r Human Brain Mapping 38:5740–5755 (2017) r

Intrinsic Functional Connectivity Variance and

State-Specific Under-Connectivity in Autism

Heng Chen,1 Jason S. Nomi,2 Lucina Q. Uddin,2 Xujun Duan,1* and

Huafu Chen 1*
1
Key laboratory for NeuroInformation of Ministry of Education, School of Life Science and
Technology and Center for Information in BioMedicine, University of Electronic Science and
Technology of China, Chengdu 610054, People’s Republic of China
2
Department of Psychology, University of Miami, Coral Gables, Florida

r r

Abstract: Autism spectrum disorder (ASD) is a neurodevelopmental condition associated with altered
brain connectivity. Previous neuroimaging research demonstrates inconsistent results, particularly in
studies of functional connectivity in ASD. Typically, these inconsistent findings are results of studies
using static measures of resting-state functional connectivity. Recent work has demonstrated that func-
tional brain connections are dynamic, suggesting that static connectivity metrics fail to capture
nuanced time-varying properties of functional connections in the brain. Here we used a dynamic func-
tional connectivity approach to examine the differences in the strength and variance of dynamic func-
tional connections between individuals with ASD and healthy controls (HCs). The variance of dynamic
functional connections was defined as the respective standard deviations of the dynamic functional
connectivity strength across time. We utilized a large multicenter dataset of 507 male subjects (209
with ASD and 298 HC, from 6 to 36 years old) from the Autism Brain Imaging Data Exchange (ABIDE)
to identify six distinct whole-brain dynamic functional connectivity states. Analyses demonstrated
greater variance of widespread long-range dynamic functional connections in ASD (P < 0.05, NBS
method) and weaker dynamic functional connections in ASD (P < 0.05, NBS method) within specific
whole-brain connectivity states. Hypervariant dynamic connections were also characterized by weaker
connectivity strength in ASD compared with HC. Increased variance of dynamic functional connec-
tions was also related to ASD symptom severity (ADOS total score) (P < 0.05), and was most promi-
nent in connections related to the medial superior frontal gyrus and temporal pole. These results
demonstrate that greater intraindividual dynamic variance is a potential biomarker of ASD. Hum Brain
Mapp 38:5740–5755, 2017. VC 2017 Wiley Periodicals, Inc.

Additional Supporting Information may be found in the online
version of this article.
Contract grant sponsor: 863 project; Contract grant number:
2015AA020505; Contract grant sponsor: Natural Science Founda-
tion of China; Contract grant numbers: 61533006 and 61673089;
Contract grant sponsor: Fundamental Research Funds for the
Central Universities; Contract grant numbers: ZYGX2014J078 and
ZYGX2015J141; Contract grant sponsor: National Institute of Men-
tal Health Career Development Award; Contract grant number:
K01MH092288; Contract grant sponsor: National Institute of Men-
tal Health; Contract grant number: R01MH107549
*Correspondence to: Hua-Fu Chen and Xujun Duan, Key labora-
tory for NeuroInformation of Ministry of Education, School of
Life Science and Technology and Center for Information in Bio-
Medicine, University of Electronic Science and Technology of
China, Chengdu 610054, People’s Republic of China. E-mails:
chenhf@uestc.edu.cn and duanxujun@uestc.edu.cn
Received for publication 8 February 2017; Revised 14 July 2017;
Accepted 30 July 2017.
DOI: 10.1002/hbm.23764
Published online 9 August 2017 in Wiley Online Library
(wileyonlinelibrary.com).

C 2017 Wiley Periodicals, Inc.

V
 r State-Related Functional Connectivity in Autism
Key words: autism spectrum disorder; state-specific under-connectivity; dynamic functional connectiv-
ity; intraindividual variance
r r
INTRODUCTION
Autism spectrum disorder (ASD) is a neurodevelopmen-
tal disorder characterized by social-communication deficits
and stereotyped behaviors [American Psychiatric Associa-
tion, 2013]. The estimates of prevalence of ASD has
increased to 1% or higher in the general population
[Developmental Disabilities Monitoring Network Surveil-
lance Year 2010 Principal Investigators and Centers for
Disease Control and Prevention, 2014]. Previous studies
have shown that ASD is related to early overgrowth of
synapses and later reduced pruning, which could be
important factors underlying recently identified atypical
static functional connectivity (sFC) patterns detected using
fMRI [Tang et al., 2014; Zhan et al., 2014].
Functional neuroimaging studies have reported reduced
sFC between brain regions in ASD [Anderson et al., 2011; Ken-
nedy and Courchesne, 2008; Khan et al., 2013] and an “under-
connectivity” hypothesis of ASD was proposed based on these
results [Just et al., 2004, 2012]. However, the sFC “under-
connectivity” patterns revealed by previous studies are not
consistent. Yerys et al. [2015] reported reduced sFC between
medial prefrontal cortex (mPFC) and precuneus/posterior
cingulate cortex (PCC), temporal pole, and pallidum; Jung
et al. [2014] reported reduced sFC between mPFC and pri-
mary motor and sensory cortices; and von dem Hagen et al.
[2013] found reduced sFC between mPFC and temporo-
parietal junction, insula, and amygdala. Several other resting-
state fMRI studies have reported over-connectivity or a mix-
ture of under- and over-connectivity in ASD [Cerliani et al.,
2015; Keown et al., 2013; Supekar et al., 2013]. The cause of
these contradictory resting-state sFC findings is still unclear,
though several attempts at reconciling these findings have
been undertaken. For example, Uddin et al. [2013] proposed
an age-related under-/over-connectivity theory that predicts
over-connectivity in children with ASD and under-
connectivity in adults with the disorder. Kitzbichler et al. uti-
lized magnetoencephalography to examine frequencies corre-
sponding to static fMRI functional connectivity [Hipp and
Siegel, 2015] and found frequency-related resting-state static
under- and over-connectivity in ASD [Kitzbichler et al., 2015].
Alaerts et al. [2016] found static over-connectivity in males
with ASD and static under-connectivity in females with ASD.
Finally, Hahamy et al. [2015] reported higher intersubject spa-
tial variance of cortical areas related to atypical sFC patterns
of individuals with ASD, and suggested that increased spatial
variance may account for the inconsistent static resting-state
functional connectivity findings.
All the previous studies reporting inconsistent findings in
ASD are based on the assumption that the resting-state
r 5741
r
fMRI signal is static during the course of data acquisition.
However, recent work suggests that fMRI signals during
the resting-state are dynamic, and that several time-varying
dynamic FC (dFC) states can be found throughout the
course of a resting-state scan [Allen et al., 2014; Hutchison
et al., 2013a,2013b; Liu and Duyn, 2013; Zalesky et al.,
2014]. Additionally, state-related dFC has also shown
improved sensitivity in classifying ASD from healthy con-
trols (HC) [Wee et al., 2016]. Other research demonstrates
that dynamic functional connectivity variance (dFCvar) bet-
ter predicts behavior than static functional connectivity
(sFC) strength in the typical population [Jia et al., 2014].
Furthermore, dFCvar shows promise as a potential bio-
marker in psychiatric disorders such as schizophrenia [Yu
et al., 2015], epilepsy [Laufs et al., 2014; Liao et al., 2014b],
and Alzheimer’s disease [Jones et al., 2012]. Thus, dFC
approaches often reveal unique patterns of connectivity that
traditional sFC methods cannot identify [Liu, et al., 2016]. It
is still unclear how the results of dFC approaches may help
to explain the inconsistent findings in sFC studies of ASD.
In this study, we utilized a large multicenter dataset to
examine differences in sFC, dFC, and dFCvar between
ASD and HC groups. First, we attempted to identify if
there were differences in dFC related to specific whole-
brain dynamic connectivity states. Second, we compared
the dFCvar between individuals with ASD and HCs. We
hypothesized that the intraindividual dFCvar in ASD
would help to explain the differences in ASD connectivity
patterns found in previous research.
MATERIALS AND METHODS
Participant Selection and Exclusion Criteria
Resting-state fMRI data were downloaded from the ABIDE
database (http://fcon_1000.projects.nitrc.org/indi/abide/)
[Di Martino et al., 2014]. Inclusion criteria were the same as
Di Martino et al.’s previous work [Di Martino et al., 2014]: (1)
only male subjects were included; (2) only sites where at least
75% subjects have a full scale IQ (FIQ) score were included;
(3) only subjects with FIQ within 2 SD 6 mean across all
ABIDE samples (108 6 15) were included; (4) only subjects
with mean framewise displacement (FD) [Power et al., 2012]
below 2 SD 1 mean (mean 5 0.23 1 SD 5 0.27 mm) were
included; (5) each subject had an anatomical image; (6) sites
with less than 10 subjects were excluded. In accordance with
these criteria, data from 763 subjects (ASD 5 360; HC 5 403)
from 17 sites remained.
To further eliminate possible confounds due to differ-
ences across subjects, we then applied a second set of
r
 r Chen et al. r

