Neural network topology in ADHD; evidence for maturational delay and default-mode network alterations
Accepted Manuscript Neural network topology in ADHD; evidence for maturational delay and default-mode network alterations T.W.P. Janssen, A. Hillebran...
Accepted Manuscript Neural network topology in ADHD; evidence for maturational delay and default-mode network alterations T.W.P. Janssen, A. Hillebrand, A. Gouw, K. Geladé, R. Van Mourik, A. Maras, J. Oosterlaan PII: DOI: Reference:
Please cite this article as: Janssen, T.W.P., Hillebrand, A., Gouw, A., Geladé, K., Van Mourik, R., Maras, A., Oosterlaan, J., Neural network topology in ADHD; evidence for maturational delay and default-mode network alterations, Clinical Neurophysiology (2017), doi: https://doi.org/10.1016/j.clinph.2017.09.004
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.
T.W.P. Janssen - Neural network topology in ADHD
Neural network topology in ADHD; evidence for maturational delay and default-mode network alterations T.W.P. Janssena, A. Hillebrandb, A. Gouwb, K. Geladéc, R. Van Mourikd, A. Marase, J. Oosterlaanc a
Corresponding author:
Vrije Universiteit Amsterdam Van der Boechorststraat 1 1081 BT Amsterdam, The Netherlands Tel.: +31 20 598 8962 E-mail: [email protected] b
VU University Medical Center
De Boelelaan 1117 1081 HV Amsterdam, The Netherlands c
Vrije Universiteit Amsterdam
Van der Boechorststraat 1 1081 BT Amsterdam, The Netherlands d
Mental Health Care Organisation Noord-Holland-Noord,
Stationsplein 138, 1703 WC Heerhugowaard, The Netherlands e
Yulius Academy
Dennenhout 1 2994 GC Barendracht, The Netherlands
1
T.W.P. Janssen - Neural network topology in ADHD
2
Highlights
Functional brain networks were explored in ADHD using source-reconstructed EEG and graph analysis.
Evidence was found for maturational delay and underlying default-mode network alterations.
This study provides new temporal and spatial insights in neural network topology in ADHD.
Abstract Objective: Attention-deficit/hyperactivity disorder (ADHD) has been associated with widespread brain abnormalities in white and grey matter, affecting not only local, but global functional networks as well. In this study, we explored these functional networks using source-reconstructed electroencephalography in ADHD and typically developing (TD) children. We expected evidence for maturational delay, with underlying abnormalities in the default mode network. Methods: Electroencephalograms were recorded in ADHD (n=42) and TD (n=43) during rest, and functional connectivity (phase lag index) and graph (minimum spanning tree) parameters were derived. Dependent variables were global and local network metrics in theta, alpha and beta bands. Results: We found evidence for a more centralized functional network in ADHD compared to TD children, with decreased diameter in the alpha band (ηp2=.06) and increased leaf fraction (ηp2=.11 and .08) in the alpha and beta bands, with underlying abnormalities in hub regions of the brain, including default mode network. Conclusions: The finding of a more centralized network is in line with maturational delay models of ADHD and should be replicated in longitudinal designs. Significance: This study contributes to the literature by combining high temporal and spatial resolution to construct EEG network topology, and associates maturational-delay and default-mode interference hypotheses of ADHD.
1. Introduction Throughout childhood and adolescence, healthy brain development is characterized by a range of neurobiological changes, such as synaptic pruning and myelination of long-distance axons (Craik and Bialystok, 2006) that ultimately lead to a matured brain that enables fast signal transduction while maintaining relatively low energy costs (Boersma et al., 2011). The organization of normal adult brain networks is described as an intermediate structure between tree extremes: (1) a locally connected, highly ordered (regular) network, (2) a random network and (3) a scale-free network, which is characterized by highly connected brain areas, or ‘hubs’ (Stam, 2014). Most graph theoretical research to date has focused on intermediate topological structures between the first two extremes, regular and random networks (Bullmore and Sporns, 2012). These so called small-world networks are optimally efficient, having a delicate balance between dense local connections and a few long-distance connections, and seem sensitive in capturing underlying neurobiological changes during normal and deviant development. Recently, EEG graph analysis studies demonstrated a shift from more random toward more regular small-world configurations with increasing age (Boersma et al., 2013, 2011; Smit et al., 2016). A useful research approach would be to map developmental disorders that are associated with developmental delay, such as attention-deficit/hyperactivity disorder (ADHD) (Shaw et al., 2007), on this changing network configuration with age. A growing literature conceptualizes ADHD from a structural and functional network perspective (Konrad and Eickhoff, 2010). Structurally, noninvasive neuroimaging methods such as magnetic resonance imaging (MRI) and diffusion tensor imaging (DTI) have revealed widespread reductions in white and grey matter volume (Valera et al., 2007) and white matter (WM) integrity in children with ADHD (van Ewijk et al., 2012). Graph analysis of WM connectivity, based on DTI, has demonstrated a less optimal WM topological organization in children with ADHD (Cao et al., 2013); although children with ADHD had a small-world brain network, this network was more regular in topology and therefore less efficient compared to typically developing (TD) children. The abnormal wiring of white matter networks may provide a crucial structural substrate that underlies abnormal functional connectivity and network topology in ADHD (Cao et al., 2014). Functional networks alterations in ADHD have been found using resting-state fMRI (rsfMRI) (Cao et
