Anomalous Network Architecture of The Resting Brain in Children
Anomalous Network Architecture of The Resting Brain in Children
Anomalous Network Architecture of The Resting Brain in Children
Author manuscript
J Fluency Disord. Author manuscript; available in PMC 2019 March 01.
Author Manuscript
Hayward, CA
3Department of Psychiatry, University of Utah, Salt Lake City, UT
Abstract
Purpose—We combined a large longitudinal neuroimaging dataset that includes children who do
and do not stutter and a whole-brain network analysis in order to examine the intra- and inter-
network connectivity changes associated with stuttering. Additionally, we asked whether whole
brain connectivity patterns observed at the initial year of scanning could predict persistent
stuttering in later years.
Methods—A total of 224 high-quality resting state fMRI scans collected from 84 children (42
stuttering, 42 controls) were entered into an independent component analysis (ICA), yielding a
Author Manuscript
number of distinct network connectivity maps (“components”) as well as expression scores for
each component that quantified the degree to which it is expressed for each child. These
expression scores were compared between stuttering and control groups’ first scans. In a second
analysis, we examined whether the components that were most predictive of stuttering status also
predicted persistence in stuttering.
disorder and provides comprehensive brain network maps that substantiate past theories
emphasizing the importance of considering situational, emotional, attentional and linguistic factors
in explaining the basis for stuttering onset, persistence, and recovery.
*
Corresponding author: Department of Psychiatry University of Michigan, Rachel Upjohn Building Rm 2541, 4250 Plymouth Rd. Ann
Arbor, MI 48109, Tel: 734-232-0300, Fax: 734-939-7868, [email protected].
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our
customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of
the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered
which could affect the content, and all legal disclaimers that apply to the journal pertain.
Chang et al. Page 2
Keywords
Author Manuscript
Stuttering; resting state fMRI; intrinsic connectivity networks; default mode network; independent
component analysis (ICA)
1. Introduction
Stuttering is a complex neurodevelopmental disorder (Smith & Weber, 2016) with a life-
span incidence estimated at 8% (Yairi & Ambrose, 2013). The etiology of stuttering remains
unclear, but growing evidence points to an interplay among genetic, epigenetic, and
environmental factors that influence brain development, which in turn affect fluent speech
production. The pathophysiological bases of stuttering likely arise during a critical period
when children undergo vigorous development in speech and language, general cognition,
Author Manuscript
motor control, and emotions. By carefully mapping the complex interactions between these
developing systems, it might be possible to explain the mechanisms of stuttering onset,
persistence, heterogeneity of symptoms, comorbidity, and severity.
Over the past two decades a large number of neuroimaging studies of stuttering have been
conducted, which have greatly added to our knowledge. These studies, however, were
mostly conducted in adults, leaving the developmental mechanisms of stuttering largely
unexplored. Moreover, studying adults risks conflating causal and compensatory
mechanisms, a point noted almost two decades ago by Ludlow (2000) in an editorial
discussing seminal neuroimaging papers: “In the adult system it may be difficult to
distinguish between mechanisms responsible for stuttering and those developed to
compensate… As new technologies emerge which are non-invasive and have improved
temporal resolution, studies in children who stutter during the critical period for speech
Author Manuscript
development may provide understanding of how this dysfunctional system emerges.” (p.
1984)
Since that time, many pediatric studies have examined both functional and structural brain
differences in young children who stutter (e.g. Arnold et al., 2011; Beal et al., 2011, 2015;
Chang et al., 2008, 2015; Chang & Zhu 2013; Choo et al., 2012; Etchell et al., 2016;
Jansson-Verkasalo et al., 2014; Kaganovic et al., 2010; Mohan & Weber-Fox 2015; Ozcan et
al., 2009; Ozge et al., 2004; Sato et al. 2011; Sowman et al. 2014; Weber-Fox et al., 2008;
Usler & Weber-Fox 2015; and for a systematic review see Etchell et al., 2016 in this special
issue). Results from this body of work are mixed. Some studies report that both children and
adults who stutter exhibit aberrant auditory-motor integration (Beal et al., 2010, 2011;
Chang & Zhu, 2013; Jansson-Verkasalo et al., 2014) and possible deficiencies in the basal-
Author Manuscript
ganglia thalamocortical loop (Lu et al., 2009, 2010; Chang & Zhu, 2013; Xuan et al., 2012),
whereas others provide conflicting findings (e.g., right hemisphere increases in structural
measures) in adults (De Nil et al., 2000; Foundas et al., 2003; Kikuchi, 2011; Preibisch et
al., 2003) but not in children (Chang et al., 2008) and an absence of differences in
lateralization of brain function during speech production in children who stutter (Sowman et
al., 2014). Convergent findings from children and adults have been considered to reflect
stuttering trait-associated differences in the brain that may be related to pathophysiology of
stuttering. Results that conflict between adults and children on the other hand, are thought to
Author Manuscript
reflect compensation- and adaptation- related changes in the adult brain that are not directly
related to pathophysiology of stuttering. While studying children who stutter is a step in the
right direction, there are three key limitations that are generally observed in this body of
work.
Riley, 2000) of all school-age stuttering children exhibit symptoms consistent with ADHD,
and even more (~58%) have been reported to exhibit clinically relevant symptoms of ADHD
as rated by parents (Donaher & Richels, 2012). Likewise, anxiety is a common comorbidity
associated with stuttering (see Kefalianos et al., 2012 for a review). A meta-analysis of over
1300 adults who stutter reported elevated levels of trait anxiety and social anxiety relative to
non-stuttering adults (Craig & Tran, 2014) and another study reported that adults who stutter
have six times greater odds of suffering from an anxiety disorder than their fluent peers
(Iverach et al., 2009). The high prevalence of comorbid disorders highlights the need to
consider other aspects of stuttering beyond those directly related to speech, such as attention
and emotion. Small sample sizes also make it difficult to examine within group factors
relating to persistence and recovery, as well as sex differences. Finally, there is also the well-
known problem of the increased risk of reporting a type II error, making it difficult to be
Author Manuscript
confident about the results that such studies provide until findings can be replicated by larger
studies and by independent research laboratories.
(Jones et al., 2014; Ntourou et al., 2013; Zengin-Bolatkale et al., 2015), temperament
(Eggers et al., 2010; Embrechts et al., 2000), inhibition (Choi et al., 2013), and attention
(Eggers et al., 2010; Karrass et al., 2006; Piispala et al., 2016), there has been very little
neuroimaging work focusing on these issues. Perhaps more crucially, the brain is not a set of
isolated regions, but rather comprises intrinsically connected networks (ICNs) that underpin
distinct but related functions (Fox et al., 2005; Seeley et al., 2007; Yeo et al., 2011). The
locationist approach is poorly suited for elucidating network-spanning alterations.
1.1.3 Static approach—Most existing imaging studies of stuttering have also examined
Author Manuscript
structural and functional differences at a single point in time. This is problematic because
stuttering is a neurodevelopmental disorder: any subtle deficiencies in stuttering are likely to
change dynamically with age and be influenced by interactions with other large-scale neural
networks subserving cognitive, motor, and emotion functions. To properly understand the
nuances of a disorder as complex as stuttering, there is a need to study brain differences in
the same individuals over an extended period of time.
difficulties make it less feasible for examining neural activity in children. For example,
young children can have considerable difficulty remaining still for the duration of a
behavioral paradigm and/or may not perform sufficiently well for their data to be usable,
leading to significant data loss. Indeed, the success rates of scanning children between the
ages of 4–6 years is so low that researchers recruit and estimated 20–40% more subjects to
make up for data loss (Yerys et al., 2009). These authors also note there are significantly
higher failure rates in clinical populations as compared to typically developing children.
To overcome these issues, some researchers have utilized passive tasks such as perceiving
rhythmic and arrhythmic tones that do not require behavioral responses (e.g., Chang et al.,
2016; Etchell et al., 2016). Another method, which will be the focus of this report, is resting
state fMRI in which participants lie awake in the scanner relaxed with their eyes open
(Thomason et al., 2011). This method is well-suited to examining brain activity in children
Author Manuscript
because it can be done in a very short period of time and does not involve performance of a
behavioral task.
Resting state fMRI (hereafter rsfMRI) analyses have enabled identification of intrinsically
connected, large-scale neural networks such as the default mode network (DMN), and this
network has substantial relevance to stuttering. The discovery of the DMN was due in part to
the observation that even when the brain was not engaged in a specific task, it continued to
consume about 20% of the body’s oxygen, demonstrating that it is neither inactive nor
simply “doing nothing.” (Fox & Raichle, 2007). This finding generated a significant amount
of interest and prompted researchers to consider what the brain was doing in the absence of
a task (Greicius et al., 2002). A seminal meta-analyses identified a set of regions that
displayed a remarkably consistent pattern: These regions were less active during cognitively
Author Manuscript
demanding tasks but more active during passive/resting control conditions (Mazoyer et al.,
2001; see also Spreng et al., 2009). This set of regions, termed the DMN (Raichle et al.,
2001), consisted of the posterior cingulate, bilateral temporal parietal junction, medial
prefrontal, bilateral superior frontal, inferior temporal, cerebellar tonsils, and bilateral
parahippocampal regions (Fox et al., 2005).
