Cerebral Cortex, September 2021;31: 4140–4150

[Link]
Advance Access Publication Date: 5 May 2021
Original Article

ORIGINAL ARTICLE

Characterizing the Network Architecture of Emotion

Regulation Neurodevelopment
João F. Guassi Moreira1 , Katie A. McLaughlin2 and Jennifer A. Silvers1
1 Department of Psychology, University of California, Los Angeles, CA 90095, USA and 2 Department of
Psychology, Harvard University, Cambridge, MA 02138, USA

Address correspondence to João F. Guassi Moreira, A191 Franz Hall, 502 Portola Plaza, Los Angeles, CA 90095-1563, USA. Email: jguassimoreira@[Link].

Abstract
The ability to regulate emotions is key to goal attainment and well-being. Although much has been discovered about
neurodevelopment and the acquisition of emotion regulation, very little of this work has leveraged information encoded in
whole-brain networks. Here we employed a network neuroscience framework to parse the neural underpinnings of emotion
regulation skill acquisition, while accounting for age, in a sample of children and adolescents (N = 70, 34 female, aged 8–17
years). Focusing on three key network metrics—network differentiation, modularity, and community number differences
between active regulation and a passive emotional baseline—we found that the control network, the default mode network,
and limbic network were each related to emotion regulation ability while controlling for age. Greater network
differentiation in the control and limbic networks was related to better emotion regulation ability. With regards to network
community structure (modularity and community number), more communities and more crosstalk between modules (i.e.,
less modularity) in the control network were associated with better regulatory ability. By contrast, less crosstalk (i.e., greater
modularity) between modules in the default mode network was associated with better regulatory ability. Together, these
findings highlight whole-brain connectome features that support the acquisition of emotion regulation in youth.

Key words: emotion regulation, cognitive reappraisal, connectome, network neuroscience, neurodevelopment

Introduction the acquisition of emotion regulation skills. This omission is

noteworthy given work in adults showing that many emotional,
Emotions promote adaptive behavior but can become detri- as well as nonemotional, processes and behaviors are sup-
mental if left unchecked (Gross, 2015), underscoring the impor- ported by whole-brain networks (Kleckner et al., 2017; Gratton
tance of emotion regulation. Emotion regulation acquisition is et al., 2018). Leveraging network neuroscience may help to fill
a protracted process that emerges across development, and is at least two major gaps in our knowledge of emotion regula-
highly consequential for well-being (Calkins, 1994; Casey et al., tion neurodevelopment. First, this approach may improve our
2008; Thompson et al., 2008). Over the past 15 years, substantial understanding of how correlated but distinct developmental
efforts have been made to examine brain regions involved in phenomena such as chronological age and emotion regulation
the development of emotion regulation, with most work focus- ability are represented in the brain. Second, neuroscientific the-
ing on activation within, or dynamics between, the amygdala, ories of emotion regulation implicitly posit dynamic interactions
ventromedial prefrontal cortex (vmPFC), and lateral prefrontal between various brain regions during active emotion regula-
cortex (lPFC) (Ernst, Pine, & Hardin, 2006; Pitskel, Bolling, Kaiser, tion (Ochsner, Silvers, & Buhle, 2012), yet only recently have
Crowley, & Pelphrey, 2011; McRae et al., 2012; Kim et al., 2013; imaging studies of emotion regulation begun to explicitly test
Gee et al., 2014; Perino, Miernicki, & Telzer, 2016; Silvers et al., this assumption (Zhang, Padmanabhan, Gross, & Menon, 2019).
2017). However, surprisingly little is known about the extent Doing so from a whole-brain network perspective will provide
to which whole-brain networks—suites of brain regions that more specific and precise evidence to be used in evaluating said
reliably activate in concert (e.g., default mode, control)—support theories.

© The Author(s) 2021. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: [Link]@[Link]
 Network Architecture of Emotion Regulation Neurodevelopment Guassi Moreira et al. 4141

Network neuroscience is a promising approach for pars- Methods

ing distinct but related developmental phenomena (Richardson
Participants
et al., 2018). The current study focused on two network prop-
erties that have been particularly important for understanding Participants for the current study were drawn from a larger
development in other domains and thus may be fruitful for sample of individuals enrolled in a longitudinal neuroimaging
pinpointing the neurodevelopment of emotion regulation abil- study about childhood maltreatment and affective neurodevel-
ity. The first property is network differentiation—the extent to opment. Youth in the current analyses were selected from a non-
which activity within a network is similar or distinct across maltreated community control group who were screened with
task states, perhaps reflecting a network’s “specialization” for the following exclusion criteria: childhood maltreatment/vio-
a given task. Examining network differentiation in combination lence, presence of a developmental disorder (e.g., autism), psy-
with behavioral assessments of emotion regulation can help chotropic medication use, and IQ < 75. To qualify for inclusion in
determine which networks become specialized to support skill the current report, participants were required to have anatomi-
development (i.e., improved emotion regulation ability). While cal images free of abnormalities or artifacts, low levels of motion
some patterns of network differentiation may correspond with during the scan (addressed below), and accompanying behav-
changes in specific skills, others that support more general ioral data from our scanning emotion regulation paradigm. Our
aspects of development may show increased specialization with final sample included 70 youth (34 female) ranging in age from
age rather than tracking with performance on a given task. 8.08 to 17.00 years (Mean age = 12.7). The study was conducted in
For instance, prior research shows greater specialization among a large, metropolitan area in the Pacific Northwest region of the
pain perception and mentalizing networks tracks with chrono- United States of America. Youth and their parents provided writ-
logical age, but not skills related to representing others’ states ten assent and consent, respectively, in accordance with the Uni-
(Richardson et al., 2018). Extrapolating to emotion regulation, versity of Washington’s Institutional Review Board. Participants
this implies that some networks might support domain-general were remunerated $75 after completion of the scan. Data and
development with chronological age (e.g., the default mode net- analysis code can be accessed on the Open Science Framework
work), whereas others specifically support emotion regulation ([Link]/bfg3d).
abilities (e.g., the salience network).
The second property is network community structure—or,
how brain networks are divided into substructures (“communi-
Experimental Design
ties”, “modules”; Sporns, 2010; Yeo et al., 2011). In the context
of emotion regulation, a network’s modules may support dif- fMRI Emotion Regulation Paradigm
ferent emotion regulation subprocesses (e.g., working memory). Participants completed a variation of a common and widely
We assessed network community structure here by examin- used emotion regulation task for youth while undergoing fMRI
ing the number of communities that emerged in a given net- (Silvers, Shu, Hubbard, Weber, & Ochsner, 2015). Participants
work between task states as well as modularity, the strength passively viewed or effortfully regulated emotional responses to
of intramodule versus intermodule connections. A network’s a series of developmentally appropriate negative and neutral
community structure (number of communities, modularity) and images (modeled after the International Affective Picture
the suite of psychological processes it supports might track System (IAPS); Lang, Bradley, & Cuthbert, 2008). Each trial
differentially with age and ability, helping reveal the manner began with an instructional cue (2 s duration), “Look” or
in which domain-general maturation (chronological age) differs “Far”. The former instructs participants to passively view a
from growing expertise in a skill (ability). Given that other lines subsequent stimulus, whereas the latter prompts participants
of research have successfully leveraged network differentiation to regulate it via a psychological distancing variant of cognitive
and community structure metrics to build increasingly accurate reappraisal. Following the cue, participants viewed a neutral
models of brain function, cognitive skills, and development or aversive stimulus. Aversive stimuli were either paired
(Baum et al., 2017; Richardson et al., 2018; Sizemore et al., 2018), with “Look” or “Far”. Neutral images were always paired
it stands to reason that such an approach may also benefit the with the “Look” cue; these images are not of interest to the
study of emotion regulation development. current study and will not be discussed further. Stimuli were
In the present study, we used the three aforementioned displayed for 6–10 s (jittered). A 5-point Likert scale was
metrics of network activity (network differentiation, number of presented thereafter to collect participants’ affective responses
communities, modularity) in conjunction with ridge regression (4 s). Lastly, a jittered ITI concluded each trial (1.5–6.5 s).
to predict regulation ability in a sample of youth who com- Prior to scanning, participants were trained extensively by
pleted an emotion regulation task during functional magnetic experimenters to ensure they understood the task and were
resonance imaging (fMRI) scanning. These three metrics were comfortable completing it, and afterwards completed a practice
chosen because they have been consistently implicated in prior run. Stimuli were purchased from a stock photography website
neuroimaging studies of social and cognitive development and ([Link] and were normalized in a sample of
have been previously used as indices of network specialization youth (120, ages 6–16, 50% female) based on valence, arousal,
(Baum et al., 2017; Richardson et al., 2018; Sizemore et al., 2018). and dominance. Stimuli and normed ratings are available
Because our questions were exploratory in nature, our study online ([Link]/43hfq/). The most negatively rated images were
was guided by open questions about how network properties selected for use in the current study. Participants completed
relate to emotion regulation ability beyond chronological age. four runs, each lasting approximately 220 s (runs ranged
However, given prior neuroimaging work implicating prefrontal- between 110 and 115 4D volumes). Each task condition (“Look”-
subcortical circuitry in emotion regulation development (Casey, Negative; “Far”-Negative; “Look”-Neutral) was presented 20
2015; Silvers et al., 2017), we generally expected that limbic and times over the four runs (5 per run). The task was programmed
prefrontal control networks were particularly likely to explain and presented in E-Prime (Psychology Software Tools, Inc.,
differences in regulatory ability. [Link]
 4142 Cerebral Cortex, 2021, Vol. 31, No. 9

