status wasassessed for all participants using the self-report Tannerstages questionnaire [Marshall and Tanner, 1968]. Twenty-five participants did not report their Tanner stage (seeTable I). There was a trend for those participants to beyounger than those who did report their Tanner stage,t(144) 5 1.85, P 5 0.066. Parents and youth gave informedconsent and assent, respectively, as approved by the localInstitutional Review Board.

Post-Scan Questionnaire

Immediately following the MRI scan, participants com-pleted a brief, semi-structured questionnaire that inquiredabout their internal experiences during the scan (providedin Supporting Information). Children indicated the percentof time they spent thinking about past, present (duringscan), or future, and their self versus others. Using a 0–6point scale (0 5 strongly disagree; 6 5 strongly agree), par-ticipants also quantitatively rated whether they thoughtabout happy or sad things during the scan. Self-reportdata were available for 66% of the study sample (n 5 97participants), with more data available in the WSU than inthe SU sample, v2(1) 5 54.77, P< 0.001. Self-report datawere used to ascribe initial functional relevance of dynam-ic connectivity states, as described below.

fMRI Data Acquisition

Imaging was performed with either a 3 T GE system(SU) or 3 T Siemens Verio scanner (WSU). Imaging proto-col and parameters were similar at the two sites. All par-ticipants completed a 6-minute resting-state scan, duringwhich they were asked to lie quietly in the scanner withtheir eyes closed. A total of 180 T2*-weighted blood-oxy-genation-level-dependent (BOLD) images were acquiredfor each participant. At SU, a custom built single-channelquadrature head coil coupled with a T2*-sensitive gradientecho spiral in/out pulse sequence was used (Glover and

r Dynamic Functional Connectivity in Children r

r 99 r

Law, 2001] with the following parameters: repetition time[TR] 5 2,000 ms; echo time [TE] 5 30 ms; flip angle 5 778;FOV [field of view] 5 220 mm, 29 slices; slicethickness 5 4.0 mm. At WSU, T2*-weighted BOLD imageswere acquired with a 12-channel head coil using echo-planar imaging with the following parameters: TR 5 2,000ms; TE 5 25 ms; flip angle 5 908; FOV 5 220 mm, 29 slices;slice thickness 5 4.0 mm.

fMRI Data Preprocessing

Preprocessing was conducted using the SPM8 softwarepackage (Statistical Parametric Mapping; http://www.fil.ion.ucl.ac.uk/spm/). The first three frames were removedto allow for signal stabilization. Images were then slice-time corrected, realigned, spatially normalized to the Mon-treal Neurological Institute (MNI) template, and smoothedusing an 8 mm full-width at half-maximum Gaussian ker-nel. Given our interest in temporal modulation, we nor-malized voxel time series (using z-score transformation) tominimize possible bias in subsequent variance-based datareduction steps [Allen et al., 2014].

Motion Screening and Spike Removal

We took several steps to reduce the potential influenceof motion-related artifact in the data. First, movement wasplotted and visually inspected for each participant usingArtRepair software (http://cibsr.stanford.edu/tools/human-brain-project/artrepair-software.html). Participantswith motion exceeding 2 mm in any direction wereexcluded from the study (n 5 11), leaving a total sample ofN 5 146. Notably, movement was relatively low for theremaining 146 participants, with translational and rotation-al mean movement 0.2 6 0.14 mm and 0.148 6 0.128, respec-tively. Translational and rotational root mean square(RMS) motion was 0.11 6 0.06 mm and 0.0028 6 0.0018,respectively. Next, as we describe below, we applied ahigh model order ICA decomposition to minimize noiseartifact [Birn et al., 2008]. Additional steps were taken toaddress potential motion-related artifact in derived compo-nent timeseries, following group ICA (“postprocessing”).In particular, high-motion frames were replaced with thebest estimate using a third-order spline fit to the clean por-tions of the timeseries, following prior work [Allen et al.,2014]. Outliers were detected based on the median abso-lute deviation, as implemented in 3DDESPIKE (http://afni.