T.W.P. Janssen - Neural network topology in ADHD
4
al., 2014; Wang et al., 2009a) and electroencephalography (EEG) (Ahmadlou et al., 2012; Liu et al., 2015), and have been interpreted as evidence for a shift towards more regular networks in ADHD in line with structural DTI-based findings (Cao et al., 2013). Cao and colleagues (2014) concluded that these findings support the developmental delay model of ADHD, considering that maturation of healthy brains follows a “local to distributed” principle. However, based on developmental EEG studies in healthy children, which show a shift from random to more regular networks during maturation, one would expect a more random network in ADHD to reflect maturational delay, rather than a more regular network (Boersma et al., 2013, 2011; Smit et al., 2016). As graph analysis is mostly applied to resting-state data, the question arises whether the default-mode network (DMN) is implicated as underlying source of topological network alterations in ADHD. The DMN is an extended network of interconnected brain areas that are associated with internally focused cognition, showing higher activity and connectivity during rest than during externally driven tasks (Buckner et al., 2008). Studies are inconsistent about the direction of abnormal connectivity in ADHD, either suggesting hyperconnectivity or hypoconnectivity of the DMN (Konrad and Eickhoff, 2010). Others argue that the DMN is undisturbed during rest, but fails to be attenuated during the transition from rest-to-task (Sonuga-Barke and Castellanos, 2007), interrupting task-related processing. Abnormal intrinsic oscillation patterns during rest have been found for task-positive neuronal network as well, such as the ventral attention network (VAN), which was furthermore hyperconnected to the DMN (Sripada et al., 2014). This hyperconnectivity may be a reflection of reduced anti-correlations between task-negative (DMN) and task-positive networks, in support of the “default network interference hypothesis”, or impaired regulation of the VAN over the DMN (Sripada et al., 2014). Although the current literature has demonstrated interesting new avenues of investigating brain connectivity and network topology in children with ADHD, several methodological issues limit the conclusions that can be drawn from these data. Firstly, most EEG-based methods use coherence to measure connectivity (Barry et al., 2011), yet coherence fails to capture intrinsic nonlinearities of brain activity, is sensitive to spurious correlations due to volume conduction/field spread and different choices for the reference electrode (van Diessen et al., 2015). Moreover, changes in coherence can be
T.W.P. Janssen - Neural network topology in ADHD
5
induced by both changes in amplitude and connectivity (Stam et al., 2007). The phase lag index (PLI) has been developed as a connectivity measure to address these issues (Stam et al., 2007). Secondly, most EEG connectivity and network studies have been performed in signal space (scalp electrodes), which makes interpretation of findings difficult due to the mixture of signals arising from spatially separated sources at a single electrode, as well as the spread of signals from a single source over multiple electrodes (van Diessen et al., 2015). Thirdly, traditional graph metrics that have been used in fMRI and EEG studies, such as clustering coefficient, shortest path length and degree distribution, are hampered by methodological issues that impede comparability between studies (Papo et al., 2016). A recent application in graph analysis of brain networks is the minimum spanning tree (MST), which may be used to achieve unbiased estimates of brain networks in ADHD (Tewarie et al., 2015). The current study was conducted to further explore neural network topology in children with ADHD, while utilizing several new developments in network analysis. We recorded dense array EEG (128 electrodes) in relatively large numbers of children with ADHD (n=42) and TD controls (n=43) and performed connectivity and graph analysis at the source level. Based on the maturational delay model of ADHD and literature on typical development (Boersma et al., 2013; Smit et al., 2016), we expected to find more centralized, star-like networks in ADHD compared to TD controls, which have been associated with random networks (Tewarie et al., 2015). Furthermore, we expected underlying alterations in hub regions of the brain, specifically in the DMN. 2. Methods 2.1 Participants Eighty-five children with ADHD and 72 TD controls met our inclusion criteria, which required an estimated full scale IQ > 80, measured with a short version of the Wechsler Intelligence Scale for Children (WISC-III; Wechsler, 1991), using the subtests Vocabulary, Arithmetic, Block Design and Picture Arrangement. Children were excluded if there was a known history of neurological conditions. Recruitment procedures have been described in a previous study: “The ADHD group was recruited through mental health outpatient facilities in the west of the Netherlands. All children obtained a clinical diagnosis of ADHD according to the DSM-IV (American Psychiatric Association