fMRI Data Acquisition adult and developmental samples suggests that task-specific
If participants were younger than 12 years of age or exhibited information—and even activity from task-specific networks—
any nervousness about scanning, they were taken to a mock MRI is privileged over domain-general information and thus may
scanner in order to familiarize them with scanning environment be informative when differentiating between task ability and
and to be trained on how to minimize head motion in the chronological age (Brody et al., 2019; Greene et al., 2018). Nodes
scanner. Additionally, in preparation for the scan, participants in the CRN were identified using the same procedure described
were packed into the head coil with an inflated, head-stabilizing in Guassi Moreira, McLaughlin, & Silvers (2019).
pillow to restrict movement.
Images were acquired at the University of Washington’s Estimating Network Connectivity Metrics
(Seattle) Integrated Brain Imaging Center on a 3 T Phillips
Achieva scanner. A 32-channel head coil was used in con- Network Connectivity via Beta-Series
junction with parallel image acquisition. A T1-weighted, In order to obtain metrics of network-level activity, we first
magnetization-prepared rapid acquisition gradient echo (MPRAGE) modeled the task data within-subjects using a Least Squares
image was acquired for registration purposes (TR = 2530 ms, Single (Mumford et al., 2014) design to yield a beta-series
TE = 1640–7040 μs, 7◦ flip angle, 256 mm2 FOV, 176 slices, 1 mm3 (Rissman et al., 2004). For each run within an individual subject,
isotropic voxels). Blood oxygenation level dependent (BOLD) a fixed-effects GLM was created for each “Far” (regulation of
signal during functional runs was recorded using a gradient- aversive stimuli) and “Look”-Negative (passive observation of
echo T2∗-weighted echoplanar imaging (EPI) sequence. A total aversive stimuli) “target trial”. Thus, a separate GLM was created
of 32 3-mm thick slices were acquired parallel to the AC-PC line for each individual target trial. Each GLM consisted of a single-
(TR = 2000 ms, TE = 30 ms, 90◦ flip angle, 256 × 256 FOV, 64 × 64 trial regressor which modeled that GLM’s target trial, a nuisance
matrix size). regressor modeling all other events in the target trial’s condition,
regressors for the other two conditions—each modeled as
they normally would be in a standard univariate analysis.
fMRI Data Preprocessing For example, the first “Far”-Negative trial for a participant
The FMRIB Software Library package (FSL, version 5.0.9; [Link] would receive its own GLM wherein it was modeled in a
[Link]) was used for preprocessing and analysis. Prior to pre- single regressor; all other “Far”-Negative trials were placed
processing, data were visually inspected for artifacts, anatom- in a separate nuisance regressor while “Look”-Negative and
ical abnormalities, and excessive head motion. Preprocessing “Look”-Neutral trials modeled in their own respective regressors.
began by using the brain extraction tool (BET) to remove non- Afterwards, a new GLM was created for the second “Far”-
brain tissue from images and estimating the extent of par- Negative trial, where it was the ‘target trial’ and modeled in
ticipant head motion by using FSL Motion Outliers to record a single-event regressor and the first trial is now modeled in
volumes that exceeded a 0.9 mm framewise displacement (FD) the separate nuisance regressor with other trials of that type.
threshold (Siegel et al., 2014). Fortunately, head motion was FSL’s extended motion parameters estimated from MCFLIRT
minimal: The average number of volumes exceeding our FD (rotation+translation, derivatives, and squares of each) and
threshold per run, per participant was 3.014 (SD = 6.09, range = 0– additional volumes that exceeded our FD = 0.9 mm threshold
30.75). Head motion as not appreciably correlated with age or were also added as regressors of noninterest to each first-level
emotion regulation ability scores (both rs < |.05|). Afterwards, we LSS GLM.
corrected for head motion by spatially realigning volumes using Following estimation of the LSS GLMs, we used parameter
MCFLIRT and then hi-pass filtered the data (100 s cutoff). We estimates from each trial-specific GLM to create linear contrast
prewhitened the data to correct for autocorrelated residuals. images comparing both trial types of interest (“Far”-Negative
& “Look”-Negative), respectively, to baseline. We then extracted
Network Definition estimates of activation for both trial types across of ROIs for
We initially began with two prospective network parcellations to all networks. This yielded two t × p × n arrays (one for each
choose from: the Schaefer 7 network parcellation (100 ROIs) and task condition), where entries to each cell represent the average
the Schaefer 17 network parcellation (400 ROIs) (Schaefer et al., parameter estimate at the t-th trial for the p-th ROI in the n-
2018). For both parcellations, spheres (4 mm radius) were drawn th network. For every network, beta-series among all its ROIs
across all peaks in order to extract data necessary to compute were correlated among each other (Spearman’s Rho), yielding
connectivity (described in the next section). A 4 mm radius two connectivity matrices (one for each trial type) per network.
was used to minimize overlap between spheres. We made the These matrices describe each network’s connectivity for a given
decision to extract data from identically sized spheres to ensure task condition (i.e., connectivity matrix), and we refer to them
that differential effects between networks were not driven by as network connectivity profiles (See Fig. 1 for an overview).
differences in volume. We ultimately selected the Schaefer 17
parcellation based on evaluating fit statistics from several dif- Network Differentiation
ferent possible models (see “Statistical Analysis” section). The first feature of network activity we examined was network
In addition to the networks in the Schaefer7 and Schae- differentiation, or the distinctness between a network’s con-
fer17 parcellations, we also included 32 ROIs from a “cognitive nectivity between two task states. We used RSA to quantify
reappraisal network” (CRN; Buhle et al., 2014). These ROIs are the degree to which the “Far”-Negative and “Look”-Negative
commonly observed when individuals engage in cognitive reap- conditions were distinct from each other in a given network (in
praisal, the emotion regulation strategy employed here. Individ- terms of connectivity).
ual nodes of the CRN are available elsewhere (see supplement
of Guassi Moreira et al., 2019). We decided to include the CRN Using RSA to Quantify Network Differentiation
because it is the most relevant network to the emotion regula- The first feature of network activity we chose to examine was
tion task incorporated in the present study. Recent work in both differentiation. In order to estimate network differentiation, we
 Network Architecture of Emotion Regulation Neurodevelopment Guassi Moreira et al. 4143