TABLE I. Participant characteristics and demographics by site

SU (n 5 73) WSU (n 5 73) P-value

Age, m (SD) 12.47 (1.88) 12.09 (2.54) 0.311Sex, females (%) 34 (46.57) 49 (67.12) 0.009Pubertal maturation, n (%) 0.313

Pre/early pubertal (Tanner stages 1–2) 25 (34.25) 30 (41.1)Mid/late pubertal (Tanner stages 3–5) 26 (35.61) 40 (54.79)Not reported 22 (30.14) 3 (4.11)

Race/Ethnicity, n (%) <0.001Caucasian 44 (60.27) 23 (31.51)African American 0 (0) 32 (43.83)Latino 3 (4.11) 3 (4.11)Asian 3 (4.11) 0 (0)Mixed race 19 (26.03) 7 (9.59)Not reported 4 (5.48) 8 (10.96)

SU Household Income, n (%)Less than $50,000 9 (12.33)$50,000–$75,000 11 (15.07)$75,000–$100,000 10 (13.69)$100,000 or more 25 (34.25)Not reported 18 (24.66)

WSU Household Income, n (%)Less than $40,000 41 (56.16)$40,000–$60,000 14 (19.18)$60,000–$80,000 5 (6.85)$80,000–$100,000 2 (2.74)$100,000 or more 8 (10.96)Not reported 3 (4.11)

Abbreviations: Stanford University, SU; Wayne State University, WSU; number, n; mean, m. P-values derived from two-sample t-testsfor age, and chi-square for sex, pubertal maturation, and race. Income comparison between sites are not available as they were mea-sured by using different ordinal scales.

r Marusak et al. r

r 100 r

nimh.nih.gov/afni). On average, 8.28 6 2.64 frames wereremoved per component timeseries, for each participant.No more than 20 average frames were replaced acrosscomponents for each participant, resulting in a minimumof 5.3 min of data. We chose to replace rather than removehigh-motion frames, as removing the frames would com-promise the subsequent sliding window approach. Thisapproach also ensures that participants contribute an equalnumber of timepoints to the analysis. Finally, the sixrealignment parameters and their temporal derivativeswere regressed out of the component timeseries [Damarajuand Calhoun, 2014b].

Group ICA and Component Identification

To facilitate comparison with FC states observed inadults, we conducted analyses similar to those used by[Allen et al., 2014]. First, to identify brain functional net-works, whole-brain data were decomposed using spatialICA as implemented in GIFT toolbox (v.3.0a; http://mialab.mrn.org/software/gift/). C 5 100 components werederived using the Infomax algorithm. Higher-order ICAyields components that correspond with known anatomi-cal and functional segmentations, and allow investigationof ICN subsystems [cf. Allen et al., 2014]. ICA results in aset of group aggregate spatial maps that are then backreconstructed into single subject space [Calhoun et al.,2001; Erhardt et al., 2011]. Each back-reconstructed compo-nent consists of a spatial z-map reflecting the network’scoherent activity across space and an associated timecourse reflecting network activity over time.

A combination of spatial template-matching and visualinspection was used to identify components correspondingto CEN, DMN, and SN. Templates were derived from pri-or ICA analyses in youth [Thomason et al., 2011; availableat www.brainnexus.com/resources/resting-state-fmri-tem-plates) and in adults [Allen et al., 2014]. Components werealso evaluated based on expectations that ICNs should belocalized primarily in grey matter, have timeseries domi-nated by low-frequency fluctuations, and low spatial over-lap with known vascular, ventricular, motion, andsusceptibility artifacts [Allen et al., 2011; Cordes et al.,2000]. Convergence across goodness of fit scores and visu-al inspection led to characterization of 52 of the 100 com-ponents as likely ICNs as opposed to physiological,movement related, or imaging artifacts. Primary analysesfocused on 25 components that corresponded with CEN,DMN, and SN. Results from the full 52 components areprovided in the Supporting Information.

Postprocessing

Postprocessing was applied to component timeseries toremove remaining noise sources, including scanner driftand movement-related artifact. This included the removalof low-frequency trends (high-frequency cutoff of 0.15 Hz),

and linear, quadratic, and cubic detrending. Movement“spikes” were addressed with outlier detection andremoval, and with nuisance regression of the six realign-ment parameters and their temporal derivatives, asdescribed above. Prior work shows that these stringentpostprocessing steps do not fundamentally alter dynamicFC structures [Allen et al., 2014].

FC Estimation

ICA-derived individual participant ICN maps were usedto estimate dynamic FC and standard static FC between com-ponents. Static FC was estimated from the timeseries matrix,as the (C 3 C) sample covariance matrix, controlling for site.Dynamic FC was estimated with a sliding window, by com-puting between-component correlation values from taperedwindowed segments. Segments were tapered by convolvinga rectangle (width 5 22 TRs 5 44 s) with a Gaussian (r 5 3TRs), and slid in steps of 1 TR, resulting in W 5 158 win-dows. Covariance of ICNs was estimated, as described previ-ously [Allen et al., 2014]. Final dynamic FC estimates foreach window were concatenated to form a C 3 C 3 W arrayrepresenting the changes in covariance (correlation) betweennetworks (components) as a function of time.