T.W.P. Janssen - Neural network topology in ADHD
6
1994) as established by a child psychiatrist. ADHD diagnosis was confirmed by parent and teacher ratings on the Disruptive Behavior Disorders Rating Scale (DBDRS; Pelham et al., 1992), which required at least one of the scores on the Inattention or Hyperactivity/Impulsivity scales to be in the clinical (>90th percentile) range for both informants. Seventy-six percent of children were naive for stimulant medication and the remaining children discontinued use of stimulants at least four weeks before testing. Children with a clinical DSM-IV diagnosis of autism spectrum disorder were excluded. The TD group was recruited through three primary schools and a sports club in the same recruitment area as the ADHD group. Control children were required to obtain normal scores on the DBDRS (<90th percentile) for both informants and to be free of any psychiatric or neurological disorder” (Janssen et al., 2016). Complete data were available for 85 children in the age range 7 to 14 years with 42 children in the ADHD group (33 boys, 9 girls) and 43 children in the TD group (32 boys, 11 girls), see Table 1. Participants were excluded from the final data analysis due to technical EEG issues (ADHD: n=10; TD: n=6), excessive eye movement, muscle or movement artefacts as detected with automatic artifact rejection (ADHD: n=25; TD: n=20) or no sufficient epoch quality (ADHD: n=8; TD: n=3), see paragraph on epoch selection. Data quality is especially paramount for connectivity-based analysis, hence our stringent selection criteria and resulting data loss. The remaining dataset (49% ADHD, 60% TD) was not different compared to the initial dataset on age, IQ and DBDRS scales.
2.2 Procedure The study was conducted according to the Declaration of Helsinki, and was approved by the local ethics committee. Parents and children aged 12 years or older signed informed-consent. The current study sample partly overlaps with a sample participating in a randomized controlled trial on the effects of neurofeedback, methylphenidate and physical exercise as treatments for ADHD (trial number: NCT01363544). The resting EEG recording was followed by the stop-signal task (30 minutes) and oddball task (20 minutes), which are described elsewhere (Janssen et al., 2016, 2015). 2.3 Behavioral assessment
T.W.P. Janssen - Neural network topology in ADHD
7
Parent and teacher reports on the Strengths and Weaknesses of ADHD symptoms and Normal behavior scale (SWAN; Arnett et al., 2013; Swanson et al., 2006) were used for correlational analyses with the primary outcome measures. The SWAN employs 18 items on a seven-point scale ranging from ‘far below average’ (3) to ‘far above average’ (-3), to allow for ratings of relative strengths as well as weaknesses on the two scales comprising the SWAN: Attention and Impulse Control. 2.4 Electrophysiological recordings EEG recording is similar to a previous study, which was described as follows: “Continuous EEG was recorded at 512Hz using the ActiveTwo Biosemi system and ActiView software (Biosemi, Amsterdam, The Netherlands) from 128 scalp electrodes according to the ABC labelling system, referenced to the active common mode and grounded to the passive driven right leg, which functions as a feedback loop to drive average potentials across electrodes to the amplifier zero. Electrooculogram (EOG) was obtained using two electrodes at the external canthi, and two electrodes at infra- and supra-orbital sides” (Janssen et al., 2016). EEG was consecutively recorded during eyesopen (EO, 3 minutes), eyes-closed (EC, 3 minutes) and task conditions (50 minutes). Only 3 minutes of EC data were used in this study, considering that this condition is more stable over sessions, easier to standardize in children and that EC is characterized by robust activity in the alpha band (van Diessen et al., 2015). Off-line analyses were performed with Brain Vision Analyzer 2 software (Brain Products, Gilching, Germany, Version 2.1). A Butterworth Zero Phase band-pass filter of 0.1-30 Hz at 48 dB/oct was applied, and scalp electrodes were re-referenced to the average of 128 electrodes. A zero phase filter was used to avoid phase distortions and the average reference was chosen to reduce confounding effects of the reference (van Diessen et al., 2015). Broken electrodes were interpolated with the spherical splines method (Perrin et al., 1989). Ocular artefacts were detected with the method of Gratton and Coles (1983) and marked in the data, which were used to inform manual epoch selection. Finally, the continuous EEG was segmented into epochs of 4096 time frames (tf; ~8 seconds of data with 512Hz). 2.5 Epoch selection