Figure 1. Overview of network activity metrics estimation. Note. “Far” and “Look” in the figure refer to the “Far”-Negative and “Look”-Negative trial types. A. Top panel
(Network Connectivity via Beta-Series) depicts the beta-series estimated for each task condition (regulate (“Far”-Negative), emotional baseline (“Look”-Negative)), for
the n-th network. Each column corresponds to the p-th ROI, each row to the t-th trial, and each entry is the average beta-value for a given ROI at a given trial. Beta-series
for each subject served as the substrates for estimating our network activity metrics of interest, depicted in the bottom row of panels. The middle panel (Network
Differentiation via RSA) depicts process of computing network differentiation. The two boxes represent connectivity profiles (matrices) for each condition for the
 4144 Cerebral Cortex, 2021, Vol. 31, No. 9

compared profiles of network connectivity when youth were matrices. An adjacency matrix summarizes a graph—each row
engaging in emotion regulation versus when they were simply and column represent the ROIs (nodes) that comprise the net-
viewing emotional stimuli using RSA on our two task-specific work, while an entry into the off-diagonal elements of the
connectivity matrices (Kriegeskorte, 2008; Kriegeskorte et al., matrix denotes the presence or absence of a connection (edge)
2008). Traditionally, RSA is a form of multivoxel pattern anal- between two nodes. It is customary to set an edge-defining thresh-
ysis that relies on the presumption that multivoxel activation old, a value at which all correlation estimates (e.g., correlation
patterns reliably contain information about a specific stimulus between two ROIs’ beta-series) at or falling below the threshold
(Kriegeskorte et al., 2009; Etzel et al., 2013). Comparison of two are set to zero, and all estimates above the threshold are set to
stimulus patterns (via correlation) in RSA can reveal the extent 1 (unweighted edges) or left at their original value (weighted
to which representations of stimuli encode unique or similar edges) (Bolt, Nomi, Yeo, & Uddin, 2017). In our case, we used
information. However, one is not bound to use RSA solely with weighted edges. We varied the edge-defining threshold as part
multivariate patterns. In fact, RSA can be used to compare of our model specification selection procedure (see Statistical
any two different representations because information about Analysis section).
pattern representations is ultimately abstracted away from the We then estimated the number of communities in each
modality in which it was initially acquired and forced into a task condition, for all networks, for all subjects using the R
common space. To this point, prior research has used RSA to igraph package’s walktrap clustering algorithm (Pons & Lat-
compare voxelwise patterns in the brain to behavioral patterns apy, 2005; Csardi & Nepusz, 2006) in the R statistical software
(Parkinson et al., 2017), behavioral patterns to other behavioral language. The walktrap clustering algorithm (Pons & Latapy,
patterns (Brooks & Freeman, 2018; Stolier et al., 2018), and, most 2005) estimates community structure in a graph using random
importantly for our purposes, representations of brain connec- walks—step-wise paces from one node to another on a graph
tivity patterns (Lee et al., 2017). We built upon and extended according to random chance. The algorithm operates on the
these prior implementations of RSA for our purposes in com- principle that the probability of taking a walk (i.e., moving
paring network connectivity states between task conditions. over) from one node to another depends on the number of
As noted above, each network and task condition, a r × shared connections between nodes. This ultimately means that
r connectivity matrix was constructed, where r is the num- walks are more likely to become “trapped” inside a commu-
ber of ROIs in the network. RSAs involved vectorizing the off- nity than a noncommunity due to the dense interconnections
diagonal elements of each task condition’s various connectivity between nodes that define community membership. The num-
matrices, Fisher transforming them (for variance stabilization), ber of a given network’s communities across both task condi-
and then correlating the elements (Spearman). This yielded a tions were extracted from the graphs and differenced (“Far”-
scalar value that describes the extent to which a given net- Negative–“Look”-Negative), resulting in metrics of community
work’s connectivity profile differs between task conditions (see structure, one per network, per participant. A higher value on
Fig. 1). An extreme high value indicates that the pattern of this metric indicates a greater number of observed communi-
connectivity across the entire network remains consistent dur- ties within a network during emotion regulation, compared to
ing “Far”-Negative (Regulate) and “Look”-Negative (Emotional passively viewing emotional stimuli.
Baseline) trials; an extreme low value (anticorrelation) indicates
a differentiated connectivity profile between the two task states.
Modularity Differences Between Network States
Our final network metric was the difference in modularity
Network Community Structure
between task conditions for a given network. Building upon
The next feature of network activity we examined was com-
the concept of community structure, a graph is said to be highly
munity structure between task conditions (e.g., when regulat-
modular if it exhibits relatively dense interconnections within
ing versus when passively viewing emotional stimuli) (Sporns,
its communities and relatively sparse connections between
2010). That is, we examined the number of communities within
them. On the other hand, a graph that is low in modularity
a network, or subclusters comprised of densely interconnected
has nodes that have relatively equal connections within
nodes within the network, as well as crosstalk (i.e., modularity)
and between their communities. In other words, modularity
between communities.
can be thought to measure the extent to which “cross-talk”
exists between the communities within a network (i.e., a
Community Number Differences Between Network States ratio between intracommunity versus intercommunity ties).
When examining community structure, we first quantified the Modularity metrics across both task states in each network
number of communities within each network as a function of across all participants were obtained using the communities
task condition. To achieve this, we used the aforementioned identified in the prior section. The same differencing was
connectivity matrices to create undirected, weighted adjacency taken (“Far”-Negative–“Look”-Negative) to obtain a metric of

