Combining fMRI during resting state and an attention bias task in children

Neuroimaging studies typically focus on either resting state or task-based fMRI data. Prior research has shown that similarity in functional connectivity between rest and cognitive tasks, interpreted as reconfiguration efficiency, is related to task performance and IQ. Here, we extend this approach from adults to children, and from cognitive tasks to a threat-based attention task. The goal of the current study was to examine whether similarity in functional connectivity during rest and an attention bias task relates to threat bias, IQ, anxiety symptoms, and social reticence. fMRI was measured during resting state and during the dot-probe task in 41 children (M = 13.44, SD = 0.70). Functional connectivity during rest and dot-probe was positively correlated, suggesting that functional hierarchies in the brain are stable. Similarity in functional connectivity between rest and the dot-probe task only related to threat bias (puncorr < .03). This effect did not survive correction for multiple testing. Overall, children who allocate more attention towards threat also may possess greater reconfiguration efficiency in switching from intrinsic to threat-related attention states. Finally, functional connectivity correlated negatively across the two conditions of the dot-probe task. Opposing patterns of modulation of functional connectivity by threat-congruent and threat-incongruent trials may reflect task-specific network changes during two different attentional processes.


Introduction
Neuroimaging studies typically focus on either resting state or taskbased fMRI 1 data. Recent studies have combined resting state and taskrelated fMRI data to gain insights on the relation between brain function and behavior (Bolt et al., 2017;J. R. Cohen & D'Esposito, 2016;Hearne et al., 2017;Ito et al., 2017;Mennes et al., 2010;Schultz and Cole, 2016). One approach of combining these two types of data is to investigate the similarity in functional connectivity between resting state and task data (Schultz and Cole, 2016). The current study extends this approach from adults to children, and from cognitive tasks to a threat-based attention task. This is important to test whether this novel analytic approach can be used in the field of developmental psychopathology.
Resting-state functional connectivity refers to temporal correlations that index brain network intrinsic activity (Fox and Raichle, 2007). Functional connectivity during rest and cognitive tasks appear to be strongly related, suggesting the presence of stable functional hierarchies ) that reflect both intrinsic, task-general, and also task-specific brain networks (Cole et al., 2014). Changes in patterns of functional connectivity from intrinsic to task-related states may reflect reconfiguration efficiency and can be measured as the similarity in (or correlation between) functional connectivity during rest and task (Schultz and Cole, 2016). When assessed during cognitive tasks, such reconfiguration efficiency relates to better task performance and, more generally, higher IQ (Schultz and Cole, 2016). We aimed to examine reconfiguration efficiency from rest to task in children, using a paradigm that has successfully uncovered correlates of anxiety, particularly among youth. The study of the reconfiguration efficiency of the emotional attention network around the mean age of onset of several anxiety disorders (e.g., specific phobia, social anxiety disorder) (de Lijster et al., 2017) might yield several new insights into the pathophysiological mechanisms of anxiety disorders. The dot-probe task examines the effects of threats on attention orienting (Bar-Haim, Lamy, Pergamin, Bakermans-Kranenburg, & van IJzendoorn, 2007). In this task, participants initially view two adjacent faces (angry-neutral, happy-neutral, or neutral-neutral) and then identify the location of a probe replacing either the angry (threat congruent) or neutral (threat incongruent) face (Macleod et al., 1986). Meta-analysis relates performance on this task, as expressed in an attention bias towards threats, to anxiety (Bar-Haim et al., 2007). Specifically, higher levels of anxiety predict shorter reaction-times during threat-congruent relative to threat-incongruent trials (threat bias (Abend et al., 2018;Bar-Haim et al., 2007);).
Several studies have used the dot-probe task in the MRI scanner to investigate mechanisms underlying threat-related attentional processes in youth, focusing on amygdala-related functional connectivity using a seedbased approach. Functional connectivity between the amygdala and several regions during threat-incongruent versus threat-congruent trials has demonstrated stability in youth when measured approximately nine weeks apart (White et al., 2016). Negative functional connectivity between the amygdala and the insula during threat versus neutral trials placed individuals with early-childhood behavioral inhibition at risk for developing internalizing symptoms in adulthood (Hardee et al., 2013). Longitudinal studies found that amygdala-PFC connectivity during threat-congruent trials moderated relations between early-childhood behavioral inhibition and anxiety symptoms in adolescence (Abend et al., 2019). Moreover, patterns of threat-related amygdala-insula functional connectivity during threat-congruent trials were found to differentiate pediatric anxiety patients from healthy controls (White et al., 2017).
Of note, prior studies have all used a seed-based approach to study amygdala connectivity during the dot-probe task. Several limitations of this literature are noted. These studies focus on different conditions within the dot-probe task (e.g. threat-incongruent versus threatcongruent trials). Furthermore, because the amygdala is only one of many regions implicated in anxiety and threat processing (Bruhl et al., 2014;Etkin and Wager, 2007;Xu et al., 2019), whole-network based analyses complement prior studies focused on isolated brain regions. Seed-based connectivity analyses in specific contrasts could be insensitive to network-level perturbations in connectivity, both quantified during resting state and during task-based fMRI. Therefore, it is important to examine functional connectivity on a whole-brain level both at rest and during threat-processing tasks to quantify the neural correlates of anxiety and threat processing. To do so, we extend prior seed-based connectivity findings by examining functional connectivity at the whole brain level, by focusing on reconfiguration efficiency from resting state to the dot-probe task.
Finally, beyond extending previous approaches to emotional-task data in children, the current study also implements two novel procedures to extend previous approaches. First, the study utilizes ICA with both restingstate and task-based data. Second, prior functional connectivity analyses with the dot-probe task used a seed-based approach to contrast connectivity in threat-congruent and threat-incongruent conditions. Beyond comparing connectivity across rest and task, the current study also employs a PPI approach within the emotional task to study modulation of functional connectivity by the two threat-related task conditions.
The current study assessed similarity in patterns of functional connectivity during rest and the dot-probe task in 13-year-old children, testing five specific hypotheses. First, we hypothesized that functional connectivity during rest correlates strongly with functional connectivity during the dot-probe task (Cole et al., 2014;Schultz and Cole, 2016). Second, we hypothesized that the degree of similarity in functional connectivity between rest and the dot-probe task (i.e., reconfiguration efficiency) relates to task performance (i.e., threat bias) (Schultz and Cole, 2016). Third, we hypothesized that reconfiguration efficiency also relates to IQ. This is because prior work finds IQ to relate to many individual-difference factors among children and adolescents, including performance in many cognitive tasks and many aspects of psychological function. Thus, IQ may also relate to efficiency in switching from resting state to task states in general, both for emotional tasks examined in the current study as well as cognitive tasks more closely linked to IQ in prior work (Schultz and Cole, 2016). Fourth, we hypothesized that reconfiguration efficiency relates to individual differences in anxiety-related measures, such as self-reported and parent-reported anxiety symptoms and socially reticent behavior. Finally, we explored whether task-evoked activity modulates functional connectivity and hypothesized that patterns of connectivity are modulated differently by threat-congruent and threat-incongruent trials of the dot-probe task.