T.W.P. Janssen - Neural network topology in ADHD
8
Epoch selection was performed in two steps: automatic and manual. First, automatic artifact rejection was applied to segments based on the following criteria: maximum allowed voltage step of 50 µV/ms, maximal peak-to-peak amplitude difference of 200 µV, maximal amplitude of ±150 µV, and minimal low activity of 0.50 µV for 100 ms intervals. If any of the 128 electrodes within an epoch contained an artefact according to these criteria, the entire epoch was rejected. Remaining epochs after the first step were subsequently manually rated for quality: (1) very good quality; no ocular (EOG), electromyographic (EMG), movement or other artefacts; (2) good quality; minimal presence of artefacts; (3) medium quality; moderate presence of artefacts; (4) poor quality; clear and strong presence of artefacts. Participants were included in the final dataset according to the following criteria: 5 epochs of very good or good quality (rated 1 or 2) with a maximum of 1 epoch of medium quality (rated 3). Epochs were chosen based on the best quality available and as close to the start of the recording as possible to minimize effects of increasing variance in vigilance (van Diessen et al., 2015), although we choose not to use a cutoff time for including epochs. 2.6 LAURA source estimation Sources underlying each selected epoch were estimated using the LAURA (Local AutoRegressive Averages) distributed linear inverse solution method (Grave de Peralta Menendez et al., 2004, 2001; Michel et al., 2004). The analysis was performed on broad-band (0.1-30Hz) data using the Cartool software by Denis Brunet (brainmapping.unige.ch/cartool). Source-based analysis was considered advantageous compared to signal-based analysis for the following reasons: (1) scalp-based network analysis may result in erroneous inferences about the underlying network topology due to volume conduction/field spread (Antiqueira et al., 2010), and (2) source-based analysis allows to associate local network metrics (such as the ‘hubness’ of regions) with brain anatomy. In a previous study we have described the method in more detail: “LAURA is a source reconstruction method that incorporates biophysical laws to obtain the optimal solution that fulfills both the observed data and bio-electromagnetic constraints. In this approach, the relationship between brain activity at one point and its neighbors is expressed in terms of a local autoregressive estimator with coefficients depending upon a power of the distance from the point (Grave de Peralta Menendez et al., 2004). Cartool software uses the L-curve method to find the optimal regularization parameter
T.W.P. Janssen - Neural network topology in ADHD
9
for a given data file (Hansen, 1992). We used the Locally Spherical Model with Anatomical Constraints (LSMAC) as lead field model, which has been shown to perform as well as more computationally intensive models like the Boundary Element Model (BEM) (Birot et al., 2014). Inverse solutions were calculated for each participant and epoch separately on a realistic head model that included 5004 equally distributed nodes within the gray matter of the Montreal Neurological Institute (MNI) transformed NIHPD pediatric brain atlas based on 7.5-13.5 years old children (Fonov et al., 2011, 2009)” (Janssen et al., 2016). Finally, 78 regions-of interest (ROIs) out of 5004 nodes were selected based on centroids in the regions of the automated anatomical labeling (AAL) atlas (Gong et al., 2009) in order to reduce the dimensionality of the data and to allow comparability with connectivity/network studies using other techniques such as MEG and fMRI (Hillebrand et al., 2016; Tewarie et al., 2015). This resulted in five epochs, each containing 78 timeseries (one for each AAL ROI) of 4096 samples (~8 seconds of data), containing intensity at each node as index of activation, where the intensity was computed as the vector norm of the source strength in tree orthogonal directions. These epochs were separately analyzed with Brainwave software (version 0.9.152.4.1) to obtain spectral, functional connectivity and graph metrics. 2.7 Spectral, functional connectivity and graph analysis Spectral, functional connectivity and graph analyses of individual epochs were performed for each frequency band separately: theta (4-8Hz), alpha (8-13Hz), and beta (13-25Hz), and averaged over 5 epochs. We refrained from analysing delta and gamma frequency bands, to limit any potential effects of electrophysiological confounds, respectively movement and electromyographic activity, that may have remained in the data despite stringent selection criteria. Connectivity was calculated using the Phase Lag Index (PLI) (Stam et al., 2007). PLI measures the strength of statistical interdependencies between pairs of time series, and was designed to reduce the effects of volume conduction/field spread by ignoring zero-lag (instantaneous) connectivity between signals. Another advantage of PLI over other connectivity methods is that this method is not sensitive for disease specific spectral changes in amplitude that have been documented in ADHD (Snyder and Hall, 2006), which can confound connectivity estimates (van Diessen et al., 2015).