nth network. Each entry to the matrix is the connectivity estimate between a given pair of ROIs. Unique off-diagonal elements were vectorized and then correlated
(Spearman’s Rho) to yield a single, scalar value of network connectivity similarity. The bottom panel (Network Community Structure via Graph Theory) depicts graphs
for each task condition for the n-th network; each node represents an ROI within a given network, whereas each edge is the connectivity between a pair of ROIs. This
step allowed us to compute the number of network communities (i.e., number of modules/communities in each network) and modularity. B. The left panel shows
two illustrative diagrams of communities, or neighborhoods of highly interconnected nodes. The left network has two communities while the right network has just
one. The right panel illustrates two networks that differ in terms of modularity, or the strength of cross-talk between different communities, represented by edges
between communities. The left network has greater modularity or less cross-talk because the connection between the two communities is weak (indicated by the
dashed line), whereas the right network has less modularity or more cross-talk because the connection between the two communities is strong (indicated by two
connections between neighborhoods, solid lines). Network connectivity matrices were thresholded to create adjacency matrices and subsequently graphs. Estimates
of community structure and modularity were compared across both task conditions for each network.
 Network Architecture of Emotion Regulation Neurodevelopment Guassi Moreira et al. 4145

modularity differences between task conditions across each Table 1 Model specifications and accompanying RMSE value from
participant’s connectome. Notably, because our focus was on the the model selection process
degree to which information about development and emotion
Specification (Parcellation, RMSE Value
regulation ability is encoded across a set of brain networks,
edge-defining threshold,
we opted to focus on within-network modularity, rather than
detrending)
between-network modularity.
Schaefer17–0.4—detrend 0.8267
Schaefer17–0.6—detrend 0.8853
Schaefer17–0.4—none 0.9299
Statistical Analysis Schaefer17–0.5—none 0.9395
Schaefer17–0.6—none 0.9455
Selecting a Parcellation and Other Modeling Decisions
Schaefer17–0.5—detrend 1.1192
We were faced with a number of different modeling decisions,
Schaefer7–0.4—none 1.3176
including choosing a parcellation (Schaefer 7 or Schaefer 17), Schaefer7–0.5—none 1.3476
an edge-defining threshold (ρ = 0.4, 0.5, 0.6), and whether to Schaefer7–0.6—none 1.5665
linearly detrend the beta-series via OLS regression (detrend, do Schaefer7–0.4—detrend 1.6607
not detrend). In order to do this, we estimated a model using Schaefer7–0.6—detrend 1.6607
all possible specifications (12 total) computed their associated Schaefer7–0.5—detrend 1.6608
root mean square error (RMSE) as a metric of prediction accuracy.
The final model was estimated using specifications that yielded Note: Parcellations tested were the 100 ROI Schaefer7 and 400 ROI Schaefer17;
edge-defining thresholds tested were at Spearman Rho values of 0.4, 0.5, and
the lowest RMSE. Notably, the RMSE was weighted by degrees of
0.6; “detrend” refers to linear detrending of all ROI betaseries via OLS prior
freedom to avoid bias toward models with more predictors. Both to analysis while “none” indicates no detrending. “RMSE” refers to root mean
the Schaefer7 and Schaefer17 networks included CRN ROIs. Only square error, for which smaller values indicate better model fit; our calculation
RMSE was computed for each model in this process, as infer- used degrees of freedom for a given model to calculate MSE instead of the overall
N. Note that cross-validation, which was used to obtain these
ence on the individual regression parameters for each model
specification was not performed. We elected to use this method
for model selection over nested cross-validation because nested
cross-validation would have required us to iteratively subset Results
our data into smaller sets whose sample size would have been Behavioral Results
inappropriate for cross-validation.
Behavioral results from our emotion regulation task have been
reported elsewhere (see Guassi Moreira et al., 2019), but we
briefly recapitulate them here for convenience. Participants’
Model Building Using Ridge Regression average negative affective ratings during the “Look”-Negative
We used ridge regression to estimate associations between net- and “Far”-Negative trials were 2.48 (SD = 0.562) and 1.95
work activity metrics and emotion regulation ability (controlling (SD = 0.499), respectively. This difference was statistically sig-
for age). Ridge regression was selected because (i) we had con- nificant (t(69) = 10.37, P < 0.001). Participants’ ability to engage in
tinuous dependent variables, and because ridge regression (ii) emotion regulation, quantified as the percent reduction in neg-
effectively estimates parameters in a model with many predic- ative affect for the regulate condition relative to the emotional
tors, (iii) handles highly correlated predictor variables, and (iv) baseline condition, improved with age (r = 0.344, P < 0.01).
most importantly, is less susceptible to overfitting and therefore
has better out of sample predictive accuracy (Murphy, 2012; Using Network Approaches to Parse Chronological Age and Emotion
McNeish, 2015). Ridge regression differs from traditional OLS Regulation Ability
regression in that its loss function contains an extra penalty Model Selection. The specified model with the lowest RMSE
term (Penalty bRidge = λbj 2 ). This has the effect of biasing, or used the Schaefer 17 parcellation, an edge-defining threshold
shrinking, coefficients toward zero, especially those with inap- of ρ = 0.4, and linearly detrended beta-series (via OLS). Model
propriately large magnitudes that drive overfitting. This added selection output is listed in Table 1.
bias, introduced via what is known as l2 regularization, helps
prevent overfitting by decreasing sample to sample variability
Predicting Emotion Regulation Ability
in regression coefficients. This produces more generalizable
models while satisfying the traditional aim of making infer- Ridge regression parameters and bootstrapped confidence inter-
ences about population parameters due to enhanced certainty vals are listed in Table 2. When predicting emotion regulation
in parameter estimates (i.e., narrower sampling distribution of ability from our network connectivity metrics, we found that
coefficient estimates). network differentiation in Control Network (subnetwork C)
Here we implemented ridge regression using the glm- and Limbic Network (subnetwork B) were each predictive of
net()function in R. Notably, ridge regression requires a tuning emotion regulation ability such that greater differentiation
parameter (λ) that controls the degree of regularization. In our between these networks’ connectivity profile during active
case, we used the [Link]()function, which finds the optimal λ regulation compared to emotional baseline was related to better
value via cross-validation. To facilitate statistical inference, we emotion regulation ability. Differences in network modularity
computed 95% bootstrapped confidence intervals (percentile) during emotion regulation compared to emotional baseline
around parameter estimates (5000 bootstrapped iterations). were also related to emotion regulation ability. Specifically,
Because the current study was exploratory in nature, we elected greater modularity in Default Mode Network (subnetwork B) was
not to correct for multiple comparisons in order to minimize the associated with increased emotion regulation ability, whereas
risk of a type II error. the opposite was true for the Control Network (subnetwork C).
 4146 Cerebral Cortex, 2021, Vol. 31, No. 9