Functionally, the DMN is associated with introspective activities that are thought to
predominate during the resting state: mind wandering, prospection, theory of mind, and
autobiographical memory (Christoff et al., 2009; Greicius et al., 2009; Mason et al., 2007;
Author Manuscript
Spreng & Grady, 2010; and see Buckner et al. 2008 for a review and Spreng et al., 2009 for
a meta-analysis). Several studies have shown that the DMN exhibits an antagonistic
relationship and strong negative correlations (anticorrelations) with “task positive” intrinsic
connectivity networks, including dorsal attention network (DAN), ventral attention networks
(VAN), and frontoparietal network (FPN) (Figure 1). Connectivity within DMN and between
DMN and task positive networks has been shown to influence behavioral performance on a
given task (e.g., Daselaar et al. 2004; Kelly et al., 2008a; Poole et al., 2016). For instance,
Poole et al. (2016) reported that greater intra-network connectivity within the DMN was
predictive of better distractor (e.g., task irrelevant information) suppression, while greater
inter-network connectivity between the DMN and attention networks was predictive of
poorer distractor suppression. The relationship between networks involved in speech
production and the DMN appears to be more complex. For instance, it has been shown that
Author Manuscript
parts of DMN can exhibit varying levels of deactivation depending on the type of words
being produced (Seghier & Price, 2012).
The DMN undergoes significant change throughout the course of development (for reviews
see Cao et al., 2016; Menon, 2013; Power et al., 2010). Whereas the DMN in adults exhibits
strong intra-hemispheric connections and long distance connections, the DMN in children is
incompletely connected (Fransson et al., 2007; Gao et al., 2009). Additionally, unlike the
adult DMN, the child DMN is characterized by a partial (Fair et al., 2008) or complete
(Kelly et al., 2009a) absence of anterior to posterior connections (see also Supekar et al.,
2010). As children age, the DMN tends to increase connectivity within its component areas
(i.e., intra-network connectivity). On the other hand, DMN shows increased segregation or
anticorrelation with other intrinsic connectivity networks (i.e. inter-network connectivity)
such as the DAN (Fox et al., 2005). Less segregation between networks that are normally
Author Manuscript
Resting state paradigms have been used in the context of stuttering (Chang & Zhu, 2013;
Ingham et al., 2012; Joos et al., 2014; Lu et al., 2012; Lu et al., 2016; Xuan et al., 2012;
Yang et al., 2016). The analysis methods used in these studies usually focused on specific
regions of interest. More recently, whole-brain connectomic methods have emerged for
better characterizing network structure and network changes in the brain (e.g. Allen et al.,
2012; Fox et al., 2005; Kessler et al., 2016; Sripada et al., 2014). These methods have yet to
be applied to childhood stuttering.
Author Manuscript
examined the brain’s intrinsic intra- and inter-network connectivity differences associated
with stuttering. Furthermore, we examined whether whole brain connectivity patterns
observed at the initial year of scanning could predict persistence or recovery from stuttering
in later years. We hypothesized that children who stutter would differ markedly from
typically developing children in 1) connectivity within the DMN and 2) connectivity
between the DMN and other intrinsic connectivity networks such as somatomotor and
attention networks, and 3) that patterns of network connectivity would predict recovery and
persistence of stuttering.
2. Methods
2.1 Participants
Participants were recruited from an ongoing longitudinal neuroimaging study of
Author Manuscript
developmental stuttering. Each participant was scanned 1 to 4 times, with each scan
occurring approximately 12 months apart. At the time of the analysis for this study, 280
rsfMRI scans from 50 children who stutter (30 boys) and 45 controls (22 boys) were
collected. Eight scans (2.9%) from eight subjects were excluded due to poor performance on
the standardized language tests administered. Forty-eight scans (17.1%) from 33 participants
were excluded due to potential contamination of movement artifacts in the rsfMRI data (the
details of the exclusion criteria are presented in the data analysis section). In the final
analysis, 224 high quality scans from 42 children who stutter (26 boys) and 42 controls (21
boys) were included. Additionally, final stuttering outcomes (persistence or recovery) could
not be determined for 3 children who stutter due to attrition after their participation in the
initial visit (e.g., family relocation); consequently the persistence logistic regression was
based on 39 children who stutter. Participants’ ages at the initial visit ranged from 3.3 to
Author Manuscript
10.8 years, and the mean age was 6.5 with a standard deviation (SD) of 1.9 (for details, see
Table 1). All procedures used in this study were approved by the Michigan State University
Institutional Review Board. All children were paid a nominal remuneration and were given
small prizes (e.g., stickers) for their participation in the experiment.
2.2 Procedure
2.2.1 Speech, Language, Cognitive, and Motor Skills Evaluation—All
participants were monolingual native North American English speakers without confirmed
diagnosis of concomitant developmental disorders (e.g., dyslexia, ADHD, learning delay,
psychiatric conditions). Such in/exclusion criteria were applied during the initial year of this
longitudinal study; however we found that in subsequent years, 14.29% of the stuttering
children went on to receive an ADHD diagnosis, 2.38% another developmental disorder, and
Author Manuscript
9.52% another psychiatric condition. The comparable percentages for controls were 2.38%,
2.38%, and 0% of that cohort, respectively. Each participant underwent careful screening to
ensure normal speech and language developmental history except for the presence of
stuttering in the experimental group. These tests included the Peabody Picture Vocabulary
Test (PPVT-4; Dunn & Dunn, 2007), Expressive Vocabulary Test (EVT-2; Williams, 2007),
Goldman-Fristoe Test of Articulation (Goldman, 2000), Wechsler Preschool and Primary
Scale of Intelligence (WPPSI-III; for children 2:6–7:3; Wechsler, 2002), and Wechsler
Abbreviated Scale of Intelligence (WASI; for children aged 7 and up; Wechsler, 1999). The
Author Manuscript
Of the total of 42 CWS entered into the study, 7 children scored below 3% SLD at the time
Author Manuscript
of their initial testing. Of these, all but one exhibited a composite SSI score that placed them
at “mild” or “moderate”. As an example, a child that scored 1.1% SLD received an SSI
composite score of 20, placing her in the “moderate” stuttering severity range. Only one
child’s SSI composite score was at “very mild” according to the SSI. SSI scores incorporate
not only stuttering frequency, but the average duration of the three longest stuttering
instances as well as any physical concomitants of stuttering. All children categorized as
stuttering needed to score at least “very mild” on the SSI, in addition to the parent concerns
and clinician reports indicating that the child indeed stutters.
Primarily based on yearly %SLD assessments and supplemented via SSI-4 scores of
children who stutter, stuttering participants were further retrospectively categorized as
recovered or persistent. Specifically, with very few exceptions (explained below), a child
Author Manuscript
was considered recovered if the %SLD score was below 3% (see Buhr & Zebrowski, 2009;
Yairi & Ambrose, 1992; 1999 for similar application of the 3% SLD criteria) and scored
“very mild” or less in severity according to SSI at the second visit or thereafter. A child was
categorized as persistent if his or her %SLD score was at or above 3% SLD at the second
visit or thereafter, and the onset of stuttering had been at least 36 months prior to his most
recent visit. Categorization of recovered and persistent status was also validated by clinician
and parental reports as stuttering symptoms can fluctuate widely from day to day for
children who stutter. If the child exhibited less than 3% SLD in his speech sample on the day
of his/her visit, but parent report indicated that child was having a very mild day compared
to his/her usual fluency level, and confirmed continued stuttering in the past 6 months, this
child was retained in the persistent group.
On the other hand, if the child exhibited above 3% SLD that mostly comprised whole-word
Author Manuscript
repetitions, exhibited very mild or less in severity according to SSI, and parent confirmed
more than 6 months of continued fluency, we considered that child to have recovered. Using
these criteria, we identified 15 children who recovered from stuttering (37 scans) and 24
children with persistent stuttering (63 scans). Three among the 42 children who stutter could
not be categorized as either persistent or recovered, as data from only their first year visit
was available due to attrition (e.g., family relocation). For controls, the inclusion criteria
were as follows: never diagnosed with stuttering, no family history of stuttering, lack of
Author Manuscript
parental concern for their child’s fluency, and the child’s %SLD score was < 3.
2.2.2 MRI data acquisition—During the rsfMRI scan, children lay supine with their eyes
open. They were instructed to remain as still as possible. Preceding the MRI scanning
session, all children were trained during a separate visit with a mock scanner to familiarize
and desensitize them to the sights and sounds of the scanner and to practice being still inside
the scanner bore (Chang & Zhu, 2015; Chang et al., 2016). To ensure that the child remained
calm and to minimize the possibility of movement, an experimenter sat by the child
throughout the duration of the scan. All MRI scans were acquired on a GE 3T Signa HDx
MR scanner (GE Healthcare) with an 8-channel head coil. During each session, 180 T1-
weighted 1-mm3 isotropic volumetric inversion recovery fast spoiled gradient-recalled
images (3D IRFSPGR) (10 min scan time), with CSF suppressed, were obtained to cover the
whole brain with the following parameters: time of echo = 3.8 ms, time 2332 ms, flip angle
Author Manuscript
= 8°, field of view = 25.6 cm × 25.6 cm, matrix size = 256 × 256, slice thickness = 1 mm,
and receiver bandwidth = 20.8 kHz.