Participants
Participants were recruited from a longitudinal study on behaviorally inhibited temperament and early childhood reticence. Data in the current report were collected when the children were approximately 13 years old. 42 children had fMRI data for both resting state and the dot-probe task. However, for one child, data were excluded because of excessive motion, yielding a final sample of 41 children (19 girls) with a mean age of 13.44 years (SD ¼ 0.70,). Resting state and taskbased data were collected on the same day for 25 children, and on different days for 16 children (resting state scan was 1-307 days after the dot-probe scan, median ¼ 101 days). Children entered the longitudinal study through two paths: one group (n ¼ 291) was selected based on temperamental reactivity at four months (Hane et al., 2008) and the other group (n ¼ 384) was randomly recruited from the community at two years of age (Jarcho et al., 2016;Michalska et al., 2019). Among the 41 children in the current report, 18 were recruited as infants and 23 were recruited at two years. All children provided written assent, and all parents provided written informed consent. The study was approved by the University of Maryland-College Park and National Institute of Mental Health institutional review boards.
Descriptive statistics appear in Table 1. Girls reported slightly more anxiety symptoms than boys, F(1,39) ¼ 3.85, p ¼ 0.057, but there were no differences between boys and girls on any of the other variables, Fs < 2.91, ps > 0.096. Children selected as infants (M ¼ 13.04, SD ¼ 0.63) were slightly younger than children recruited at two years (M ¼ 13.75, SD ¼ 0.61), F(1,39) ¼ 13.55, p ¼ 0.001. Children who did both scans on the same day (M ¼ 13.64, SD ¼ 0.65) were slightly older than children who did the scans on different days (M ¼ 13.14, SD ¼ 0.69), Note: social reticence scores for two participants were missing. SD ¼ standard deviation; ABV ¼ attention bias variability; SCARED ¼ Screen for Child Anxiety Related Disorders; C ¼ child-reported; P ¼ parent-reported; std DVARS ¼ standardized spatial standard deviation of the data after temporal differencing; DVARS ¼ original spatial standard deviation of the data after temporal differencing; FD ¼ framewise displacement. See main text for definitions of bias and bias variability F(1,39) ¼ 5.49, p ¼ 0.024. No other variables differed between children recruited as infants or at two years or between children who did both scans on same or different days, Fs < 3.20, ps > 0.082. The main analyses were corrected for age, sex, participant recruitment source (i.e., during infancy or at age two years), and time between scans. SD ¼ standard deviation; ABV ¼ attention bias variability; SCARED ¼ Screen for Child Anxiety Related Disorders; C ¼ child-reported; P ¼ parent-reported; std DVARS ¼ standardized spatial standard deviation of the data after temporal differencing; DVARS ¼ original spatial standard deviation of the data after temporal differencing; FD ¼ framewise displacements. See main text for definitions of bias and bias variability.