TABLE I. Participant demographics

ASD (n 5 209) HC (n 5 298) Group comparisons (P value)

Age 16.5 6 6.2 16.8 6 6.2 0.5642a

Site 3 group interaction - - 0.9642b
Full scale IQ 111 6 13 110 6 11 0.7191a
Site 3 group interaction - - 0.8502b
Mean FD (mm) 0.14 6 0.1 0.14 6 0.1 0.7219a
Site 3 group interaction - - 0.8341b
High head motion timepointsc 14.99 6 20.78 14.77 6 21.26 0.9091a
ADI-R
Social score 20 6 5 - -
Communication score 16 6 4 - -
RRB score 663 - -
ADOS
Total score 12 6 4 - -
Social score 863 - -
Communication score 461 - -
RRB score 261 - -

ADI-R, Autism Diagnostic Interview-Revised; ADOS, Autism Diagnostic Observation Schedule; RRB, Restricted and Repetitive Behaviors.
a
two sample t-test.
b
ANOVA.
c
Defined as the number of the timepoints whose FD > 0.5 mm and the preceding timepoint and the following two timepoints.

exclusion criteria: (1) 104 subjects with high levels of head
motion were excluded (maximum motion >2 mm or 28 rota-
tion, or more than 50% of frames with high FD; time points
whose FD was larger than 0.5 mm, along with the preceding
timepoint and following two timepoints, were defined as
high head motion time points); (2) 15 subjects were excluded
due to incomplete cortical coverage in the scan; (3) 9 subjects
with an age above 3 SD 1 mean across the samples were
excluded; (4) to create a well-matched dataset between the
HC group and the ASD group, we applied a data driven
algorithm that maximized the P values of the group differ-
ence of age, FIQ and FD (using two sample t tests). As the
following analysis would compare the relationship of age
and dFCvar between HC and ASD, to exclude the influence
of the site interaction effect, we also maximized the P values
of the interaction effects between sites and diagnostic
groups of age, FIQ and FD (using ANOVA), 76 subjects
were excluded. It is important to note that the maximization
of a p value is not to create a “best-matched” dataset, but
rather as a heuristic to create datasets that are not signifi-
cantly different from each other. (5) The UCLA (the second
dataset), KKI, and OLIN datasets were excluded for having
<5 ASD or HC subjects. Finally a well-matched dataset of
507 subjects from 14 sites was constructed (Table I; subject
IDs are listed in Supporting Information).
fMRI Data Preprocessing
All resting-state fMRI data were preprocessed using the
Statistical Parameter Mapping 8 toolbox (SPM8 v6313,
http://www.fil.ion.ucl.ac.uk/spm) and the Data Processing
Assistant for Resting-State fMRI toolbox (DPARSF advanced
edition v4.1_160415, http://rfmri.org/DPARSF) [Chao-Gan
and Yu-Feng, 2010]. The first 10 images were discarded to
ensure steady-state longitudinal magnetization. Temporal
and head motion correction was applied on the remaining
images, which were then warped into standard Montreal
Neurological Institute (MNI) space at the resolution of
3 3 3 3 3 mm3. All normalized images were smoothed with
a Gaussian kernel (full width at half maximum 5 8 mm) and
then detrended. Signals from white matter, cerebrospinal
fluid, and 24 rigid body motion parameters were regressed
out of the data [Friston et al., 1996]. The white matter and
cerebrospinal fluid masks were obtained from the tissue
probability maps in the SPM toolbox. We did not apply
global signal regression [Fox et al., 2009; Saad et al., 2012].
Finally a bandpass filter (0.01–0.1 Hz) was applied on the
regressed time series [Liu, et al., 2013a]. Scrubbing [Power
et al., 2012] was not applied in our study as it interrupts the
continuity of the time series. FD [Power et al., 2012] was cal-
culated to avoid group differences in head motion and was
defined as
FDi 5jDdix j1jDdiy j1jDdiz j1rjDai j1rjDbi j1rjDgi j
where the Ddix represents the displacement of x axis at time-
point i and is similar with other parameters of Ddiy , Ddiz ,
Dai , Dbi , and Dgi . The r is the radius of 50 mm, which is the
approximate mean distance from the center of the head to
the cortex. No significant difference in mean FD between
ASD and HC was found (P 5 0.7219, two-sample t test).
Static FC Analysis
To parcellate the brain into regions-of-interest (ROIs),
the automated anatomical labeling (AAL) template

r 5742 r
 r State-Related Functional Connectivity in Autism
[Tzourio-Mazoyer et al., 2002] was used. We discarded the
ROIs located in the cerebellum; 90 ROIs remained. The
averaged preprocessed fMRI signals in these 90 ROIs were
extracted. To construct the sFC network we calculated the
person correlation between each pair of these 90 averaged
signals. The FC values were then transformed to Z values
according to a Fisher-Z transformation:
 
1 pffiffiffiffiffiffiffiffiffiffi 11r
Z5 n23ln
2 12r
where n represents the number of timepoints and r repre-
sents the correlation coefficients between regions.
A two-sample t test was used to detect the connections
whose sFC strength is different between ASD and HC with
age, mean FD, FIQ, and site (using dummy coding scheme)
as nuisance covariates. Next, a network-based statistic
(NBS) method was applied on the resulting P values (con-
nection P < 0.05, cluster P < 0.05, 1000 loops). The NBS
method identifies clusters of atypical connections and then
determines whether the cluster size is significantly larger
than chance using permutation testing [Zalesky et al., 2010].
Dynamic FC Analysis
We utilized a Flexible Least Squares (FLS) algorithm
which is embedded in the DynamicBC toolbox (Dynam-
icBCv1.1 http://restfmri.net/forum/DynamicBC) to calcu-
late dFC [Liao et al., 2014a]. The FLS is based on a time-
varying parameter regression model:
yðtÞ5xðtÞbðtÞ1lðtÞ
where yðtÞ and xðtÞ are the time series of two ROIs, lðtÞ is
the approximation error. The bðtÞ is the coefficient to esti-
mate the covariance of the two ROIs and reflects the con-
nectivity between x and y at time t. To obtain the bðtÞ at
each time point, a cost function was defined as
Cðb; l; TÞ5l  r2D ðb; T Þ1r2M ðb; TÞ
where r2M ðb; T Þ is the sum of squared residual measure-
ment errors;
X
T
r2M ðb; T Þ5 ðyðtÞ2xðtÞbðtÞÞ2
t51
that satisfy yðtÞ2xðtÞbðtÞ  0. r2D ðb; T Þ, and is the sum of
the squared residual dynamic error:
X
T21
r2D ðb; T Þ5 ðbðt11Þ2bðtÞÞT ðbðt11Þ2bðtÞÞ
t51
that regulates the evolving speed of the vector of coeffi-
cients bðtÞ (bðt11Þ2bðtÞ  0). l is the weighting parameter
to control these two parts of the cost function. Here the
default value of 100 was used in this study [Liao et al.,
r
2014a]. By adding the regulation r2D ðb; TÞ to the cost func-
tion, bðtÞ was obtained not only by the signals at timepoint
t, but also using information from other timepoints. Finally
the dFC strength at each time points were defined as bðtÞ.
State-Specific dFC Strength Differences Between
ASD and HC
We next examined differences of dFC strength for each
functional connection in each whole-brain functional con-
nectivity state between ASD and HC as follows: (1) we
combined all dynamic FC networks of all subjects, result-
ing in a dFC matrix of 4005 3 95810 (4005 refers to the
number of possible connections between the ROIs of the
dFC network (90 3 (90 2 1)/2); 95810 refers to the num-
ber of timepoints across all subjects meaning there were
95810 dFC matrices across all subjects); (2) a k-means clus-
tering method was applied to divide the 95810 dFC matri-
ces into 6 clusters (i.e., 6 whole-brain connectivity states),
this procedure was repeated 10 times to avoid local min-
ima. Previous studies have reported the optimal number
of dFC clusters ranging from 5 to 7 [Allen et al., 2014;
Damaraju et al., 2014; Nomi et al., 2016, 2017; Rashid et al.,
2014], here we chose 6 and a reproducibility analysis was
done using 5 and 7 (Supporting Information, Figs. S1 and
S2). As our study is a multisite study, to avoid the effect
of different dFC baselines of each sites, we used the corre-
lation coefficient k-means clustering algorithm [Damaraju
et al., 2014].
The k-means was computed on the whole dataset to
ensure that the k-means produced clusters representative of
(1)
ASD and HC individuals. We repeated the k-means cluster-
ing on only the HC dataset and found almost no difference
of the 6 cluster centers between the whole dataset and the
HC dataset. The final six state-specified dFC whole-brain
networks for each subject were obtained by back-
reconstructing individual subject mean dFCs for each state.
As the dFC strength is normally distributed and for ease
(2) of calculation, we then utilized two-sample t tests to com-
pare dFC strength between the two groups within all 6
states using age, mean FD, FIQ, and site as nuisance cova-
riates. The NBS method was applied on the resulting P
values (connection P < 0.05, cluster P < 0.05, 500 loops).
(3)
Comparison of dFCvar Between ASD and HC
Participants
For each connection, we defined dFCvar as the standard
deviation of the dFC connectivity correlations across time
(Fig. 1). For each subject, the dFCvar matrix that identifies
(4)
the variability of all dFC estimates across the entire
resting-state scan was calculated. As the dFCvar values do
not represent a normal distribution, a permutation test
was applied to test whether the dFCvar between individu-
als with ASD and HC was significantly different for each
possible dFC connection. For each dFC connection, the
r 5743 r
 r Chen et al. r

