University of Amsterdam
Graduate School of Psychology
R
ESEARCH
M
ASTER
’
S
P
SYCHOLOGY
–
T
HESIS
Age-related changes in subcortico-subcortical and cortico-subcortical
resting-state functional connectivity across adolescence:
a longitudinal study
Thesis – Final Version
Date: May 24, 2016
Student
Name
B. (Bianca) Westhoff
Student ID number
10000662
Supervisors
Within Research Master
prof. dr. M.E.J. (Maartje) Raijmakers
Specialisation
Developmental Psychology
External supervisor
dr. A.C.K. (Anna) van Duijvenvoorde
Second assessor
dr. I. (Ingmar) Visser
Page 2 of 28
Abstract
Development of the adolescent brain and behaviour has been described by the imbalance model; it has an integrated circuit-based perspective and suggests changing connections within and between cortical and subcortical networks, which are interactive rather than orthogonal. A powerful approach to study development of functional connectivity, independent of task-demands, is resting-state functional connectivity (rsFC). In this study, we aimed to investigate the developmental trajectories (i.e., age-related changes; linear, quadratic, or cubic) of rsFC strength (1) within a subcortical network including the nucleus accumbens, caudate, amygdala, and hippocampus, (SUB-SUB), and (2) between these subcortical and cortical regions, particularly the insula, ventral medial PFC (vmPFC), and ACC (CORT-SUB); we collected data in a large sample with a wide continuous age range (N=255, 8-28 years) on three time points, separated by two years. We hypothesized (1) a peak in SUB-SUB rsFC strength in adolescence, indicating changed affective influences during adolescence, and (2) monotonic age-related changes in CORT-SUB rsFC, indicating increased cognitive control over time. We observed linear rsFC increases between amygdala and hippocampus (SUB-SUB). The vmPFC showed linear rsFC increases with the caudate and hippocampus, and the insula-amygdala rsFC also increased linearly across adolescence (CORT-SUB). Although we observed age-related rsFC changes, we also observed large between-subject differences. Future studies should look into which (trait like / developmentally stable) factors explain between-subject variability in rsFC.
Page 3 of 28
1. Introduction
Adolescence is a period of change, in behaviour as well as neural systems (Somerville, Jones, & Casey, 2010). We now know the brain develops and matures much longer than previously thought – well into our early twenties (Sowell, Thompson, Holmes, Jernigan, & Toga, 1999) The astonishing changes the brain undergoes throughout development, coincide with changes in affect and cognition across adolescence (Khundrakpam, Lewis, Zhao, Chouinard-Decorte, & Evans, 2016). For instance, adolescence is seen as a period of increased sensation seeking, risk-taking, and reward sensitivity, behaviours that can lead to negative outcomes such as substance abuse, injuries and death
(Somerville et al., 2010; Spear, 2000; Steinberg, 2008; Tottenham et al., 2009). Yet, it is also a time when mood and mental disorders show peak prevalence.
Over the years several neurobiological models for understanding adolescent behaviour have emerged (Casey, 2015). For example the influential dual-systems model, suggesting that the subcortical system (‘affective/reward system’, e.g., nucleus accumbens (NAcc)) matures at a faster pace than cortical prefrontal regions (‘cognitive control system’) (Casey, Jones, & Somerville, 2011; Somerville et al., 2010; Steinberg et al., 2008). As such, the subcortical affective system is relatively hyperactive, causing a period of increased risk-taking and reward sensitivity during adolescence. Indeed, neuroimaging studies have repeatedly confirmed the subcortical hyperactivity (typically NAcc and amygdala) in adolescents compared to children and adults (Hare et al., 2008; Richards, Plate, & Ernst, 2013; Somerville, Hare, & Casey, 2011). The imbalance model on the other hand, proposes that the affective and cognitive systems are not orthogonal but rather dynamic and interactive (Casey, Galván, & Somerville, 2016; Casey, 2015). That is, changes in behaviour during adolescence are explained by changing connections within and between networks, rather than changes in (activation of) specific brain regions. The specific trajectories (e.g., monotonic or quadratic increases) are, however, not specified by the model. A recently proposed plausible mechanism for the shift in balance between control and affect from childhood to adulthood is a fine-tuning of circuits, serially from subcortico-subcortical (SUB-SUB) to cortico-subcortical (CORT-SUB) to cortico-cortical (CORT-CORT) connectivity (Casey et al., 2016). Specifically, they propose a hierarchical fine-tuning where first local SUB circuitry develops in early adolescence; once SUB-SUB signaling becomes robust, it provokes strengthening of distal CORT-SUB-SUB circuits (late adolescence). This SUB development in turn serves as a building block for successive CORT-CORT functional development in young adulthood. The changes in connectivity strength across development would parallel the dynamic changes in behaviour as observed in adolescence.
In line with the dual-systems model, so far most neuroimaging studies investigating adolescent behaviour and brain development, have mainly focused on changes in functional
Page 4 of 28 activation of either subcortical or cortical brain regions across age (Richards et al., 2013). However, as supported by the imbalance model, it is also important to study the development of functional connectivity (FC) as it yields additional information about the functioning of the brain by revealing which networks communicate and to what extent (Casey, 2015). Some studies have looked into FC by performing task-based FC analyses for the subcortical and prefrontal control regions. For example, Somerville et al., (2011) showed that FC between the ventral striatum (VS) and dorsal striatum (DS) strengthened in adolescence compared to childhood or adulthood, and that prefrontal-striatal FC increases linearly across age groups. These age-related regional changes in circuit connectivity were associated with developmental differences in behaviour (i.e., diminished impulse control in adolescence). Furthermore, the study by Hare et al., (2008), found that participants with poorer amygdala habituation had weaker FC between the amygdala and the ventral prefrontal cortex (PFC), which was interpreted as decreased top-down downregulation of emotional responses by the PFC. Taken together, these examples demonstrate that subcortical and prefrontal cortical brain regions are functionally connected (i.e., share and integrate information during task-performance) and that FC between those regions changes across development.
Thus, there is evidence that age-related changes of SUB-SUB and CORT-SUB FC manifest across childhood into adulthood, conforming the imbalance model. Yet, these studies have only tested developmental FC changes in a limited set of brain regions, and it has moreover been shown that FC differs as a function of specific task-based contexts (Cho et al., 2013; Richards et al., 2013). To overcome these problems, one can study the FC of different brain regions in the absence of task demands by using resting-state functional magnetic resonance imaging (rs-fMRI) (Ernst, Torrisi, Balderston, Grillon, & Hale, 2014; Satterthwaite & Baker, 2015). This tool measures the intrinsic functional connectivity, or resting-state functional connectivity (rsFC), between brain regions by assessing synchronous, spontaneous low frequency fluctuations in the blood oxygen level dependent (BOLD) signal (Biswal, Yetkin, Haughton, & Hyde, 1995; Ernst et al., 2014; Van Duijvenvoorde, Achterberg, Braams, Peters, & Crone, 2015). It should be noted that regions with strong rsFC are not necessarily structurally connected as well, and vice versa. rs-fMRI is non-invasive and involves minimal instruction and low attention demands as the subjects are awake but resting, which is especially beneficial when studying children (Ernst et al., 2014; Satterthwaite & Baker, 2015). As such, rs-fMRI is a powerful approach to study age-related changes in neural network organization.