Table 2 Ridge regression results predicting emotion regulation ability

Term Parameter Estimate 95% CI Term Parameter Estimate 95% CI

sim CoNA −0.466 [−0.179, 0.079] mod SaVANA 1.088 [−0.015, 0.239]
sim CoNB 0.345 [−0.058, 0.195] mod SaVANB 0.003 [−0.127, 0.105]
sim CoNC −1.646 [−0.341, −0.036] mod SMNA 0.856 [−0.042, 0.186]
sim DMNA 0.215 [−0.089, 0.14] mod SMNB −0.121 [−0.151, 0.1]
sim DMNB −0.118 [−0.13, 0.144] mod TPN 0.462 [−0.069, 0.176]
sim DMNC 0.835 [−0.007, 0.254] mod VCN 0.111 [−0.096, 0.126]
sim DANA 0.117 [−0.112, 0.126] mod VPN −0.208 [−0.118, 0.109]
sim DANB −1.078 [−0.245, 0.015] mod CRN 0.389 [−0.073, 0.172]
sim LiNA −0.174 [−0.126, 0.117] com CoNA 0.481 [−0.11, 0.179]
sim LiNB −1.354 [−0.256, −0.022] com CoNB 1.528 [0.017, 0.239]
sim SaVANA −0.195 [−0.14, 0.097] com CoNC 0.443 [−0.114, 0.144]
sim SaVANB 0.041 [−0.111, 0.12] com DMNA 0.445 [−0.081, 0.192]
sim SMNA −0.284 [−0.135, 0.105] com DMNB −0.110 [−0.156, 0.116]
sim SMNB 0.220 [−0.093, 0.153] com DMNC −0.903 [−0.252, 0.038]
sim TPN −0.762 [−0.211, 0.051] com DANA 0.135 [−0.101, 0.159]
sim VCN −0.014 [−0.137, 0.129] com DANB 0.805 [−0.111, 0.177]
sim VPN 0.330 [−0.096, 0.127] com LiNA 0.365 [−0.083, 0.188]
sim CRN −0.734 [−0.222, 0.06] com LiNB 0.372 [−0.087, 0.173]
mod CoNA −0.739 [−0.179, 0.055] com SaVANA 0.564 [−0.104, 0.148]
mod CoNB 0.312 [−0.115, 0.131] com SaVANB 0.851 [−0.05, 0.211]
mod CoNC −1.597 [−0.333, −0.018] com SMNA −0.021 [−0.173, 0.084]
mod DMNA 0.462 [−0.075, 0.143] com SMNB 0.062 [−0.167, 0.119]
mod DMNB 1.971 [0.055, 0.305] com TPN 0.748 [−0.039, 0.227]
mod DMNC −0.429 [−0.157, 0.071] com VCN −0.914 [−0.227, 0.026]
mod DANA 0.381 [−0.066, 0.151] com VPN 0.868 [−0.027, 0.197]
mod DANB 0.163 [−0.112, 0.163] com CRN 0.465 [−0.076, 0.197]
mod LiNA −0.559 [−0.193, 0.087] AGE 1.922 [0.042, 0.301]
mod LiNB 0.310 [−0.115, 0.120] − − −

Note: “sim” refers to network differentiation (obtained via RSA); “mod” refers to differences in modularity (regulation – baseline); “com” refers to differences in
community structure, i.e., the number of communities (regulation – baseline). Greater differentiation is indicated by relatively lower similarity scores (i.e., higher
similarity indicates that a network’s connectivity is less differentiated between task states). CoN refers to control network; DMN refers to default mode network; DAN
refers to dorsal attention network; LiN refers to limbic network; SaVAN refers to salience/ventral attention network; SMN refers to somatomotor network; TPN refers
to temporal–parietal network; VCN refers to visual center network; VPN refers to visual peripheral network; CRN refers to cognitive reappraisal network. Letters next
to each network indicate subnetwork. 95% confidence intervals ref lect bootstrapped percentile intervals (5000 bootstraps). Predictors whose 95% CI did not include
zero, thus indicating statistical significance, are bolded.

Only community differences in Control Network (subnetwork between 4.49% and 9.84% of the variance in emotion regulation
B) were related to emotion regulation ability, such that more ability, while age uniquely accounted for 4.65% of the variance in
communities during regulation versus emotional baseline were emotion regulation ability (Table 3). Because these are squared
associated with better emotion regulation ability. Notably, all of semipartial correlations, these estimates of variance accounted
these effects were significant while statistically adjusting for all for reflect statistical adjustment of the other significant terms
terms in the model, including age. Age was directly associated in the model.
with emotion regulation ability after statistically adjusting for
all network terms.
Given that multiple subnetworks of Control Network showed Discussion
significant associations with emotion regulation ability, we We set out to interrogate the neurodevelopment of emotion
decided to examine the correlation between Control Network regulation using network neuroscience by parsing brain network
subnetwork predictors. While the use of ridge regression features that contribute to correlated but distinct developmental
theoretically obviates concerns about collinearity among processes (chronological age and acquisition of emotion
predictors, it does not address the degree to which these regulation abilities). This work builds upon past network
subcomponents represent distinct entities. Fortunately, none neuroscience studies of development to better understand task-
of the Control Network subnetworks were highly correlated specific aspects of maturation (Rudolph et al., 2017). We found
across the three metrics (differentiation rs: 0.26–.40; modularity that emotion regulation ability showed significant associations
rs: −.07–.26; community rs: 0.11–.38), indicating that these with whole-brain network metrics in the control network,
subnetworks are related but distinct units, in the context of default mode network, and the limbic network. Generally,
emotion regulation. because greater network differentiation and modularity in
Last, in order to determine the variance in regulation ability these networks was associated with better regulation ability,
that was uniquely predicted by the aforementioned network it suggests these networks develop specialized features that
metrics as well as age, we calculated squared semipartial lend themselves to better implementing emotion regulation.
correlations. Network metrics each uniquely accounted for Importantly, these combined network properties explained
 Network Architecture of Emotion Regulation Neurodevelopment Guassi Moreira et al. 4147