Figure 1.
Flowchart of the dynamic FC analysis. [Color figure can be viewed at wileyonlinelibrary.com]

FIQ, age, head motion measured by mean FD [Power
et al., 2012], and site (using a dummy coding scheme)
were regressed out of the dFCvar before the permutation
test. An NBS method was utilized to find the connections
with atypical dFCvar in ASD (connection P < 0.05, cluster
P < 0.05, 1000 loops). Because the NBS method only identi-
fied increased dFCvar in ASD compared to HC (as
opposed to decreased dFCvar), subsequent analysis
focused on these “hypervariant” connections.
Spatial Distribution of Hypervariant Connections
As the NBS method aims to identify atypical clusters that
are significantly larger than random cases and results in clus-
ters connecting many brain regions, we identified regions
with the most atypical connections. To do so, we first
counted the amount of atypical connections for each region
and then identified the regions with an amount of signifi-
cantly hypervariant connections above the mean 1 3 SD
(regions with the amount of connections greater than the
mean 1 1 SD and 2 SD are also listed in Fig. 3B).
We sorted the length of all connections defined by the
AAL template and defined the 33% shortest connections as
short-range connections, the 33% longest connections as
long-range connections, and the remaining 33% connec-
tions as middle-range connections. The length of
connection between two regions was defined by using the
Euclidean distance between the centers of the two regions.
We counted the amount of long-range, middle-range and
short-range connections in the atypical clusters to reflect
the length distribution of the connections with atypical
dFCvar. To examine how hypervariant connections were
significantly associated with the type of connections (i.e.,
short, mid-range, long-range), we treated the distribution
of the rate of long-range hypervariant connections as a
Binomial distribution with the successful rate of 1/3 and
the total trials as the total number of the hypervariant FCs,
then we transferred the rate into P values based on the
cumulative distribution function. For these three types of
hypervariant connections, we then identified the regions
with the most atypical connections.
Partial Correlation Analysis Between Group,
dFCvar, and sFC Strength
Previous studies have shown that the dFCvar and sFC
strength are significantly anticorrelated [Thompson and
Fransson, 2015, 2016], demonstrating that the altered sFC
strength in the ASD group may drive atypical dFCvar in
this study. To determine the relationship between dFC
hypervariance and sFC under-connectivity in ASD, a par-
tial correlation analysis was done to explore the difference

r 5744 r
 r State-Related Functional Connectivity in Autism
of dFCvar between HC and ASD while controlling the sFC
strength. We extracted the mean dFCvar value of the con-
nections which showed hypervariant and the mean sFC
strength value of the connections which showed undercon-
nected. Then a partial correlation analysis was done
between the mean dFCvar and group with mean sFC
strength as covariant. (As the group variable is discrete, a
Spearman correlation coefficient was reported.)
Relationship Between sFC, dFC, dFCvar, and
Severity of ASD
A network contingency (NC) analysis was applied to
explore the connections whose dFCvar values were related to
the severity of ASD (measured by the ADOS total score)
which tested whether the population of severity-related con-
nections is larger than one would expect by chance in the
hypervariant cluster [Sripada et al., 2014]. The steps are as fol-
lows: (1) Correlation step: for each connection in the hyper-
variant cluster, Pearson correlation coefficients were
calculated between the dFCvar and ADOS total score. The
number of significantly positively correlated connections
(P < 0.05) was summed. During the correlation step, the age,
mean FD, FIQ, sites, and group (ASD or HC) were first
regressed out; (2) Permutation step: the subjects were shuf-
fled to create a random dataset, then the Pearson correlation
coefficients between the dFCvar of each connection in the
hypervariant cluster and ADOS total score were calculated in
the shuffled dataset. The number of significantly positively
correlated connections (P < 0.05) was summed. This step was
repeated 1000 times and 1000 significantly positively corre-
lated connections at random were obtained. The P value of
the hypervariant cluster was calculated by the times in which
the random correlated connections were more than the actual
correlated connections divided by 1000. Then we determined
the regions related to the connections associated with ASD
symptom severity. Subjects with no ADOS total score were
excluded during the analysis, leaving 148 subjects.
We also tested the relationship between the dFCvar of
the hypervariant cluster and subscores of the ADOS and
ADI-R in ASD. The NC analysis was applied on the 3
ADOS subscores (ADOS communication scores of 131 sub-
jects, ADOS social scores of 131 subjects, and ADOS RRB
scores of 118 subjects). ADI-R scores were included in the
correlation analysis (ADI-R social scores of 150 subjects,
ADI-R communication scores of 152 subjects, and ADI-R
RRB score of 152 subjects).
These correlation analyses were repeated using sFC and
dFC to determine if they demonstrated a stronger relation-
ship with ASD symptom severity compared with dFCvar.
Age-Related Analysis of dFCvar and dFC
Strength
To determine the relationship between the widespread
hypervariant FCs and age, we calculated the correlation
r 5745
r
coefficients between the mean dFCvar and dFC strength of
the hypervariant cluster and age in both the HC and ASD
group with FIQ, mean FD, site (using a dummy coding
scheme), and group (ASD or HC) as nuisance covariates.
We also examined the relationship between whole-brain
mean dFCvar and dFC strength, and age. Whole-brain
mean dFCvar and whole-brain dFC was calculated as the
mean values of all dFCvar and dFC strengths across all
time points.
Reproducibility Analyses
To determine the reproducibility of the current findings,
we conducted a leave-one-site-out cross validation
(LOSOCV). The procedure of LOSOCV is similar to LOOCV
[Li et al., 2015; Liu, et al., 2013b], instead of leaving one
subject out one time, we excluded one site at a time and
repeated the analysis 14 times. Results from this analysis
are presented in Supporting Information, Table S2.
As head motion is an important factor affecting the
dFCvar, we repeated the analysis by adding a despiking
(AFNI_16.2.16, https://afni.nimh.nih.gov/) step between
the detrend and regress steps and then compared the
dFCvar values between ASD and HC.
A recent study has demonstrated that the overall
strength of static FC may lead to spurious differences in
other functional connectivity metrics [van den Heuvel
et al., 2017]. To examine whether the difference of dFCvar
between ASD and HC is caused by changes in the overall
network organization pattern, we compared several graph
theoretical metrics between ASD and HC (detailed proce-
dures are listed in Supporting Information). These metrics
included global efficiency (defined as the reciprocal of
shortest length), local efficiency (defined as the reciprocal
of cluster coefficiency), and modularity. These were calcu-
lated for the static FC matrices and the variance of these
three metrics based on the dynamic FC matrices. We
repeated the dFCvar analysis with these graph metrics as
nuisance covariates [Ji et al., 2017]. Results from this analy-
sis are presented in Supporting Information.
The mu parameter of FLS method controls the continu-
ity of the dFC strength across timepoints which might
influence the dFCvar values [Liao et al., 2014a]. To explore
the influence of the mu parameter to our dFCvar results,
we repeated the dFCvar analysis using the mu parameter
of 50 and 200.
RESULTS
Static Functional Connectivity
A cluster of 1127 connections with reduced sFC strength
in ASD was detected by the NBS method (connection
P < 0.05, cluster P < 0.05). The connections were mainly
between prefrontal and temporal regions (Fig. 2). No
r
 r Chen et al. r