The development of rsFC has been investigated in a few cross-sectional studies with typically developing adolescents (Uddin, Supekar, & Menon, 2010; Van Duijvenvoorde et al., 2015). For example, Van Duijvenvoorde et al. (2015) used rs-fMRI in individuals between 8 to 25-years, in which
Page 5 of 28 they tracked whole-brain rsFC from a DLPFC seed and a NAcc seed. Results showed linear changes in rsFC within and across networks associated with cognitive control (DLPFC) and affective-motivational processes (NAcc), which were associated with behavioural measures (i.e., learning performance and winning pleasure). Furthermore, a cross-sectional study by Fareri et al. (2015) tested whole-brain rsFC from a VS seed in individuals between 4 and 23-years. Results showed adolescent-specific decreases in rsFC between the VS and posterior hippocampus (SUB-SUB), and between the VS and mid/posterior insula (CORT-SUB), whereas rsFC between the VS and medial PFC (CORT-SUB) increased linearly with age. Thus, these studies demonstrated age-effects on SUB-SUB and CORT-SUB FC across adolescence, supporting the integrated circuit-based perspective of the imbalance model. However, no study to date combined rsFC between different subcortical structures and their rsFC to PFC regions in a comprehensive study.
Moreover, to study developmental trajectories of rsFC, a longitudinal design is crucial because, considering individual differences, the individual developmental trajectories yield more information about change compared to single time points (i.e., cross-sectional designs) (Braams, van Duijvenvoorde, Peper, & Crone, 2015; Crone & Elzinga, 2015). That is, a longitudinal design for testing development is imperative, since (1) it profits increased power to detect group-level changes, and (2) allows for testing changes on the intra-individual level (Mills & Tamnes, 2014). However, to our knowledge no studies have investigated development of rsFC with a longitudinal design (Ernst et al., 2014). Hence, what is currently unknown are developmental trajectories (i.e., age-related changes) of rsFC in SUB-SUB, CORT-SUB and CORT-CORT circuits across childhood and adolescence, i.e, no studies to date have comprehensively tested the imbalance model.
In the current study, we therefore aimed to investigate the age-related development of rsFC strength (1) within a subcortical network including the NAcc, caudate nucleus, amygdala, and hippocampus (SUB-SUB), and (2) between these subcortical and cortical regions, particularly the insula, vmPFC and ACC (CORT-SUB), in a wide age range (8-28 years; N=619 scans) by collecting data on three time points (T1,T2,T3; separated by two years). This selection of brain regions is based on the dual-systems model, imbalance model, and previous studies on the adolescent brain (e.g., Casey, Jones, & Hare, 2008; Fareri et al., 2015; Somerville et al., 2010; Van Duijvenvoorde et al., 2015). In accordance with the imbalance and dual-systems model, we hypothesized age-related changes in SUB-SUB rsFC strength, with a peak in adolescence, indicating a changed involvement in affective influences during adolescence. The direction of age-related changes (i.e., negative or positive coupling) of SUB-SUB and CORT-SUB networks across adolescence, is not yet specified in earlier rsFC studies. However, there are indications for both a heightened rsFC between subcortical structures
Page 6 of 28 particularly during adolescence. Regarding the CORT-SUB rsFC, we hypothesized that the subcortical regions generally show monotonic age-related increases in rsFC with cortical regions, indicating a relative increase in cognitive control during adolescence.
2. Methods
2.1 Participants
The current study was part of a large longitudinal study - referred to as Braintime - conducted at Leiden University, The Netherlands. Participants were recruited through local schools and advertisements, and provided written informed consent for the study (for participants <18 years: parental consent and participant assent) at every time point. Participants were screened for MRI contra indications and had no (history of) neurological or psychiatric disorders (at T1; however, 38 participants had been diagnosed on a later time point). All anatomical MRI scans were reviewed by a radiologist from the radiology department of the Leiden University Medical Center, who did not report anomalous findings. On each time point, participants received an endowment for participation in a larger scale study (≥18 years: €60; 12-17 years: €30; <12 years: €20). The study and its procedures were approved by institutional review board of the Leiden University Medical Centre.
At T1, MRI data were collected from 299 participants (Meanage = 14.15 years; SDage = 3.56;
Rangeage = 8.01–25.95 years; 143 males), who were invited for T2 approximately 2 years later from
T1 (Meantime-difference = 1.99 years; SDtime-difference = 0.10; Rangetime-difference = 1.66-2.47 years). Thirteen
participants indicated they could not or did not want participate again, and due to braces another 32 participants could not participate in the MRI session. Hence, at T2 MRI data were collected from 254 participants (Meanage = 16.11 years; SDage = 3.66; Rangeage = 9.92-26.62 years; 122 males). All
participants were invited for T3, approximately 2 years from T2 (Meantime-difference = 2.02 years; SD time-difference = 0.09; Rangetime-difference = 1.62-2.35 years). In total, 56 participants did not want to or could
not participate in the MRI session (e.g., due to braces), hence 244 participants participated in the MRI session (Meanage =18.13 years; SDage = 3,67; Rangeage = 11.90-28.70 years; 117 males).
Participants were excluded from the analyses if either the resting-state scan, high-resolution scan, or T1-weighted anatomical scan was missing or failed due to technical errors (T1: n=5; T2: n=1; T3: n=6). In total 38 participants had been diagnosed with neurological or psychiatric disorders (e.g., depression, ADHD/ADD, anxiety disorder) at T2 and/or T3: their data from all three time points were excluded from the analyses. Dyslexia, however, was not considered as an exclusion criterion as we did not expect it would influence resting-state connectivity of the selected ROIs. Furthermore, participants were excluded for having excessive head motion (≥2mm translation or ≥2˚ rotation in any direction), having ≥10 volumes (out of 140) with more than 0.5 mm movement between two
Page 7 of 28 frames (framewise displacement, FD, Power et al., 2014), or having ≥10 volumes (out of 140) that are REFRMS outliers (i.e., RMS intensity difference of volume N to the reference volume, exceeding the threshold of 75th percentile + 1.5*interquartile range). FD and REFRMS outliers were established using the motion outlier tool implemented in FSL version 5.0.4 (FMRIB’s Software Library, http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/, Smith et al., 2004). Based on these motion exclusion criteria, we excluded 33 participants at T1, 23 participants at T2, and 11 participants at T3. The final sample consisted of 619 observations from 255 participants. Specifically, 223 participants at T1 (111 males; Meanage = 14.10, SDage = 3.56; range = 8.21–25.95 years), 197 participants at T2 (97 males;
Meanage = 16.25, SDage = 3.41; range = 10.26–26.36 years), and 199 participants at T3 (94 males;
Meanage = 17.99, SDage = 3.48; range = 12.00–28.50 years) (see Table 1).