Dynamic States and Clustering

To assess the frequency and structure of reoccurring FCstates, we applied k-means clustering (using the L1 dis-tance function) to windowed covariance matrices. Onlycovariances between the 25 ICNs corresponding withCEN, DMN, and SN were used in the clustering analysis,resulting in 25 3 (25 2 1)/2 5 300 features. We determinedthe number of FC states (or “clusters”) to be six using theelbow criterion of the cluster validity index, which is com-puted as the ratio between within-cluster distances tobetween-cluster distance. Cluster “centrotypes” are thenused as starting points to cluster all the dynamic FC data.To examine the structure of dynamic FC states across allsubjects, we evaluated group-level FC states.

FC Variability and Zones of Instability

FC estimates between some ICNs exhibit greater tempo-ral variability than others. We estimated temporal variabil-ity in FC by calculating the standard deviation across thesliding window correlation. Larger standard deviationsindicate more variable (or less stable) functional connec-tions between ICNs. “Zones of instability” (ZOI) aredefined as connections that are more variable over time[Allen et al., 2014].

Relation of Dynamic FC to Internal Thought

To ascribe functional significance to observed resting-state dynamic FC states, we evaluated subject-specific FC

r Dynamic Functional Connectivity in Children r

r 101 r

data for correspondence with thought content reported inthe post-scan questionnaire. In particular, we investigated(1) mean dwell time, or how long a participant is in eachstate; and/or (2) fraction of (total) time spent in each state.We evaluated the correspondence of the FC measures withthe percent of time youth thought about: (1) self- versusothers; (2) past, present, or future; and (3) had positiveversus negative thoughts. We also examined the effects ofage on dwell time and fraction of time spent in each state.To further control for possible contaminating effects ofsubject motion, we included subject-wise RMS motion as anuisance regressor in follow-up analyses.

Age Effects

We tested for age-related change in (1) temporal vari-ability between ICNs, (2) dwell time and fraction of timespent in each FC states, and (3) number of state transitionsduring the scan.

RESULTS

Core Neurocognitive Networks

To identify ICNs, we first performed group-level ICAon resting-scan scans of 146 youth. We identified 52 com-ponents as likely ICNs, and focused on 25 of those thatcorrespond with subsystems of DMN, SN, and CEN (Fig.1). Static and dynamic connectivity results for the full 52components are provided in Supporting Information.Overall, derived ICNs are similar to those observed in pre-vious high model order ICA decompositions [Allen et al.,2014; Hutchison and Morton, 2015; Smith et al., 2009].Component information and spatial maps are provided inSupporting Information.

Static FC

Figure 2 displays standard static FC, or FC computedover the entire scan length and averaged over participants,for the CEN, DMN, and SN subsystems. Patterns of FCare consistent with previous findings of positive correla-tion within ICNs, particularly within DMN, and negativecorrelation between DMN and CEN subsystems [Buckneret al., 2008; Thomason et al., 2008].

Dynamic FC

Using sliding windows and k-means clustering, weidentified six highly structured FC states that re-occurredthroughout the scan and across participants. FC states arearranged in order of emergence in Figure 3. State 1 is char-acterized by strong within-SN FC, anticorrelation betweenSN and DMN, and sparse positive FC between CEN andDMN subsystems. State 1 is unique in that connectivitywithin DMN is weak, including weak connectivity of the

precuneus ICN with other DMN subsystems. This state isobserved less frequently than are other detected states, butnonetheless represents a FC pattern that diverges substantiallyfrom the mean. State 2, which accounts for greater than 30% ofwindows, is characterized by weak connectivity among allICNs. This state of diffuse integration resembles a FC statepreviously observed in adults, which may reflect the averageof a number of additional states that are not sufficiently dis-tinct or frequent to be separated [see Allen et al., 2014].

Following State 2, State 3 appears most frequently andshows strong synchrony (i.e., positive FC) within bothCEN and DMN, and between CEN and DMN. State 4 ischaracterized by synchrony within posterior DMN net-works, which also displays strong negative correlationswith select CEN components (e.g., right inferior parietallobe). State 5 is characterized by a state of hyperconnectiv-ity; this is the only state with positive FC between nearlyall DMN and CEN subsystems. State 6 is characterized bysynchrony within DMN, and strong anticorrelationbetween DMN and CEN subsystems. In State 6, there isstrong integration of posterior insula and middle cingulatecortex ICNs of SN. Across participants, we did not observea clear increase or decrease in the frequency of any stateover the course of the scan.