Dot-probe task
The dot-probe task has been used extensively to measure attention bias to threat (Abend et al., 2019;Bar-Haim et al., 2007;Macleod et al., 1986). In each trial of this task, a pair of faces was shown (either angry-neutral, happy-neutral, or neutral-neutral; see Fig. 1). After 500 m s, an arrow (i.e., probe) appeared at the location of one of these faces. The locations of the faces and arrows were counterbalanced across the task. Participants had to press a button as quickly as possible to indicate the direction of the arrow. In congruent trials the probe was presented at the location of the emotional face, in incongruent trials the probe was presented at the location of the neutral face. The task consisted of 48 threat-congruent, 48 threat-incongruent, 48 happy-congruent, 48 happy-incongruent, and 96 neutral-neutral trials across four blocks of 4:15 min. The task was programmed and administered with E-Prime (Psychology Software Tools, Pittsburgh, PA) and faces from the NimStim set were used (Tottenham et al., 2009).

IQ
IQ was estimated when children were approximately 10 years old using the vocabulary and matrix reasoning subscales of the Wechsler Abbreviated Scale of Intelligence (Wechsler, 1999). IQ has been shown to have high long-term stability (Canivez and Watkins, 1998;Watkins and Smith, 2013).

Anxiety symptoms
Trait anxiety symptoms were measured using the SCARED, a reliable child-and parent-report questionnaire with 42 items (Birmaher et al., 1997;Muris et al., 2004). We used the SCARED total score for self-report and parent-report separately, replacing up to two missing responses in the total score with the subject's mean value for the other items. Children and parents filled out the questionnaire separately within a year of the dot-probe scan (M ¼ 99.92 days, SD ¼ 108.29, range ¼ À1 to 348 days), with 31 (75.6%) children and parents completing the SCARED within 6 months. Test-retest reliability of the child-and parent-reported SCARED total score is moderate to good over a period of three to six months (Behrens et al., 2019;Birmaher et al., 1997;Boyd et al., 2003;Haley et al., 2011). Over a period of five years, most anxiety symptoms slightly decrease, whereas social anxiety symptoms remain relatively stable and generalized anxiety symptoms slightly increase in adolescent girls (Hale et al., 2008).

Social reticence
Social reticence reflects shy and anxiously avoidant behavior during early childhood (Degnan et al., 2014), and predicts later anxiety symptoms and related neural measures (Jarcho et al., 2016;Michalska et al., 2019). It was measured using a composite score of maternal-report questionnaires and behavioral observations collected at 2, 3, 4, 5, and 7 years of age (Jarcho et al., 2016;Michalska et al., 2019). The Social Fear subscale from the Toddler Behavioral Assessment Questionnaire (Goldsmith, 1996) was used at 2 and 3 years, and the Shyness subscale from the Children's Behavior Questionnaire (Rothbart et al., 2001) was used at 4, 5 and 7 years. Socially reticent behavior was observed during free play, cleanup and social problem-solving interactions with an unfamiliar, age-matched peer at 2, 3, 4, 5 and 7 years (Degnan et al., 2011(Degnan et al., , 2014. All subscales and behavior scores were standardized within timepoint and averaged together.

fMRI data collection
fMRI data were collected during resting state and the dot-probe task on a 3T MR750 General Electric scanner (Waukesha, Wisconsin, USA) with a 32-channel head coil. For resting state, 180 functional image volumes with 47 contiguous interleaved axial slices were collected with a T2*-weighted echo-planar sequence . For each of the four blocks of the dot-probe task, 111 functional image volumes with 47 contiguous interleaved axial slices were collected with a T2*-weighted echo-planar sequence (TR ¼ 2300 m s, TE ¼ 25 m s, flip angle ¼ 50, FOV ¼ 240 mm, matrix ¼ 96 Â 96 in plane resolution 2.5 Â 2.5 Â 3 mm). In addition, a high-resolution T1-weighted whole-brain volumetric scan was acquired during each scan session, with a high-resolution magnetization prepared gradient echo sequence (MPRAGE; TE ¼ min full; TI ¼ 425 m s; flip angle ¼ 7; FOV ¼ 256 mm; matrix ¼ 256 Â 256; in plane resolution ¼ 1 Â 1x1mm).