Intelligence quotient (IQ) was estimated on T1 with the subsets ‘similarities’ and ‘block design’ of the Wechsler Intelligence Scale for Adults (WAIS-III) or the Wechsler Intelligence Scale for Children, third edition (WISC-III; Wechsler, 1974). All estimated Mean scores were in the normal range (MeanIQ=110, SDIQ=10.42, range = 80-143).
Table 1. Number of participants after exclusion, per age on each time point. Indicated for the total
sample, as well as for split group 1 and split group 2. The data in split group 1 were used to select connections with an age effect; these selected connections were retested for age effects in split group 2.
Age T1 T2 T3
Total group 1 Split group 2 Split Total Split group 1 group 2 Split Total Split group 1 group 2 Split
8 9 4 5 0 0 0 0 0 0 9 16 4 12 0 0 0 0 0 0 10 19 7 12 7 3 4 0 0 0 11 22 7 15 12 5 7 0 0 0 12 26 9 17 16 6 10 11 4 7 13 32 11 21 18 5 13 11 3 8 14 23 9 14 22 7 17 20 6 14 15 17 5 12 26 9 17 18 5 13 16 16 4 12 21 8 13 23 7 16 17 17 7 10 19 7 14 26 10 16 18 0 0 0 18 6 12 18 6 12 19 7 3 4 17 6 11 18 7 11 20 9 4 5 0 0 0 16 5 11 21 0 0 0 7 3 4 17 7 10 22 7 2 5 6 3 3 2 0 2 23 1 0 1 0 0 0 5 3 2 24 1 0 1 6 2 4 7 3 4 25 1 0 1 1 0 1 0 0 0 26 0 0 0 1 0 1 5 2 3 27 0 0 0 0 0 0 1 0 1 28 0 0 0 0 0 0 1 0 1 TOTAL 223 76 147 197 70 127 199 68 131
Page 8 of 28
2.2 MRI Data Acquisition
Neuroimaging was conducted using a 3.0 Tesla 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 resting-state scans were acquired at the beginning of a fixed imaging protocol, with T2*-weighted echo-planar imaging (EPI) (140 volumes; 38 slices; sequential acquisition; time repetition (TR)=2200 ms, time echo (TE)=30 ms; flip angle=80°; field of view (FOV)=220x220x114.67 mm; slice thickness=2.75 mm); two additional dummy scans preceded the scan to allow for equilibration of T1 saturation effects. Participants were instructed to lie still with their eyes closed, without falling asleep. We assessed alertness by asking the participants right after completion of the resting-state scan.
For registration purposes, we additionally obtained a high-resolution T2*-weighted gradient EPI scan (84 slices; TR=2200 ms; TE=30 ms; flip angle=80°; FOV=220x220x168 mm; in-plane resolution=1.96x1.96; slice thickness=2 mm), and a T1-weighted anatomical scan (140 slices; TR=9.76 ms; TE=4.59 ms; flip angle=8°; FOV=224×177.33×168 mm; in-plane resolution=0.875x0.875 mm; slice thickness=2 mm), at the end of a fixed imaging protocol which included functional tasks.
At T1, the participants were acclimated to the scanner environment in a mock scanner. At T2 this was done only if the participants wanted to, but at T3 a mock scanner was not used as we assumed sufficient experience and given the increasing age of the participants.
2.3 fMRI Data Preprocessing
The RS fMRI 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 5mm full-width-at-half-maximum 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, v2.1). Next, the RS fMRI scans of an individual were registered to the corresponding high-resolution EPI image (6 DOF), which in turn were registered to the T1-weighted anatomical image using the integrated version of boundary based registration (BBR) to improve the accuracy of functional-to-structural space registration. Lastly, the images were registered to standard MNI-152 space using FNIRT (FMRIB’s Nonlinear Imaging Registration Tool; 12 DOF, warp resolution 10 mm).
Page 9 of 28
2.4 Motion Correction
Head motion is undesirable in fMRI studies (e.g., Friston, Williams, Howard, Frackowiak, & Turner, 1996), but especially in rsFC MRI studies head motion is problematic as it may overestimate short-distance correlations and underestimate long-short-distance correlations (Power, Barnes, Snyder, Schlaggar, & Petersen, 2012; Van Dijk, Sabuncu, & Buckner, 2012). In particular developmental samples are susceptible for this confound as head motion is highly related to subject age
(Satterthwaite et al., 2013).
To minimize the degree of motion, participants were emphatically told multiple times each session not to move during scanning, and we additionally used stabilization measures (i.e., foam pads, snug wrapping the lower body with sheets). Participants were excluded from the analyses if they had nevertheless moved too much. That is, if they had excessive head motion (≥2mm translation or more than 2˚ rotation in any direction), ≥10 volumes (out of 140) with more than 0.5 mm movement between two frames (framewise displacement, FD, Power et al., 2014), or ≥10 volumes (out of 140) that are REFRMS outliers (i.e., RMS intensity difference of volume N to the reference volume, exceeding the threshold of 75th percentile + 1.5*interquartile range).
Furthermore, we denoised the preprocessed RS data of the included participants with FIX (FMRIB's ICA-based Xnoiseifier, version 1.06) using the included standard training dataset (threshold 15) (Griffanti et al., 2014; Salimi-Khorshidi et al., 2014). FIX classifies ICA components and automatically removes the noise components (e.g., result of motion) from the resting-state timeseries.
2.5 Regions of Interest (ROI)
Regions of interest (ROIs) were selected from the Harvard–Oxford probabilistic anatomical brain atlas (cortical and subcortical) in FSL. In this atlas, each brain region consists of a probability map, where each voxel is assigned a probability of being part of that region. We selected the cortical structures Insular Cortex (Insula), Cingulate Gyrus Anterior Division (ACC), and the vmPFC which was a combination of the Frontal Medial Cortex and Subcallosal Cortex. For the subcortical regions, we combined the left and right part of a region into one bilateral ROI. We selected the following subcortical ROIs: Accumbens (NAcc), Amygdala, Caudate, and Hippocampus. We binarized all seven masks, only allowing voxels with a probability of ≥0.5.
For each participant separately, the ROIs were transformed to subject space by FNIRT-based dewarping these standard-space masks into the native dimensions of the BET-applied RS fMRI volumes. Next, the individual RS timeseries were extracted from each ROI separately, calculated as the mean timeseries of all non-zero voxels within that ROI.
Page 10 of 28 Figure 1. Regions of interest. Shown are the unthresholded ROIs from the Harvard Oxford atlas on an MNI
standard brain (coronal, sagittal, and axial view). The left side of the image corresponds to the right side of the brain.