Table 3 Semipartial correlations between select brain network met- number of network communities in Control Network (relative
rics and emotion regulation ability to baseline) was related with better regulatory ability. It is
noteworthy that prior work found that increasing structural
Predictor Squared
modularity of the Control Network is associated with increased
Semipartial
cognitive skills across age (Baum et al., 2017). In tandem, these
Correlation
results suggest that greater functional subdivision in the Control
Age 4.65% Network (more communities) is related to better regulatory
Control Network differentiation 7.18% ability, but only when functional modules are engaging in
Limbic Network differentiation 4.64% enough “cross-talk” (less modularity). Thus, the Control Network
Control Network modularity 4.49% best supports developing emotion regulation abilities when it
Default Mode Network modularity 9.84% is characterized by ready communication between specialized
Control Network community number 5.37% modules. Similar to the Default Mode Network findings, the
specific subnetwork of the Control Network involved most
Note: Semipartial correlations ref lect unique variance that each predictor in the
table explains in emotion regulation ability after partialling out shared variance
consistently in our study was comprised of individual ROIs
with the other predictors. In other words, they are Pearson correlations between that have been implicated in visuospatial imagery and self-
emotion regulation ability and the residuals of a given predictor by the others. referential processing (Cavanna & Trimble, 2006). These are
They are reported here in a percent metric to facilitate interpretation of unique
two psychological processes that are heavily involved in the
variance accounted for by each predictor.
distancing variant of reappraisal we employed and further
suggest functional specialization supports fine-tuning of the
a substantial amount of the variance in emotion regulation cognitive skills needed to engage in reappraisal.
ability over and above age. These findings have implications for Overall, our findings speak to the nature of neurodevelop-
our basic understanding of the neurodevelopment of emotion ment. Specifically, they suggest that maturation is driven by
regulation, and how brain networks may broadly change and changes in network states. Rather than viewing neural network
interact across cognitive and emotional development. development as monolithic, this perspective suggests that dif-
In the present study, we were able to explain variance in emo- ferent psychological states—for example, regulating emotions
tion regulation ability using network activity metrics (cumula- versus not regulating emotions—elicit unique and discrete con-
tive variance explained by brain predictors after controlling for figurations of activity and connectivity within a single network
age = 36.17%), even after accounting for age (variance explained (Shine et al. 2019a, 2019b), and that the differences between
after controlling for brain predictors = 4.65%). These results these states varies as a function of age and skill development.
suggest that task-based network activity encodes information While some network dimensions may be more consequential
about skill development associated with a given task rather than for general maturation (and thus, are likely to correlate strongly
domain general features of development (i.e., age). We found with age), others are likely more impactful for shaping task-
that greater differentiation in the Control and Limbic Networks specific development (for example, those we observed here to be
were associated with increased emotion regulation ability. These related to regulation ability). Our data preclude us from verifying
results are directly in line with what is posited by dominant this possibility, yet future work equipped with more data from a
neurobiological models of cognitive emotion regulation func- variety of different sources could successfully test this.
tions (Ochsner et al., 2012; Etkin, Büchel, & Gross, 2015). These Our results highlight how studying whole-brain networks
models posit that bottom-up affective signals generated in may inform contemporary neuroscientific models of emotion
limbic regions are modulated by top-down cognitive processes regulation. Current accounts of emotion regulation posit that
in prefrontal cortical areas. Our data support this notion by individual differences in emotion regulation ability are driven by
showing that youths with better regulation skills show more activity among a set of brain regions that comprise a dedicated
differentiated patterns of connectivity in the Control and Limbic network for a given regulatory strategy (Ochsner et al., 2012),
Networks. That is, connectivity in these networks looks different while other research has only recently begun to consider the
when regulating than when passively experiencing emotions. role of canonical whole-brain networks as meaningful biological
We also found that differences in Default Mode and Control units (Zhang et al., 2019). Although the notion of dedicated net-
Network modularity were associated with regulation ability, works for emotion regulation strategy is enticing, our research
albeit in different ways. Greater modularity in the Default Mode fails to find support that a Cognitive Reappraisal Network is
Network during regulation, relative to baseline, was related to associated with individual differences in emotion regulation
better ability, whereas the opposite trend was found with Con- ability. This may be because the network itself does not consti-
trol Network. This is consistent with recent neurodevelopmental tute a meaningful biological unit, despite prior evidence indi-
findings showing that increasing functional subdivision of the cating otherwise, or because variability in both network activity
Default Mode Network facilitated better mentalizing abilities on and regulation ability were truncated from an otherwise full
a theory of mind task (Richardson et al., 2018), as well as clinical range in our sample (i.e., we only sampled a limited range of
work linking enhanced cross-network connectivity and activity the population values of network activity metrics). Yet another
in the Default Mode Network with emotion regulation disorders possibility is that Cognitive Reappraisal Network activity during
such as Major Depression ( Sheline et al., 2009; Liu et al., 2012). reappraisal may actually emerge as an extension or special
Because cognitive reappraisal relies on higher order executive form of another network (e.g., the Control Network) whose core
functions such as self-referential thought that are widely activity might be more tightly linked with regulatory ability.
considered to be supported by Default Mode Network (Buckner
& Carroll, 2007), it is possible that the anatomical substrates
Limitations and Future Directions
of the network segregate to support functional specialization.
While the opposite trend with the Control Network can appear This study is accompanied by several limitations. First,
initially contradictory, it is worth noting that evincing a greater although our sample size is fairly large for task-based pediatric
 4148 Cerebral Cortex, 2021, Vol. 31, No. 9