Behavioral analysis
Selection of dot-probe trials to be included in analyses followed the same criteria used by Abend et al. (2019). Trials were included only if the response to the probe was correct and if RT was between 150 and 2000 m s and less than 2.5 standard deviations away from the participant's mean RT. In line with prior studies (Abend et al., 2019;White et al., 2017), task data were considered valid if accuracy was !70%; all 41 participants surpassed this threshold. Task performance was measured by calculating threat bias scores (mean RT in threat-incongruent trials minus mean RT in threat-congruent trials), with higher scores reflecting an attention bias towards threat (that is, slower responding in threat-incongruent trials). Happy bias scores were also calculated (mean RT in happy-incongruent minus mean RT in happy-congruent trials).
In addition, secondary analyses examined ABV, an index of temporal variability in attention allocation previously found to relate to anxiety symptoms and brain function (Abend et al., 2019;Iacoviello et al., 2014). In a moving-window analysis, threat bias scores (as described above) were calculated for 10 successive angry-neutral trials. The standard deviation of all these threat bias scores within one participant was calculated, and then divided by the overall mean RT of the participant. The procedure was repeated for happy-neutral trials to calculate happy ABV scores (Abend et al., 2019). 2.6. fMRI preprocessing Quality of fMRI data was assessed using the MRI Quality Control tool (MRIQC) (Esteban et al., 2017). Data of one participant were excluded, because this subject was an outlier based on AFNI's outlier ratio and because visual inspection showed excessive movement. The first 4 TRs from each scan were removed to account for non-steady-state data. fMRI data from resting state and the dot-probe task were preprocessed using fMRIPrep (Esteban et al., 2019) including the following steps: skull stripping, correction for head-motion parameters, slice time correction, co-registration to corresponding structural image (boundary-based registration with 9 degrees of freedom), spatial normalization to MNI space (nonlinear registration), and ICA-AROMA (Pruim et al., 2015). All spatial transformations in fMRIPrep are done in one step, to allow for a single-interpolation resampling of volumes.
The following nuisance variables were regressed out from both resting state and dot-probe time series: motion parameters (translation in x, y and z directions, plus rotations around these axes), cerebral spinal fluid signal, white matter signal, original and standardized spatial standard deviation of the data after temporal differencing (DVARS and std DVARS), framewise displacement (Power et al., 2012), 6 temporal and anatomical noise components calculated with CompCor (Behzadi et al., 2007), cosine variables, and ICA-AROMA components that were classified as related to head motion. The data for both rest and task were high-pass filtered (0.04 Hz). Correlations between motion variables and the behavioral variables are shown in Table 2.

ICA and dual regression
ICA was run on the resting state data using Multivariate Exploratory Linear Optimized Decomposition into Independent Components (MELODIC). The model order was selected using Bayesian methods described in Beckmann and Smith (2004); 20 components were automatically identified. After visual inspection, three were removed as they were comprised of regions occupied by cerebrospinal fluid or other non-brain tissue. The remaining 17 components were selected for further analyses (Fig. 2). Naming of these components was based on putative functions related to their topography; we also used previous literature on resting-state imaging as reference (Beckmann et al., 2005;Damoiseaux et al., 2006;Smith et al., 2009). Time series for these 17 components were extracted from resting state and from dot-probe data (in native resolution) using dual regression in FSL (Filippini et al., 2009). Thus, for each independent component, a time course for rest and another for the dot-probe task were obtained. In addition, for the dot-probe task, task events (threat-congruent, threat-incongruent, happy-congruent, happy-incongruent, and neutral trials and trials with errors) were convolved using a canonical hemodynamic response function (HRF) approach with Glover's HRF (Glover, 1999) and were regressed out from the fMRI time course.