Figure 2.
Decreased sFC strength in ASD. (A) The mean sFC networks of HC and ASD groups; (B)
decreased sFCs in ASD, red points represent connections with lower sFC strength in ASD com-
pared to HC. [Color figure can be viewed at wileyonlinelibrary.com]

cluster of connections with increased sFC strength in ASD
was found.
Dynamic FC Analysis
We divided all dFC networks into 6 clusters that repre-
sent 6 dynamic brain network states. As shown in Figure
5, the connectivity patterns of the 6 state-related dFC net-
works were different: in State 1, subcortical regions are
highly connected to other regions; in State 2, the most con-
nected regions are located in calcarine, cuneus, and supe-
rior occipital gyrus; in State 3, the regions with most
connections are located in orbitalfrontal cortex and inferior
parietal lobule; in State 4, the most connected regions are
medial superior orbitalfrontal cortex and some posterior
regions such as posterior cingulate cortex and angular
gyrus; in State 5, the most connected region is rolandic
operculum; in State 6, the hippocampus, parahippocam-
pus, and temporal pole are highly connected.
State-Specific dFC Strength Differences Between
ASD and HC
We compared the dFC networks within 6 states between
ASD and HC. Under-connectivity was found in dFC states
1 (428 connections), 2 (535 connections), 4 (770 connec-
tions), and 6 (474 connections) (connectivity level P < 0.05,
cluster level P < 0.05). Additionally, the under-connectivity
pattern in States 1, 2, and 4 were different from the under-
connectivity pattern in the sFC network: compared to the
sFC results, State 1 decreased dFCs were more connected
to the motor regions located in the posterior part of the
frontal lobe, State 2 decreased dFCs were less connected to
temporal lobes, and State 4 decreased dFCs were more
connected to the occipital lobes. These results showed that
under-connectivity patterns are manifested differently in a
dFC compared to a sFC analysis. These analyses were also
repeated using 5 and 7 states (Supporting Information,
Figs. S1 and S2).
To ensure that the clustering was not biased by group,
we repeated the clustering on only the HC dataset, and
compared the cluster centers between the whole dataset
and the HC dataset. We found that the mean correlation
coefficient of the corresponding state between the whole
dataset centers and HC dataset centers was 0.9959, indicat-
ing that the clustering was not biased by one group.
Comparison of dFCvar Between ASD and HC
Participants
Using a permutation test and the NBS method (connec-
tivity level P < 0.05, cluster level P < 0.05), a cluster con-
taining 278 hypervariant connections with higher dFCvar
in ASD was revealed, which connected almost all brain
regions (79 out of 90 ROIs). No hypovariant connections
were found with the NBS method (Fig. 3A). Additionally,
the connections within the hypervariant cluster were asso-
ciated with prefrontal and temporal regions. The most
hypervariant connections were related to the right medial
superior frontal gyrus (Fig. 3B).
By comparing the dFC strength between ASD and HC
of the hypervariant connections in ASD, we found that the
hypervariant connections showed under-connectivity in
state 2 (P 5 2.77e-4, Cohen’s d 5 0.39), 3 (P 5 0.0304,
Cohen’s d 5 0.2), 4 (P 5 6.14e-5, Cohen’s d 5 0.42), and 6
(P 5 0.0054, Cohen’s d 5 0.35) of the dFC networks. These
results demonstrate that hypervariant connections in ASD
are associated with dFC under-connectivity.

r 5746 r
 r State-Related Functional Connectivity in Autism
Figure 3.
Hypervariant dFCs in ASD. (A) Hypervariant connections in
ASD. Red blocks represent hypervariant connections in ASD
and blue blocks represent connections with no significant differ-
ence of dFCvar between HC and ASD. The numbers below rep-
resent the numbers of atypical connections with corresponding
regions divided by the number of total atypical connections. (B)
Hypervariant regions. The length of the bars represents the
number of hypervariant connections of the corresponding
regions. Only regions with hypervariant connections greater
than the mean 1 SD are shown. Abbreviations of the regions
Spatial Distribution of Hypervariant Connections
As shown in Figure 4, there were 153 long-range hyper-
variant connections out of the total 278 hypervariant con-
nections (P 5 1.44e-13, based on a Binomial distribution,
153/278), indicating that there were more long-range con-
nections in the hypervariant cluster. By analyzing the three
types of hypervariant connections, we found that the long-
range hypervariant connections were associated with fron-
tal regions, especially the medial superior frontal gyrus.
Middle-range hypervariant connections were associated
with temporal regions such as the temporal pole, and
short-range hypervariant connections were associated with
regions such as the medial superior frontal gyrus and the
left anterior cingulate gyrus.
Relationship Between sFC, dFC, dFCvar, and
Severity of ASD
Using the NC method, the dFCvar values of 31 hyper
variant connections were found significantly positively
correlated with ADOS total score. The population was
significant larger than random cases (permutation
P 5 0.002). These connections are mostly connected with
r 5747
r
are listed in Supporting Information. (C) Severity-related hyper-
variant connections. The hypervariant connections significantly
positively correlated with ADOS total score in ASD. The size of
nodes represents the number of connections. Red regions rep-
resent the areas with the most connections. PFR, prefrontal
regions; FR, other regions of frontal lobe; PR, parietal regions;
TR, temporal regions; OR, occipital regions; SUB, subcortical
regions [Wang et al., 2007]. [Color figure can be viewed at
wileyonlinelibrary.com]
the right medial superior frontal gyrus and the left mid-
dle temporal pole (both regions have 7 connections). The
distribution of the hypervariant connections whose
dFCvar values were related to the severity of ASD is
shown in Figure 3C.
By testing the sub scores of the ADOS, we found 27 hyper-
variant connections whose dFCvar values were related to
the ADOS communication score (permutation P 5 0.016), 16
connections related to the ADOS social score (permutation
P 5 0.067), and 17 connections related to the ADOS RRB
score (permutation P 5 0.102). We did not find significant
connections correlated with the 3 ADI-R subscores (ADI-R
social: 7 connections, permutation P 5 0.957; ADI-R commu-
nication: 3 connections, permutation P 5 0.999; ADI-R RRB:
9 connections, permutation P 5 0.897). Although the num-
bers of connections related to the ADOS communication
score was significantly larger than random cases, the P value
did not survive Bonferroni correction.
Significant sFCs whose strength were correlated with
ADOS total and communication scores were also found;
however, the P values did not survive Bonferroni correc-
tion. No significant dFCs whose state-related dFC strength
was correlated with ADOS scores were found. Detail
results are reported in Supporting Information, Table S1.
r
 r Chen et al. r

Figure 4.
Length distribution of hypervariant connections in ASD. (A) atypical connections. (B) Counts of long-range, middle-range,
Overview of the long-range, middle-range, and short-range and short-range hypervariant connections in ASD. Abbreviations
hypervariant connections in ASD: blue lines represent long- of the regions are listed in Supporting Information. [Color figure
range atypical connections, green lines represent middle-range can be viewed at wileyonlinelibrary.com]
atypical connections, and yellow lines represent short-range

Correlation of dFC and dFCvar With Age
As shown in Figure 6A, both the dFCvar of whole brain
and hypervariant FCs decrease with age in the HC group
(whole brain: r2 5 0.036, P 5 0.0001; hypervariant cluster:
r2 5 0.0484, P 5 0.00014) but not in the ASD group
(r2 5 0.0036, P 5 0.4; hypervariant cluster: r2 5 0.0009,
P 5 0.71), indicating atypical development of whole brain
FC variance in individuals with ASD. The dFC strength of
whole brain and hypervariant FCs increased with age in
the HC group (whole brain: r2 5 0.04, P 5 0.00057;
hypervariant cluster: r2 5 0.036, P 5 0.001) but not in the
ASD group (whole brain: r2 5 0.00016, P 5 0.57; hypervar-
iant cluster: r2 5 0.00016, P 5 0.57) (Fig. 6B).
Partial Correlation Analyses
After controlling the sFC strength, we found significant
correlation between dFCvar and group (rho 5 0.2551,
P 5 5.9e-9), which showed that the difference of dFCvar
between HC and ASD was not caused by the difference of
sFC strength.

r 5748 r
 r State-Related Functional Connectivity in Autism r

Figure 5.
State-specific difference of the dFC network between ASD and
HC. (A) Top row represents the dFC networks at 6 states. (B)
Second row represents the hubs of the dFC networks at 6
states. For display, we defined the significant connections as the
connections whose absolute strength is greater than the
mean 1 SD, and hub regions as the nodes whose degree is
above mean 1 2 SD. (C) The third row represents the distribu-
tion of the decreased connections in ASD for each state while
red represents the distribution of decreased dFC in ASD for
each state. For display, we also compared the sFC between ASD
and HC (blue). No decreased dFC in ASD at States 3 and 5
were found. The values of the radar map are the ratio of the
decreased connections related to corresponding regions divided
by the total decreased connections. (D) Fourth row represents
the dFC strength of the hypervariant connections. Red bars rep-
resent the dFC strength for the ASD group and blue bars repre-
sent the dFC strength for the HC group. * represents P < 0.05
(Bonferroni corrected). Bi, bilateral; SOG, superior occipital
gyrus; ORBmid, middle frontal gyrus, orbital part; IPL, inferior
parietal; ORBsupmed, superior frontal gyrus, medial orbital;
PCC, posterior cingulate gyrus; ANG, angular; ROL, rolandic
operculum; HIP, hippocampus; PHG, parahippocampus; TPOmid,
temporal pole, middle temporal gyrus. [Color figure can be
viewed at wileyonlinelibrary.com]