neuroimaging, and is twice the median cell size of human fMRI Brooks JA, Freeman JB. 2018. Conceptual knowledge predicts the
studies (Poldrack et al., 2017), it is nevertheless somewhat small representational structure of facial emotion perception. Nat.
in comparison to other network neuroscience studies (e.g., Hum. Behav . doi: 10.1038/s41562-018-0376-6.
Baum et al., 2017; DuPre & Spreng, 2017). Further, although Buckner RL, Carroll DC. 2007. Self-projection and the brain.
cross-sectional research is an adequate starting point for Trends Cogn Sci. 11(2):49–57. doi: 10.1016/[Link].2006.11.004.
characterizing the neurodevelopment of emotion regulation, Calkins S. 1994. Origins and outcomes of individual differences
future research would benefit from longitudinal data to examine in emotion regulation. Monographs of the Society for Research in
how the trajectory of each network’s activity change across Child Development. 59(2–3):53–72.
time and relate to changes in emotion regulation ability. Casey BJ. 2015. Beyond simple models of self-control to
Additionally, amid growing concerns about the reliability circuit-based accounts of adolescent behavior. Annu. Rev.
of computerized tasks intended to measure psychological Psychol. 66:295–319. doi: 10.1146/annurev-psych-010814-
processes and individual differences (Elliott et al., 2019; Enkavi 015156.
et al., 2019), we note here that the behavioral and neural test– Casey BJ, Jones RM, Hare TA. 2008. The adolescent brain. Ann. N.
retest reliability of the emotion regulation task used here has Y. Acad. Sci. 1124:111–126. doi: 10.1196/annals.1440.010.
not yet been established. A final future direction involves testing Cavanna AE, Trimble MR. 2006. The precuneus: a review of
whether the predictive properties of the networks studied its functional anatomy and behavioural correlates. Brain.
here are relevant for other emotional states, or even across 129(3):564–583.
different types of emotion regulation (e.g., extinction learning; Csardi G, Nepusz T. 2006. The igraph software package for
expressive suppression). Addressing these concerns under complex network research. Inter Journal, Complex Sy. 1695.
further empirical scrutiny will help lend further confidence DuPre E, Spreng RN. 2017. Structural covariance networks across
in the conclusions raised by our results. the lifespan, from 6-94 years of age. Network Neuroscience.
In sum, this study is the first to our knowledge that examines 1–38. doi: 10.1162/NETN_a_00016.
how the developing brain supports the acquisition of emotion Elliott ML, Knodt AR, Ireland D, Morris ML, Ramrakha S, Sison
regulation, a central feature of lifespan mental health. These ML, Hariri AR. 2019. Poor test-retest reliability of task-fMRI:
results are the first show how canonical whole-brain networks new empirical evidence and a meta-analysis. BioRxiv. 919:
comprise the functional architecture of emotion regulation skill 1–31.
acquisition across youth. Enkavi AZ, Eisenberg IW, Bissett PG, Mazza GL, Mackinnon DP,
Marsch LA, Poldrack RA. 2019. Large-scale analysis of test –
retest reliabilities of self-regulation measures. Proc. Natl. Acad.
Funding Sci. 116(12):5472–5477. doi: 10.1073/pnas.1818430116.
Ernst M, Pine DS, Hardin M. 2006. Triadic model of the
This research was supported by Grant R01-MH103291 from the neurobiology of motivated behavior in adolescence. Psy-
National Institute of Mental Health award to K.A.M. Preparation chol Med. 36(3):299–312. doi: 10.1017/S0033291705005891.
of the article was supported by a National Science Foundation Triadic.
Graduate Research Fellowship (NSF-GRF) to J.F.G.M. (Fellow ID: Etkin A, Büchel C, Gross JJ. 2015. The neural bases of
2016220797). emotion regulation. Nat. Rev. Neurosci.. 16(11):693–700. doi:
10.1038/nrn4044.
Etzel JA, Zacks JM, Braver TS. 2013. Searchlight analysis:
Notes promise, pitfalls, and potential. Neuro Image. 78:261–269. doi:
10.1016/[Link].
We thank the members of the Harvard Stress & Development lab
Gee DG, Gabard-Durnam L, Telzer EH, Humphreys KL, Goff
for their help in data collection. Conflict of Interest: None declared.
B, Shapiro M, Tottenham N. 2014. Maternal buffering of
human amygdala-prefrontal circuitry during childhood but
not during adolescence. Psychol. Sci. 25(11):2067–2078. doi:
References 10.1177/0956797614550878.
Baum GL, Ciric R, Roalf DR, Betzel RF, Moore TM, Shinohara Gratton C, Laumann TO, Nielsen AN, Greene DJ, Gordon EM,
RT, Satterthwaite TD. 2017. Modular segregation of struc- Gilmore AW, Petersen SE. 2018. Functional brain networks
tural brain networks supports the development of execu- are dominated by stable group and individual factors, not
tive function. Curr Biol. 27(11):1561–1572.e8. doi: 10.1016/[Link]. cognitive or daily variation. Neuron. 98(2):439–452.e5. doi:
2017.04.051. 10.1016/[Link].2018.03.035.
Bolt T, Nomi JS, Yeo BTT, Uddin LQ. 2017. Data-driven extrac- Greene AS, Gao S, Scheinost D, Constable RT. 2018. Task-induced
tion of a nested model of human brain function. J. Neurosci. brain state manipulation improves prediction of individual
0323–0317. doi: 10.1523/JNEUROSCI.0323-17.2017 traits. Nat. Commun. 9(1):1–13.
Buhle JT, Silvers JA, Wager TD, Lopez R, Onyemekwu C, Kober Gross JJ. 1998. The emerging field of emotion regulation:
H, Weber J, Ochsner KN. 2014. Cognitive reappraisal of emo- an integrative review. Rev. Gen. Psychol. 2(5):271–299. doi:
tion: a meta-analysis of human neuroimaging studies. Cereb. 10.1017.S0048577201393198.
Cortex. 24(11), 2981–2990. Gross JJ. 2015. Emotion regulation: current status
Brody GH, Yu T, Nusslock R, Barton AW, Miller GE, Chen and future prospects. Psychol. Inq. 26(1):1–26. doi:
E, Holmes C, McCormick M, Sweet LH. 2019. The protec- 10.1080/1047840X.2014.940781.
tive effects of supportive parenting on the relationship Guassi Moreira JF, McLaughlin KA, Silvers JA. 2019. Spatial and
between adolescent poverty and resting-state functional temporal cortical variability track with age and affective
brain connectivity during adulthood. Psychological Science. experience during emotion regulation in youth. Dev. Psychol.
30(7), 1040–1049. 55(9):1921–1937.
 Network Architecture of Emotion Regulation Neurodevelopment Guassi Moreira et al. 4149