Functional connectivity and clustering
Full and partial correlations between each pair of the 17 components were calculated as measurements of functional connectivity, converted to a z-score using Fisher's r-to-z transformation, and assembled into connectivity matrices. For each subject, this generated one functional connectivity matrix for resting state and another for the dot-probe task. Since these matrices are symmetric, for convenience we chose to retain only the upper triangular part (above the main diagonal) for partial correlations, leaving the lower triangular part (below the main diagonal) to hold the full (not partial) correlations, thus avoiding redundancies when generating figures. These steps were conducted using customized FSLNets scripts (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSLNets) . To assess the similarity between these two matrices, the partial correlations above the main diagonal of each matrix were unwrapped into a vector, and the Pearson's correlation between the vector for resting state and the vector for the dot-probe task was calculated and converted to a z-score using Fisher's r-to-z transformation (Schultz and Cole, 2016). 2 These steps were performed for each participant, thereby generating one similarity value for each participant.
For visualization purposes, functional connectivity matrices were averaged across participants. To visualize the temporal proximity between the 17 ICA components, these components were clustered using hierarchical nearest neighbor with Ward's linkage as implemented in FSLNets.
To investigate whether task-evoked activity would modulate functional connectivity, two additional functional connectivity matrices, one for the threat-congruent and another for the threat-incongruent trials, were computed using a PPI strategy (Friston et al., 1997;O'Reilly et al., 2012). Instead of computing the correlation in the time series between all 17 components, we performed a regression analysis with the time series of one component modulated by the time series for threat-congruent trials (i.e., an interaction term) predicting the time series of another component, while including the modeled time series for threat-congruent trials as nuisance. This analysis assessed whether functional connectivity between these two components would be modulated by threat-congruent trials. This was done for all 17 components predicting all other components, resulting in a matrix similar to the FC matrix. As with the analyses for rest and task data, Pearson's correlation between these two matrices for each participant were calculated, and then subjected to the Fisher's r-to-z-transformation.

Between-subject statistical analysis
Group analyses assessed whether similarity in functional connectivity during rest and dot-probe was related to task performance (threat bias) as well as IQ, anxiety symptoms, and social reticence. These analyses used Note: *p < 0.05, uncorrected. ABV ¼ attention bias variability; SCARED ¼ Screen for Child Anxiety Related Disorders; C ¼ child-reported; P ¼ parent-reported; std DVARS ¼ standardized spatial standard deviation of the data after temporal differencing; DVARS ¼ original spatial standard deviation of the data after temporal differencing; FD ¼ framewise displacement.
2 Note that at this stage, investigations that use both resting state and taskbased fMRI do not require these two modalities to have the same spatial or temporal resolution.
A. Harrewijn et al. NeuroImage 205 (2020) 116301 nonparametric permutation inference using PALM (Winkler et al., 2014). Five designs were tested, each including similarity as the dependent variable, an intercept, nuisance variables, and an independent variable of interest (Table 3). Only the independent variable of interest differed per design, such that each design tested how the similarity in functional connectivity during rest and dot-probe relates to threat bias, IQ, child-reported anxiety symptoms, parent-reported anxiety symptoms, and social reticence. We tested both positive and negative effects (i.e., contrasts) for each of these five independent variables of interest, e.g. a positive effect of threat bias on similarity and a negative effect of threat bias on similarity in design 1. For the first design (testing threat bias), we also tested the intercept, that is, whether the correlation between resting state and dot-probe functional connectivity matrices was significantly different than zero. FWER was used in PALM to correct non-parametrically for multiple testing across the five designs (Winkler et al., 2016). A final set of exploratory nonparametric permutation tests in PALM was performed on six different dependent variables: similarity, distance measures (correlations between two vectors transformed to Euclidean distances), functional connectivity matrix during rest (full and partial correlations), and functional connectivity matrix during task (full and partial correlations). Consistent with the previous five designs, these analyses included each an intercept and the following nuisance variables: age, sex, participant recruitment source (i.e., during infancy or at age two years), and time between scans. We tested twelve different designs, each with the nuisance variables and one extra independent variable of interest: threat bias, IQ, child-reported anxiety symptoms, parent-reported anxiety symptoms, social reticence, interaction between sex and childreported anxiety symptoms, interaction between sex and parentreported anxiety symptoms, absolute threat bias, threat ABV, happy bias, absolute happy bias, and happy ABV (Supplementary Table 1). The p-values were corrected non-parametrically for multiple testing across the six independent variables and twelve designs in PALM using FWER (Winkler et al., 2016). The main analysis with five designs and the exploratory analyses with twelve designs were repeated for the functional connectivity matrices based on PPI analysis. Fig. 2. Spatial maps for each of the 17 independent components. Their putative functional interpretation based on previous studies is as follows: a ¼ right dorsal visual stream, b ¼ lateral visual cortical areas, c ¼ left dorsal visual stream, d ¼ sensory-motor system, e ¼ medial visual cortical areas, f ¼ auditory system, g ¼ visuo-spatial system, h ¼ executive control.

Table 3
Overview of the five designs that were tested using nonparametric permutation tests.