Reproducibility Analyses and modularity calculated on the sFC matrices between

In the LOSOCV analysis, the result of higher dFCvar in
ASD repeated 10 out of 14 times, other results were repeated
14/14 (Supporting Information, Table S2). The correlation
results were repeated 8/8 (Supporting Information, Table S3).
After adding the despiking step, we still found a hypervar-
iant cluster of 322 connections in ASD group, indicating that
the hypervariant connections in ASD group are not likely to
be caused by head motion (Supporting Information, Fig. S3).
We did not find any significant difference of the AUC
(area under curve) for global efficiency, local efficiency,
ASD and HC. The AUC for the variance of global effi-
ciency, the variance of local efficiency, and the variance of
modularity calculated on dFC matrices also showed no
significant differences between ASD and HC. The ASD
group still showed dFC hypervariance even when these
graph theoretical metrics were used as nuisance covariates.
Detailed results can be found in Supporting Information
(Fig. S4 and Table S4).
The dFCvar analysis result did not show any major dif-
ference when using the mu parameter of FLS method as
50 and 200 (Supporting Information, Fig. S5).

r 5749 r
 r Chen et al. r

Figure 6.
Different relationships between FC variance and age and in ASD represent the linear fitting curve of mean dFC strength in ASD
and HC. (A) Relationship between FC variance and age in ASD while the blue lines represent the HC. The upper and lower
and HC. The FC variance of the whole-brain and hypervariant dashed lines represent the linear fitting curve of mean dFC
FCs showed significant negative correlation with age in the HC strength plus and minus the dFCvar in ASD and HC. [Color fig-
group, but not in the ASD group. (B) Diagrammatic drawing of ure can be viewed at wileyonlinelibrary.com]
the dFC development curve in ASD and HC. Red solid lines

Consideration of the SNR of the Hypervariant 4005 connections in HC group, which indicates a minimal
Connections in ASD effect of the SNR on hypervariant connections.

As the dFCvar was defined as the standard deviation of

dFC strength across time, it may be affected by the SNR of
connections, especially the weak connections with low
SNR. We calculated the mean sFC strength of the hyper-
variant connections and found the mean sFC strength of
the hypervariant connections was above 71% of the total
DISCUSSION
In this study based on a large sample, we utilized dFC
analyses to demonstrate differences in dynamic functional
connections between ASD and HC. These differences were