Figure 1.

Core neurocognitive networks. Group independent component

analysis was used to parcellate the brain into intrinsic connectivity

networks (ICNs). See Supporting Information Table S1 for color

key and more detailed information on each component ICN. [Col-

or figure can be viewed at wileyonlinelibrary.com.]

r Marusak et al. r

r 102 r

Zones of Instability

We observed greater FC variability between some ICNpairs than others (see Figure 4). Connections with thegreatest variability, so-called “ZOIs” [Allen et al., 2014],were observed between DMN and CEN components, andwithin CEN. ZOIs frequently involved the right inferiorparietal lobe of CEN and the precuneus of DMN.

Relation to Post-Scan Questionnaire

To ascribe initial functional significance to observed re-occurring resting-state connectivity states, we correlatedsubject-specific dynamic FC values with self-reports ofspontaneous thought immediately following the scan. Fourout of the six observed FC states (2, 4, 5, and 6) were signifi-cantly correlated with dimensions of internal thought. Meandwell time in State 2 was positively correlated with self-

focused thought, r(97) 5 0.22, P 5 0.031, and negatively cor-related with thoughts about others, r(95) 5 20.309, P 5 0.002.Participants who reported more positive thoughts spent alarger fraction of time in State 4, r(93) 5 0.273, P 5 0.008. Par-ticipants who reported thinking more about others spent alarger fraction of time and remained longer in State 5,r(95) 5 0.235, P 5 0.022, and r(95) 5 0.259, P 5 0.011, respec-tively. Finally, participants who reported thinking moreabout the past spent a larger fraction of time and remainedlonger in State 6, r(96) 5 0.266, P 5 0.009, and r(96) 5 0.266,P 5 0.009. All associations remained significant when RMShead motion was included as a nuisance regressor.

Age Effects

Increased age was associated with increased overall tem-poral variability between neurocognitive networks,

Figure 2.

Static functional connectivity (FC) of core neurocognitive net-

works in children. FC was averaged for all participants for the

entire length of the resting-state scan to produce the static corre-

lation matrix. Intrinsic connectivity networks (ICNs) are labeled

with their corresponding component number. See Supporting

Information Table S1 for more detailed information on each com-

ponent. Abbreviations: left, L; right, R; middle frontal gyrus, MiFG;

inferior parietal lobe, IPL, superior temporal gyrus, STG; inferior

frontal gyrus, IFG; anterior cingulate cortex, ACC; posterior cin-

gulate cortex, PCC; angular gyrus, AG; parahippocampal gyrus,

PHG; middle temporal gyrus, MTG; anterior insula, aInsula; poste-

rior insula, pInsula; middle cingulate cortex, MCC. [Color figure

can be viewed at wileyonlinelibrary.com.]

r Dynamic Functional Connectivity in Children r

r 103 r

Figure 3.

Dynamic functional connectivity (FC) states in children. Dynamic FC states are derived using k-

means clustering. Total number and percentage of occurrences is listed above each state. Intrin-

sic connectivity networks (ICNs) are divided into central executive (CEN), default mode (DMN)

and salience (SN) sub-networks. See Figure 1 and Supporting Information Table S1 for abbrevia-

tions and more detailed information on each component. [Color figure can be viewed at

wileyonlinelibrary.com.]

r(146) 5 0.172, P 5 0.038. Age was positively associatedwith the fraction of time spent in State 4, r(146) 5 0.198,P 5 0.017, and negatively associated with the fraction oftime spent in State 2, r(146) 5 20.267, P 5 0.001. Theseassociations remained significant when RMS head motionwas included as a nuisance regressor. Age was not signifi-cantly associated with the number of state transitions,r(146) 5 0.084, P 5 0.314.

DISCUSSION

Interactions within and between core neurocognitivenetworks, that is, CEN, DMN, SN, are considered criticalfor behavioral shifts and adaptive processes. Emergingdata suggest that network FC is not stationary but rather,fluctuates, giving rise to highly structured patterns of FCthat reoccur over time [Allen et al., 2014; Chang and Glov-er, 2010]. Here, we characterize dynamic FC states in a rel-atively large sample of youth (n 5 146) from variedsociodemographic backgrounds. We found six highlystructured patterns of FC that deviate, in part, from static

FC averaged over the entire resting-state scan. The overallamount of time children spent or how long they remained(i.e., dwell time) in particular FC states corresponded withdifferent types of self-reported mental activity during thescan. Although the number of state transitions did notchange with age, we found age-related increases in thevariability of FC among core neurocognitive networks, anddifferences in the time children remained in particular FCstates. We discuss these results in turn, and suggest thatcontinued research in this area will increase our under-standing of how the functional architecture of the brain isestablished in early life, forming the backbone for theemergence of complex cognitive processes. Research inthis area may also illuminate the role of dynamic FC inneurodevelopmental disorders, an idea supported byrecent findings in adults with psychiatric disorders [e.g.,Damaraju et al., 2014a].

