A three‐wave longitudinal study of subcortical–cortical resting‐state connectivity in adolescence: Testing age‐ and puberty‐related changes

Abstract Adolescence is the transitional period between childhood and adulthood, characterized by substantial changes in reward‐driven behavior. Although reward‐driven behavior is supported by subcortical‐medial prefrontal cortex (PFC) connectivity, the development of these circuits is not well understood. Particularly, while puberty has been hypothesized to accelerate organization and activation of functional neural circuits, the relationship between age, sex, pubertal change, and functional connectivity has hardly been studied. Here, we present an analysis of resting‐state functional connectivity between subcortical structures and the medial PFC, in 661 scans of 273 participants between 8 and 29 years, using a three‐wave longitudinal design. Generalized additive mixed model procedures were used to assess the effects of age, sex, and self‐reported pubertal status on connectivity between subcortical structures (nucleus accumbens, caudate, putamen, hippocampus, and amygdala) and cortical medial structures (dorsal anterior cingulate, ventral anterior cingulate, subcallosal cortex, frontal medial cortex). We observed an age‐related strengthening of subcortico‐subcortical and cortico‐cortical connectivity. Subcortical–cortical connectivity, such as, between the nucleus accumbens—frontal medial cortex, and the caudate—dorsal anterior cingulate cortex, however, weakened across age. Model‐based comparisons revealed that for specific connections pubertal development described developmental change better than chronological age. This was particularly the case for changes in subcortical–cortical connectivity and distinctively for boys and girls. Together, these findings indicate changes in functional network strengthening with pubertal development. These changes in functional connectivity may maximize the neural efficiency of interregional communication and set the stage for further inquiry of biological factors driving adolescent functional connectivity changes.

Generalized additive mixed model procedures were used to assess the effects of age, sex, and self-reported pubertal status on connectivity between subcortical structures (nucleus accumbens, caudate, putamen, hippocampus, and amygdala) and cortical medial structures (dorsal anterior cingulate, ventral anterior cingulate, subcallosal cortex, frontal medial cortex). We observed an age-related strengthening of subcortico-subcortical and cortico-cortical connectivity. Subcortical-cortical connectivity, such as, between the nucleus accumbens-frontal medial cortex, and the caudate-dorsal anterior cingulate cortex, however, weakened across age. Modelbased comparisons revealed that for specific connections pubertal development described developmental change better than chronological age. This was particularly the case for changes in subcortical-cortical connectivity and distinctively for boys and girls. Together, these findings indicate changes in functional network strengthening with pubertal development. These changes in functional connectivity may maximize the neural efficiency of interregional communication and set the stage for further inquiry of biological factors driving adolescent functional connectivity changes.