Page 11 of 28
2.6 Nuisance Signal Regression
Signals from white matter (WM) and cerebrospinal fluid (CSF) can be confounding effects that result in overestimated rsFC strength. Therefore, we removed WM and CSF signals from the RS timeseries, as these signals primarily reflect noise from non-neural origin (e.g., scanner instabilities, physiological effects) and are largely independent from BOLD signal fluctuations in gray matter
(Windischberger et al., 2002). Also global signal was removed from the timeseries to reduce influence of artifacts caused by physiological processes (i.e., cardiac and respiratory fluctuations) 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 grey matter). Next, these masks were eroded by one voxel (3x3x3mm) to minimize potential partial volume effects, and FLIRT-based transformed into functional subject space. The resulting masks were multiplied by the RS fMRI, and the individual WM and CSF timeseries were calculated as the mean signal of all non-zero voxels in its respective mask. Global signal timeseries were calculated in native space with FSL’s fslmeants as the average signal across all non-zero voxels in the brain. WM, CSF, and global signal timeseries were used as temporal covariates and removed from the RS timeseries of each ROI at the individual participant level through linear regression in MATLAB (R2011b).
2.7 Testing (patterns of) Age-Related Change of RS Connectivity
Once each ROI’s RS timeseries for each individual were cleared from nuisance (WM, CSF, global signal), the Pearson correlations between all 7 ROIs’ timeseries were calculated and Fisher Z transformed in MATLAB (R2011B). This resulted in 21 correlations – indicating the strength of the connection between the two respective ROIs – of which the developmental trajectories could be determined. However, testing age effects of all 21 connections would require a very strict multiple comparison correction, potentially inducing false negatives. To bypass this, and because not all connections may be as relevant, we used a test-retest procedure. That is, we used one part of the sample to select connections that were retested for age-effects in the other part of the sample. As such, per age category (8-10, 11-14, 15-17, 18-25 years of age at T1) we randomly assigned participants into two groups with a 33:67 ratio, while assuring an equal male-to-female ratio. A 33:67 ratio was most optimal as all age categories were sufficiently represented in each split group, and this ratio has shown to be more powerful than a 50:50 split (De Vos, Huizenga, Waldorp, Grasman, & Weeda, 2010).
Page 12 of 28 The following analyses have been performed separately for split group 1 (T1: n=76; 39 males; Meanage=14.02, SDage=3.44; range=8.50–22.79 years; T2: n=70; 33 males; Meanage=16.23, SDage=3.43;
range=10.26–24.83 years; T3: n=68; 33 males; Meanage=18.23, SDage=3.49; range=12.30–26.80 years),
and split group 2 (T1: n=147; 72 males; Meanage=14.14, SDage=3.63; range=8.21-25.95 years; T2:
n=127; 64 males; Meanage=16.25, SDage=3.41; range=10.27-26.36 years; T3: n=131; 61 males;
Meanage=17.86, SDage=3.49; range=12.00-28.50 years). The descriptive of age of the final sample is
listed in Table 1.
Developmental trajectories of the RS connections were tested using multilevel modeling (also known as mixed models, hierarchical linear modeling, or random effects modeling), in the nlme package in R version 3.2.5 (Pinheiro, Bates, DebRoy, & Sarkar, 2013; R Core Team, 2008). Multilevel modeling is recommended for longitudinal data, as it takes the dependency of the data (i.e., time points are nested within participants) into account.
The model fitting procedure started with a null model that included a fixed and a random intercept, which allows individual differences in starting points. Subsequently, three models that test the grand mean trajectory of age were created by adding three polynomial terms (linear, quadratic, and cubic). Linear effects of age indicate a monotonic change over age, a quadratic effect indicates an adolescent-specific effect (either a peak or trough in rsFC strength around adolescence relative to children and adults), and lastly cubic effects of age indicate an adolescent emergent effect of age where there is an increase or decrease during adolescence but stability in childhood and adulthood
(Casey, 2015). The following code was used to fit a linear model in R:
library(nlme),
ModelName <-lme(Connection ~ poly(Age,1), data = dataset, random = ~ 1|Subject, control = list(maxIter = 1000, msMaxIter = 1000, niterEM = 1000), method="ML")
To fit a quadratic model, the poly term “poly(Age,1)” is replaced by “poly(Age,2)”, and for a cubic term it is replaced by “poly(Age,3)”. As with the poly-function age is mean-centered, the fixed intercept represents the grand mean rsFC at the mean age of the sample. The models that differ one degree of freedom (df) were compared (i.e., null vs linear, linear vs quadratic, quadratic vs cubic) using a loglikelihood test. The model with the significantly best model fit was selected.
Model fitting (linear, quadratic, cubic) was done first for all 21 connections in split group 1, with a significance level set at α=0.1 (no multiple comparisons correction). Connections that showed a significant age effect were selected (n=12) and retested for split group 2 (α=0.05) while correcting for multiple comparisons by limiting the false discovery rate with the Benjamini–Hochberg
Page 13 of 28 procedure (FDR; Benjamini & Hochberg, 1995). The specific shape of the age effects in split group 1 was not taken into account as these tests were only used for selection purposes.
Finally, we tested whether adding a random slope of age improved model fit of the best-fitting age models (see for a similar procedure Braams et al., 2015; Peters, Van Duijvenvoorde, Koolschijn, & Crone, 2016). A random slope of age allows for testing the extent of between-subject variation (i.e., individual differences) in the age-effect. The significance of the random term was determined using a loglikelihood ratio test (model with vs without random slope), with a significance level of 0.05. Post-hoc, we used loglikelihood ratio tests (α=0.05, FDR multiple comparisons correction) to test whether adding a fixed effect of gender significantly improved model fit of the best-fitting age models, i.e., whether gender explains – additionally to age – variation in rsFC.
For connections that were tested in split group 2, we determined the intraclass correlations (ICC) between T1, T2, and T3, as a measure of homogeneity of the data. ICCs were modeled with a two-way mixed model with absolute agreement, using IBM SPSS Statistics for Windows version 23. All ICC values (MeanICC=0.431, rangeICC=0.309-0.555) indicated sufficient (i.e., ICC>0.1) nesting of
observations within individuals, confirming the necessity of multilevel modeling (Lee, 2010). The average ICC values are reported in Table 2.
3. Results
3.1 Average resting-state functional connectivity
For split group 2, we calculated average rsFC strength (i.e., correlations) of each connection for the three time points separately; see Figure 1 for correlation matrices. These matrices do not show striking differences between the three time points. In general, the selected ROIs are positively functionally connected, i.e., average r > 0 (MeanrsFC,T1 = 0.19, SDrsFC,T1 = 0.25, rangersFC,T1 = -0.13 –
0.92; MeanrsFC,T2 = 0.21, SDrsFC,T2 = 0.25, rangersFC,T2 = -0.11 – 0.92; MeanrsFC,T3 = 0.21, SDrsFC,T3 = 0.27,
Page 14 of 28 Figure 1. Correlation matrices: average correlations between ROIs, indicating rsFC strength, per time point. Correlations are for splitgroup 2 only, averaged over
participants; connections indicated with the green squares were selected from split group 1, and retested for age effects in split group 2. Only the lower part of the matrices is displayed.