Our dynamic FC analyses identified six reoccurring FCstates that depart significantly from the static FC pattern.These findings, together with prior research in adult andpediatric samples [Allen et al., 2014; Hutchison and Mor-ton, 2015; Qin et al., 2015], caution against a labeling

Figure 4.

Variability in connections over time. Larger standard deviations indicate more variable, or less

stable, connections over the course of the measurement period. See Figure 1 and Supporting

Information Table S1 for abbreviations and more detailed information on each component. [Col-

or figure can be viewed at wileyonlinelibrary.com.]

r Dynamic Functional Connectivity in Children r

r 105 r

scheme in which ICNs are treated as singular and stableentities. Rather, ICNs show significant flexibility in func-tional coordination with other brain systems. For instance,established “core” SN regions, such as anterior insula andanterior cingulate cortex, exhibited synchronous activityonly in certain states (1, 5). Similarly, precuneus ICNs(DMN) exhibited affiliations with the DMN module inonly some states (3–6), similar to prior observations inadults [Allen et al., 2014]. As Allen and colleagues pointout, variable connectivity between ICNs may inform ques-tions regarding the inclusion of the precuneus in the DMN[see Buckner et al., 2008].

We observed particularly variable connections betweenthe CEN and DMN. These networks are highly integratedin the adult brain [Cole et al., 2010], suggesting that het-erogeneous and integrative function between the CEN andDMN is established earlier in life, at least in childhood.Within these networks, connections involving the precu-neus of DMN and the inferior parietal lobe of CEN werehighly variable. These regions are among the most globallyconnected in the brain [Cole et al., 2010] and are character-ized as core hubs of integration [Hagmann et al., 2008].The inferior parietal lobe also emerged as a ZOI in adults[Allen et al., 2014], suggesting a stable, critical role for thisICN during childhood and adulthood, which may rest onits ability to flexibly associate with other ICNs. The precu-neus, in contrast, did not emerge as a prominent ZOI inadults [Allen et al., 2014]. This is striking given that theprecuneus is flexibly associated with other DMN subsys-tems in both children and adults, but a high level of inte-gration across other neurocognitive networks was observedin children, here, but not in adults. This disparity suggeststhat the precuneus ICN has a privileged role in the DMN,particularly in early life [Yang et al., 2014].

Characterizing how dynamic FC evolves over the courseof development should yield insight into neural factorsinvolved in the range of increasingly complex cognitivefunctions that emerge across childhood and adolescence.For instance, we found age-related increases in the vari-ability of network coupling, which is consistent with priorelectrophysiological findings [McIntosh et al., 2008]. Great-er functional neural variability with age may afford greatercognitive and behavioral flexibility [Hutchison and Mor-ton, 2015; McIntosh et al., 2008]. In line with the priordynamic FC study in 9–32 year olds [Hutchison and Mor-ton, 2015], we found no significant relation between ageand the number of FC state transitions during rest. Nota-bly, age was positively associated with the number ofstates expressed during a cognitive control task [Hutchisonand Morton, 2015]. Thus, what appears to be changingwith age for self-generated cognition during rest is thetype of FC states that are expressed, rather than the fre-quency of transitioning between different states. For exter-nally directed cognition, in contrast, there does appear tobe more rapid switching between FC states with age.Together, these findings suggest that developmental

changes in the functional neural backbone underlyinginternally versus externally directed cognitive processesare distinct and may thus require several complementarymeasures.

Four of the six dynamic FC states covaried with retro-spective self-reports of thought content during the scan.Amount of self-focused thought correlated with frequencyof FC State 2, which was characterized by weak and dif-fuse connectivity between all ICNs. A state of weak FCwas also observed in adults in the Allen et al. [2014]study, in which the authors speculate that this state signi-fies the average of a large number of additional states thatare not sufficiently distinct or frequent to be separated.Based on our findings, it is possible that one or more ofthese diffuse FC states are involved in self-referential proc-essing in children, which is known to take several formsand may recruit both separate and overlapping brain sys-tems [Farb et al., 2007; Moran et al., 2009]. In contrast toState 2, State 6 was associated with thinking about others.This state was characterized by synchrony within DMN,and anticorrelation between DMN and CEN. This observa-tion is consistent with the extensive apparent overlapbetween the DMN and regions of the “social brain” [Marset al., 2012].