K E Y W O R D S
adolescence, functional connectivity, longitudinal, pubertal development, resting-state 1 | INTRODUCTION Adolescence is a transitional period linking childhood and adulthood, and is accompanied by long-lasting, largely asynchronous brain changes in both cortical and subcortical brain regions. It is particularly relevant to consider these transformations not only in relation to neural structural or localized activation changes, but also in relation to functional connectivity changes in the adolescent brain (Casey, 2015;Crone & Dahl, 2012;Stevens, 2016). Relative to studies examining structural (Tamnes et al., 2017) and functional (Telzer et al., 2018) brain development, very few studies have examined longitudinal functional connectivity changes. The goal of this study was therefore to analyze within and between subcortical-cortical connectivity in participants ages 8-29 years, using a three-wave longitudinal design covering 5 years for each individual.
Central to adolescent brain development is a change in the neural motivational circuitry (Doremus-Fitzwater & Spear, 2016;Ernst, 2014;Telzer, 2016;, which may lead to an increased drive for reward and enhanced affective responses during adolescence, and may create vulnerabilities for developing psychopathology (Paus, Keshavan, & Giedd, 2008). The ventral striatum, and particularly the nucleus accumbens, is considered a key structure for reward processing. This structure is extensively connected to both cortical and subcortical structures supporting motivated behavior (Alexander, Crutcher, & DeLong, 1990;Haber & Knutson, 2010), through looped cortical-subcortical connections. The medial prefrontal cortex (PFC) is densely connected to the ventral striatum and suggested to be a crucial regulator of reward-directed behavior. In addition, the amygdala and hippocampus are also central regions for respectively affective processing (e.g., Scherf, Smyth, & Delgado, 2013), processing aversive stimuli (Ernst, 2014), and memory (e.g., Davidow, Foerde, Galván, & Shohamy, 2016) in adolescents, and are often coactivated with the medial PFC. Thus, the ventral striatum together with nuclei in the amygdala, parts of the hippocampus, and the medial PFC, form a larger circuitry that modulate responses to salient stimuli and drive reward learning and decision-making.
This study examined the functional coupling between key reward regions using resting-state (RS) functional magnetic resonance imaging (fMRI), which provides an important framework for investigating functional systems in the organization of the adolescent developing brain (Ernst, Torrisi, Balderston, Grillon, & Hale, 2015;Uddin, Supekar, & Menon, 2010), considering the minimal experimental demands. Previous RS studies already observed developmental changes in functional connectivity between subcortical regions and prefrontal circuitry. Whereas the functional coupling between the amygdala and medial PFC has been found to increase (Gabard-Durnam et al., 2014) or show minimal changes (Peters, Peper, Duijvenvoorde, Braams, & Crone, 2016) across adolescence, studies also found a developmental decrease in connectivity strength for other subcortical-cortical connections, such as connectivity between the ventral striatum and PFC (Fareri et al., 2015;Padmanabhan, Lynn, Foran, Luna, & O'Hearn, 2013;Porter et al., 2015;van Duijvenvoorde, Achterberg, Braams, Peters, & Crone, 2016). This decrease in functional coupling between subcortical and prefrontal circuitries has been interpreted as a maturation of brain networks, and linked to a developmental decrease in risky behavior and reward valuation (van Duijvenvoorde, Achterberg, et al., 2016) across adolescence, but also to individual differences in risky behavior. For instance, greater functional coupling between the ventral striatum and PFC has been related to an earlier onset of substance use in adolescence (Weissman et al., 2015), and a family history of alcoholism (Cservenka, Casimo, Fair, & Nagel, 2014). However, these studies used different age samples, cross-sectional designs, and focused on single connections. Therefore, it remains to be determined how functional connectivity changes within and between several subcortical-cortical connections. Moreover, it has often been assumed that heightened subcortical reactivity is related to pubertal onset (Braams, van Duijvenvoorde, Peper, & Crone, 2015;Pfeifer et al., 2011), and pubertal development is suggested to be the maturational process driving developmental changes in reward regions, accelerating typical developing trajectories (Blakemore, Burnett, & Dahl, 2010;Crone & Dahl, 2012;Schulz & Sisk, 2016;Vijayakumar, Op de Macks, Shirtcliff, & Pfeifer, 2018). Only a handful of studies tested the influence of pubertal development on subcortical-cortical functional connectivity (Fareri et al., 2015;Peters, Jolles, van Duijvenvoorde, Crone, & Peper, 2015). These studies highlighted that higher pubertal hormone concentrations were linked to a decrease in subcortical-prefrontal connectivity strength as seen across typical adolescent development. However, to date, no study has examined the relative contributions of age and puberty on functional connectivity changes.
In sum, a reorganization of subcortical-cortical circuitry in adolescence is integral to adolescent development. However, few studies have yet examined connectivity in regions of the adolescent reward circuitry in a comprehensive maturational perspective. Here, we related subcorticalcortical circuitry to age and pubertal development in a three-wave longitudinal sample (8-29 years). Longitudinal accelerated designs consider individual trajectories, thereby allowing for a more accurate estimate of developmental change. RS functional connectivity was examined between a set of subcortical and cortical structures of interest, which included the ventral and dorsal striatum, putamen, hippocampus, amygdala, and all atlas-based anatomical regions of the medial PFC (subcallosal cortex, ventral medial PFC, anterior cingulate cortex [ACC]). We expected a decoupling between subcortical and medial PFC regions with age, which may be particularly driven by pubertal-related changes. Given the large sample size and intensive longitudinal measurements, this study also allowed us to examine sex differences and age by sex interactions in RS connectivity change.