Page 15 of 28
3.2 Age-related changes of subcortico-subcortical (SUB-SUB) rsFC
To test our age-effects we first ran our multilevel models on split group 1 (see Table 1) to select connections of interest. From split group 1, two SUB-SUB connections were selected: amygdala-hippocampus and amygdala-caudate. These two connections were retested in split group 2 for age-related changes in rsFC. P-values and decrease in deviance (i.e., increase of model fit) of each model of age-effects in split group 2, as well as the intercepts and coefficients (rsFC strength) of the fixed effect of age for the significant models are shown in Table 2.
Only for the amygdala-hippocampus rsFC an age-related change was replicated in split group 2: a linear increase across age was observed (see Figures 2, 3, and Table 2). There were no significant individual differences in the effect of age (random slope of age, p>0.05). We did not observe significant adolescent-specific (quadratic) or adolescent-emergent (cubic) age-related effects for rsFC in SUB-SUB connections. Post-hoc loglikelihood ratio tests revealed no main effects of gender (fixed effect of gender, p>0.05).
3.3 Age-related changes of cortico-subcortical (CORT-SUB) rsFC
In split group 1, ten CORT-SUB connections showed age-effects, which were selected for retest in split group 2: both the insula and vmPFC showed age effects with the NAcc, amygdala, hippocampus, and caudate; the ACC showed age effects with de caudate and hippocampus.
For these 10 selected CORT-SUB connections, we retested in split group 2 whether rsFC between cortical and subcortical structures changes across age. The vmPFC revealed a significant linear age-related increase in functional connectivity with the caudate and hippocampus (see Figures 2, 3, and Table 2). The rsFC of the insula with the amygdala also increased linearly with age. No significant adolescent-specific (quadratic) or adolescent-emergent (cubic) age-related changes were observed. For neither of the connections with an age effect, individual differences were found (random slope of age, p>0.05). Post-hoc loglikelihood ratio tests revealed that adding a fixed main effect of gender improved model fit only for the hippocampus-amygdala connection (𝜒𝜒(1)2 =5.72, p=0.017), but this effect did not survive multiple comparisons correction (i.e., FDR).
3.4 Age effects on rsFC, but FDR correction not survived
A few CORT-SUB and SUB-SUB connections showed significant age-related changes of rsFC in both split groups, but the effects were not strong enough to survive the (FDR) multiple comparison correction in split group 2 (indicated by dotted lines in Figure 2). Linear decreases were observed between the insula and caudate, and between the vmPFC and NAcc. Between the amygdala and
Page 16 of 28 caudate the rsFC increased linearly across age. Finally, the ACC-hippocampus connection showed an adolescent-specific (quadratic) increase of rsFC across age.
Figure 2. Schematic overview of resting-state connections that show an age effect in both split groups. The
single red lines indicate a linear decrease over age, the single green lines indicates a linear increase over age, and the double green line indicates a quadratic increase over age. Solid lines represent significant effects of age on the RS connectivity in split group 2, FDR corrected for multiple comparisons; dotted lines are used for connections that did not survive the FDR correction.
Page 17 of 28 Table 2. Models of linear, quadratic, and cubic age effects on rsFC tested in split group 2. For each tested connection in split group 2, p-values and decrease in deviance (Δ
deviance, i.e., increase of model fit relative to the model with 1 df less) of each model, as well as the intercepts, and coefficients (β) of the fixed effect of age for the significant models are indicated. Age is mean centered by the poly function, and intercept and β values are correlation values. Intraclass correlations indicate sufficient nesting of observations within individuals. Best fitting models are indicated in boldface type. Models that did not survive multiple comparison correction are indicated with an asterisk.
Page 18 of 28 Figure 3. Plots representing the relation between age and resting-state functional connectivity. These plots are for visualization purposes only and display extracted
parameter estimates (Z values) of rsFC. Individual subjects are represented by individual (black) lines; subjects measured only once are represented by points. The solid blue lines indicate the predicted values based on the optimal model, the dotted blue lines represent 95% confidence interval.
Page 19 of 28
4. Discussion
In this longitudinal study we aimed to examine developmental trajectories (i.e., age-related changes) of functional connectivity within and between subcortical and cortical neural circuits, in a task-free manner across childhood, adolescence, and early adulthood. Therefore we tested for different rsFC trajectories (linear, quadratic, and cubic) (1) within a subcortical network including the NAcc, caudate, amygdala, and hippocampus, (SUB-SUB), and (2) between these subcortical and cortical regions, particularly the insula, vmPFC and ACC (CORT-SUB); we collected data in large sample with a wide continuous age range (N=255, 8-28 years) on three time points, separated by two years. Correlation matrices (Figure 1) show, in general, positive functional connectivity between ROIs. However, our main interest was to test for linear and non-linear change in rsFC across age. We observed age-related linear increases for SUB-SUB and CORT-SUB connections; there were no individual differences in the effect of age. Post-hoc tests revealed that – in addition to age – gender did not explain variation in rsFC. The findings are discussed in more detail in the next paragraphs.
4.1 Age-related changes in intrinsic functional connectivity (rsFC)
In the current study, for SUB-SUB connections we found that rsFC between the hippocampus and amygdala changed across age in a linear manner. We expected that – in general – SUB-SUB connections would show adolescent-specific rsFC increases (peak in rsFC strength at adolescence), instead of monotonic increases. However, for this specific connection this developmental trajectory is not surprising, since it is known from animal studies that the amygdala and hippocampus interact in emotional situations (Phelps, 2004). That is, the amygdala modulates encoding and storage hippocampal-dependent memories, and the hippocampus can influence the amygdala response in emotional situations. These results suggest that this interaction between emotion and memory improves across development rather than showing a peak at a certain age period. Unexpectedly, no other SUB-SUB rsFC age effects were found.
We observed handful CORT-SUB connections with age-related changes: vmPFC showed linear rsFC increases with the caudate and hippocampus, and also insula-amygdala rsFC increased linearly with age. These developmental trajectories were in line with our hypothesis that CORT-SUB connections would generally show monotonic age-related increases, indicating a relative increase in cognitive control through cortical top-down projections. Both the hippocampus and caudate are regions involved in motivation and learning (Shohamy, 2011) and it has indeed been shown that vmPFC exerts top-down control on these regions, e.g., during decision-making (e.g., Gluth, Sommer, Rieskamp, & Büchel, 2015; Wunderlich, Dayan, & Dolan, 2012). A recent rsFC study with emotionally dysregulated youth demonstrated an inverse relationship between rsFC between amygdala-insula
Page 20 of 28 rsFC and emotional dysregulation; this suggests that functional coupling between these regions is necessary for emotion processing and regulation (Bebko et al., 2015). Together, these findings agree with the current findings, suggesting that the coupling between these regions increases across typical development for proper behavioural functioning.