Children who reported more positive thoughts spent alarger amount of time in State 4, and children whoreported more thoughts about the past spent a largeramount of time and stayed longer in State 6. States 4 and6 were both characterized by synchrony within DMN, andanticorrelation between DMN and CEN. The DMN hasbeen implicated in both past and positive thinking [Gorgo-lewski et al., 2014]. Notably, fraction of time spent in State4 increased with increasing age, whereas State 6 did notshow age-related effects. State 4 also differed from State 6in that within-DMN correlations were more heavily con-centrated among posterior DMN ICNs, and DMN-CENnegative correlations involved a more restricted number ofCEN ICNs. Thus, while similar networks were involved inboth positive and past-related thoughts, the pattern of FCwithin- and between networks differed. Investigation ofdynamic FC patterns may elucidate how similar neuralsubstrates are involved in an array of mental activities.Overall, associations with self-generated thought centeredalmost exclusively on DMN connections rather than SNand CEN. This is likely due to our focus on characterizingFC dynamics during the resting state, which is highlylinked to DMN activity, and complements prior workcomparing FC dynamics during task and rest [Hutchisonand Morton, 2015].

It is important to consider several factors in interpretingour findings. The first factor involves the quantity of dataavailable for each participant. Although we were able toachieve a relatively large and heterogeneous sample bycombining data across two sites, each participant contrib-uted a modest amount (6 min) of resting-state data. Lon-ger scan times should improve estimates of FC variability

r Marusak et al. r

r 106 r

by allowing patterns of connectivity to reoccur severalmore times. This should also provide greater specificity inidentifying FC patterns that correspond with individualdifferences in reports of self-generated mental activity. Sec-ond, although we took several experimental (pretraining)and analytic steps (ICA-based denoising, motion outlierremoval, motion regression) to ensure that dynamic FCpatterns were driven by ICNs rather than noise, fastersampling times and concurrent physiological recordingsare needed to rule out possible influences of time-varyingnoise. A third consideration is the use of a retrospectivequestionnaire to inquire about thought content during thescan. Given the unconstrained nature of resting-state, it isdifficult to interrogate mental states without disruptingongoing spontaneous processes. Innovative approaches areemerging that should offer greater precision in mappingcognitive states to neural FC states [Christoff et al., 2009;O’Callaghan et al., 2015]. Even so, there may be importantindividual differences that emerge in truly unconstrainedstates. Here, we found associations with thought contentdimensions on a post-scan questionnaire, providing cluesabout the relevance of observed dynamic FC states forongoing mental activity in children.

CONCLUSIONS

Using a sliding window analysis, we characterizedresting-state FC dynamics among core neurocognitive net-works in 146 children. We found six FC states that reoc-curred over time and were consistent with andsignificantly expanded on standard static FC averagedacross the experiment. Providing initial insight into theirfunctional roles, some FC states corresponded with contentof self-generated thought reported by children, supportinga link between brain dynamics and underlying mentalexperience. In addition, some connections were more vari-able over time, including inferior parietal lobe and precu-neus. These regions exhibited membership in multiple-large scale systems, suggesting a unique role in adaptiveprocesses in early life. With age, there were differences inthe amount of time children expressed certain FC states,and age-related increases in the variability in FC amongcore neurocognitive networks, suggesting greater neuralcomplexity. Continued characterization of the developingchronnectome should yield new insights into the matura-tion of human brain networks that support higher-levelcognitive processes, and new avenues for identifyingmechanisms that underlie neurodevelopmental disorders.

ACKNOWLEDGMENTS

The content is solely the responsibility of the authors anddoes not necessarily represent the official views of thefunding agencies.The authors thank Allison Li, Clara Zundel, Farah Sheikh,Tarek Bazzi, Nataliya Bukavyn, Sameen Jaffry, Allesandra

Iadipaolo, and Farrah Elrahal at Wayne State Universityand Emily Dennis at Stanford University for assistance inparticipant recruitment and data collection, and the chil-dren and families who generously shared their time.

REFERENCES

Allen EA, Erhardt EB, Damaraju E, Gruner W, Segall JM, SilvaRF, et al. (2011): A baseline for the multivariate comparison ofresting-state networks. Front Syst Neurosci 5:2.

Allen EA, Damaraju E, Plis SM, Erhardt EB, Eichele T, CalhounVD (2014): Tracking whole-brain connectivity dynamics in theresting state. Cereb Cortex 24:663–676.