Functional connectivity matrices
Three clusters of correlated networks emerged from the data in both resting state and dot-probe functional connectivity matrices: one cluster encompassing default-mode, executive control, visual, and auditory networks; a second encompassing visual networks; and a third encompassing sensory-motor and auditory networks (see Figs. 3 and 4 for the FC matrices with z-statistics, and see Supplementary Figures 1 and 2 for the FC matrices with rs). The cluster encompassing sensory-motor and auditory networks was the same during resting state and the dot-probe task. The only difference in clustering between the resting state and the dot-probe task was that two networks (left dorsal visual stream [component 3] and left/right dorsal visual stream [component 15]) were part of the visual cluster during resting state and part of the defaultmode, executive control, visual and auditory cluster during the dotprobe task. 3 Treated separately, patterns of correlations during either resting state or the dot-probe task were unrelated to threat bias, IQ, anxiety symptoms, social reticence or any other measures from secondary regression analyses.

Similarity between resting state and the dot-probe task
The first PALM analysis testing the intercept showed that the average similarity between functional connectivity matrices during resting state and the dot-probe task was significantly different than zero across participants, with an average r-to-z ¼ 0.37, test statistic t ¼ 20.67, p FWER < 0.001, two-tailed. After controlling for all designs tested, the main PALM analysis with five designs showed that similarity between resting state and the dot-probe task was not related to threat bias, IQ, anxiety symptoms, or social reticence (Table 4). However, when we did not correct for multiple testing, similarity between resting state and the dot-probe task was positively related to threat bias, partial r ¼ 0.32, p uncorrected ¼ 0.03, p FWER ¼ 0.23 (corrected for age, sex, participant recruitment source, and time between scans; Fig. 5) 4,5 . This demonstrates that enhanced reconfiguration efficiency relates to greater bias towards threat in children. We Fig. 3. Functional connectivity matrix during resting state. Full correlations (transformed to zstatistics) are presented below the diagonal, partial correlations (transformed to z-statistics) are presented above the diagonal. Three clusters emerged: (a) visuo-spatial system, executive control and some visual and auditory networks (in red); (b) visual networks (in green); and (c) sensory-motor and auditory networks (in blue). Components with more similar time courses appear closer together. The threshold for unique colors in the dendrogram was set at 0.75. Note: numbers correspond to the 17 components from Fig. 2. 3 We reran the same analysis stream for examining the similarity between resting state and the dot-probe task without component 3 and 15, to investigate whether the association with threat bias would be stronger if only the stable networks were considered. The results, however, remained the same. 4 There was no difference in similarity between resting state and the dotprobe task between participants who did the scans on the same day versus different days, F(1, 39) ¼ 0.52, p ¼ 0.48. Furthermore, there was no correlation between similarity and the number of days between scans, r ¼ 0.06, p ¼ 0.69. We repeated the analyses including only children who performed both scans on the same day (n ¼ 25). The correlation between similarity and threat bias retained the same direction but was no longer significant, potentially due to lower power, r ¼ 0.20, t ¼ 1.20, p FWER ¼ 0.76, p uncorrected ¼ 0.12. 5 When analyses were repeated with the thresholded (0.5) MELODIC output with the left and right amygdala added to the functional connectivity matrices, the correlation between similarity and threat bias did not remain significant.
calculated the characteristic path length (Rubinov and Sporns, 2010) for both resting state and the dot-probe task, to test whether this effect was driven by differences in rest or task functional connectivity. The relation between threat bias and characteristic path length during both rest and task was not significant, respectively r ¼ 0.27, p ¼ 0.093 and r ¼ 0.27, p ¼ 0.087. This suggests that the relation between threat bias and reconfiguration efficiency was not driven by either resting state or the dot-probe task, but rather a combination. The exploratory PALM analysis with twelve designs showed no additional significant effects (see Supplementary Table 2). Finally, to test whether one of the clusters drove the association between similarity and threat bias, we calculated similarity between resting state and the dot-probe task for each cluster separately (excluding components 3 and 15 because they belonged to different clusters in rest versus task data). The relation between similarity and threat bias was not significant within each of these clusters, suggesting that the observed correlation is not driven by any of these three clusters, and therefore, may reflect global reconfiguration. Fig. 4. Functional connectivity matrix during the dot-probe task. Full correlations (transformed to zstatistics) are presented below the diagonal, partial correlations (transformed to z-statistics) are presented above the diagonal. Three clusters emerged: (a) visuo-spatial system, executive control and some visual and auditory networks (in red); (b) visual networks (in green); and (c) sensory-motor and auditory networks (in blue). Components with more similar time courses appear closer together. The threshold for unique colors in the dendrogram was set at 0.75. Note: numbers correspond to the 17 components (based on resting state data) from Fig. 2.   Fig. 5. Effect of similarity between resting state and dot-probe task on threat bias.