| Participants
The current study was part of BrainTime, a longitudinal study from Leiden University, Leiden, the Netherlands. Participants were recruited through local schools and advertisements and provided written informed consent for the study at every time point (participant assent and parental consent in case of minors). Participants were screened for MRI contraindications and had no neurological or psychiatric disorders at time point 1 (T1). All anatomical MRI scans were reviewed by a radiologist and no anomalous findings at any of the time points were reported. At each time point, participants received an endowment for participation in a larger scale study. The RS data presented here were collected as the first scan of the BrainTime experimental protocol examining affective and cognitive development via the use of task-based functional neuroimaging. The study and its procedures were approved by the institutional review board of the Leiden University Medical Center. Cross-sectional RS analyses have previously been reported in van Duijvenvoorde, Achterberg, et al. (2016) for nucleus accumbens-prefrontal connectivity, and two data waves have previously been reported in Peters, Peper, et al. (2016)  Exclusion from further analyses occurred due to a number of reasons. First, participants were excluded when either the RS scan, highresolution scan, or T1-weighted anatomical scan was missing or failed due to technical errors (T1: n = 5, T2: n = 2; T3: n = 4). Second, participants were excluded from all time points if they were diagnosed with a neurological or psychiatric disorder (e.g., depression, Attention-Deficit Hyperactivity Disorder, Attention-Deficit Disorder, anxiety disorder) at T2 and/or T3 (n = 21). Third, participants were excluded when excessive head motion was detected (T1: n = 38; T2: n = 23; T3: n = 10). Motion exclusion was based on having ≥2 mm translation or more than 2 rotation in any direction, having ≥10 volumes (with more than 0.5 mm movement between two frames (framewise displacement, FD, Power et al., 2014), and/or having ≥10 volumes that are reference RMS outliers (i.e., Root mean square intensity difference of volume N to the reference volume, exceeding the threshold of 75th percentile + 1.5 × interquartile range). FD and reference RMS outliers were established using the motion outlier tool implemented in FMRIB Software Library (FSL) version 5.0.4 (http://fsl.fmrib.ox.ac.uk/fsl/ fslwiki/, Smith et al., 2004).
Intelligence quotient (IQ) was estimated at T1 using the subsets "similarities" and "block design" and at T2 using the subsets "Vocabulary" and "Picture Completion" of the Wechsler Intelligence Scale for Adults (WAIS-III) or the Wechsler Intelligence Scale for Children, third edition (WISC-III; Wechsler, 1974). All estimated IQ scores were in the normal range on T1 (M IQ = 109.9, SD IQ = 10.7, range = 80-143) and T2 (M IQ = 108.4, SD IQ = 10.3, range = 80-148) and were not significantly related to age at either time point (all ps > .4) for included subjects.

| Pubertal stage
Stage of physical pubertal maturation was assessed at each time point with the Pubertal Development Scale (PDS) for participants under 18 years of age (Petersen et al. 1988). This self-report questionnaire contains questions concerning secondary sexual characteristics. The

| MRI data acquisition
Neuroimaging was conducted using a 3.0 T Philips Achieva MRI scanner with a standard whole-head coil. The same scanner and settings were used for all participants and at all three time points. The RS scans were acquired as the first scan of a fixed-order imaging protocol, with

| fMRI data preprocessing
The RS functional data were preprocessed using FEAT (fMRI Expert Analysis Tool; v6.00), part of FSL (Smith et al., 2004). Preprocessing of the RS data included motion correction (MCFLIRT; Jenkinson, Bannister, Brady, & Smith, 2002), slice timing correction (regular down), brain extraction (BET), spatial smoothing with a 5 mm full-width-at-halfmaximum Gaussian kernel, and high-pass temporal filtering with a cutoff point of 100 s. The high-resolution EPI images and T1-weighted anatomical images were brain-extracted (BET). Next, the RS fMRI scans of an individual were registered to the corresponding highresolution EPI image (6 DOF), which in turn were registered to the T1-weighted anatomical image using the integrated version of boundary-based registration to improve the accuracy of functional-tostructural space registration. Finally, the images were registered to standard MNI-152 space using FNIRT (FMRIB's Nonlinear Imaging Registration Tool; 12 DOF, warp resolution 10 mm).