2.2.3 rsfMRI—Thirty-six contiguous 3-mm axial slices were collected with a gradient-echo
EPI sequence (7 min) in an interleaved order with TE = 27.7 ms, TR = 2.5 sec, flip angle =
80°, FOV = 22 cm, matrix size = 64×64, and 164 time points, with the first four data points
discarded.
2.3 Analyses
2.3.1 Resting state fMRI data analysis
2.3.1.1 Preprocessing and Connectome Generation: Resting state functional images were
Author Manuscript
corrected for slice acquisition times and realigned to the first volume to correct for motion
using AFNI (https://afni.nimh.nih.gov/afni). Each data set was de-noised using a procedure
that uses spatial independent component analysis (sICA) and pattern recognition algorithms
to automatically separate and remove motion and physiological noise from signals
originating from neuronal activity (Xu et al., 2014). This method has been demonstrated to
be effective in removing artifact during continuous speech production as well as pediatric
resting-state fMRI scans. The de-noised data sets were normalized to a standard space
(ICBM152) using diffeomorphic image registration algorithm (DARTEL; Ashburner, 2007)
implemented in SPM12 (http://www.fil.ion.ucl.ac.uk/spm/software/spm12/). Finally, images
were smoothed with a Gaussian kernel of 6 mm full width at half maximum.
After the above preprocessing, connectomes were generated as described in our previous
work (Kessler et al., 2014; 2016; Sripada et al., 2014). Briefly, first the data were linearly
Author Manuscript
detrended. Then we carried out regressions to remove nuisance effects from each voxel’s
time series including 24 motion parameters (6 original parameters estimated during
realignment above plus their first derivatives and quadratic terms for each of these), and the
first 5 principal components of each of CSF and white matter extracted from subject-specific
masks using the segmented images from preprocessing above (Behzadi et al., 2007). Next,
we band-pass filtered the data in the 0.01–0.10 Hz range to isolate the low-frequency signals
of interest for resting state analysis. To further reduce the potential adverse effects of
motions, motion scrubbing was performed via removal of individual frames from the times
Author Manuscript
series that had a framewise displacement value greater than 0.5 (Power et al., 2012).
Additionally, any scan which had more than half its time points removed by scrubbing was
excluded from the analysis. Spatially averaged time series were extracted from each of 1101
ROIs placed in a regular 12-mm grid throughout the brain. Finally, Pearson’s correlation
coefficients were calculated pairwise between time courses for each of the ROIs, producing
a cross-correlation map with 605,550 non-redundant entries, and the values were
transformed with Fisher’s r-to-z transform yielding a matrix of z-values for each subject.
et al., 2014; 2016). In brief, the steps are as follows (see Figure 2). Regression-based
cleansing across subjects is used to remove nuisance variation (linear and quadratic mean
framewise displacement, handedness, and IQ) from each connection of the connectome.
Although IQ, PPVT, and EVT were significantly lower in the persistent group relative to
controls, we only included IQ in the regression model because these three measures were
highly correlated (r > 0.7). After PCA-based dimensionality reduction (standard in ICA)
with model order set at 15 (chosen heuristically to correspond with previous work [Kessler
et al., 2014; 2016]), the FastICA algorithm was applied to these reduced data (Hyvärinen,
1999) to obtain component source maps and subject-specific expression scores (i.e., how
much a given component is expressed in each subject). Components were thresholded at |z|
>3 for display and nodes were grouped into the intrinsic connectivity network they belong to
(based on the network parcellation of Yeo et al., 2011). The full, thresholded component
Author Manuscript
maps are presented in the Appendix. Stability was assessed with ICASSO (Himberg et al.,
2004) which was run 1000 times and indicated that most components were stable: 12 of the
15 components had high Iq ranging from 0.9149–.9973, two of the remaining three
components (11,14) ranged from 0.8433–0.8485, and the remaining component (13) was
substantially less reliable than the others (Iq= 0.7196).
Of note, we utilized the full dataset of scans (224) to increase the power of the ICA
algorithm to identify reliable components. However, as mentioned in the introduction our
primary goal for this study was to identify early markers that could predict later stuttering
persistence, so the analyses discussed below only used data from the earliest scan from each
of the 84 included participants. Additionally, given that the sign of components coming out
of ICA is arbitrary, we chose to display all components such that increasing expression of
Author Manuscript
2.3.1.3 Generating circle graph visualizations: The visualizations in Figures 3–5 examine
the ICA component effects within pairs of networks. The procedure for generating these is
discussed at length in Kessler et al. (2015). Briefly, the width of the arcs linking regions is
Author Manuscript
proportional to the number of implicated connections between those regions, such that more
heavily implicated regions will be connected by wider arcs. To enhance the readability of the
circle graphs, arcs that represent <1% of the internetwork connections were omitted. In
addition, any subregions in which both left and right sides participated in <1% of per-graph
connections for all visualizations were omitted from graphs.
predictors. This model was compared to a model containing sex alone as a predictor to
determine if the component expressions significantly improved the model. In a subsequent
logistic regression model, using only those components that differentiated stuttering from
controls (component expressions from the first model with p<0.05), we further examined
whether component expressions predicted persistence of stuttering in later years among
children who stutter. For this second regression we included component expressions for
significant components as well as sex and age (linear and quadratic terms) as predictors,
given their relevance to persistent stuttering (e.g., boys, and older children among children
who stutter, are more likely to exhibit persistent stuttering; Yairi & Ambrose, 1999). This
model was compared to a model with just sex and age (linear and quadratic) as predictors to
determine if component expression significantly improved the model. Finally, in order to
determine whether the effects we observed in the above analyses were specific to stuttering
Author Manuscript
and not driven by comorbid attention or other psychiatric conditions, we performed two
additional subset analyses. In each case we repeated the joint ICA component identification
and subsequent regression models for stuttering and persistence when 1) removing all
subjects with a comorbid ADHD diagnosis or subclinical attention problems (N=7) and 2)
removing all subjects with any comorbid psychiatric diagnosis (N=9).
3. Results
3.1 Participant demographic data
Children with persistent stuttering (“persistent”), those recovered from stuttering
(“recovered”), and typically developing fluent controls did not differ in mean age,
socioeconomic status, or sex ratio. Compared with controls, the persistent group scored
Author Manuscript
significantly lower on IQ, PPVT, and EVT (p < 0.05; Table 1), but the recovered group did
not differ significantly from controls in any of these tests. Persistent and recovered groups
differed significantly on the GFTA. Because of this group difference, we added GFTA as a
nuisance variable in subsequent regression analyses (see below). The two stuttering groups
did not differ on any other standardized test score, and did not differ in average
chronological age or sex ratio. Stuttering severity measures were not included in the model
because severity was used to define persistent and recovered groups. As expected, the
at subsequent visits (p<0.01), indicating the divergence of symptoms between the stuttering
groups with progressing age.
3.2 Intrinsic connectivity of large scale brain networks associated with stuttering that
predict stuttering persistence
Our logistic regression model identified four components (5,7,8,15) that were significantly
predictive of stuttering diagnosis (ps<0.05). This overall regression model (all 15
components plus sex) was significantly better than the sex alone model with a likelihood
ratio test (χ215 = 36.02, p=0.0018). Of these, component 15 was found to be highly
significantly related to nuisance variation (regression model with component expression as
outcome and handedness, IQ, and linear and quadratic mean framewise displacement as
predictors, F4,79 = 68.26, p <2.2×10−16), despite the pre-ICA cleansing, so it was dropped
Author Manuscript
from the following model. We restricted our subsequent analyses to predict recovery/
persistence with a logistic regression with the remaining 3 component expressions (5,7,8) as
predictors. This overall model with component expressions plus sex and linear and quadratic
age was significantly better than the model with just sex and linear and quadratic age (χ23 =
22.96, p = 0.000046), with component 5 being individually significant (p = 0.035). This
component predicted whether a child would persist or recover in a direction consistent with
the prediction of stuttering/non-stuttering status (i.e., increased expression of this component
was associated with both presence and persistence of stuttering).
As noted in the behavioral analysis, there was a significant difference between persistent and
recovered children in GFTA scores. Because of this, we ran a second logistic regression
adding GFTA as a predictor to our previous model to determine if our conclusions would
change. Inclusion of GFTA did not significantly increase the predictive value of this model
Author Manuscript
Connectivity between the DMN and FPN showed overall increases, with notable
Author Manuscript
involvement of the bilateral dmPFC, Ang, posterior cingulate cortex (PCC), and lateral
temporal lobe (LTL) of the DMN and their connections to the lateral parietal lobule, superior
frontal cortex, and dorsal lateral prefrontal cortex of the FPN (Figure 3B). In terms of inter-
network connectivity between DMN and attention networks (DAN, VAN), the results
showed both increases and decreases. For connections with DAN, there were hyper-
connectivities involving the left inferior parietal lobe (IPL) and its connections with the
bilateral angular gyrus and bilateral lateral temporal lobe (LTL), whereas there was vast
hypo-connectivity involving the left MTG and its connections to the bilateral dmPFC,
Author Manuscript
bilateral LTL, and PCC (Figure 3C). The connectivity between DMN and VAN showed
overall greater hypo-connectivities, especially involving the PCC. However, there was
increased connectivity between left anterior insula (aINS) and right dmPFC, between right
aINS and left oIFG, and between bilateral Ang and the SMG (Figure 3D). For this
component (5), increased component expression was significantly correlated with greater
risk of persistent stuttering (Z = 2.11, p = 0.035).