T.W.P. Janssen - Neural network topology in ADHD
10
PLI connectivity matrices were subsequently used to construct functional networks using graph theory. For this purpose, the 78 ROIs were used as nodes and the functional connections between nodes as edges. In this study, we reconstructed the minimum spanning tree (MST) graph, for which we used 1/PLI as input. The MST is an acyclic sub-network that connects all nodes of weighted, undirected connectivity matrices, using Kruskal’s algorithm (Kruskal, 1956). This results in a graph containing (mainly) the strongest connections of the original network that can be considered as backbone of the functional network and which is unique if the link weights are unique (Stam et al., 2014). The MST addresses several methodological issues of conventional graph analyses (van Wijk et al., 2010), such as confounding effects of alterations in connection strength and link density (Tewarie et al., 2015). From the individual MSTs both global and local network metrics can be derived. Global network metrics can be informative about the integration and segregation of the entire network, while local network metrics can specify the level of importance of individual nodes within the network. We limited global network analyses to MST parameters (1) diameter, (2) leaf fraction, and (3) tree hierarchy (Th). Diameter is defined as the longest shortest path between any two nodes in the MST, and leaf fraction is the fraction of nodes with a degree of one in the MST, with degree being the number of links for a given node (Tewarie et al., 2015), see Figure 1 for an illustration of various MST parameters. The lower limit of the number of leafs in a MST is two, and the upper bound (m) is the number of nodes (n) minus one (in our network 77). Diameter is inversely related to leaf number. A small diameter and high leaf number are characteristic for a star-like topology (centralized network), and large diameter and low leaf number are characteristic for a line-like topology (decentralized network). An optimal tree requires a small diameter, but without overloading central nodes within the network (hubs), which is quantified with Th (Boersma et al., 2013; Tewarie et al., 2015). As a local network metric we used betweenness centrality (BC), which is a ‘hub’ measure based on the fraction of shortest paths that run through a node. It has been shown that important hub regions in the brain have high BC (Bullmore and Sporns, 2012). We restricted BC analysis to frequency bands where the global network parameters differed between groups. Although not a network parameter, local relative power spectra for theta, alpha and beta were calculated as well and compared between groups,
T.W.P. Janssen - Neural network topology in ADHD
11
to validate the distributed source localization solutions, and to assess whether BC differences depend on power differences despite using PLI and the MST.
2.8 Statistical analysis Statistical analyses were performed with SPSS 23 (IBM, 2015). Significance was assumed if p<.05 (two-tailed). Demographic data were compared between groups with one-way ANOVA or χ2 test with Fisher exact correction. Twelve separate General Linear Model (GLM) ANOVAs were used to test for group differences between TD and ADHD in mean PLI, MST diameter, leaf fraction and Th for theta, alpha and beta frequency bands. For the main outcomes, mean difference and 95% confidence interval [95% CI] are reported. Effect sizes are reported as partial eta-squared (ηp2), with effects interpreted as small (.01), medium (.06) or large (.14). Within each frequency band, the false discovery rate (FDR) criterion (q=.05), using the Benjamini-Hochberg procedure, was applied to control for multiple comparisons. Pearson correlations were calculated between connectivity/network metrics and IQ and age. To reduce the number of statistical tests, correlations between connectivity/network measures and SWAN scales were limited to measures with significant group differences. BC and relative power were explored for group differences using 5000 F-test permutations to account for skewed data, for each of the 78 nodes and visualized with MATLAB (The Mathworks Inc., Natick, Massachusetts, USA). Significance was assumed if p<.01 (two-tailed).
3. Results 3.1 Group characteristics and data quality Table 1 summarizes the group characteristics. Groups did not differ on age or gender. As expected, IQ was lower in the ADHD group. Groups did not differ in data quality, with no differences in epoch quality (rated 1-4, ADHD: mean=1.31; TD: mean=1.38), F(1,83)=0.63, p=.430, or number of interpolated channels (ADHD: mean=4.17; TD: mean=3.51), F(1,83)=1.98, p=.163. 3.2 Global network metrics
T.W.P. Janssen - Neural network topology in ADHD
12
Table 2 and Figure 2 show the main results of the study. The ADHD group demonstrated decreased MST diameter in the alpha band, mean difference(ADHD-TD)=-.009, 95%CI=[-.017, -.001], with a medium effect size, ηp2=.06. Furthermore, the ADHD group showed increased MST leaf fraction in the alpha band, mean difference(ADHD-TD)=.014, 95%CI=[.005, .023], with a medium/large effect size, ηp2=.11, and an increased MST Th in the alpha band, mean difference(ADHD-TD)=.010, 95%CI=[.002, .017], with a medium effect size, ηp2=.08. An increase in MST leaf fraction was found in the beta band as well; mean difference(ADHD-TD)=.017, 95%CI=[.005, .029], with a medium effect size, ηp2=.08. A strong correlation was found between IQ and MST leaf fraction in the theta band, only for the TD group, reflecting lower mean leaf fraction with increasing IQ. Another correlation between IQ and PLI in the beta band changed to non-significant when removing one outlier, and was therefore not considered a genuine effect. For the behavioral measures, moderate and significant correlations were found between MST diameter in the alpha band and the SWAN Attention scale according to parents for the ADHD and TD groups. The signs of these correlations were opposite for the groups, with a negative association for ADHD, in line with the ANOVA results, reflecting decreasing diameter with increasing inattention problems. In contrast, TD children demonstrated a positive association, reflecting decreasing diameter with increasing attentional skills. Figure 2 denotes that regression lines of the TD and ADHD groups cross near average attentional skills on the x-axis. Exploratory hierarchical linear regression analysis was performed to test whether a quadratic function of the SWAN Attention scale could explain alpha diameter better than a linear function for all children. Although a linear function did not fit the data, F(1,83)=2.71, p=.104, R2=.03, adding a quadratic function significantly improved the model, F(1,83)=6.71, p=.002, R2=.14, with R2 change of .11, F(1,83)=10.40, p=.002. At last, MST leaf fraction in the beta band and SWAN Attention scale according to parents, correlated positively for the ADHD group.