| Motion correction
Head motion is undesirable in all fMRI studies (e.g., Friston, Williams, Howard, Frackowiak, & Turner, 1996), and especially so for RS stud- Developmental samples are particularly susceptible for this confound, given that head motion is highly related to subject age (Satterthwaite et al., 2013. To minimize motion, subjects were trained with a mock-scanning procedure, were reminded several times during the session not to move during scanning, and head motion was restricted using foam padding. We applied a strict exclusion criterion (see Section 2.1) based on absolute motion, an FD cutoff of 0.5 mm on ≥10 volumes, and/or reference RMS outliers on ≥10 volumes. Although the mean FD was relatively low on each time point (see Table 1 Table 1). Note that for these cleaned time series, all included participants adhered even to a more stringent FD cutoff of 0.2 mm on ≥2 volumes.

| Nuisance signal regression
In addition to motion artifacts, signals from white matter (WM) and cerebrospinal fluid (CSF) can be confounding effects that result in overestimated RS connectivity strength. These signals primarily reflect noise from non-neural origin (e.g., scanner instabilities, physiological effects) and are largely independent from Blood Oxygenation Level-Dependent (BOLD) signal fluctuations in gray matter (Windischberger et al., 2002). Global signal was also removed from the time series to reduce influence of artifacts caused by physiological processes (i.e., cardiac and respiratory fluctuations), vigilance level (Liu, Nalci, & Falahpour, 2017), and scanner drifts (Fox & Raichle, 2007).
WM and CSF masks were obtained using FAST (FMRIB's Automated Segmentation Tool), which segments the T1-weighted anatomical scan into different tissue types (WM, CSF, and gray matter).
These maps were then FLIRT-based transformed into functional sub-

| Regions of interest
Regions of interest (ROIs) were selected from the Harvard-Oxford probabilistic anatomical brain atlas (subcortical and cortical) in FSL, with a thresholded probability of ≥0.5. Although based on anatomical parcellation, the Harvard-Oxford atlas is an often used and well-known atlas in functional brain analyses. Given our focus on age-related change in connectivity between frontal midline and subcortical structures, we included four cortical midline structures in this atlas (subcallosal cortex, frontal medial cortex, and cingulate gyrus anterior division) that spanned a ventral to dorsal cortical midline. Considering the extent and functional specificity of the ACC, this anatomical structure was divided in a more posterior-dorsal and a more anterior-ventral part with a cutoff of y = 30 based on Bush, Luu, and Posner (2000) (see for a similar segregation Achterberg et al., 2018). We included five bilateral subcortical ROIs: nucleus accumbens, caudate, putamen, amygdala, and hippocampus. For each subcortical region, bilateral masks were combined into one ROI for further analyses. ROIs are visualized in Figure 1. For each participant, the ROIs were transformed to subject space and the mean individual RS time series were extracted from each ROI separately.

| Experimental design and statistical analysis
Age-related change in longitudinal data sets is often assessed with polynomial growth models including age as linear, quadratic, or cubic regressor, while controlling for the repeated nature of the data. However, a limitation of these models is that they assume age-related changes follow this restricted set of growth models. Additionally, these models may not be optimal to compare groups that show different developmental trajectories (Vijayakumar, Op de Macks, et al., 2018). Thus, we used a distinct class of models called generalized additive mixed models (GAMMs) to characterize age-and sex effects, pubertal-related effects, and behavioral effects on RS functional connectivity. All models were run using the mgcv package (Wood, 2011) in R (R Core Team, 2017; https://www.r-project.org/  (Wood, 2017) and p-values (Wood, 2013). Moreover, GAMM models are well suited for our developmental sample and accelerated longitudinal design, as this model accounts for within-subject dependence and differences in developmental time points at which participants entered the study (Alexander-Bloch et al., 2014;Harezlak, Ryan, Giedd, & Lange, 2005;Wierenga et al., 2018).
Our research questions followed the following model-fit procedures. First, to assess age-and sex-related change in functional connectivity we used a model-building procedure assessing (a) the developmental age-related pattern over the whole group; (b) a main effect of sex; and (c) differences in developmental trajectories between sexes. These models were compared to test which model provided the best fit for each connection. In short, first a simple age model of formula (1)  Results showed that PDS score was described by a main effect of sex and spline effect of age. That is, pubertal development manifested differently for boys and girls and increased with age (see Figure 2).
To assess effects of puberty on RS connectivity, models were assessed separately for boys and girls, given that puberty had different timings in males and females (see Wierenga et al., 2018 for an example on structural brain development and pubertal development).
We first examined significant effects of PDS by creating a model including a smooth PDS term (Equation 5).
Testing the robustness of these PDS findings, we compared the BIC The p-values of fixed effects in all best-fitting models were corrected for multiple comparisons with a Bonferroni-Holm correction and evaluated at p < .05.
As a measure of homogeneity of the data, we determined the intraclass correlations (ICCs) for each RS connection. ICCs were computed by estimating a null model (Equation 8) with maximum likelihood across all data points and dividing the variance in intercept by the sum of the variance in intercept and residual variance. ICCs are listed in Table 2 and range from very poor (<.1) to poor (<.41) (Cicchetti, 1994