Birn RM, Murphy K, Bandettini PA (2008): The effect of respira-tion variations on independent component analysis results ofresting state functional connectivity. Hum Brain Mapp 29:740–750.

Buckner RL, Andrews-Hanna JR, Schacter DL (2008): The brain’sdefault network: Anatomy, function, and relevance to disease.Ann N Y Acad Sci 1124:1–38.

Bunge SA, Wright SB (2007): Neurodevelopmental changes inworking memory and cognitive control. Curr Opin Neurobiol17:243–250.

Calhoun VD, Adali T, Pearlson GD, Pekar JJ (2001): A method formaking group inferences from functional MRI data using inde-pendent component analysis. Hum Brain Mapp 14:140–151.

Calhoun VD, Miller R, Pearlson G, Adali T (2014): The chronnec-tome: Time-varying connectivity networks as the next frontierin fMRI data discovery. Neuron 84:262–274.

Chang C, Glover GH (2010): Time-frequency dynamics of resting-state brain connectivity measured with fMRI. NeuroImage 50:81–98.

Christoff K, Gordon AM, Smallwood J, Smith R, Schooler JW(2009): Experience sampling during fMRI reveals default net-work and executive system contributions to mind wandering.Proc Natl Acad Sci U S A 106:8719–8724.

Cole MW, Pathak S, Schneider W (2010): Identifying the brain’smost globally connected regions. NeuroImage 49:3132–3148.

Cordes D, Haughton VM, Arfanakis K, Wendt GJ, Turski PA,Moritz CH, et al. (2000): Mapping functionally related regionsof brain with functional connectivity MR imaging. AJNR Am JNeuroradiol 21:1636–1644.

Damaraju E, Calhoun VD (2014b). Impact of head motion on ICA-derived functional connectivity measures. Biennial Conferenceon Resting State Brain Connectivity: Boston, MA.

Damaraju E, Allen EA, Belger A, Ford JM, McEwen S, MathalonDH, et al. (2014a): Dynamic functional connectivity analysisreveals transient states of dysconnectivity in schizophrenia.NeuroImage Clin 5:298–308.

Delamillieure P, Doucet G, Mazoyer B, Turbelin MR, Delcroix N,Mellet E, et al. (2010): The resting state questionnaire: Anintrospective questionnaire for evaluation of inner experienceduring the conscious resting state. Brain Res Bull 81:565–573.

Dosenbach NU, Nardos B, Cohen AL, Fair DA, Power JD, ChurchJA, et al. (2010): Prediction of individual brain maturity usingfMRI. Science (New York, NY) 329:1358–1361.

Erhardt EB, Rachakonda S, Bedrick EJ, Allen EA, Adali T,Calhoun VD (2011): Comparison of multi-subject ICA methodsfor analysis of fMRI data. Hum Brain Mapp 32:2075–2095.

Fair DA, Dosenbach NU, Church JA, Cohen AL, Brahmbhatt S,Miezin FM, et al. (2007): Development of distinct control

r Dynamic Functional Connectivity in Children r

r 107 r

networks through segregation and integration. Proc Natl AcadSci U S A 104:13507–13512.

Falk EB, Hyde LW, Mitchell C, Faul J, Gonzalez R, Heitzeg MM,et al. (2013): What is a representative brain? Neurosciencemeets population science. Proc Natl Acad Sci U S A 110:17615–17622.

Farb NA, Segal ZV, Mayberg H, Bean J, McKeon D, Fatima Z,et al. (2007): Attending to the present: Mindfulness meditationreveals distinct neural modes of self-reference. Social CognAffect Neurosci 2:313–322.

Filler MS, Giambra LM (1973): Daydreaming as a function of cue-ing and task difficulty. Percept Mot Skills 37:503–509.

Fransson P, Aden U, Blennow M, Lagercrantz H (2011): The func-tional architecture of the infant brain as revealed by resting-state fMRI. Cereb Cortex 21:145–154.

Glover GH, Law CS (2001): Spiral-in/out BOLD fMRI forincreased SNR and reduced susceptibility artifacts. MagnReson Med 46:515–522.

Gorgolewski KJ, Lurie D, Urchs S, Kipping JA, Craddock RC,Milham MP, et al. (2014): A correspondence between individu-al differences in the brain’s intrinsic functional architectureand the content and form of self-generated thoughts. PloS One9:e97176.

Hagmann P, Cammoun L, Gigandet X, Meuli R, Honey CJ,Wedeen VJ, et al. (2008): Mapping the structural core ofhuman cerebral cortex. PLoS Biol 6:e159.