3.4 Component 7 showed decreased connectivity between the dorsal attention and visual
networks for stuttering children
Component 7 (Figure 4) showed overall decreased intra-network connectivity in the VN
(Figure 4A) and between the VN-DAN that was associated with stuttering. The bulk of the
hypoconnectivities occurring between the VN and DAN involved the left MTG of the DAN
Author Manuscript
(Figure 4B). Connectivity between VN and DMN was primarily decreased, with the
exception of connections involving DMN nodes left PCC and left dmPFC (Figure 4C).
3.5 Component 8 showed increased inter-network connectivity between DMN and VAN,
and mixed results between SMN and VAN; however SMN -DAN connectivity largely
decreased in stuttering
In component 8 (Figure 5), the inter-network connectivity between DMN and VAN was
predominantly heightened, except for connections between the left PCC-bilateral insula,
bilateral SMA and right insula-bilateral ventromedial prefrontal cortex (vmPFC), where
there were hypo-connectivities between the two networks (Figure 5A). Connectivity
between DMN and SMN showed decreases involving the PCC node of the DMN; the
connectivity involving the SMN node left STG however showed increased connectivity with
Author Manuscript
many DMN regions (Figure 5B). Inter-network connectivity involving the SMN and
attention networks showed interesting and contrasting patterns. The SMN-VAN exhibited
predominantly hyper-connectivity among primarily SMA and insula but hypo-connectivity
between precuneus and all SMN regions (Figure 5C), whereas the SMN-DAN connectivity
was overall decreased (Figure 5D).
alone (χ215 = 45.54, p=0.000062). In the persistence model, component 5 was still
individually significant, and the overall model with expressions, sex, and linear and
quadratic age was better than the model with sex and linear and quadratic age alone (χ23 =
24.54, p=0.000019).
For the analysis excluding all participants with a comorbid psychiatric diagnosis all of the
components that were significant in the original stuttering diagnosis model remained
significant. The model with component expressions and sex was still significantly better than
the model with sex alone (χ215 = 41.96, p=0.00023). In the persistence model, component 5
Author Manuscript
was no longer individually significant, however the overall model with expressions, sex, and
linear and quadratic age was still significantly better than the model with sex and linear and
quadratic age alone (χ23 = 24.19, p=0.000023).
4. Discussion
Abnormal connectivity within and between intrinsic connectivity networks (ICNs) is
implicated in a host of neurodevelopmental disorders such as autism (Washington et al.,
2014), attention-deficit/hyperactivity disorder (Sripada et al., 2014; Kessler et al., 2016),
developmental dyslexia (Schurz et al., 2015), and other psychopathologies such as anxiety
disorders and schizophrenia (see Broyd et al. 2009; Menon 2011; Whitfield-Gabrieli & Ford,
2012 for reviews; and see also Songua-Barke & Castellanos, 2007). Here we provide the
first evidence supporting the presence of anomalous ICN architecture in stuttering children.
Author Manuscript
The current study used whole brain ICA to compare inter and intra-network connectivity in a
large sample of children who do and do not stutter and who were each scanned multiple
times over a number of years. We found that children who stutter exhibit abnormal
connectivity within default mode network (DMN) as well as between DMN and other ICNs
(see Figure 6 for a summary). In addition, we showed these network changes predict
whether stuttering subsequently resolves or persists. Based on these results we provide an
updated view of possible pathomechanisms associated with childhood stuttering.
ICNs (Hagmann et al., 2008). Interestingly, recent work has demonstrated that administering
the dopamine agonist L-DOPA reduces connectivity between right caudate and PCC as well
as other regions within the DMN (Kelly et al., 2009b). There is some theoretical (Alm,
2004) and empirical (Civier et al., 2013; Wu et al., 1997) evidence that stuttering is
associated with excess levels of dopamine in the basal ganglia. Taken together, these studies
suggest the possibility that the aberrant connectivity of PCC found in the current study may
be related to dopamine levels. Although we did not examine dopamine levels, or directly
examine connections with the putamen, we observed reduced connectivity between PCC and
inferior frontal gyrus (pars orbitalis), lateral temporal lobe, and dmPFC within the DMN in
children who stutter and those who persist in stuttering. Our results therefore point to a role
for PCC in stuttering, which in turn might be explained in terms of altered levels of
dopamine.
Author Manuscript
The DMN undergoes development during childhood that generally leads to increased intra-
network connectivity among DMN nodes, while increasing segregation with other ICNs
(Fair et al., 2007, Kessler et al., 2016; see Menon, 2013 for a review). The overall decreased
intra-network connectivity within the DMN in the present study (along with other inter-
network patterns found in component 5) was associated with stuttering in general as well as
stuttering persistence. This suggests that coherent development of DMN may be
compromised in children who stutter. Aberrant development of the DMN may affect how
Author Manuscript
this network interacts with other networks, leading to inappropriate DMN activity that
interferes with task-related functions (Songua-Barke & Castellanos, 2007) such as speech
motor control.
4.2 Anomalous connectivity between the DMN and other ICNs is predictive of persistent
stuttering
Connectivity between DMN and other ICNs, such as DAN, was able to predict whether a
child would exhibit persistent stuttering (versus recover) in later years. Hyper-connectivity
of the left inferior parietal lobule (IPL) in DAN with multiple nodes of the DMN is
particularly interesting. Simonyan and Fuertinger (2015) showed that the IPL and the
cerebellum are significantly more active during speech production than at rest and concluded
that these regions facilitated the transition from rest to speech. One explanation for the
Author Manuscript
heightened connectivity found in stuttering children between left IPL and DMN may be
related to inefficient transitioning from rest to speech. Indeed, people who stutter often have
difficulty initiating speech movements as indicated by smaller motor evoked potentials in the
left tongue motor representation of the motor cortex (Neef et al., 2015) and elevated motor
thresholds as measured from the hand area of the motor cortex (Busan et al., 2013;see also
Alm et al., 2013) relative to their fluent peers. It is also interesting to note that
administration of dopamine antagonists increases the degree of anticorrelation of DMN and
the inferior parietal regions (Cole et al., 2013). This may have particular relevance for
stuttering. Dopamine antagonists, such as Olanzapine (Lavid et al., 1999; Maguire, 2004),
Risperidone (Maguire et al., 2000) as well as other drugs (Stager et al., 2005), have been
used to treat stuttering to enhance fluent speech (see for review Maguire et al. 2012). The
current data provide potential brain network-based explanations for how dopamine
Author Manuscript
antagonists may help achieve fluent speech in some speakers who stutter (i.e., they facilitate
the transition of rest to an active state for the speech production network).
We also observed prominent hypo-connectivity between DMN and DAN, most of which
involved left middle temporal gyrus (MTG), an important hub within DAN. The MTG
undergoes significant changes between the ages of 3 and 5 (Xiao et al., 2016) during which
time there is a decrease in the dominance of the right hemisphere and a shift to more
bilateral organization. This is also the period when stuttering typically emerges (Yairi &
Ambrose, 1992). MTG is a critical region involved in semantic access and association,
including word retrieval and translating semantics (word meaning) to phonology (speech
sounds; Price, 2012). The MTG is more active when attending to speech compared to
passively listening to the same stimuli (Hugdahl et al., 2003). The right and left MTG (along
with the STG, left IFG, and insula) are associated with lexical and non-lexical lip reading
Author Manuscript
(Paulesu et al., 2003). Interestingly, in a group analysis, MTG was the only region of the
brain to exhibit common activation between viewing silent speech and the auditory
perception of speech (Hall et al., 2005). It is also more active when viewing moving rather
than stationary lips (Calvert & Campbell, 2003), indicating that it has a role in visual
attention to speech related stimuli. The aberrant connectivity of this region and DMN
perhaps suggests deficiencies in attention processes that enable perception of subtle
orofacial movements during the course of speech development in children who stutter.
We observed increased connectivity between DMN and two task positive networks, FPN and
Author Manuscript
DAN, in component 5, and this component was found to predict persistent stuttering. These
findings can be explained in terms of the default network interference model (Songua-Barke
& Castellanos, 2007). During childhood, nodes within DMN increase connectivity with age,
and at the same time DMN segregates from (i.e., exhibits increasing anticorrelations with)
task-positive ICNs (Fair et al., 2009; Supekar et al., 2009). This segregation appears to
support efficiency of task-positive networks, including those supporting attention and motor
task performance. Diminished segregation might allow intrusion of DMN activity that
causes inefficient functioning of task-positive processes (Zou et al., 2013) and behavioral
variability (Kelly et al., 2008; Poole et al., 2016).
4.3 Anomalous connectivity within and between ICNs associated with stuttering
regardless of eventual persistence or recovery
4.3.1 A dissociation between the dorsal and ventral attention networks and
their interaction with the somatomotor network—Another interesting result from
the current study concerned inter-relationships between SMN, DAN, and VAN (component
8). Specifically, we observed an overall decrease in connectivity between SMN and DAN,
but an increase in connectivity between SMN and VAN, the latter especially involving the
anterior insula. This is intriguing because these attention networks are associated with
different forms of attention control. Whereas DAN is associated with goal driven behavior
Author Manuscript
and top-down control crucial for auditory attention, VAN is linked with attention to
infrequent or unexpected events (Vossel et al., 2014). The dissociation of the two attention
networks in their relationship with SMN suggests an imbalance in how attention processes
regulate speech motor control. In the context of speech acquisition, such an imbalance might
take the form of invalid or behaviorally irrelevant cues taking precedence over more
important valid cues in guiding speech movements. For example, a child who stutters may
pay more attention to facial expression or gestures than either voice or lip movement (we
elaborate on this point below).