| Age-related change in RS functional connectivity
We assessed age-related changes in RS connectivity using GAMM (see Table 3). After correction for multiple comparisons (Bonferroni-Holm), significant positive effects of age were observed for subcorticalsubcortical connectivity between the amygdala, putamen, and hippocampus, between the putamen, nucleus accumbens, and hippocampus, and between the hippocampus and nucleus accumbens (Figure 4a), all showing a strengthening of functional connectivity with age. Cortico-cortical connections also showed a significant strengthening of connectivity with age between the frontal medial cortex-ventral ACC, frontal medial cortex-dorsal ACC, and between the subcallosal cortex-ventral ACC ( Figure 4b). Age-related strengthening of connectivity was observed in subcortical-cortical connectivity, between the hippocampus-subcallosal cortex, caudate-subcallosal cortex, and nucleus accumbens-ventral ACC ( Figure 4c). Finally, age-related decreases were observed only for subcortical-cortical connectivity (Figure 4d), particularly between the caudate-dorsal ACC, hippocampus-dorsal ACC, nucleus accumbensfrontal medial cortex, nucleus accumbens-subcallosal cortex, and putamen-frontal medial cortex. Visualization of significant age splines showed a relatively linear developmental pattern for all connections, with a few subcortical-cortical connections leveling off in adolescence or young adulthood (see Figure 4).
For two connections, the best-fitting model included a main effect of sex. That is, functional connectivity between hippocampus-amygdala increased with age and was greater for males than females (see Figure 4 plotted in red/blue, and Table 3). Functional connectivity between the hippocampus-ventral ACC was not dependent on age, but was greater for females than males. None of the models showed a best fit for model (3) including age by sex interaction terms.