Hutchison RM, Morton JB (2015): Tracking the brain’s functionalcoupling dynamics over development. J Neurosci 35:6849–6859.

Hutchison RM, Morton JB (2016): It’s a matter of time: Reframingthe development of cognitive control as a modification of thebrain’s temporal dynamics. Dev Cogn Neurosci 18:70–77.

Hutchison RM, Womelsdorf T, Gati JS, Everling S, Menon RS(2013): Resting-state networks show dynamic functional con-nectivity in awake humans and anesthetized macaques. HumBrain Mapp 34:2154–2177.

Luna B, Padmanabhan A, O’Hearn K (2010): What has fMRI toldus about the development of cognitive control through adoles-cence? Brain Cogn 72:101–113.

Mars RB, Neubert FX, Noonan MP, Sallet J, Toni I, Rushworth MF(2012): On the relationship between the “default modenetwork” and the “social brain”. Front Human Neurosci 6:189.

Marshall WA, Tanner JM (1968): Growth and physiological devel-opment during adolescence. Annu Rev Med 19:283–300.

McIntosh AR, Kovacevic N, Itier RJ (2008): Increased brain signalvariability accompanies lower behavioral variability in devel-opment. PLoS Comput Biol 4:e1000106.

Menon V (2011): Large-scale brain networks and psychopatholo-gy: A unifying triple network model. Trends Cogn Sci 15:483–506.

Menon V (2013): Developmental pathways to functional brain net-works: Emerging principles. Trends Cogn Sci 17:627–640.

Moran JM, Heatherton TF, Kelley WM (2009): Modulation of corti-cal midline structures by implicit and explicit self-relevanceevaluation. Soc Neurosci 4:197–211.

O’Callaghan C, Shine JM, Lewis SJ, Andrews-Hanna JR, Irish M(2015): Shaped by our thoughts–a new task to assess spontane-ous cognition and its associated neural correlates in the defaultnetwork. Brain Cogn 93:1–10.

Qin J, Chen SG, Hu D, Zeng LL, Fan YM, Chen XP, et al. (2015):Predicting individual brain maturity using dynamic functionalconnectivity. Front Human Neurosci 9:418.

Rashid B, Damaraju E, Pearlson GD, Calhoun VD (2014): Dynamicconnectivity states estimated from resting fMRI Identify differ-ences among Schizophrenia, bipolar disorder, and healthy con-trol subjects. Front Human Neurosci 8:897.

Sakoglu U, Pearlson GD, Kiehl KA, Wang YM, Michael AM,Calhoun VD (2010): A method for evaluating dynamic func-tional network connectivity and task-modulation: Applicationto schizophrenia. Magma (New York, NY) 23:351–366.

Sherman LE, Rudie JD, Pfeifer JH, Masten CL, McNealy K,Dapretto M (2014): Development of the default mode and cen-tral executive networks across early adolescence: A longitudi-nal study. Dev Cogn Neurosci 10:148–159.

Shirer WR, Ryali S, Rykhlevskaia E, Menon V, Greicius MD(2012): Decoding subject-driven cognitive states with whole-brain connectivity patterns. Cereb Cortex 22:158–165.

Smith SM, Fox PT, Miller KL, Glahn DC, Fox PM, Mackay CE,et al. (2009): Correspondence of the brain’s functional architec-ture during activation and rest. Proc Natl Acad Sci U S A 106:13040–13045.

Thomason ME, Chang CE, Glover GH, Gabrieli JD, Greicius MD,Gotlib IH (2008): Default-mode function and task-induceddeactivation have overlapping brain substrates in children.Neuroimage 41:1493–1503.

Thomason ME, Dennis EL, Joshi AA, Joshi SH, Dinov ID, ChangC, et al. (2011): Resting-state fMRI can reliably map neural net-works in children. NeuroImage 55:165–175.

Thomason ME, Grove LE, Lozon TA, Jr., Vila AM, Ye Y, Nye MJ,et al. (2015): Age-related increases in long-range connectivityin fetal functional neural connectivity networks in utero. DevCogn Neurosci 11:96–104.

Uddin LQ, Supekar KS, Ryali S, Menon V (2011): Dynamic recon-figuration of structural and functional connectivity across coreneurocognitive brain networks with development. J Neurosci31:18578–18589.

Yang Z, Chang C, Xu T, Jiang L, Handwerker DA, Castellanos FX,et al. (2014): Connectivity trajectory across lifespan differenti-ates the precuneus from the default network. NeuroImage 89:45–56.

r Marusak et al. r

r 108 r