3.3 Local network metrics
T.W.P. Janssen - Neural network topology in ADHD
13
Figures 3 and 4 show (1) the relative power for theta, alpha and beta bands, and (2) betweenness centrality for alpha and beta bands, respectively, for each of the 78 nodes, separately for TD and ADHD, and statistical parametric maps of group comparisons. Both groups showed similar distributions of theta, alpha and beta power, with maximal power at central medial brain regions for theta (cingulate gyrus, pre-SMA), posterior medial and lateral areas for alpha (cuneus, precuneus, lateral occipital cortex), and posterior medial areas for beta (cuneus, posterior cingulate). Children with ADHD showed higher theta power in the left fronto-orbital cortex, F(1,83)=7.199 to 10.929, p=.009 to .001, right inferior frontal gyrus pars triangularis, F(1,83)=7.770, p=.007, and right temporal pole, F(1,83)=6.990, p=.010. Furthermore, children with ADHD demonstrated higher alpha power in the right temporal pole, F(1,83)=6.990 to 9.643, p=.010 to .003. Betweenness centrality (BC) values showed again similar distributions for both groups. For alpha, a network with relatively high BC was found in the cuneus, precuneus, lateral occipital cortex and angular gyrus. For beta, approximately the same network was found as for alpha, extending to the posterior cingulate gyrus. Children with ADHD demonstrated decreased BC in the left cuneus in the alpha band compared to TD children, F(1,83)=9.350, p=.003, and in the left anterior cingulate/medial frontal orbital area in the beta band, F(1,83)=7.156, p=.009. 3.4 Exploratory analyses It has been suggested that the alpha1 (8-10Hz) and alpha2 (10-13Hz) band constitute components with different topographies and functional roles (Bazanova and Vernon, 2014; Petsche et al., 1997). Post-hoc analyses were therefore performed to further extent BC findings in the alpha band by analysing alpha1 and alpha2 separately. Significance was assumed if p<0.05 (two-tailed). Lower BC was found for children with ADHD in the right precuneus for both alpha1 and alpha2 bands, see Figure 5. However, several additional group effects were dependent on the specific alpha band. For alpha1, the ADHD group showed higher BC in the right precentral gyrus (primary motor cortex), and lower BC in the right inferior frontal gyrus pars triangularis and rolandic operculum, and bilateral
T.W.P. Janssen - Neural network topology in ADHD
14
cuneus. For alpha2, children with ADHD showed higher BC in right orbitofrontal, anterior cingulate and left superior frontal cortex, while lower BC was found in left orbital and middle frontal cortex.