Riecker, 2010), which is highly relevant for phonation for speech and singing (Zarate et al.,
2010). More relevant to its role in VAN, the anterior insula has been implicated in salience
processing (i.e., detecting salient stimuli in the external environment) and in regulating shifts
between introspective and extrospective modes of attention (Menon et al., 2010; Sridharan et
al., 2008). Hyper-connectivity between anterior insula and SMN may therefore contribute to
inefficiencies in utilizing attention during speech motor control in children. For example,
aberrant salience attribution by anterior insula could produce excessive distractibility by
The observed increased connectivity involving VAN is in line with a large body of work
demonstrating that children who stutter frequently meet the diagnostic criteria for attention
disorders (Riley & Riley, 2000) or meet referral criteria for additional evaluation (Donaher
& Richels, 2012) and have difficulty orienting attention (Eggers et al., 2012). Notably,
orienting attention depends on one’s ability to select information from competing sensory
inputs (Lepsien & Nobre, 2006) and is relevant to salience processing (Uddin, 2015). More
generally, abnormal functioning of VAN corroborates a large body of work demonstrating
that children who stutter exhibit poorer performance on cognitive tasks involving attention
regulation (Felsenfeld et al., 2010; Piispala et al., 2016), inhibitory control (Eggers et al.,
2013; see also Piispala et al., 2016), adaptability to change (Anderson et al., 2003), emotion
(Jones et al., 2014; Karrass et al., 2006), and dual task conditions (Bosshardt, 2006; Smits-
Author Manuscript
Bandstra & De Nil, 2009; Vasic & Wijnen, 2005). The current data largely support the view
based on previous behavioral studies that have found that children who stutter may have a
tendency to focus on less important sensory cues, in addition to being more sensitive to
interference from increased attentional demands (Bajaj, 2007; Bosshardt, 2006). The current
findings involving VAN may explain some of these observations, which could elucidate the
basis for transient disruption in speech production that are affected by aberrant attention
processes in children who stutter.
speak, the child attempts to generate motor movements to hit an auditory target (Hickok,
2012). This process relies heavily on auditory perception. This is evidenced by detrimental
effects of hearing difficulty on the intelligibility of speech in children (e.g. Tobey et al.,
1991) and deterioration in speech intelligibility following adult onset deafness (Waldstein,
1990). But auditory information is not the only sensory input; learning to speak also relies
on visual perception of articulatory movements. Weikum et al., (2007) reported that 4–6
month old infants can discriminate between languages (French versus English) using visual
cues only (i.e., silent videos showing articulations). Electrophysiological work shows that
visual cues not only enhance auditory cues (e.g., Crosse et al., 2015; see Peelle & Sommers,
2015 for a review), but that they actually precede them (Strelnikov et al., 2015). In line with
this reasoning, Venezia et al. (2016) propose that “exposure to visual speech during
acquisition of speech production establishes the neural circuitry linking visually-perceived
Author Manuscript
gestures to the speech motor system.” If visual information is used to facilitate speech
acquisition, aberrant function of VN could interfere with this process. We speculate that a
deficit in VN could render a child vulnerable to developing stuttering by introducing subtle
deficiencies in linking visually perceived speech movements and auditory stimuli, in turn
affecting his or her ability to produce fluent speech. However, since this network component
was associated with a risk for stuttering and not strongly with persistent stuttering, any such
deficit is most likely subtle and transient.
We report the first whole brain connectomic study of stuttering, based on the largest ever
longitudinally acquired rsfMRI dataset from children who do and do not stutter. The present
findings support the view that stuttering is a complex neurodevelopmental disorder with
multiple factors that dynamically interact to influence symptom onset, persistence, and
recovery. Our results are particularly interesting in light of theoretical models that attempt to
explain heterogeneity of stuttering and fluctuations in symptom severity over the course of a
day, weeks, or months. A number of models of stuttering—the conditioned disintegration
theory (Brutten & Shoemaker, 1967), the demands and capacities model (Starkweather &
Gottwald, 1990), the dual diathesis stress model (Conture & Walden, 2012; Walden et al.,
2012), the variable release threshold hypothesis (Brocklehurst et al., 2013), as well as other
multifactorial models of stuttering (e.g. Packman, 2012; Smith, 1999) — all emphasize the
importance of situational, emotional, attentional and linguistic factors that impinge upon a
Author Manuscript
vulnerable speech motor system that may ultimately lead to moments of disfluent speech.
The results from our study seem to corroborate these ideas and demonstrate that anomalous
connectivity among networks that support attention, motor, perception, and emotion not only
place individuals at risk for stuttering, but may influence whether a child persists or recovers
from stuttering.
To date, most studies examining the neural bases of stuttering have focused on cortical and
subcortical areas that have an established relationship in supporting speech production
processes. These studies have led to insights into “neural signatures of stuttering” (Brown et
al., 2004), stuttering trait (differences that are present in speakers who stutter regardless of
speaking condition), and state (differences between stuttered versus fluent speaking in
speakers who stutter) that could be linked to anomalous brain structure and function (Belyk
et al., 2015; Budde et al., 2014; Neef et al., 2015). A limitation in taking such localization
Author Manuscript
indicating this is a common risk factor regardless of eventual clinical outcome. There were
important findings on inter-network connectivity involving the SMN and other networks,
however, which provide new insights into neural mechanisms that differentiate those
children who go on to develop persistent stuttering, versus those that recover.
More specifically, the present results point to aberrant connectivity of major ICNs, including
those supporting internally directed mentation (DMN), attention and executive control
(VAN, DAN, FPN), sensory/motor processes (SMN), and vision (VN) in stuttering children.
Author Manuscript
Furthermore, the intra-network connectivity in DMN and its connections with attention and
executive control networks predicted persistent stuttering. These results suggest that while
stuttering children in general have abnormal connectivity in networks relevant to speech
motor control (e.g., SMN), persistent stuttering seem to be associated with connectivity
differences that primarily involve DMN and its connectivity with attention and executive
control networks. According to the default network interference model (Songua-Barke &
Castellanos, 2007), DMN intrudes on task positive networks and adds variability in
performance of externally directed tasks. Normalized connectivity involving DMN and
attention, executive control networks may mean that, relative to persistent children, these
functions are better differentiated from DMN in recovered children. In future studies, these
ideas will be explored in more detail in order to reveal large-scale brain connectivity
differences that explain heterogeneity and variability of stuttering symptoms, and to delve
Author Manuscript
deeper into the connectivity changes associated with recovery. Finally, these findings may
help guide development of mechanism based treatment approaches in the future.
components that differentiated between stuttering and fluent groups. However, we did not
interrogate the effects of other behavioral measures in modulating network connectivity
across development. Instead, the goal for this initial investigation was to identify early
markers (based on the initial scan of each child) that could predict later stuttering
persistence. In future studies, we are planning to acquire more than twice the number of
scans reported in this study, which will allow us to conduct detailed analyses of the
biological factors (sex, age) and other behavioral and treatment factors that could potentially
affect longitudinal trajectories of change.
The determination of whether a child was recovered was based on consideration of multiple
data points, including speech sample analysis as well as parent and clinician reports. For
example, to be considered “recovered”, children must have been exhibiting fluent speech for
6 or more months, in addition to having less than 3% SLD in their speech samples on the
Author Manuscript
day of the assessment (and at or lower than “very mild” on SSI composite rating). These
assessments took place every year for 3 years. This means that we had the opportunity to
reassess the children’s fluency and revisit our judgment about persistence or recovery at any
additional later time point. For the majority of the children, at least 2 additional time points
after the initial visit were available for us to determine persistence vs. recovery, and thus we
are confident about the assignment of children into persistent versus recovery groups.
However, there is a small chance that some children in the persistent group could still grow
out of stuttering, or that some children in the recovered group could relapse back into
Author Manuscript
stuttering. While the chances that a change in persistent or recovery status could occur 3–4
years post stuttering-onset (which is the case for the majority of our children) is slim, we
nevertheless plan to monitor these children in future years to confirm persistence versus
recovery.
It is also possible that there is individual variability within the group of stuttering children in
terms of connectivity among DMN and other ICNs, as was observed in this study. These
differences can be explored in conjunction with detailed consideration of behavioral data
across development, with the goal of finding objective network connectivity markers that
predict individual trajectories. Identification of such network markers may lead to better
diagnosis and the development of individualized treatments. We expect that such efforts
would significantly contribute to better understanding of the complex neural mechanisms
associated with childhood stuttering onset, persistence, and recovery.
Author Manuscript
Supplementary Material
Refer to Web version on PubMed Central for supplementary material.