Considering previous studies examining rsFC in a developmental sample, we expected to find more connections to show an age effect. For example, Fareri et al., (2015) found age-related rsFC changes of the ventral striatum (NAcc) with the hippocampus, vmPFC, amygdala, and insula, whereas we did not find any age effects with the NAcc. Also Van Duijvenvoorde et al., (2015) found age-related rsFC changes in connections where we did not observe an age effect, e.g., for the NAcc-ACC, NAcc-hippocampus, and NAcc-vmPFC rsFC. However, these studies took a different in approach for examining development of rsFC, which could explain the discrepancies between our findings. That is, these studies are cross-sectional and moreover tested rsFC changes across age exploratively (whole-brain rsFC tracking from a seed). We on the other hand, tested a priori connections in a longitudinal dataset; only with a longitudinal design one can address developmental neural changes within individuals which is more informative for development.
4.2 Current findings: implications for an integrated circuit-based model
Development of the adolescent brain and behaviour has been described by the imbalance model; it has an integrated circuit-based perspective and suggests changing connections within and between cortical and subcortical networks, which are interactive rather than orthogonal. The development is thought to pass hierarchically from SUB-SUB, to CORT-SUB, to CORT-CORT connections through fine-tuning of the connections (Casey et al., 2016; Casey, 2015). In the current study we tested developmental trajectories of such connections. In line with the model, we did find age-effects on rsFC, but not to the extent the model proposes: we only found one SUB-SUB, three CORT-SUB, and zero CORT-CORT connections with age-related changes in rsFC. The SUB-SUB connections are thought to develop in early adolescence, an age period sufficiently present in our sample; more age effects would have been expected. The same applies for CORT-CORT connections, thought to develop in young adulthood. In sum, even though the age range of the current study covered indeed early adolescence to young adulthood, and we tested relevant connections, the results do not exactly harmonize the expectations of the model.
A possible explanation for the deviance between the current results and the model, is that some connections may have been (mainly) established at a younger age (i.e., before early adolescence), and therefore do not show age effects in the current sample. Hence, it would be of interest to test a wider age-range. Previous studies have, however, observed age-related changes in rsFC after 8 years (e.g., Fareri
et al., 2015; Van Duijvenvoorde et al., 2015), and also structural connectivity has been shown to be still
Page 21 of 28
Indicative of relatively stable rsFC across the current age-range are the ICC values. That is, the tested connections have large ICCs – indicating large stability within participants but relatively large variances between individuals – but showed no or only small age effects. This suggests that not age but other factors underlie rsFC differences between participants. Possibly, puberty, genetic factors, or environmental factors may influence development of rsFC, as they are known to influence neural development (Peper et al., 2011; Wiggins et al., 2012).
Also, some methodological decisions we made may have influenced results and should be taken into account: they are described in the next section.
4.3 Considerations
The current study has a homogeneous sample, regarding e.g., ethnicity (97% Caucasian) and IQ, and did not contain participants with a psychiatric or neurological disorder. This homogenous healthy sample allowed examining typical rsFC development across adolescence. It would be of interest for future studies, however, to also study rsFC in a rather heterogeneous sample or atypically developing sample: possible differences could yield a better understanding of rsFC development.
It is commonly acknowledged that head motion is undesirable in fMRI studies (e.g., Friston et al.,
1996). However, recent studies have demonstrated that especially in rsFC MRI studies motion is
problematic: small in-scanner motion may lead to overestimation of short-distance correlations and underestimation of long-distance correlations (Power et al., 2012; Van Dijk et al., 2012). Consequently, decreases in rsFC over age could rather reflect decreases in head motion over age instead of decreases in functional connectivity. In the current study we have taken important steps to account for possible motion confounds. First, we applied stringent exclusion criteria for motion (2mm), as well as micromovements using 2 metrics for evaluating the extent of micromovements (FD, REFRMS). We furthermore used FIX to remove noise components from the resting-state timeseries of included participants. It should be noted however, that the included standard training dataset we used for FIX may not be optimal for a developmental sample. As a result, not all noise components may have been removed from the data.
Another source of noise, apart from motion, may come from physiological artefacts (cardiac and respiratory fluctuations) (Fox & Raichle, 2007). We did not record additional measures such as heart rate or respiration, and moreover toolboxes to generate physiological regressors are not optimized for developmental samples (physiology changes throughout development) (Beall, 2010). In line with previous studies (Fareri et al., 2015; Gabard-durnam et al., 2014) we therefore regressed out the mean global signal at the subject-level instead. As such, we accounted for physiological artefacts, but also for scanner drifts (Fox & Raichle, 2007). We are aware, however, of the ongoing debate in the literature regarding this global signal regression in rs-fMRI. There are concerns for using global signal regression since it is thought to induce anticorrelations between regions (e.g., Murphy, Birn, Handwerker, Jones, & Bandettini, 2009).
Page 22 of 28
On the other hand there is evidence that such negative correlations may be present anyway, also when global signal is not regressed out (Keller et al., 2013). Alternatively, one could use partial correlations where you correct for the other included ROIs (Schouten et al., 2016). However, for the current study this was not a valid alternative for global signal regression, as we only used a limited number and a specific set of regions. Moreover, our average correlations between regions were mainly positive, which seems to suggest that anticorrelations have not been induced by the global signal regression in the current study.
4.4 Conclusions
In sum, the current study shows age-related monotonic increases in rsFC across adolescence within and between subcortical and cortical networks. This is in line with the imbalance model which suggests changes in connections across adolescence. The results, however, do not exactly harmonize the model’s expectations: we found fewer connections with an age effect than one would expect based on the model. Moreover, the found age-effects were rather small, as indicated by the low coefficient values, whereas the model accounts age as the driving factor for changes in functional connectivity. In addition, we observed large between-subject differences in rsFC; therefore we suggest that maybe not age, but other factors (e.g., pubertal, genetic, or environmental factors) explain rsFC development and variability between individuals. A future direction will be to examine such factors, and how they relate to (adolescent) behaviour. It may also be of interest to additionally test for a direct anatomical connection between brain regions, by combining functional as well as structural connectivity indices (Dennis &
Page 23 of 28
5. References
Beall, E. B. (2010). Adaptive cyclic physiologic noise modeling and correction in functional MRI. Journal of Neuroscience Methods, 187(2), 216–228. doi:10.1016/j.jneumeth.2010.01.013 Bebko, G., Bertocci, M., Chase, H., Dwojak, A., Bonar, L., Almeida, J., … Phillips, M. L. (2015).
Decreased amygdala-insula resting state connectivity in behaviorally and emotionally dysregulated youth. Psychiatry Research - Neuroimaging, 231(1), 77–86.
doi:10.1016/j.pscychresns.2014.10.015
Benjamini, Y., & Hochberg, Y. (1995). Controlling the false discovery rate - a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society. Series B (Methodological), 57(1), 289–300.