4. Discussion The developing brain is increasingly conceptualized and studied from a network perspective, both during typical and pathological maturation, such as in ADHD. In the current cross-sectional study, we aimed to further elucidate functional network alterations in children with ADHD as measured with high-density EEG at the source level, while utilizing new methodological developments in network analysis. Globally, children with ADHD demonstrated lower diameter and higher leaf fraction and tree hierarchy in the alpha band, which indicates a shift toward a more centralized, star-like network topology compared to typically developing (TD) children. This kind of topology has been associated with more random networks in conventional graph analysis (Tewarie et al., 2015). Comparable results were obtained in the beta band, showing higher leaf fractions in children with ADHD. Locally, children with ADHD demonstrated higher theta power in the left orbitofrontal cortex, and the right inferior frontal gyrus pars triangularis and for both theta and alpha in the right temporal pole, while reduced betweenness centrality (BC) in ADHD was confined to the left cuneus in the alpha band and left medial orbital frontal cortex in the beta band. The main results are in line with our hypothesis of maturational delay in ADHD and studies on healthy development using comparable methodology (Boersma et al., 2013; Smit et al., 2016). However, we are aware that the results seem to deviate from other network studies in ADHD (Ahmadlou et al., 2012; Liu et al., 2015; Wang et al., 2009b) that indicated a shift towards a more regular functional network instead of a more random network in children with ADHD. Several factors may explain these disparate findings between the current study and these former studies. Firstly, both the applied connectivity (PLI versus SL/FSL) and graph (MST versus conventional) methods differ. Unlike synchronization likelihood (SL), PLI is relatively insensitive to volume conduction/field spread and amplitude differences (Stam et al., 2007), which is especially important in EEG studies. Moreover, MST allows for unbiased estimates of network topology, whereas group differences
T.W.P. Janssen - Neural network topology in ADHD
15
obtained with conventional network analyses can go in any direction, depending on the choices made during the analysis (Tijms et al., 2013; van Wijk et al., 2010). And although MST parameters correlate with conventional metrics, the nature of this correlation depends on the underlying, unknown, network model (Tewarie et al., 2015). Secondly, previous EEG studies were relatively underpowered (n=12), which decreases the positive predictive value (PPV), which is the probability that a ‘positive’ finding reflects a true effect (Button et al., 2013). Thirdly, the current analyses were performed at the EEG source instead of sensor level, which is more likely to provide a faithful reconstruction of the network topology (Antiqueira et al., 2010). Fourthly, EEG and rs-fcMRI are differently sensitive to respectively fast and slow time scales, and fMRI provides an indirect measure of neuronal activity, unlike EEG. Functional networks reconstructed on the basis of rs-fcMRI may therefore more closely reflect gross underlying structural networks (Honey et al., 2007), or even vasculature (Webb et al., 2013). Our findings of a more centralized, star-like network topology in children with ADHD compared to TD children may be explained as evidence for maturational delay. Boersma and colleagues (2013) prospectively followed 227 healthy children from age 5 to 7, using the same PLI/MST method as used in the current study (but applied at the sensor level). They found increasing diameter and decreasing leaf fraction with age in the alpha band, suggesting a shift from a more starlike network towards a more line-like network. Importantly, they arrived at the same conclusion with conventional graph analysis, which showed a shift from a more random towards a more regular topology (Boersma et al., 2011). When considering development of network topology as a continuum ranging from random to regular, based on our findings, children with ADHD may lack behind their typically developing peers, with decreased diameter, increased leaf fraction and tree hierarchy in the same alpha band. That is, children with ADHD show a more random network compared to TD children. This hypothesis remains to be tested in a longitudinal design to distinguish between developmental delay and developmental deviation, especially since no age effects were demonstrated in our samples. Additional evidence for the clinical relevance of MST network metrics was obtained in our analysis relating global network metrics to continuous measures of ADHD. These associations were in
T.W.P. Janssen - Neural network topology in ADHD
16
line with the shift towards a star-like topology in ADHD: worse attention skills were related to lower diameter in the alpha band and higher leaf fraction in the beta band. An unexpected finding was the reverse correlation in the TD group for the diameter in the alpha band. For all children, the relation was best described as an inverted u-shape, comprising low alpha diameters for children with either poor or strong attention skills and high alpha diameters for children with average attentional skills. It could be speculated that hyper- and hypo-focused attention are two extreme ends of performance, with intermediate attention functions as optimal trade-off between focused attention and attentional flexibility. How attention and network parameters are related remains an interesting question. It is rather striking that the results of this and others MST studies (Boersma et al., 2013) suggest that younger, ADHD, and less intelligent children, all demonstrate a more random network topology. Although some of these results are opposite to results found in conventional graph studies (possible reasons for these differences have been discussed above), the direction of these effects seem consistent with each other. In fact, maturation, ADHD symptoms and intelligence are conceptually and empirically related. ADHD is characterized by