Acknowledgments
The authors wish to thank all the children and parents who participated in this study. We also thank Kristin Hicks
for her assistance in participant recruitment, behavioral testing, and help with MRI data collection, Scarlett Doyle
for her assistance in MRI data acquisition, and Ashley Larva for her assistance in speech data analyses. This study
was supported by the National Institute on Deafness and other Communication Disorders (NIDCD)
(R01DC011277) to SC and the Matthew Smith Stuttering Research Fund to SC. The content is solely the
responsibility of the authors and does not necessarily represent the official views of the NIDCD or the National
Institutes of Health.
Author Manuscript
Abbreviations
ADHD Attention deficit hyperactivity disorder
VN visual network
References
Author Manuscript
Ackermann H, Riecker A. The contribution(s) of the insula to speech production: a review of the
clinical and functional imaging literature. Brain Structure and Function. 2010; 214(5–6):419–433.
DOI: 10.1007/s00429-010-0257-x [PubMed: 20512374]
Allen EA, Damaraju E, Plis SM, Erhardt EB, Eichele T, Calhoun VD. Tracking whole-brain
connectivity dynamics in the resting state. Cerebral cortex. 2012 bhs352.
Alm PA. Stuttering and the basal ganglia circuits: a critical review of possible relations. Journal of
communication disorders. 2004; 37(4):325–369. [PubMed: 15159193]
Alm PA, Karlsson R, Sundberg M, Axelson HW. Hemispheric lateralization of motor thresholds in
relation to stuttering. PloS one. 2013; 8(10):e76824. [PubMed: 24146930]
Ambrose NG, Yairi E. Normative disfluency data for early childhood stuttering. Journal of Speech,
Language, and Hearing Research. 1999; 42(4):895–909.
Anderson JD, Pellowski MW, Conture EG, Kelly EM. Temperamental characteristics of young
children who stutter. Journal of Speech, Language, and Hearing Research. 2003; 46(5):1221–1233.
Arndt J, Healey EC. Concomitant disorders in school-age children who stutter. Language, Speech, and
Author Manuscript
21095231]
Beal DS, Lerch JP, Cameron B, Henderson R, Gracco VL, De Nil LF. The trajectory of gray matter
development in Broca’s area is abnormal in people who stutter. Frontiers in human neuroscience.
2015; 9
Behzadi Y, Restom K, Liau J, Liu TT. A component based noise correction method (CompCor) for
BOLD and perfusion based fMRI. Neuroimage. 2007; 37(1):90–101. [PubMed: 17560126]
Belyk M, Kraft SJ, Brown S. Stuttering as a trait or state - an ALE meta-analysis of neuroimaging
studies. European Journal of Neuroscience. 2014; doi: 10.1111/ejn.12765
Bosshardt HG. Cognitive processing load as a determinant of stuttering: Summary of a research
programme. Clinical Linguistics & Phonetics. 2006; 20(5):371–385. [PubMed: 16728334]
Brocklehurst PH, Lickley RJ, Corley M. Revisiting bloodstein’s anticipatory struggle hypothesis from
a psycholinguistic perspective: A variable release threshold hypothesis of stuttering. Journal of
communication disorders. 2013; 46(3):217–237. [PubMed: 23631932]
Brown S, Ingham RJ, Ingham JC, Laird AR, Fox PT. Stuttered and fluent speech production: an ALE
meta-analysis of functional neuroimaging studies. Human Brain Mapping. 2015; 25(1):105–117.
Author Manuscript
DOI: 10.1002/hbm.20140
Broyd SJ, Demanuele C, Debener S, Helps SK, James CJ, Sonuga-Barke EJ. Default-mode brain
dysfunction in mental disorders: a systematic review. Neuroscience & biobehavioral reviews.
2009; 33(3):279–296. [PubMed: 18824195]
Brutten EJ, Shoemaker DJ. The modification of stuttering. 1967Buckner RL, Andrews-Hanna JR,
Schacter DL. The brain’s default network. Annals of the New York Academy of Sciences. 2008;
1124(1):1–38. [PubMed: 18400922]
Buckner RL, Andrews-Hanna JR, Schacter DL. The brain’s default network. Annals of the New York
Academy of Sciences. 2008; 1124(1):1–38. [PubMed: 18400922]
Author Manuscript
Buhr A, Zebrowski P. Sentence position and syntactic complexity of stuttering in early childhood: A
longitudinal study. Journal of fluency disorders. 2009; 34(3):155–172. [PubMed: 19948270]
Budde KS, Barron DS, Fox PT. Stuttering, induced fluency, and natural fluency: A hierarchical series
of activation likelihood estimation meta-analyses. Brain and Language. 2014; 139(C):99–107.
2014. DOI: 10.1016/j.bandl.2014.10.002 [PubMed: 25463820]
Busan P, D’Ausilio A, Borelli M, Monti F, Pelamatti G, Pizzolato G, Fadiga L. Motor excitability
evaluation in developmental stuttering: a transcranial magnetic stimulation study. Cortex. 2013;
49(3):781–792. [PubMed: 22225881]
Cai S, Tourville JA, Beal DS, Perkell JS, Guenther FH, Ghosh SS. Diffusion imaging of cerebral white
matter in persons who stutter: evidence for network-level anomalies. Frontiers in human
neuroscience. 2014; 8:54. [PubMed: 24611042]
Calhoun VD, Adali T, Giuliani NR, Pekar JJ, Kiehl KA, Pearlson GD. Method for multimodal analysis
of independent source differences in schizophrenia: combining gray matter structural and auditory
oddball functional data. Human brain mapping. 2006a; 27(1):47–62. [PubMed: 16108017]
Author Manuscript
Calhoun VD, Adalı T, Kiehl KA, Astur R, Pekar JJ, Pearlson GD. A method for multitask fMRI data
fusion applied to schizophrenia. Human brain mapping. 2006b; 27(7):598–610. [PubMed:
16342150]
Calhoun VD, Liu J, Adalı T. A review of group ICA for fMRI data and ICA for joint inference of
imaging, genetic, and ERP data. Neuroimage. 2009; 45(1):S163–S172. [PubMed: 19059344]
Calvert GA, Campbell R. Reading speech from still and moving faces: the neural substrates of visible
speech. Journal of Cognitive Neuroscience. 2003; 15(1):57–70. [PubMed: 12590843]
Cao M, Huang H, Peng Y, Dong Q, He Y. Toward Developmental Connectomics of the Human Brain.
Frontiers in neuroanatomy. 2016; 10
Chang SE, Chow H, Wieland EA, McAuley JD. Relation between functional connectivity and rhythm
discrimination in children who do and do not stutter. NeuroImage: Clinical. 2016; 12:442–450.
DOI: 10.1016/j.nicl.2016.08.021 [PubMed: 27622141]
Chang SE, Erickson KI, Ambrose NG, Hasegawa-Johnson MA, Ludlow CL. Brain anatomy
differences in childhood stuttering. Neuroimage. 2008; 39(3):1333–1344. [PubMed: 18023366]
Author Manuscript
Chang SE, Zhu DC, Choo AL, Angstadt M. White matter neuroanatomical differences in young
children who stutter. Brain. 2015; 138(3):694–711. [PubMed: 25619509]
Chang SE, Zhu DC. Neural network connectivity differences in children who stutter. Brain. 2013;
136(12):3709–3726. [PubMed: 24131593]
Choi D, Conture EG, Walden TA, Lambert WE, Tumanova V. Behavioral inhibition and childhood
stuttering. Journal of fluency disorders. 2013; 38(2):171–183. [PubMed: 23773669]
Choo AL, Chang SE, Zengin-Bolatkale H, Ambrose NG, Loucks TM. Corpus callosum morphology in
children who stutter. Journal of Communication Disorders. 2012; 45(4):279–289. [PubMed:
22516042]
Christoff K, Gordon AM, Smallwood J, Smith R, Schooler JW. Experience sampling during fmri
reveals default network and executive system contributions to mind wandering. Proceedings of the
National Academy of Sciences. 2009; 106(21):8719–8724.
Civier O, Bullock D, Max L, Guenther FH. Computational modeling of stuttering caused by
impairments in a basal ganglia thalamo-cortical circuit involved in syllable selection and initiation.
Brain and language. 2013; 126(3):263–278. [PubMed: 23872286]
Author Manuscript
Cole DM, Beckmann CF, Oei NY, Both S, van Gerven JM, Rombouts SA. Differential and distributed
effects of dopamine neuromodulations on resting-state network connectivity. Neuroimage. 2013;
78:59–67. [PubMed: 23603346]
Conture E, Walden T. Dual diathesis-stressor model of stuttering. Theoretical issues of fluency
disorders. 2012:94–127.
Craig A, Tran Y. Trait and social anxiety in adults with chronic stuttering: Conclusions following
meta-analysis. Journal of Fluency Disorders. 2014; 40:35–43. [PubMed: 24929465]
Crosse MJ, Butler JS, Lalor EC. Congruent visual speech enhances cortical entrainment to continuous
auditory speech in noise-free conditions. The Journal of Neuroscience. 2015; 35(42):14195–
Author Manuscript
Goldman R. Fristoe m. Goldman-Fristoe test of articulation, 2. 2000Gordon EM, Lee PS, Maisog JM,
Foss-Feig J, Billington ME, VanMeter J, Vaidya CJ. Strength of default mode resting-state
Author Manuscript
connectivity relates to white matter integrity in children. Developmental science. 2011; 14(4):738–
751. [PubMed: 21676094]
Greicius MD, Krasnow B, Reiss AL, Menon V. Functional connectivity in the resting brain: a network
analysis of the default mode hypothesis. Proceedings of the National Academy of Sciences. 2003;
100(1):253–258.