| Pubertal development changes in RS functional connectivity
To assess pubertal developmental changes in RS functional connectivity, a set of GAMMs were run, only including participants between ages 8 and 18-years old, during which PDS changes are most pronounced. Given the difference in pubertal timing between sex, all models were run separately for boys and girls (see for a similar approach Wierenga et al., 2018).
For boys, subcortical-cortical connectivity between the nucleus accumbens-frontal medial cortex and putamen-frontal medial cortex related negatively to PDS scores, with increasing PDS significantly decreasing RS connectivity (similarly to what was observed for age) (see Figure 5; Table 4). Additionally, connectivity between the hippocampusamygdala (similar to age effects) and subcallosal cortex-dorsal ACC (not observed for age) was significantly positively related to PDS score, with increasing PDS being associated with increased RS connectivity. The nucleus accumbens-frontal medial cortex and subcallosal cortex-dorsal ACC connectivity showed a better fit for a PDS-only than an age-only model when comparing BICs. When including age at baseline and PDS in the same model, only the hippocampus-amygdala connectivity showed a significant effect of PDS on RS connectivity over and above age at baseline (p = .002).
For girls, subcortical-cortical connectivity between the hippocampusdorsal ACC (also observed for age), caudate-dorsal ACC (also observed for age), and putamen-frontal medial cortex (also observed for age) was significantly negatively related to PDS scores, with increasing PDS related to decreased RS connectivity (see Figure 5; factor for nonlinear age patterns (Crone & Dahl, 2012;Ladouceur, 2012;Pfeifer et al., 2011), an important question was whether development would be better described by pubertal development than age.
For this purpose, we made use of a three-wave accelerated longitudinal data set with boys and girls across different stages of pubertal development, who completed RS scans at each time point.
We observed several key findings. First, a quarter of the subcorticalcortical connections we investigated weakened with age. The decrease in subcortical-cortical connectivity was unique, given that functional connectivity between subcortical and between cortical regions only strengthened with age. Second, for several key connections the decrease in subcortical-cortical connectivity was better described by pubertal F I G U R E 4 Spaghetti plots indicating a significant fitted line of age on top of the raw longitudinal data for (a) cortico-cortical connections (b) and subcortical-cortical connections (c,d). Location of region of interest (ROI) is indicated schematically as dots, visualized with the BrainNet viewer (Xia, Wang, & He, 2013, http://www.nitrc.org/projects/bnv/). Green lines between ROIs indicate age-related increases, and red lines indicate age-related decreases. If there was a sex difference in functional connectivity development, males are plotted in blue and females in red in spaghetti plots (only the case for hippocampus-amygdala connectivity) [Color figure can be viewed at wileyonlinelibrary.com] development than age, suggesting that puberty may be one of the mechanisms that initiate change in subcortical-cortical development. The discussion is organized alongside these main findings.

| Age-related changes in functional connectivity
To examine functional connectivity patterns in adolescence, we specified regions as part of subcortical and medial cortical reward regions typically implicated in functional reward processing and motivated behavior (Doremus-Fitzwater & Spear, 2016;Haber & Knutson, 2010;Telzer, 2016;van Duijvenvoorde, Achterberg, et al., 2016). Our results showed, first, that connectivity patterns were highly stable across time points at the group level, with excellent within-sample replication of positive and negative connectivity patterns across three time points. Second, the within-individual stability was relatively low, with ICCs ranging between 0.08 and 0.33, although most being at least above .1 (e.g., Ordaz, Foran, Velanova, & Luna, 2013). Compared to fMRI ICCs, these values are comparable for neural activity in subcortical brain regions (Braams et al., 2015;Herting, Gautam, Chen, Mezher, & Vetter, 2017;Schreuders et al., 2018) and highlight that patterns were more consistent at the group than the individual level. One interpretation of low test-retest reliability (such as the ICC) is that there is a poor consistency of functional connectivity. However, for studies with relatively longer delays between time points and younger populations, a low ICC may also reflect development over time . To distinguish which of these effects contributes most to these ICC results, future studies should further examine the test-retest reliability of RS scans in developmental populations, preferably also including shorter durations between scans.
A main goal in this study was whether RS functional connectivity changed over the course of adolescent development. Results showed that the patterns of change were dependent on the specific connection that was studied. More specifically, we observed an age-related decline in functional connectivity between the nucleus accumbens, the putamen, and more ventral regions of the medial PFC, (i.e., subcallosal medial cortex and frontal medial cortex). These findings fit well with prior cross-sectional reports (Fareri et al., 2015;van Duijvenvoorde, Achterberg, et al., 2016), and have been interpreted as a more independent functioning of these networks involved in affective-motivational processes. Decreases in functional connectivity with age were also observed for structures such as the caudate with the dorsal ACC, which concurs with a more dorsal-to-ventral divide of striatal connectivity (Di Martino et al., 2008;Porter et al., 2015). T A B L E 4 Generalized additive mixed models examining effects of PDS and age for males. For each connection, model fits are shown for (1) PDS-only models (2)  within cortical regions across development. Although GAMM is a descriptive test of age effects, inspection of the best-fitting age curves ( Figure 4) from our models indicates that for specific connections developmental changes in connectivity follow a nonlinear pattern, with changes leveling off in early adulthood. This "leveling off" was particularly pronounced for decreases in connectivity between the putamen-frontal medial cortex, caudate-dorsal ACC, and hippocampus-dorsal ACC. This may suggest most prominent changes in adolescence within this decoupling of these functional networks.
Together, these findings may be interpreted as change in local versus distributed networks in which specific functional networks both strengthen and weaken across adolescence. Previous work also showed that structural networks become more segregated across development (Baum, Ciric, Roalf, Betzel, et al., 2017), and that BOLD dimensionality decreases with age (Kundu et al., 2018). One suggested interpretation is that these changes may maximize the neural efficiency of interregional communication (Stevens, 2016). Future studies may complement our RS functional connectivity findings by using more explorative analyses such as graph-theory models together with age-and pubertal development, allowing examination of metrics of brain organization and neural efficiency on a whole-brain basis.