r 5750 r
 r State-Related Functional Connectivity in Autism
mainly represented by dynamic under-connectivity and
increased variability in specific whole-brain connectivity
states in ASD. Importantly, dFC under-connectivity patterns
within specific whole-brain dynamic connectivity states
were different from the under-connectivity patterns found
in the sFC analysis. This demonstrates that atypical func-
tional connectivity can be transient in nature and related to
specific temporal windows of a resting-state fMRI scan that
may not be captured by traditional sFC approaches.
This study also found increased variability of dFC connec-
tions (i.e., hypervariant connections: dFCvar) in ASD. The
results also demonstrated that hypervariant ASD dFCvar
was significantly related to ADOS scores but not ADI-R
scores, while there was no relationship between ASD symp-
tom severity with sFC or dFC. Further analyses demon-
strated that long-range hypervariant ASD connections were
associated with frontal regions, especially the medial supe-
rior frontal gyrus, middle-range hypervariant ASD connec-
tions were associated with temporal regions such as the
temporal pole, and short-range hypervariant ASD connec-
tions were associated with regions such as the medial supe-
rior frontal gyrus and the left anterior cingulate gyrus.
Finally, an age-related analysis showed atypical develop-
ment of dynamic brain properties in ASD compared to HC.
State-Related Under-Connectivity in ASD
Earlier dominant views suggested that the ASD brain is
best characterized by static long-range under-connectivity
[Anderson et al., 2011; Just et al., 2004, 2007]. However, the
static under-connectivity pattern in ASD has not been consis-
tently observed across studies [Jung et al., 2014; von dem
Hagen et al., 2013; Yerys et al., 2015]. More recent studies
have reported mixed- or over-connectivity in static FC ASD
research [Picci et al., 2016; Supekar et al., 2013]. This study uti-
lized a state-related dynamic FC analysis to demonstrate
dynamic under-connectivity and increased individual-level
dFCvar in a large sample of individuals with ASD. However,
dynamic under-connectivity was not present in all substates,
and the under-connectivity patterns in each substate were dif-
ferent from each other and the sFC under-connectivity
results. The results demonstrate that under-connectivity can
be found in different temporal windows throughout the
resting-state scan that may be hidden by traditional sFC
approaches. This result suggests that under-connectivity in
ASD may also be state-related, and specific states within
resting-state fMRI datasets may affect the FC results in ASD.
Additionally, the results demonstrate that there were no tran-
sient periods of hyperconnectivity in individuals with ASD
that could have been masked by traditional sFC approaches.
Greater dFCvar in ASD and Relationship With
Long-Distance Connections
Previous studies showed that dynamic connections are
more prevalent between separated brain regions [Chang
r 5751
r
and Glover, 2010]. Long-range connections often connect
hub regions and construct the “backbone” of the brain net-
work to minimize wiring and energy costs [Bassett and
Bullmore, 2006; van den Heuvel et al., 2012]. Our results
show that hypervariant connectivity in ASD tends to occur
in long-range connections, which is consistent with multi-
ple prior imaging studies reporting long-range sFC deficits
in ASD [Just et al., 2004, 2012; Kikuchi et al., 2015]. The
long-range hypervariant connectivity in ASD may cause
unstable transmission between major brain hubs in indi-
viduals with ASD [Rudie et al., 2012].
By calculating the ratio of hypervariant connections
between different brain areas, we found that the hypervar-
iant connections in ASD were mainly associated with the
frontal and temporal lobes. This trend is consistent with the
early overgrowth found for different brain regions in ASD
[Chomiak and Hu, 2013; Courchesne, 2004; Schumann et al.,
2010; Stoner et al., 2014]. The early overgrowth theory posits
that early postnatal brain overgrowth and later delayed
growth are core neurobiological mechanisms underlying
ASD [Schumann et al., 2010]. For HC, high-level frontal and
temporal association cortices mature later than low-level
sensory processing areas such as the occipital lobe. How-
ever, there is evidence that frontal and temporal lobe matu-
ration occurs earlier in ASD compared to HC [Chomiak and
Hu, 2013], leading to early excitatory synaptic overgrowth
and reduced synaptic pruning [Nelson and Valakh, 2015;
Tang et al., 2014]. As inhibition is believed to contribute to
sharpening the selectivity of excitatory brain responses, the
excess of excitation and loss of inhibition could result in
enhanced “noise” of brain regions and imprecise brain activ-
ity [Nelson and Valakh, 2015]. This theory was also sup-
ported by a recent multisite study that reported higher brain
signal fluctuation of frontal and temporal regions in ASD as
measured by fractional amplitude of low-frequency fluctua-
tion (fALFF) analyses [Di Martino et al., 2014], indicating
greater brain signal oscillation of these two regions in ASD.
As frontal and temporal lobes are two critical regions relat-
ing to higher cognitive, social, and communication functions
[Chomiak and Hu, 2013], the hypervariance of the dFCs
related to these two regions may disturb the transmission of
cognitive and social information in the ASD brain, contribut-
ing to the core symptoms of ASD.
Additionally, we found that the medial superior frontal
gyrus (Fig. 3B) was associated with the most hypervariant
connections. The medial superior frontal gyrus is a key
region of the anterior default mode network that is related
to self-referential mental activity [Gusnard et al., 2001].
Interestingly, we found that this region was associated
with long- and short-range hypervariant connections in
ASD (Fig. 4B). This result is similar to a previous sFC
study that reported reduced long- and short-range func-
tional connectivity strength of medial frontal regions in
ASD [Long et al., 2016].
Middle-range hypervariant connections were associated
the most with the temporal pole (Fig. 4B). Anatomical and
r
 r Chen et al.
functional studies have suggested that the temporal pole is
important in autobiographical memory, face recognition,
and linguistic integration [Dupont, 2002]. The medial fron-
tal and temporal pole regions are both parts of the default
mode network, which have previously been identified as
key atypical circuits in ASD [Jung et al., 2014; Washington
et al., 2014].
Relationship Between dFCvar and ASD
Symptom Severity
This study found that dFCvar, but not sFC or dFC
strength, was associated with ASD symptom severity. This
builds on previous work demonstrating that dynamic FC
approaches can better explain behavior in the typical pop-
ulation [Jia et al., 2014]. This study also found that only
symptom severity scores from the ADOS but not the ADI-
R was associated with dFCvar. This may be due to the dif-
ferent measurements included in the ADOS and ADI-R.
The ADOS focuses on the current states of individuals
with ASD [Lord et al., 1989] while the ADI-R scores are a
combination of current behavior and historical behavior at
a certain point in time [Le Couteur et al., 1989]. Our study
showed a significant correlation between age and dFCvar,
suggesting that dFCvar may not reflect the historical part
of ADI-R scores, resulting in the insignificant correlation
between dFCvar and ADI-R scores.
Hypervariant connections related to symptom-severity
were also associated with the right medial superior frontal
gyrus and left middle temporal pole. These results further
support the importance of these DMN regions in atypical
ASD neurobiology.
Atypical Developmental Course of the FC
Variance/Strength in ASD
The age-related analysis showed that in the HC group,
dFCvar decreases with age while dFC strength increases
with age. In the ASD group, these two developmental
trends were not significant. We speculate that this devel-
opmental course is related to synaptic development. Post-
natal synaptic development consists of formation and
pruning. In early ages, the formation exceeds pruning,
producing excitatory synapses for the assembly of neural
circuits. Subsequently, the pruning outpaces formation
during the developmental course from childhood through
adolescence [Tang et al., 2014], which suggests that the
density of dendritic spines reaches a peak in early child-
hood and then decreases across late childhood, adoles-
cence, and adult periods [Peter et al., 2011]. The altered
developmental course of the dFCvar variance and strength
in ASD might be related to atypical synaptic development
in ASD. Previous studies have reported that dendritic
spine density was slightly higher in childhood ASD com-
pared with HC, but greatly higher in adolescent ASD com-
pared with HC. In HC, dendritic spines decrease by about
r 5752
r
45% from childhood to adolescence, but only by 16% in
individuals with ASD [Tang et al., 2014], implying a net
pruning deficit in ASD. At early ages, the overgrowth of
synapses provides more information transmission channels
and results in greater flexibility and strength of FC in
ASD. Subsequently, altered synaptic pruning may result in
delayed and weak changes of these dFC properties.
In this study, we observed a large overlap between the
HC and ASD state-related dFC at early ages. In the adults,
we can see less variance and stronger strength of FCs in
HC, and less overlap of the state-related FC between HC
and ASD. This trend fits the static under-connectivity find-
ings in adults with ASD [Jung et al., 2014; Rausch et al.,
2016; von dem Hagen et al., 2013].
Although there are significant correlations between dFC
and age (Fig. 6), the effect size of the relationships are
small. Several factors might contribute to the small effect
size: (1) our study used a multisite dataset and the vari-
ance of different rating criteria of the ADOS might reduce