Greicius MD, Supekar K, Menon V, Dougherty RF. Resting-state functional connectivity reflects
structural connectivity in the default mode network. Cerebral cortex. 2009; 19(1):72–78. [PubMed:
18403396]
Hagmann P, Cammoun L, Gigandet X, Meuli R, Honey CJ, Wedeen VJ, Sporns O. Mapping the
structural core of human cerebral cortex. PLoS Biol. 2008; 6(7):e159. [PubMed: 18597554]
Hall DA, Fussell C, Summerfield AQ. Reading fluent speech from talking faces: typical brain networks
and individual differences. Journal of Cognitive Neuroscience. 2005; 17(6):939–953. [PubMed:
15969911]
Hickok G. Computational neuroanatomy of speech production. Nature Reviews Neuroscience. 2012;
Author Manuscript
Atypical central auditory speech-sound discrimination in children who stutter as indexed by the
mismatch negativity. Journal of fluency disorders. 2014; 41:1–11. [PubMed: 25066139]
Jones RM, Conture EG, Walden TA. Emotional reactivity and regulation associated with fluent and
stuttered utterances of preschool-age children who stutter. Journal of communication disorders.
2014; 48:38–51. [PubMed: 24630144]
Joos K, De Ridder D, Boey RA, Vanneste S. Functional connectivity changes in adults with
developmental stuttering: a preliminary study using quantitative electroencephalography. Frontiers
in human neuroscience. 2014; 8
Kaganovich N, Wray AH, Weber-Fox C. Non-linguistic auditory processing and working memory
update in pre-school children who stutter: an electrophysiological study. Developmental
neuropsychology. 2010; 35(6):712–736. [PubMed: 21038162]
Karrass J, Walden TA, Conture EG, Graham CG, Arnold HS, Hartfield KN, Schwenk KA. Relation of
emotional reactivity and regulation to childhood stuttering. Journal of communication disorders.
2006; 39(6):402–423. [PubMed: 16488427]
Kefalianos E, Onslow M, Block S, Menzies R, Reilly S. Early stuttering, temperament and anxiety:
Author Manuscript
contributions of the ventral and dorsal posterior cingulate cortex to cognitive control. The Journal
of Neuroscience. 2011; 31(9):3217–3224. [PubMed: 21368033]
Lepsien J, Nobre AC. Cognitive control of attention in the human brain: Insights from orienting
attention to mental representations. Brain research. 2006; 1105(1):20–31. [PubMed: 16729979]
Ludlow CL, Rosenberg J, Salazar A, Grafman J, Smutok M. Site of penetrating brain lesions causing
chronic acquired stuttering. Annals of Neurolology. 1987; 22(1):60–66.
Ludlow CL. Stuttering: dysfunction in a complex and dynamic system. Brain. 2000; 123(10):1983–
1984. [PubMed: 11004116]
Lu C, Ning N, Peng D, Ding G, Li K, Yang Y, Lin C. The role of large-scale neural interactions for
developmental stuttering. Neuroscience. 2009; 161(4):1008–1026. [PubMed: 19364522]
Lu C, Peng D, Chen C, Ning N, Ding G, Li K, Yang Y, Lin C. Altered effective connectivity and
anomalous anatomy in the basal ganglia-thalamocortical circuit of stuttering speakers. Cortex.
2010; 46(1):49–67. [PubMed: 19375076]
Lu C, Chen C, Peng D, You W, Zhang X, Ding G, Deng X, Yan Q, Howell P. Neural anomaly and
Author Manuscript
reorganization in speakers who stutter a short-term intervention study. Neurology. 2012; 79(7):
625–632. [PubMed: 22875083]
Lu C, Long Y, Zheng L, Shi G, Liu L, Ding G, Howell P. Relationship between speech production and
perception in people who stutter. Frontiers in human neuroscience. 2016; 10
Maguire GA, Riley GD, Franklin DL, Gottschalk LA. Risperidone for the treatment of stuttering.
Journal of clinical psychopharmacology. 2000; 20(4):479–482. [PubMed: 10917410]
Maguire GA, Riley GD, Franklin DL, Maguire ME, Nguyen CT, Brojeni PH. Olanzapine in the
treatment of developmental stuttering: a double-blind, placebocontrolled trial. Annals of Clinical
Psychiatry. 2004; 16(2):63–67. [PubMed: 15328899]
Maguire GA, Yeh CY, Ito BS. Overview of the diagnosis and treatment of stuttering. Journal of
Experimental & Clinical Medicine. 2012; 4(2):92–97.
Mason MF, Norton MI, Van Horn JD, Wegner DM, Grafton ST, Macrae CN. Wandering minds: the
default network and stimulus-independent thought. Science. 2007; 315(5810):393–395. [PubMed:
17234951]
Max L, Guenther FH, Gracco VL, Ghosh SS, Wallace ME. Unstable or insufficiently activated internal
Author Manuscript
Menon V. Large-scale brain networks and psychopathology: a unifying triple network model. Trends in
cognitive sciences. 2011; 15(10):483–506. [PubMed: 21908230]
Author Manuscript
[PubMed: 27498891]
Poole VN, Robinson ME, Singleton O, Degutis J, Milberg WP, McGlinchey RE, Esterman M. Intrinsic
functional connectivity predicts individual differences in distractibility. Neuropsychologia. 2016;
86:176–182. [PubMed: 27132070]
Power JD, Fair DA, Schlaggar BL, Petersen SE. The development of human functional brain networks.
Neuron. 2010; 67(5):735–748. [PubMed: 20826306]
Power JD, Barnes KA, Snyder AZ, Schlaggar BL, Petersen SE. Spurious but systematic correlations in
functional connectivity MRI networks arise from subject motion. Neuroimage. 2012; 59(3):
2142–2154. [PubMed: 22019881]
Preibisch C, Neumann K, Raab P, Euler HA, von Gudenberg AW, Lanfermann H, Giraud AL.
Evidence for compensation for stuttering by the right frontal operculum. Neuroimage. 2003;
20(2):1356–1364. [PubMed: 14568504]
Price CJ. A review and synthesis of the first 20 years of PET and fMRI studies of heard speech, spoken
language and reading. NeuroImage. 2012; 62(2):816–847. DOI: 10.1016/j.neuroimage.
2012.04.062 [PubMed: 22584224]
Author Manuscript
Raichle ME, MacLeod AM, Snyder AZ, Powers WJ, Gusnard DA, Shulman GL. A default mode of
brain function. Proceedings of the National Academy of Sciences. 2001; 98(2):676–682.
Riley, G. Stuttering severity instrument. 4th. Austin, TX: Pro-Ed; 2009. SSI-4.
Riley GE, Riley J. A revised component model for diagnosing and treating children who stutter.
Contemporary Issues in Communication Sciences and Disorders. 2000; 27:188–199.
Roebroeck A, Formisano E, Goebel R. Mapping directed influence over the brain using Granger
causality and fMRI. NeuroImage. 2005; 25(1):230–242. [PubMed: 15734358]
and theory of mind, and their relationship to the default mode network. Journal of cognitive
neuroscience. 2010; 22(6):1112–1123. [PubMed: 19580387]
Sridharan D, Levitin DJ, Menon V. A critical role for the right fronto-insular cortex in switching
between central-executive and default-mode networks. Proceedings of the National Academy of
Sciences. 2008; 105(34):12569–12574.
Sripada CS, Kessler D, Angstadt M. Lag in maturation of the brain’s intrinsic functional architecture in
attention-deficit/hyperactivity disorder. Proceedings of the National Academy of Sciences. 2014;
111(39):14259–14264.
Stager SV, Calis K, Grothe D, Bloch M, Berensen NM, Smith PJ, Braun A. Treatment with
medications affecting dopaminergic and serotonergic mechanisms: Effects on fluency and
anxiety in persons who stutter. Journal of Fluency Disorders. 2005; 30(4):319–335. [PubMed:
16246409]
Starkweather CW, Gottwald SR. The demands and capacities model II: Clinical applications. Journal
of Fluency Disorders. 1990; 15(3):143–157.
Strelnikov K, Foxton J, Marx M, Barone P. Brain prediction of auditory emphasis by facial expressions
Author Manuscript
Supekar K, Uddin LQ, Prater K, Amin H, Greicius MD, Menon V. Development of functional and
structural connectivity within the default mode network in young children. Neuroimage. 2010;
Author Manuscript
Walden TA, Frankel CB, Buhr AP, Johnson KN, Conture EG, Karrass JM. Dual diathesis-stressor
model of emotional and linguistic contributions to developmental stuttering. Journal of abnormal
child psychology. 2012; 40(4):633–644. [PubMed: 22016200]
Waldstein RS. Effects of postlingual deafness on speech production: implications for the role of
auditory feedback. The Journal of the Acoustical Society of America. 1990; 88(5):2099–2114.
[PubMed: 2269726]
Washington SD, Gordon EM, Brar J, Warburton S, Sawyer AT, Wolfe A, Gaillard WD. Dysmaturation
of the default mode network in autism. Human brain mapping. 2014; 35(4):1284–1296.