Biswal, B., Yetkin, F. Z., Haughton, V. M., & Hyde, J. S. (1995). Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magnetic Resonance in Medicine : Official Journal of the Society of Magnetic Resonance in Medicine / Society of Magnetic Resonance in Medicine, 34(4), 537–541. doi:10.1002/mrm.1910340409
Braams, B. R., van Duijvenvoorde, A. C. K., Peper, J. S., & Crone, E. A. (2015). Longitudinal Changes in Adolescent Risk-Taking: A Comprehensive Study of Neural Responses to Rewards, Pubertal Development, and Risk-Taking Behavior. Journal of Neuroscience, 35(18), 7226–7238. doi:10.1523/JNEUROSCI.4764-14.2015
Casey, B. J. (2015). Beyond Simple Models of Self-Control to Circuit-Based Accounts of Adolescent Behavior. Annual Review of Psychology, 66(July 2014), 295–319. doi:10.1146/annurev-psych-010814-015156
Casey, B. J., Galván, A., & Somerville, L. H. (2016). Beyond simple models of adolescence to an integrated circuit-based account: A commentary. Developmental Cognitive Neuroscience, 17, 128–130. doi:10.1016/j.dcn.2015.12.006
Casey, B. J., Jones, R. M., & Hare, T. A. (2008). The adolescent brain. Annals of the New York Academy of Sciences, 1124, 111–26. doi:10.1196/annals.1440.010
Casey, B. J., Jones, R. M., & Somerville, L. H. (2011). Braking and Accelerating of the Adolescent Brain. Journal of Research on Adolescence, 21(1), 21–33. doi:10.1111/j.1532-7795.2010.00712.x Cho, Y. T., Fromm, S., Guyer, A. E., Detloff, A., Pine, D. S., Fudge, J. L., & Ernst, M. (2013). Nucleus
accumbens, thalamus and insula connectivity during incentive anticipation in typical adults and adolescents. NeuroImage, 66, 508–521. doi:10.1016/j.neuroimage.2012.10.013
Page 24 of 28 Crone, E. a., & Elzinga, B. M. (2015). Changing brains: how longitudinal functional magnetic
resonance imaging studies can inform us about cognitive and social-affective growth trajectories. Wiley Interdisciplinary Reviews: Cognitive Science, 6(1), 53–63.
doi:10.1002/wcs.1327
De Vos, F., Huizenga, H. M., Waldorp, L. J., Grasman, R. P. P. P., & Weeda, W. D. (2010). Two-stage hypothesis testing procedures can enhance statistical power in fMRI analysis.
Dennis, E. L., & Thompson, P. M. (2014). Mapping connectivity in the developing brain. International Journal of Developmental Neuroscience, 32(7), 41–57. doi:10.1016/j.ijdevneu.2013.11.005 Ernst, M., Torrisi, S., Balderston, N., Grillon, C., & Hale, E. A. (2014). fMRI Functional Connectivity
Applied to Adolescent Neurodevelopment. Annual Review of Clinical Psychology, 11(1), 150112144717005. doi:10.1146/annurev-clinpsy-032814-112753
Fareri, D. S., Gabard-Durnam, L., Goff, B., Flannery, J., Gee, D. G., Lumian, D. S., … Tottenham, N. (2015). Normative development of ventral striatal resting state connectivity in humans. NeuroImage, 118, 422–437. doi:10.1016/j.neuroimage.2015.06.022
Fox, M. D., & Raichle, M. E. (2007). Spontaneous fluctuations in brain activity observed with functional magnetic resonance imaging. Nat Rev Neurosci, 8(9), 700–711. doi:nrn2201 [pii]\n10.1038/nrn2201
Friston, K. J., Williams, S., Howard, R., Frackowiak, R. S., & Turner, R. (1996). Movement-related effects in fMRI time-series. Magnetic Resonance in Medicine : Official Journal of the Society of Magnetic Resonance in Medicine / Society of Magnetic Resonance in Medicine, 35(3), 346–355. doi:DOI 10.1002/mrm.1910350312
Gabard-durnam, L. J., Flannery, J., Goff, B., Gee, D. G., Humphreys, K. L., Telzer, E., … Tottenham, N. (2014). The development of human amygdala functional connectivity at rest from 4 to 23 years : A cross-sectional study (Author’s personal copy). NeuroImage, 95, 193–207.
Gluth, S., Sommer, T., Rieskamp, J., & Büchel, C. (2015). Effective Connectivity between Hippocampus and Ventromedial Prefrontal Cortex Controls Preferential Choices from Memory. Neuron, 86(4), 1078–1090. doi:10.1016/j.neuron.2015.04.023
Griffanti, L., Salimi-Khorshidi, G., Beckmann, C. F., Auerbach, E. J., Douaud, G., Sexton, C. E., … Smith, S. M. (2014). ICA-based artefact removal and accelerated fMRI acquisition for improved resting state network imaging. NeuroImage, 95, 232–247. doi:10.1016/j.neuroimage.2014.03.034 Hare, T. a., Tottenham, N., Galvan, A., Voss, H. U., Glover, G. H., & Casey, B. J. (2008). Biological
Page 25 of 28 Substrates of Emotional Reactivity and Regulation in Adolescence During an Emotional Go-Nogo Task. Biological Psychiatry, 63(10), 927–934. doi:10.1016/j.biopsych.2008.03.015
Jenkinson, M., Bannister, P., Brady, M., & Smith, S. (2002). Improved Optimization for the Robust and Accurate Linear Registration and Motion Correction of Brain Images. NeuroImage, 17(2), 825– 841. doi:10.1006/nimg.2002.1132
Keller, C. J., Bickel, S., Honey, C. J., Groppe, D. M., Entz, L., Craddock, R. C., … Mehta, A. D. (2013). Neurophysiological investigation of spontaneous correlated and anticorrelated fluctuations of the BOLD signal, 33(15), 6333–6342. doi:10.1523/JNEUROSCI.4837-12.2013.Neurophysiological Khundrakpam, B. S., Lewis, J. D., Zhao, L., Chouinard-Decorte, F., & Evans, A. C. (2016). Brain
connectivity in normally developing children and adolescents. NeuroImage, 134, 192–203. doi:10.1016/j.neuroimage.2016.03.062
Lee, V. E. (2010). Using Hierarchical Linear Modeling to Study Social Contexts : The Case of School Effects. Educational Psychologist, 35(2), 125–141. doi:10.1207/S15326985EP3502
Mills, K. L., & Tamnes, C. K. (2014). Methods and considerations for longitudinal structural brain imaging analysis across development. Developmental Cognitive Neuroscience, 9, 172–190. doi:10.1016/j.dcn.2014.04.004
Murphy, K., Birn, R. M., Handwerker, D. A., Jones, T. B., & Bandettini, P. A. (2009). The impact of global signal regression on resting state correlations: Are anti-correlated networks introduced? Neuroimage, 44(3), 893–905. doi:10.1016/j.neuroimage.2008.09.036.
Peper, J. S., van den Heuvel, M. P., Mandl, R. C. W., Pol, H. E. H., & van Honk, J. (2011). Sex steroids and connectivity in the human brain: A review of neuroimaging studies.