the correlation coefficients; (2) ASD is a spectrum disorder
which is thought to be caused by a multitude of factors.
Such heterogeneity is characteristic of ASD [Amaral, 2011].
Thus, intersubject variance of ASD might also affect the
group correlation analysis, especially in large multisite
datasets.
Previous dFC ASD Research
This study adds to a growing literature utilizing dFC
approaches to investigating atypical brain function in
ASD. Recently, Falahpour et al. [2016] utilized a sliding-
window targeted ROI dFC approach in children, adoles-
cents, and adults using publicly available data (ABIDE)
and in-house data to investigate functional connections of
the default mode network (DMN), salience network (SN),
along with the amygdala and thalamus. A mediation anal-
ysis demonstrated that sFC under-connectivity in ASD
was driven by increased dFCvar when compared to HC
individuals. Wee et al. [2016] utilized a whole-brain ROI
dynamic approach to investigate if a support vector
machine classification can reliably index ASD compared
HC dynamic connections. Instead of a sliding-window
approach, Wee et al. utilized k-means clustering on the
entire time-series from all ROIs to identify different clus-
ters (sets of ROI connections) of functional connectivity
patterns using a short-time sparse-regressions with Pear-
son correlation (ST-SR-PC) weighting. A support vector
machine (SVM) algorithm was then utilized to demon-
strate functional connectivity estimates from frontal, parie-
tal, occipital, temporal, and subcortical structures across
all clusters using the ST-SR-PC method could better differ-
entiate ASD from HC participants than functional connec-
tivity estimates using just ST, SR, or PC methods
individually.
This study compliments these previous studies by utiliz-
ing a different type of dFC methodology. Additionally,
r
 r State-Related Functional Connectivity in Autism
although this study demonstrates increased ASD dFCvar
just as in Falahpour, there are a number of important dif-
ferences between the two studies. First, this study extends
on Falahpour et al. by utilizing a whole-brain cortical-sub-
cortical parcellation (without cerebellum) instead of tar-
geted ROIs to demonstrate differences in regions other
than the DMN, SN, amygdala, and thalamus. Second,
Falahpour et al. focused solely on the mediation effect of
dFCvar on sFC in ASD while this study focused on (1)
how dFCvar is related to dFC; (2) how dFC, dFCvar, and
sFC is related to age and symptom severity in ASD; and
(3) how differences in dFCvar are related to the distance
of the functional connections in ASD. Thus, although this
study compliments Falahpour et al. by also demonstrating
hypervariant dynamics, this study also provides important
novel information such as (1) increased dFCvar is more
prevalent in long- compared to mid- or short-range con-
nections, (2) increased dFCvar goes down with age in HC,
but not in ASD, and (3) only dFCvar (not dFC or sFC) is
related to symptom severity, and only with ADOS mea-
sures but not with ADI-R measures.
CONCLUSIONS
In summary, this study demonstrates that differences in
ASD functional connections extend beyond traditional sFC
approaches to dFC approaches where differences in
dynamic functional connections and the variability of such
connections in specific whole-brain connectivity states
exist. This suggests that differences in ASD FC can also be
transient in addition to being stationary as identified in
traditional sFC ASD research. Furthermore, differences in
dFCvar demonstrate that differences in FC may be related
to the variability of functional connections. This study also
demonstrates that dFCvar better explains differences in
ASD symptom severity compared with sFC and dFC.
Finally, the age-related analysis showed an atypical devel-
opmental course of dFCvar and dFC strength in ASD.
These novel results demonstrate how a dFC approach may
offer a more nuanced identification of neurobiological defi-
cits in ASD compared to traditional sFC approaches.
CONFLICTS OF INTEREST
All authors declared no conflicts of interest.
REFERENCES
Alaerts K, Swinnen S, Wenderoth N (2016): Sex differences in
Autism: A resting-state fMRI investigation of functional brain
connectivity in males and females. Soc Cogn Affect Neurosci.
Allen EA, Damaraju E, Plis SM, Erhardt EB, Eichele T, Calhoun
VD (2014): Tracking whole-brain connectivity dynamics in the
resting state. Cereb Cortex 24:663–676.
r 5753
r
Amaral DG (2011): The promise and the pitfalls of autism
research: An introductory note for new autism researchers.
Brain Res 1380:3–9.
American Psychiatric Association. (2013): The Diagnostic and Sta-
tistical Manual of Mental Disorders: DSM 5. USA: Bookpoint.
p 991.
Anderson JS, Druzgal TJ, Froehlich A, DuBray MB, Lange N,
Alexander AL, Abildskov T, Nielsen JA, Cariello AN,
Cooperrider JR, Bigler ED, Lainhart JE (2011): Decreased inter-
hemispheric functional connectivity in autism. Cereb Cortex
21:1134–1146.
Bassett DS, Bullmore E (2006): Small-world brain networks. Neu-
roscientist 12:512–523.
Cerliani L, Mennes M, Thomas RM, Di Martino A, Thioux M,
Keysers C (2015): Increased functional connectivity between
subcortical and cortical resting-state networks in autism spec-
trum disorder. JAMA Psychiatry 72:767–777.
Chang C, Glover GH (2010): Time-frequency dynamics of resting-
state brain connectivity measured with fMRI. NeuroImage 50:
81–98.
Chao-Gan Y, Yu-Feng Z (2010): DPARSF: A MATLAB toolbox for
“Pipeline” data analysis of resting-state fMRI. Front Syst Neu-
rosci 4:13.
Chomiak T, Hu B (2013): Alterations of neocortical development
and maturation in autism: Insight from valproic acid exposure
and animal models of autism. Neurotoxicol Teratol 36:57–66.
Courchesne E (2004): Brain development in autism: Early over-
growth followed by premature arrest of growth. Mental Retard
Develop Disabil Res Rev 10:106–111.
Damaraju E, Allen EA, Belger A, Ford JM, McEwen S, Mathalon
DH, Mueller BA, Pearlson GD, Potkin SG, Preda A, Turner JA,
Vaidya JG, van Erp TG, Calhoun VD (2014): Dynamic func-
tional connectivity analysis reveals transient states of dyscon-
nectivity in schizophrenia. NeuroImage Clin 5:298–308.
Developmental Disabilities Monitoring Network Surveillance Year
2010 Principal Investigators and Centers for Disease Control
and Prevention (2014): Prevalence of autism spectrum disorder
among children aged 8 years-autism and developmental dis-
abilities monitoring network, 11 sites, United States, 2010. Mor-
bidity and mortality weekly report. Surveillance summaries
(Washington, DC: 2002), 63:1.
Di Martino A, Yan CG, Li Q, Denio E, Castellanos FX, Alaerts K,
Anderson JS, Assaf M, Bookheimer SY, Dapretto M, Deen B,
Delmonte S, Dinstein I, Ertl-Wagner B, Fair DA, Gallagher L,
Kennedy DP, Keown CL, Keysers C, Lainhart JE, Lord C, Luna
B, Menon V, Minshew NJ, Monk CS, Mueller S, Muller RA,
Nebel MB, Nigg JT, O’Hearn K, Pelphrey KA, Peltier SJ, Rudie
JD, Sunaert S, Thioux M, Tyszka JM, Uddin LQ, Verhoeven JS,
Wenderoth N, Wiggins JL, Mostofsky SH, Milham MP (2014):
The autism brain imaging data exchange: Towards a large-
scale evaluation of the intrinsic brain architecture in autism.
Mol Psychiatry 19:659–667.
Dupont S (2002): Investigating temporal pole function by func-
tional imaging. Epileptic Disord 4(Suppl 1):S17–S22.
Falahpour M, Thompson WK, Abbott AE, Jahedi A, Mulvey ME,
Datko M, Liu TT, Muller RA (2016): Underconnected, but not
broken? Dynamic functional connectivity MRI shows under-
connectivity in autism is linked to increased intra-individual
variability across time. Brain Connect 6:403–414.
Fox MD, Zhang D, Snyder AZ, Raichle ME (2009): The global sig-
nal and observed anticorrelated resting state brain networks.
J Neurophysiol 101:3270–3283.
r
 r Chen et al.
Friston KJ, Williams S, Howard R, Frackowiak RS, Turner R
(1996): Movement-related effects in fMRI time-series. Magn
Reson Med 35:346–355.
Gusnard DA, Akbudak E, Shulman GL, Raichle ME (2001): Medial
prefrontal cortex and self-referential mental activity: Relation
to a default mode of brain function. Proc Natl Acad Sci USA
98:4259–4264.
Hahamy A, Behrmann M, Malach R (2015): The idiosyncratic
brain: Distortion of spontaneous connectivity patterns in
autism spectrum disorder. Nat Neurosci 18:302–309.
Hipp JF, Siegel M (2015): BOLD fMRI correlation reflects
frequency-specific neuronal correlation. Curr Biol 25:1368–1374.
Hutchison RM, Womelsdorf T, Allen EA, Bandettini PA, Calhoun
VD, Corbetta M, Della Penna S, Duyn JH, Glover GH,
Gonzalez-Castillo J, Handwerker DA, Keilholz S, Kiviniemi V,
Leopold DA, de Pasquale F, Sporns O, Walter M, Chang C
(2013a): Dynamic functional connectivity: Promise, issues, and
interpretations. NeuroImage 80:360–378.
Hutchison RM, Womelsdorf T, Gati JS, Everling S, Menon RS
(2013b): Resting-state networks show dynamic functional con-
nectivity in awake humans and anesthetized macaques. Hum
Brain Mapp 34:2154–2177.
Ji GJ, Yu Y, Miao HH, Wang ZJ, Tang YL, Liao W (2017):
Decreased network efficiency in benign epilepsy with centro-
temporal spikes. Radiology 283:186–194.
Jia H, Hu X, Deshpande G (2014): Behavioral relevance of the dynam-
ics of the functional brain connectome. Brain Connect 4:741–759.
Jones DT, Vemuri P, Murphy MC, Gunter JL, Senjem ML,
Machulda MM, Przybelski SA, Gregg BE, Kantarci K,
Knopman DS, Boeve BF, Petersen RC, Jack CR Jr (2012): Non-
stationarity in the “resting brain’s” modular architecture. PLoS
One 7:e39731.
Jung M, Kosaka H, Saito DN, Ishitobi M, Morita T, Inohara K,
Asano M, Arai S, Munesue T, Tomoda A, Wada Y, Sadato N,
Okazawa H, Iidaka T (2014): Default mode network in young
male adults with autism spectrum disorder: Relationship with
autism spectrum traits. Mol Autism 5:35.
Just MA, Cherkassky VL, Keller TA, Kana RK, Minshew NJ (2007):
Functional and anatomical cortical underconnectivity in autism:
Evidence from an FMRI study of an executive function task and
corpus callosum morphometry. Cereb Cortex 17:951–961.
Just MA, Cherkassky VL, Keller TA, Minshew NJ (2004): Cortical