Kim P, Evans GW, Angstadt M, Ho SS, Sripada CS, Swain JE, transparent and reproducible neuroimaging research. Nat.
Phan KL. 2013. Effects of childhood poverty and chronic Rev. Neurosci. 18:115–126. [Link]
stress on emotion regulatory brain function in adult- Pons P, Latapy M. 2005. Computing communities in large net-
hood. Proc Natl Acad Sci U S A. 110:18442–18447. doi: works using random walks. In: International Symposium on
10.1073/pnas.1308240110. Computer and Information Sciences. Berlin, Heidelberg: Springer,
Kleckner IR, Zhang J, Touroutoglou A, Chanes L, Xia C, Simmons pp. 284–293.
WK, Feldman Barrett L. 2017. Evidence for a large-scale brain Richardson H, Lisandrelli G, Riobueno-Naylor A, Saxe
system supporting allostasis and interoception in humans. R. 2018. Development of the social brain from age
Nat. Hum. Behav. 1(5). doi: 10.1038/s41562-017-0069. three to twelve years. Nat. Commun. 9(1):1027. doi:
Kriegeskorte N. 2008. Representational similarity analysis – con- 10.1038/s41467-018-03399-2.
necting the branches of systems neuroscience. Front. Syst. Rissman J, Gazzaley A, D’Esposito M. 2004. Measuring functional
Neurosci. 2:1–28. doi: 10.3389/neuro.06.004.2008. connectivity during distinct stages of a cognitive task. Neu-
Kriegeskorte N, Mur M, Ruff DA, Kiani R, Bodurka J, Esteky H, roImage. 23(2):752–763. doi: 10.1016/[Link].2004.06.035.
Bandettini PA. 2008. Matching categorical object representa- Rudolph MD, Miranda-Domínguez O, Cohen AO, Breiner K,
tions in inferior temporal cortex of man and monkey. Neuron. Steinberg L, Bonnie RJ, Fair DA. 2017. At risk of being
60(6):1126–1141. doi: 10.1016/[Link].2008.10.043. risky: the relationship between “brain age” under emotional
Kriegeskorte N, Simmons WK, Bellgowan PS, Baker CI. 2009. Cir- states and risk preference. Dev. Cogn. Neurosci. 24:93–106. doi:
cular analysis in systems neuroscience: the dangers of dou- 10.1016/[Link].2017.01.010.
ble dipping. Nat Neurosci. 12(5):535–540. doi: 10.1038/nn.2303. Siegel JS, Power JD, Dubis JW, Vogel AC, Church JA, Schlaggar
Lang PJ, Bradley MM, Cuthbert BN. 2008. International affec- BL, Petersen SE. 2014. Statistical improvements in functional
tive picture system (IAPS): affective ratings of pictures magnetic resonance imaging analyses produced by censor-
and instruction manual. Tech. Rep A-8. doi: 10.1016/[Link]. ing high-motion data pointspreference. Hum. Brain Mapp.
2006.03.016. 35(5):1981–1996.
Lee TH, Miernicki ME, Telzer EH. 2017. Families that fire together Schaefer A, Kong R, Gordon EM, Laumann TO, Zuo X, Holmes
smile together: resting state connectome similarity and AJ, Yeo BTT. 2018. Local-global parcellation of the human
daily emotional synchrony in parent-child dyads. NeuroImage. cerebral cortex from intrinsic functional connectivity MRI.
152:31–37. doi: 10.1016/[Link].2017.02.078. Cereb. Cortex. 25:3095–3114. doi: 10.1093/cercor/bhx179.
Liu C, Ma X, Li F, Wang Y, Tie C, Li S, Wang C. 2012. Regional homo- Sheline YI, Barch DM, Price JL, Rundle MM, Vaishnavi SN,
geneity within the default mode network in bipolar depres- Snyder AZ, Raichle ME. 2009. The default mode network
sion: a resting-state functional magnetic resonance imaging and self-referential processes in depression. Proceedings
study. PLoS One. 7(11):1–7. doi: 10.1371/[Link].0048181. of the 33rd International Conference on. Mach. Learn. 106(6):
McNeish DM. 2015. Using lasso for predictor selection and 1942–1947.
to assuage overfitting: a method long overlooked in Shine JM, Breakspear M, Bell PT, Martens KAE, Shine R,
behavioral sciences. Multivariate Behav Res. 50:471–483. Koyejo O, Poldrack RA. 2019a. Human cognition involves
doi: 10.1080/00273171.2015.1036965. dynamic integration of neural activity and neuromodula-
McRae K, Gross JJ, Weber J, Robertson ER, Sokol-Hessner P, tory systems. Nat Neurosci. 22:289–299. doi: 10.1038/s41593-
Ray RD, Ochsner KN. 2012. The development of emotion 018-0312-0.
regulation: an fMRI study of cognitive reappraisal in chil- Shine JM, Hearne LJ, Breakspear M, Poldrack RA, Mattingley
dren, adolescents and young adults. Soc. Cogn. Affect. Neurosci. JB, Cocchi L. 2019b. The low-dimensional neural
7(1):11–22. doi: 10.1093/scan/nsr093. architecture of cognitive complexity is related to
Mumford JA, Davis T, Poldrack RA. 2014. The impact of study activity in medial thalamic nuclei. Neuron. 104:1–7. doi:
design on pattern estimation for single-trial multivariate 10.1016/[Link].2019.09.002.
pattern analysis. NeuroImage. 103:130–138. doi: 10.1016/j. Silvers JA, Insel C, Powers A, Franz P, Helion C, Martin
neuroimage.2014.09.026. RE, Ochsner KN. 2017. vlPFC – vmPFC – amygdala
Murphy KP. 2012. Machine learning: A probabilistic perspective. interactions underlie age-related fifferences in cognitive
Cambridge (Ma): The MIT Press. regulation of emotion. Cereb. Cortex. 27(7):3502–3514.
Ochsner KN, Silvers JA, Buhle JT. 2012. Review doi: 10.1093/cercor/bhw073.
and evolving model of the cognitive control of Silvers JA, McRae K, Gabrieli JDE, Gross JJ, Remy KA, Ochsner
emotion. Ann. N. Y. Acad. Sci. (1251):E1–E24. doi: KN. 2012. Age-related differences in emotional reactivity,
10.1111/[Link]. regulation, and rejection sensitivity in adolescence. Emotion.
Parkinson C, Kleinbaum AM, Wheatley T. 2017. Spontaneous 12(6):1235–1247. doi: 10.1037/[Link]-Related.
neural encoding of social network position. Nat. Hum. Behav. Silvers JA, Shu J, Hubbard AD, Weber J, Ochsner KN. 2015. Con-
1(5):1–7. doi: 10.1038/s41562-017-0072. current and lasting effects of emotion regulation on amyg-
Perino MT, Miernicki ME, Telzer EH. 2016. Letting the good times dala response in adolescence and young adulthood. Dev. Sci.
roll: adolescence as a period of reduced inhibition to appet- 18(5):771–784. doi: 10.1111/desc.12260.
itive social cues. Soc. Cogn. Affect. Neurosci. 11(11):1762–1771. Sizemore AE, Giusti C, Kahn A, Vettel JM, Betzel RF, Bassett
doi: 10.1093/scan/nsw096. DS. 2018. Cliques and cavities in the human connectome. J.
Pitskel NB, Bolling DZ, Kaiser MD, Crowley MJ, Pelphrey Comput. Neurosci. 44:115–145.
KA. 2011. How grossed out are you? The neural Sporns O. 2010. Networks of the Brain. Cambridge, MA: MIT Press.
bases of emotion regulation from childhood to Stolier RM, Hehman E, Keller MD, Walker M, Freeman JB. 2018.
adolescence. Dev. Cogn. Neurosci. 1(3):324–337. doi: The conceptual structure of face impressions. Proceedings of
10.1016/[Link].2011.03.004. the National Academy of Sciences. doi: 10.31234/[Link]/d9ha7.
Poldrack RA, Baker CI, Durnez J, Gorgolewski KJ, Matthews PM, Thompson R, Lewis MD, Calkins S. 2008. Reassessing emotion
Munafò M, Yarkoni T. 2017. Scanning the horizon: towards regulation. Child Dev. Perspect. 2(3):124–131.
 4150 Cerebral Cortex, 2021, Vol. 31, No. 9

Troy AS, Wilhelm FH, Shallcross AJ, Mauss IB. 2010. Seeing the human association cortex. Cereb. Cortex. 25(10):3654–3672.
silver lining: cognitive reappraisal ability moderates the rela- doi: 10.1093/cercor/bhu217.
tionship between stress and depressive symptoms. Emotion. Zhang Y, Padmanabhan A, Gross JJ, Menon V. 2019. Development
10(6):783–795. doi: 10.1037/[Link]. of human emotion circuits investigated using a big-data ana-
Yeo BTT, Krienen FM, Eickhoff SB, Yaakub SN, Fox PT, Buck- lytic approach: stability, reliability, and robustness. J. Neurosci.
ner RL. 2015. Functional specialization and flexibility in 39(36):7155–7172.