Table 4
Results of the main PALM analysis with five designs testing the effect of threat bias, IQ, anxiety symptoms and social reticence on similarity between resting state and the dot-probe task. Note: rs are corrected for the nuisance variables in the design (age, sex, participant recruitment source, and time between scans).

PPI analyses
The functional connectivity matrices based on PPI analyses are shown in Figs. 6 and 7 (displaying z-statistics, and see Supplementary Figures 3  and 4 for the PPI matrices with rs). No significant clusters emerged from the data in either the threat-congruent or threat-incongruent conditions. The first PALM analysis testing the intercept showed that the average similarity between functional connectivity matrices as modulated by threat-congruent and threat-incongruent conditions was significantly different than zero across participants, with an average r-to-z ¼ À0.11, test statistic t ¼ À5.28, p FWER < 0.001, two-tailed. The degree to which functional connectivity among brain regions was modulated by the task effects was not related to threat bias, IQ, anxiety symptoms, social reticence, or any other measures from exploratory analyses. The main PALM analysis with five designs showed that the similarity of task-related connectivity modulation between the threat-congruent and threatincongruent conditions was unrelated to threat bias, IQ, anxiety symptoms, and social reticence ( Table 5). The exploratory PALM analysis with twelve designs showed no additional significant effects (see Supplementary Table 3).

Discussion
The goal of the current study was to examine whether similarity in functional connectivity during rest and the dot-probe task relates to threat bias, IQ, anxiety symptoms and social reticence in children. The study generated three main findings. First, functional connectivity during rest and the dot-probe task was positively correlated. Second, levels of similarity between rest and the dot-probe task related to threat bias, but not to IQ, anxiety symptoms or socially reticent behavior. However, this finding should be interpreted with caution since the association did not remain significant when correcting for all tested designs. Finally, functional connectivity based on PPI analyses was negatively correlated between threat-congruent and threat-incongruent conditions, though such task-related modulation in connectivity did not relate to measures of threat bias nor anxiety.
Functional connectivity during rest and the dot-probe task was positively correlated, suggesting that functional hierarchies in the brain are stable (Cole et al., 2014;Smith et al., 2009). Our finding extends similar findings from previous studies focusing on cognitive tasks in adults (Cole et al., 2014;Schultz and Cole, 2016;Smith et al., 2009) to an emotional task in children. In addition, we have shown that these functional hierarchies in the brain can also be found using a resting-state ICA approach which has not been used in previous studies.
Similarity in functional connectivity during rest and the dot-probe task was positively related to threat bias, although this did not survive correction for multiple testing. This pattern of findings suggests that children with a greater bias towards threat possess greater efficiency in switching from resting state to the dot-probe task. This might further imply that, in children with threat bias, the brain during rest could Fig. 6. Functional connectivity matrix (transformed to z-statistics) with results from PPI analyses focusing on threat-congruent trials. Components with more similar time courses appear closer together. The threshold for unique colors in the dendrogram was set at 0.75. Note: numbers correspond to the 17 components (based on resting state data) from Fig. 2. already be preconfigured to a threat-bias state. This could be related to anxiety symptoms, as threat-related attentional biases are related to anxiety (Bar-Haim et al., 2007). Approximately half of the participants were drawn from infants selected for heightened risk for psychopathology, but this greater efficiency in switching from resting state to a threat-bias state was not related to anxiety. However, the participants were relatively young and had few anxiety symptoms, which typically increase in mid-adolescence. Future studies examining children with anxiety disorders are needed to clarify if children with anxiety disorders are more preconfigured to a threat-bias state and the clinical implications of these findings.
The current study demonstrates comparable effects, in terms of correlations between similarity and task performance, for an emotional task to the effects for cognitive tasks used in prior research. Importantly, in the current study, the correlation between reconfiguration efficiency and attention bias would be classified as a small-to-medium effect size, using criteria from J. Cohen (1988), and was only significant before correction for multiple testing. Similarly, Schultz and Cole (2016) detected associations between reconfiguration efficiency and measures of task performance across three cognitive tasks that also would be classified as small-to-medium. The current focus on an emotional instead of cognitive tasks may explain other differences in the observed findings between the current study and Schultz and Cole (2016). For example, Schultz and Cole (2016) related reconfiguration efficiency for cognitive tasks to both performance on cognitive tasks and IQ, the two of which were correlated (Schultz and Cole, 2016). In the current study, threat bias during the dot-probe task was not related to IQ (r ¼ 0.13, p ¼ 0.43), possibly explaining why there was no correlation between reconfiguration efficiency and IQ in the current study.
Variations in brain parcellations, such as anatomical templates or ICA-based functional networks, can influence the number of statistical tests performed in a study (Eickhoff et al., 2018). We did not use the same approach as Schultz and Cole (2016) because they used a template