Psychoneuroendocrinology, 36(8), 1101–1113. doi:10.1016/j.psyneuen.2011.05.004 Peters, S., Van Duijvenvoorde, A. C. K., Koolschijn, P. C. M. P., & Crone, E. A. (2016). Longitudinal
development of frontoparietal activity during feedback learning: Contributions of age,
performance, working memory and cortical thickness. Developmental Cognitive Neuroscience, 19, 211–222. doi:10.1016/j.dcn.2016.04.004
Phelps, E. A. (2004). Human emotion and memory: Interactions of the amygdala and hippocampal complex. Current Opinion in Neurobiology, 14(2), 198–202. doi:10.1016/j.conb.2004.03.015 Pinheiro, J., Bates, D., DebRoy, S., & Sarkar, D. (2013). NLME: Linear and nonlinear mixed effects
models. R package version 3.1-122, http://CRAN.R-project.org/package=nlme.
Page 26 of 28 systematic correlations in functional connectivity MRI networks arise from subject motion. NeuroImage, 59(3), 2142–2154. doi:10.1016/j.neuroimage.2011.10.018
Power, J. D., Mitra, A., Laumann, T. O., Snyder, A. Z., Schlaggar, B. L., & Petersen, S. E. (2014).
Methods to detect, characterize, and remove motion artifact in resting state fMRI. NeuroImage, 84, 320–341. doi:10.1016/j.neuroimage.2013.08.048
R Development Core Team (2008). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.
Richards, J. M., Plate, R. C., & Ernst, M. (2013). A systematic review of fMRI reward paradigms used in studies of adolescents vs. adults: The impact of task design and implications for understanding neurodevelopment. Neuroscience & Biobehavioral Reviews, 37(5), 976–991.
doi:10.1016/j.neubiorev.2013.03.004
Salimi-Khorshidi, G., Douaud, G., Beckmann, C. F., Glasser, M. F., Griffanti, L., & Smith, S. M. (2014). Automatic denoising of functional MRI data: Combining independent component analysis and hierarchical fusion of classifiers. NeuroImage, 90, 449–468.
doi:10.1016/j.neuroimage.2013.11.046
Satterthwaite, T. D., & Baker, J. T. (2015). How can studies of resting-state functional connectivity help us understand psychosis as a disorder of brain development? Current Opinion in Neurobiology, 30(October 2014), 85–91. doi:10.1016/j.conb.2014.10.005
Satterthwaite, T. D., Wolf, D. H., Ruparel, K., Erus, G., Elliott, M. A., Eickhoff, S. B., … Gur, R. C. (2013). Heterogeneous impact of motion on fundamental patterns of developmental changes in functional connectivity during youth. NeuroImage, 83, 45–57.
doi:10.1016/j.neuroimage.2013.06.045
Schouten, T. M., Loitfelder, M., de Vos, F., Seiler, S., van der Grond, J., Lechner, A., … Rombouts, S. A. R. B. (2016). Combining anatomical, diffusion, and resting state functional magnetic resonance imaging for individual classification of mild and moderate Alzheimer’s disease. NeuroImage: Clinical, 11, 46–51. doi:10.1016/j.nicl.2016.01.002
Shohamy, D. (2011). Learning and motivation in the human striatum. Current Opinion in Neurobiology, 21(3), 408–414. doi:10.1016/j.conb.2011.05.009
Smith, S. M., Jenkinson, M., Woolrich, M. W., Beckmann, C. F., Behrens, T. E. J., Johansen-Berg, H., … Matthews, P. M. (2004). Advances in functional and structural MR image analysis and
Page 27 of 28 doi:10.1016/j.neuroimage.2004.07.051
Somerville, L. H., Hare, T., & Casey, B. J. (2011). Frontostriatal maturation predicts cognitive control failure to appetitive cues in adolescents. Journal of Cognitive Neuroscience, 23(9), 2123–34. doi:10.1162/jocn.2010.21572
Somerville, L. H., Jones, R. M., & Casey, B. J. (2010). A time of change: Behavioral and neural correlates of adolescent sensitivity to appetitive and aversive environmental cues. Brain and Cognition, 72(1), 124–133. doi:10.1016/j.bandc.2009.07.003
Sowell, E. R., Thompson, P. M., Holmes, C. J., Jernigan, T. L., & Toga, a W. (1999). In vivo evidence for post-adolescent brain maturation in frontal and striatal regions. Nature Neuroscience, 2(10), 859–861. doi:10.1038/13154
Spear, L. P. (2000). The adolescent brain and age-related behavioral manifestations. Neuroscience & Biobehavioral Reviews, 24(4), 417–463. doi:10.1016/S0149-7634(00)00014-2
Steinberg, L. (2008). A Social Neuroscience Perspective on Adolescent Risk-Taking. Developmental Review : DR, 28(1), 78–106. doi:10.1016/j.dr.2007.08.002
Steinberg, L., Albert, D., Cauffman, E., Banich, M., Graham, S., & Woolard, J. (2008). Age differences in sensation seeking and impulsivity as indexed by behavior and self-report: evidence for a dual systems model. Developmental Psychology, 44(6), 1764–78. doi:10.1037/a0012955
Tottenham, N., Tanaka, J. W., Leon, A. C., McCarry, T., Nurse, M., Hare, T. A., … Nelson, C. (2009). The NimStim set of facial expressions: judgments from untrained research participants. Psychiatry Research, 168(3), 242–9. doi:10.1016/j.psychres.2008.05.006
Uddin, L. Q., Supekar, K., & Menon, V. (2010). Typical and atypical development of functional human brain networks: insights from resting-state FMRI. Frontiers in Systems Neuroscience, 4(May), 21. doi:10.3389/fnsys.2010.00021
Van Dijk, K. R. A., Sabuncu, M. R., & Buckner, R. L. (2012). The influence of head motion on intrinsic functional connectivity MRI. NeuroImage, 59(1), 431–8. doi:10.1016/j.neuroimage.2011.07.044 Van Duijvenvoorde, A. C. K., Achterberg, M., Braams, B. R., Peters, S., & Crone, E. A. (2015). Testing a dual-systems model of adolescent brain development using resting-state connectivity analyses. NeuroImage. doi:10.1016/j.neuroimage.2015.04.069
Wiggins, J. L., Bedoyan, J. K., Peltier, S. J., Ashinoff, S., Carrasco, M., Weng, S. J., … Monk, C. S. (2012). The impact of serotonin transporter (5-HTTLPR) genotype on the development of resting-state functional connectivity in children and adolescents: A preliminary report. NeuroImage, 59(3),
Page 28 of 28 2760–2770. doi:10.1016/j.neuroimage.2011.10.030
Windischberger, C., Langenberger, H., Sycha, T., Tschernko, E. M., Fuchsjäger-Mayerl, G.,
Schmetterer, L., & Moser, E. (2002). On the origin of respiratory artifacts in BOLD-EPI of the human brain. Magnetic Resonance Imaging, 20(8), 575–582.
doi:10.1016/S0730-725X(02)00563-5
Wunderlich, K., Dayan, P., & Dolan, R. J. (2012). Mapping value based planning and extensively trained choice in the human brain. Nature Neuroscience, 15(5), 786–791. doi:10.1038/nn.3068