| Age-versus puberty-related changes in functional connectivity
Previous studies focusing on functional reactivity have often suggested that pubertal development may advance or enhance growth trajectories of brain development, although most studies up to now focused on brain structural development (e.g., Herting & Sowell, 2017;Vijayakumar, Mills, Alexander-Bloch, Tamnes, & Whittle, 2018) or cross-sectional functional connectivity development (Fareri et al., 2015). Here, we observed that a number of subcortical-cortical connections sensitive to developmental change were better described by (self-reported) pubertal development than age. That is, in boys we observed that the decrease in connectivity between the nucleus accumbens with the frontal medial cortex was better described by Note that when testing sex differences, we observed only few differences in functional connectivity between boys and girls. That is, amygdala-hippocampus connectivity, a connection most prominently influenced by sex, was stronger for boys than for girls, and hippocampal-ventral ACC connectivity was stronger for girls than for boys. Previous research on sex differences has mostly been done in adults (e.g., Alarcón, Cservenka, Rudolph, Fair, & Nagel, 2015;Kogler et al., 2016), and has shown higher hippocampal and/or amygdala connectivity in females than males, possibly related to their better memory performance (Gur & Gur, 2016). The sex differences we observed here may be an interesting starting point, but should be interpreted with caution until replicated, given the lack of consistent findings in prior developmental samples.

| Limitations
RS connectivity is inherently susceptible to effects of motion, which can have a marked influence on developmental findings. In the current study, we have taken steps to account for such possible confounds (see Satterthwaite et al., 2017). Specifically, we first excluded people above our set motion threshold. Further, we included realignment parameters, tissue-specific signals, global signal regression, and a denoising procedure based on an independent components analyses (FSL FIX). One possible concern is that this latter denoising procedure has mainly been applied in adult populations. However, the use of these denoising techniques may be especially helpful in cleaning relatively noisy data from such specific populations, and can improve signal and analysis quality. Applying control analyses on motion confounds in our cleaned time series indicated that FD was minimalized across all time points, supporting its use in the current data set. Another concern is the inclusion of global-signal regression. Regressing out global signal may reduce noise from physiological measures such as heart rate and respiration (Chen et al., 2012;Power et al., 2015) and differences in vigilance and arousal (Liu et al., 2017), yet it has been found to induce negative correlations and spurious results (Satterthwaite et al., 2013).
A second limitation of our analyses may be that we used a set of anatomical ROIs based on the Harvard-Oxford atlas. Using anatomical ROIs versus a functional brain atlas (such as the areal atlas of Power et al., 2011) may be less sensitive in detecting age-(or puberty) related change. Parcellation studies in adults have identified that structural atlases suffer from lower homogeneity than functional parcellations (Craddock, James, Holtzheimer, Hu, & Mayberg, 2012;Gordon et al., 2014). This is particularly so for atlases that use large structural regions such as the AAL (Gordon et al., 2014), while more fine-grained structural atlases such as Brodmann areas seem to perform better. The advantage of structural atlases is that they are highly standardized, and used typically in both developmental functional imaging studies (e.g., Achterberg et al., 2018) and RS studies (e.g., Fareri et al., 2015;Stevens, 2016). Here, we used an anatomical atlas that may allow for more specific cortical regions by choosing a standardized structural probabilistic atlas, namely the Harvard-Oxford atlas. Nonetheless, future studies should consider comparing these functional and structural approaches in a developmental perspective.

| CONCLUSION
The current study used RS functional connectivity in a large longitudinal sample to understand developmental changes in connectivity between and within subcortical and medial prefrontal regions. These