activation and synchronization during sentence comprehension
in high-functioning autism: Evidence of underconnectivity.
Brain 127:1811–1821.
Just MA, Keller TA, Malave VL, Kana RK, Varma S (2012): Autism
as a neural systems disorder: A theory of frontal-posterior
underconnectivity. Neurosci Biobehav Rev 36:1292–1313.
Kennedy DP, Courchesne E (2008): The intrinsic functional organi-
zation of the brain is altered in autism. NeuroImage 39:
1877–1885.
Keown CL, Shih P, Nair A, Peterson N, Mulvey ME, Muller RA
(2013): Local functional overconnectivity in posterior brain
regions is associated with symptom severity in autism spec-
trum disorders. Cell Rep 5:567–572.
Khan S, Gramfort A, Shetty NR, Kitzbichler MG, Ganesan S, Moran
JM, Lee SM, Gabrieli JD, Tager-Flusberg HB, Joseph RM,
Herbert MR, Hamalainen MS, Kenet T (2013): Local and long-
range functional connectivity is reduced in concert in autism
spectrum disorders. Proc Natl Acad Sci USA 110:3107–3112.
Kikuchi M, Yoshimura Y, Hiraishi H, Munesue T, Hashimoto T,
Tsubokawa T, Takahashi T, Suzuki M, Higashida H, Minabe Y
r 5754
r
(2015): Reduced long-range functional connectivity in young
children with autism spectrum disorder. Soc Cogn Affect Neu-
rosci 10:248–254.
Kitzbichler MG, Khan S, Ganesan S, Vangel MG, Herbert MR,
Hamalainen MS, Kenet T (2015): Altered development and
multifaceted band-specific abnormalities of resting state net-
works in autism. Biol Psychiatry 77:794–804.
Laufs H, Rodionov R, Thornton R, Duncan JS, Lemieux L,
Tagliazucchi E (2014): Altered FMRI connectivity dynamics in
temporal lobe epilepsy might explain seizure semiology. Front
Neurol 5:175.
Le Couteur A, Rutter M, Lord C, Rios P, Robertson S, Holdgrafer
M, McLennan J (1989): Autism diagnostic interview: A stan-
dardized investigator-based instrument. J Autism Develop Dis-
ord 19:363–387.
Li HJ, Xu Y, Zhang KR, Hoptman MJ, Zuo XN (2015): Homotopic
connectivity in drug-naive, first-episode, early-onset schizo-
phrenia. J Child Psychol Psychiatry Allied Discip 56:432–443.
Liao W, Wu GR, Xu Q, Ji GJ, Zhang Z, Zang YF, Lu G (2014a):
DynamicBC: A MATLAB toolbox for dynamic brain connec-
tome analysis. Brain Connect 4:780–790.
Liao W, Zhang Z, Mantini D, Xu Q, Ji GJ, Zhang H, Wang J,
Wang Z, Chen G, Tian L, Jiao Q, Zang YF, Lu G (2014b):
Dynamical intrinsic functional architecture of the brain during
absence seizures. Brain Struct Funct 219:2001–2015.
Liu X, Duyn JH (2013): Time-varying functional network informa-
tion extracted from brief instances of spontaneous brain activ-
ity. Proc Natl Acad Sci USA 110:4392–4397.
Liu F, Guo W, Liu L, Long Z, Ma C, Xue Z, Wang Y, Li J, Hu M,
Zhang J, Du H, Zeng L, Liu Z, Wooderson SC, Tan C, Zhao J,
Chen H (2013a): Abnormal amplitude low-frequency oscilla-
tions in medication-naive, first-episode patients with major
depressive disorder: a resting-state fMRI study. Journal of
affective disorders 146:401–406.
Liu F, Guo W, Fouche J-P, Wang Y, Wang W, Ding J, Zeng L, Qiu
C, Gong Q, Zhang W (2013b): Multivariate classification of
social anxiety disorder using whole brain functional connectiv-
ity. Brain Structure and Function 1–15.
Liu F, Wang Y, Li M, Wang W, Li R, Zhang Z, Lu G, Chen H
(2016): Dynamic functional network connectivity in idiopathic
generalized epilepsy with generalized tonic-clonic seizure.
Human brain mapping.
Long Z, Duan X, Mantini D, Chen H (2016): Alteration of func-
tional connectivity in autism spectrum disorder: Effect of age
and anatomical distance. Sci Rep 6:26527.
Lord C, Rutter M, Goode S, Heemsbergen J, Jordan H, Mawhood
L, Schopler E (1989): Autism diagnostic observation schedule:
A standardized observation of communicative and social
behavior. J Autism Develop Disord 19:185–212.
Nelson SB, Valakh V (2015): Excitatory/inhibitory balance and circuit
homeostasis in autism spectrum disorders. Neuron 87:684–698.
Nomi JS, Farrant K, Damaraju E, Rachakonda S, Calhoun VD,
Uddin LQ (2016): Dynamic functional network connectivity
reveals unique and overlapping profiles of insula subdivisions.
Hum Brain Mapp.
Nomi JS, Vij SG, Dajani DR, Steimke R, Damaraju E, Rachakonda
S, Calhoun VD, Uddin LQ (2017): Chronnectomic patterns and
neural flexibility underlie executive function. NeuroImage 147:
861–871.
Peter P, Cahill ME, Jones KA, Jon-Eric VL, Woolfrey KM (2011):
Dendritic spine pathology in neuropsychiatric disorders. Nat
Neurosci 14:285–293.
r
 r State-Related Functional Connectivity in Autism
Picci G, Gotts SJ, Scherf KS (2016): A theoretical rut: Revisiting
and critically evaluating the generalized under/over-connectiv-
ity hypothesis of autism. Develop Sci 19:524–549.
Power JD, Barnes KA, Snyder AZ, Schlaggar BL, Petersen SE (2012):
Spurious but systematic correlations in functional connectivity MRI
networks arise from subject motion. NeuroImage 59:2142–2154.
Rashid B, Damaraju E, Pearlson GD, Calhoun VD (2014): Dynamic
connectivity states estimated from resting fMRI Identify differ-
ences among Schizophrenia, bipolar disorder, and healthy con-
trol subjects. Front Hum Neurosci 8:897.
Rausch A, Wei Z, Haak KV, Mennes M, Hermans EJ, Oort EV,
Wingen GV, Beckmann CF, Buitelaar JK, Groen WB (2016):
Altered functional connectivity of the amygdaloid input nuclei
in adolescents and young adults with autism spectrum disor-
der: A resting state fMRI study. Mol Autism 7:1–13.
Rudie JD, Brown JA, Beck-Pancer D, Hernandez LM, Dennis EL,
Thompson PM, Bookheimer SY, Dapretto M (2012): Altered
functional and structural brain network organization in autism.
NeuroImage Clin 2:79–94.
Saad ZS, Gotts SJ, Murphy K, Chen G, Jo HJ, Martin A, Cox RW
(2012): Trouble at rest: How correlation patterns and group
differences become distorted after global signal regression.
Brain Connect 2:25–32.
Schumann CM, Bloss CS, Barnes CC, Wideman GM, Carper RA,
Akshoomoff N, Pierce K, Hagler D, Schork N, Lord C,
Courchesne E (2010): Longitudinal magnetic resonance imag-
ing study of cortical development through early childhood in
autism. J Neurosci 30:4419–4427.
Sripada C, Kessler D, Fang Y, Welsh RC, Prem Kumar K,
Angstadt M (2014): Disrupted network architecture of the rest-
ing brain in attention-deficit/hyperactivity disorder. Hum
Brain Mapp 35:4693–4705.
Stoner R, Chow ML, Boyle MP, Sunkin SM, Mouton PR, Roy S,
Wynshaw-Boris A, Colamarino SA, Lein ES, Courchesne E
(2014): Patches of disorganization in the neocortex of children
with autism. N Engl J Med 370:1209–1219.
Supekar K, Uddin LQ, Khouzam A, Phillips J, Gaillard WD,
Kenworthy LE, Yerys BE, Vaidya CJ, Menon V (2013): Brain
hyperconnectivity in children with autism and its links to
social deficits. Cell Rep 5:738–747.
Tang G, Gudsnuk K, Kuo SH, Cotrina ML, Rosoklija G, Sosunov A,
Sonders MS, Kanter E, Castagna C, Yamamoto A, Yue Z, Arancio
O, Peterson BS, Champagne F, Dwork AJ, Goldman J, Sulzer D
(2014): Loss of mTOR-dependent macroautophagy causes
autistic-like synaptic pruning deficits. Neuron 83:1131–1143.
Thompson WH, Fransson P (2015): The mean-variance relation-
ship reveals two possible strategies for dynamic brain connec-
tivity analysis in fMRI. Front Hum Neurosci 9:398.
Thompson WH, Fransson P (2016): On stabilizing the variance of
dynamic functional brain connectivity time series. Brain Con-
nect 6:735–746.
r 5755
r
Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F,
Etard O, Delcroix N, Mazoyer B, Joliot M (2002): Automated
anatomical labeling of activations in SPM using a macroscopic
anatomical parcellation of the MNI MRI single-subject brain.
NeuroImage 15:273–289.
Uddin LQ, Supekar K, Menon V (2013): Reconceptualizing func-
tional brain connectivity in autism from a developmental per-
spective. Front Hum Neurosci 7:458.
van den Heuvel MP, de Lange SC, Zalesky A, Seguin C, Yeo BT,
Schmidt R (2017): Proportional thresholding in resting-state
fMRI functional connectivity networks and consequences for
patient-control connectome studies: Issues and recommenda-
tions. NeuroImage 152:437–449.
van den Heuvel MP, Kahn RS, Goni J, Sporns O (2012): High-cost,
high-capacity backbone for global brain communication. Proc
Natl Acad Sci USA 109:11372–11377.
von dem Hagen EA, Stoyanova RS, Baron-Cohen S, Calder AJ
(2013): Reduced functional connectivity within and between
‘social’ resting state networks in autism spectrum conditions.
Soc Cogn Affect Neurosci 8:694–701.
Wang K, Liang M, Wang L, Tian L, Zhang X, Li K, Jiang T (2007):
Altered functional connectivity in early Alzheimer’s disease: A
resting-state fMRI study. Hum Brain Mapp 28:967–978.
Washington SD, Gordon EM, Brar J, Warburton S, Sawyer AT,
Wolfe A, Mease-Ference ER, Girton L, Hailu A, Mbwana J,
Gaillard WD, Kalbfleisch ML, VanMeter JW (2014): Dysmatu-
ration of the default mode network in autism. Hum Brain
Mapp 35:1284–1296.
Wee CY, Yap PT, Shen D (2016): Diagnosis of autism spectrum
disorders using temporally distinct resting-state functional
connectivity networks. CNS Neurosci Therap 22:212–219.
Yerys BE, Gordon EM, Abrams DN, Satterthwaite TD, Weinblatt
R, Jankowski KF, Strang J, Kenworthy L, Gaillard WD, Vaidya
CJ (2015): Default mode network segregation and social defi-
cits in autism spectrum disorder: Evidence from non-
medicated children. NeuroImage Clin 9:223–232.
Yu Q, Erhardt EB, Sui J, Du Y, He H, Hjelm D, Cetin MS,
Rachakonda S, Miller RL, Pearlson G, Calhoun VD (2015):
Assessing dynamic brain graphs of time-varying connectivity
in fMRI data: Application to healthy controls and patients
with schizophrenia. NeuroImage 107:345–355.
Zalesky A, Fornito A, Bullmore ET (2010): Network-based statistic:
Identifying differences in brain networks. NeuroImage 53:
1197–1207.
Zalesky A, Fornito A, Cocchi L, Gollo LL, Breakspear M (2014):
Time-resolved resting-state brain networks. Proc Natl Acad Sci
USA 111:10341–10346.
Zhan Y, Paolicelli RC, Sforazzini F, Weinhard L, Bolasco G,
Pagani F, Vyssotski AL, Bifone A, Gozzi A, Ragozzino D,
Gross CT (2014): Deficient neuron-microglia signaling results
in impaired functional brain connectivity and social behavior.
Nat Neurosci 17:400–406.
r