Table 5
Results of the main PALM analysis with five designs testing the effect of threat bias, IQ, anxiety symptoms and social reticence on similarity of modulation by task-related connectivity in threat-congruent and threat-incongruent trials. derived from data in adults (Power et al., 2011). In addition, a paucity of time points in our data precluded partial correlation computations for 264 regions. For these reasons we performed an ICA on the resting state data to reduce the data to 17 networks. Beyond reducing the number of parameters, ICA has other key advantages: it reliably identifies resting state networks that are interpretable (Damoiseaux et al., 2006), that resemble discrete functional units (Beckmann et al., 2005), that have spatial correspondence with task-related activity , and that are heritable (Glahn et al., 2010). Group ICA, as was used here, captures spatial patterns that are consistent across participants, being less prone to subject-specific variability or artefacts than other methods. Of note, our results for threat bias associations were non-significant when applying another parcellation (Yeo et al., 2011) complemented with additional, subcortical regions derived from FreeSurfer. A final set of unique observations arose from the PPI analyses. These analyses showed that levels of event-related modulation of functional connectivity were negatively correlated for the threat-congruent and threat-incongruent trials. These two trial types represent events where the spatial location of threats and probes lie in respectively similar and opposite hemi-fields. Such negative correlations could arise from the unique psychological demands placed on children from distinct event types, i.e. maintaining versus redirecting attention. Both previous studies and our current findings reveal positive correlations between functional connectivity during rest and tasks, suggesting that task-related connectivity consists of both intrinsic, task-general, and task-specific network activity (Cole et al., 2014). The negative correlation between threat-congruent and threat-incongruent could be interpreted as reflecting the deployment of task-specific changes in network connectivity, related to the distinct attentional demands of the threat-congruent and threat-incongruent trials. By examining these task-specific changes, the current study adds to the repertoire of available tools for understanding the potential clinical significance of individual differences in functional connectivity.
This study showed that similarity in functional connectivity during rest and the dot-probe task was related to threat bias in children (although, as noted, this correlation did not survive correction for multiple testing), extending previous research on cognitive tasks in adults to an emotional task in children. Several limitations should be noted, however. First, the sample size was relatively small. Second, the sample consisted of mostly typically developing children, only half of whom were selected for risk for anxiety, thus there may have been insufficient variability to precisely estimate associations with anxiety symptoms. In addition, some participants completed the SCARED up to a year apart from the MRI scans. Although the SCARED has shown acceptable testretest reliability (Behrens et al., 2019;Birmaher et al., 1997;Boyd et al., 2003;Haley et al., 2011), these estimates of test-retest reliability mainly arise in studies using time windows that do not extend beyond six months. Future research in larger samples of anxious patients with concurrent assessment of symptoms is needed to test if similarity in functional connectivity during rest and the dot-probe task is related to anxiety symptoms. Third, fMRI during resting state and the dot-probe task was not measured on the same day in all children. Even though resting state functional networks have high reproducibility over 3.5 years, some changes can be seen over time (Choe et al., 2015;Poldrack et al., 2015). Therefore, future research should attempt to collect resting state and task data on the same day. We did find a positive correlation between functional connectivity during rest and task, so future studies should indicate if this similarity could be seen as a trait measure or how it develops over time. Fourth, we used a canonical HRF approach to regress out task activations instead of a FIR task regression. A recent paper (Cole et al., 2019) has shown that task FC estimates may be consistently and spuriously influenced by task activations, and that the FIR approach is the best way to correct for this. We were not able to apply this FIR approach because we would not have had enough degrees of freedom for the analysis. Future studies should consider the possibility of FIR modelling for the removal of task effects.
To conclude, functional connectivity during rest and the dot-probe task was positively correlated, confirming previous findings of stable functional hierarchies in the brain. Reconfiguration efficiency was related to threat bias if we did not correct for multiple testing. Children with a bias towards threat might be more prone to process threat, as they can more efficiently switch from rest to threat processing. Modulation of functional connectivity by task-evoked activity during threat-congruent and threat-incongruent trials was negatively correlated, possibly reflecting task-specific network changes during two different attentional processes. We applied a novel analytic method for combining resting state and task-related fMRI data to an emotional paradigm in children. The next step in applying this method in the field of developmental psychopathology would be to test between-group differences in a larger sample of children with and without anxiety disorders.

Declarations of competing interest
None.