[PubMed: 23334984]
Weber-Fox C, Spruill JE, Spencer R, Smith A. Atypical neural functions underlying phonological
processing and silent rehearsal in children who stutter. Developmental Science. 2008; 11(2):321–
337. [PubMed: 18333985]
Wechsler, D. Wechsler Abbreviated Scale of Intelligence. Psychological Corporation; 1999.
Wechsler, WD. Weschler preschool and primary scale of intelligence–iii. San Antonio: Psychological
Corporation; 2002.
Author Manuscript
Xiao Y, Zhai H, Friederici AD, Jia F. The development of the intrinsic functional connectivity of
default network subsystems from age 3 to 5. Brain imaging and behavior. 2016; 10(1):50–59.
Author Manuscript
[PubMed: 25759285]
Xu Y, Tong Y, Liu S, Chow HM, AbdulSabur NY, Mattay GS, Braun AR. Denoising the speaking
brain: toward a robust technique for correcting artifact-contaminated fMRI data under severe
motion. Neuroimage. 2014; 103:33–47. [PubMed: 25225001]
Xuan Y, Meng C, Yang Y, Zhu C, Wang L, Yan Q, Lin C, Yu C. Resting-state brain activity in adult
males who stutter. PloS one. 2012; 7(1):e30570–e30570. [PubMed: 22276215]
Yairi E, Ambrose N. A Longitudinal Study of Stuttering in Children: A Preliminary Report. Journal of
Speech, Language, and Hearing Research. 1992; 35(4):755–760.
Yairi E, Ambrose NG. Early Childhood Stuttering: Persistency and Recovery Rates. Journal of Speech,
Language, and Hearing Research. 1999; 42(5):1097–1112.
Yairi E, Ambrose N. Epidemiology of stuttering: 21st century advances. Journal of Fluency Disorders.
2013; 38(2):66–87. DOI: 10.1016/j.jfludis.2012.11.002 [PubMed: 23773662]
Yang Y, Jia F, Siok WT, Tan LH. Altered functional connectivity in persistent developmental stuttering.
Scientific reports. 2016; 6
Author Manuscript
Yeo BT, Krienen FM, Sepulcre J, Sabuncu MR, Lashkari D, Hollinshead M, Fischl B. The
organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal
of neurophysiology. 2011; 106(3):1125–1165. [PubMed: 21653723]
Yerys BE, Jankowski KF, Shook D, Rosenberger LR, Barnes KA, Berl MM, Ritzl EK, VanMeter J,
Vaidya CJ, Gaillard WD. The fmri success rate of children and adolescents: typical development,
epilepsy, attention deficit/hyperactivity disorder, and autism spectrum disorders. Human brain
mapping. 2009; 30(10):3426–3435. [PubMed: 19384887]
Zarate JM. The neural control of singing. Frontiers in Human Neuroscience. 2013; 7(237):1–12. DOI:
10.3389/fnhum.2013.00237 [PubMed: 23355817]
Zengin-Bolatkale H, Conture EG, Walden TA. Sympathetic arousal of young children who stutter
during a stressful picture naming task. Journal of fluency disorders. 2015; 46:24–40. [PubMed:
26296616]
Zou Q, Ross TJ, Gu H, Geng X, Zuo XN, Hong LE, Gao JH, Stein EA, Zang YF, Yang Y. Intrinsic
resting-state activity predicts working memory brain activation and behavioral performance.
Author Manuscript
Highlights
Author Manuscript
Figure 1.
A–F. Resting state intrinsic connectivity networks (ICNs) examined in this study. A) Default
mode network (DMN), B) Frontoparietal network (FPN), C) Dorsal attention network
(DAN) D) Ventral attention network (VAN), E) Somatomotor network (SMN), F) Visual
network (VN). Modified from Lee MH, Hacker CD, Snyder AZ, Corbetta M, Zhang D, et al.
(2012) Clustering of Resting State Networks. PLoS ONE 7(7): e40370.
doi:10.1371/journal.pone.0040370.
Author Manuscript
Author Manuscript
Figure 2.
Schematic of ICA analysis stream used in this study. Resting state functional connectomes
based on 224 scans (from 84 unique participants) entered into a joint ICA, which parses the
connectomes into several cohesive components. The components reflect patterns of intra-
Author Manuscript
children who stutter. The density curves in the bottom sections of the figure are schematic
Author Manuscript
hypothetical results, not based on real data. See text for more details.
Author Manuscript
Author Manuscript
Author Manuscript
Figure 3.
ICA Component 5. This component is more expressed in children who stutter, and predicted
stuttering persistence. Red lines represent increased connectivity between regions, while
blue lines represent decreased connectivity. The width of the arcs is proportional to the
number of aberrant connections between regions. With the exception of PCC, all regions are
presented bilaterally, with the left ROI first followed by the right ROI (counter-clockwise
from the top of the figure). DMN: Default Mode Network; VAN: Ventral Attention Network:
DAN: Dorsal Attention Network; FPN: Frontoparietal Network; see Table 5 for region
abbreviations.
Author Manuscript
Figure 4.
ICA Component 7. This component is more expressed in children who stutter than those
who do not. VN: Visual Network; DAN: Dorsal Attention Network; DMN: Default Mode
Network. See methods and Figure 3 caption for full details.
Author Manuscript
Figure 5.
ICA Component 8. This component is more expressed in children who stutter. DAN: Dorsal
Attention Network; SMN: Somatomotor Network; VAN: Ventral Attention Network; DMN:
Default Mode Network. See methods and Figure 3 caption for full details.
Author Manuscript
Figure 6.
A summary of ICN relationships associated with stuttering and persistence. All lines
indicate those connections expressed to a greater extent in stuttering children; the thick lines
represent those connections further associated with stuttering persistence. Blue lines indicate
decreased connectivity found in stuttering children relative to controls; red lines, increased
connectivity. Black lines indicate anomalous connectivity in children who stutter relative to
Author Manuscript
controls, showing both increases and decreases among specific nodes of each network.
Those networks with blue border represent those networks with significant intra-network
connectivity decreases in children who stutter relative to controls.
Author Manuscript
Table 1
a
Scores significantly lower in stuttering than controls (p<0.05)
b
Scores significantly higher in persistent than recovered (p<0.05)
SD = standard deviation; SES = socioeconomic status; IQ = intelligent quotient; PPVT = Peabody Picture vocabulary Test; EVT = Expressive
Vocabulary Test; GFTA-2 = Goldman-Fristoe Test of Articulation; %SLD = percentage of stuttering-like disfluency; SSI-4 = Stuttering Severity
Instrument Edition 4
Author Manuscript
Author Manuscript
Table 2
%SLD at the initial visit 6.7 (4.4) 1.1–22.5 4.9 (3.0) 1.8–12
%SLD at the final visit 6.0 (4.8)b 1.8–17.9 1.6 (0.8) 0.7–3.2
Author Manuscript
SSI-4 at the initial visit 21.2 (5.9)b 13–37 16.4 (5.3) 8–26
SSI-4 at the final visit 19.9 (7.9)b 11–39 9.6 (2.2) 6–13
a
Scores significantly lower in persistent than controls (p<0.05)
b
Scores significantly higher in persistent than recovered (p<0.05)
c
Scores significantly higher in recovered than persistent (p<0.05)
d
Scores significantly lower in recovered than controls (p<0.05)
SD = standard deviation; SES = socioeconomic status; IQ = intelligent quotient; PPVT = Peabody Picture Vocabulary Test; EVT = Expressive
Vocabulary Test; GFTA-2 = Goldman-Fristoe Test of Articulation; SSI-4 = Stuttering Severity Instrument Edition 4; %SLD= percent stuttering-like
disfluencies
Author Manuscript
Author Manuscript
Table 3
Regression coefficients, statistics, and odds ratios for the logistic regression predicting stuttering diagnosis
Author Manuscript
from the 15 component expressions derived from the ICA analysis. Components 5, 7, 8, and 15 were
significantly related to stuttering diagnosis (ps<0.05).
Table 4
Regression coefficients, statistics, and odds ratios for the logistic regression predicting stuttering persistence
Author Manuscript
from the 3 retained significant components from the stuttering diagnosis regression. Component 5 was
significantly related to stuttering persistence in a consistent direction (i.e. increased expression of component 5
predicted both presence and persistence of stuttering, ps<0.05).
Table 5
List of intrinsic connectivity networks (ICN) and abbreviations for subregions used in figures, arranged by
Author Manuscript
ICN.
Table 6
Regression statistics for the logistic regressions predicting stuttering diagnosis and persistence in the original models (N=84/39) versus two subset
analyses without 1) ADHD comorbidity (N=77/34); and 2) any psychiatric comorbidity (N=75/32).
Chang et al.
Stuttering Diagnosis
5 2.193 0.028* 2.259 0.024* 2.208 0.027*
7 2.005 0.045* 2.311 0.021* 2.275 0.023*
8 2.370 0.018* 2.480 0.013* 2.595 0.0095*
15 2.544 0.011* 2.995 0.0027* 2.563 0.010*
Stuttering Persistence
5 2.110 0.035* 2.072 0.038* 1.879 0.060
7 1.792 0.073 1.861 0.063 1.732 0.083
8 −1.754 0.080 −1.137 0.255 −1.105 0.269