a delay of cortical maturation (Shaw et al., 2007), and children with ADHD have on average lower IQs (Frazier et al., 2004). Note, however, that not all constructs were interrelated in this study. The second aim of this study was to explore local network characteristics in the brain, for which we choose betweenness centrality (BC) as index for the relative importance of a node in a network (i.e. its role as a ‘hub’). High BC in alpha and beta bands was found primarily in default mode network (DMN) areas such as the precuneus, posterior cingulate, and angular gyrus, but also in the cuneus. These results are corroborated by combined EEG/fMRI studies, which demonstrated positive associations between DMN activity and EEG alpha and beta power (Mantini et al., 2007), and more specifically between DMN and upper alpha 2 EEG synchronization (Jann et al., 2009). Group differences were restricted to decreased BC in the left cuneus in the alpha band, and anterior cingulate/medial frontal orbital cortex in the beta band, in children with ADHD. Gong et al. (2009) identified the cuneus in a DTI study as one of several important hubs in the brain based on BC using the same 78 region parcellation. The cuneus and medial frontal areas are part of the visual and DMN systems respectively, which have been implicated in ADHD (Cortese et al., 2012). With exploratory
T.W.P. Janssen - Neural network topology in ADHD
17
analyses, we split the alpha band in a lower and upper band, and computed BC. Interestingly, group differences were again localized in areas of high BC as reported by Gong et al. (2009), including DMN regions that are strongly active in the alpha band (Jann et al., 2009; Mantini et al., 2007). Children with ADHD showed reduced BC in the right precuneus for both alpha1 and alpha2, which is a central node in the DMN (Buckner et al., 2008). This finding is in line with hypoconnectivity accounts of the DMN in ADHD (Castellanos et al., 2008). Hypoconnectivity may result in reduced capability for modulation of DMN activity that in turn leads to DMN interference with task-related networks (Sonuga-Barke and Castellanos, 2007). Our results, however, indicate a more complicated pattern of both reduced BC (middle, inferior and orbital frontal gyri, precuneus), and increased BC (superior frontal, precentral and anterior cingulate gyri) in ADHD. Some limitations of this study should be mentioned. Firstly, although we intended to address methodological concerns of previous connectivity studies by using PLI and MST at the EEG source level, it remains challenging to fully integrate our findings with the literature. Future studies with comparable methodology are needed. Secondly, unplanned exploratory analyses of alpha1 and alpha2 bands have to be interpreted with caution and need replication. Thirdly, evidence for the involvement of the DMN is indirect and should ideally be contrasted with an active task condition, or with modularity/independent components analyses (ICA) to demonstrate strong interconnectivity between these regions (Buckner et al., 2008). Fourthly, LAURA source imaging is an estimation of local brain activity and may contain localization errors, although dense scalp coverage (128 electrodes) and validated stability and reliability of LAURA (Pascual-Marqui et al., 2009) help to increase confidence in the current findings. Lastly, PLI is sensitive to epoch length; however, PLI in source space is stable at approximately 6 seconds epoch length, which makes the 8-second epoch lengths in the current study sufficient (Fraschini et al., 2016). Furthermore, MST is based on the relative ranking of the connectivity values, and would therefore not be affected by a bias introduced by epoch length. Strengths of the study include relatively large samples sizes, improved methodology, and stringent quality control and selection of EEG traces. In conclusion, findings of this study support a growing body of evidence for abnormal brain connectivity in children with ADHD. We found evidence for a more centralized, star-like network
T.W.P. Janssen - Neural network topology in ADHD
18
configuration in ADHD compared to TD children, which may be an indication of developmental delay. This hypothesis should be further tested longitudinally. In addition, previously identified hub regions of the brain, including DMN areas, were affected in children with ADHD. Together, these network alterations may lead to suboptimal information processing and behavioral symptoms of ADHD. This study contributes to the literature by combining high temporal and spatial resolution to construct EEG network topography, and associates maturational-delay and default-mode interference hypotheses of ADHD.
Acknowledgements We like to thank all participating children and families for their contribution, as well as all research interns for their valuable support. Furthermore, we would like to thank the participating centers of child and adolescent psychiatry: Yulius Academie, Groene Hart ziekenhuis, Lucertis, Alles Kits, GGZ Delfland, Maasstad ziekenhuis, RIAGG Schiedam, Kinderpraktijk Zoetermeer, Albert Schweitzer ziekenhuis, Groos Mentaal Beter Jong, ADHD behandelcentrum, GGZ inGeest and PuntP. This research was funded by the Netherlands Organization for Health Research and Development (ZonMw): 157 003 012. ZonMw funded the trial, but had no role in the data analysis, manuscript preparation or decision to publish.
Conflict of interest The authors have declared that they have no conflicts of interest in relation to this study.
T.W.P. Janssen - Neural network topology in ADHD
19
References Ahmadlou M, Adeli H, Adeli A. Graph Theoretical Analysis of Organization of Functional Brain Networks in ADHD. Clin EEG Neurosci 2012;43:5–13. doi:10.1177/1550059411428555. Antiqueira L, Rodrigues FA, van Wijk BCM, Costa L da F, Daffertshofer A. Estimating complex cortical networks via surface recordings-A critical note. Neuroimage 2010;53:439–49. doi:10.1016/j.neuroimage.2010.06.018. Arnett AB, Pennington BF, Friend A, Willcutt EG, Byrne B, Samuelsson S, et al. The SWAN Captures Variance at the Negative and Positive Ends of the ADHD Symptom Dimension. J Atten Disord 2013;17:152–62. doi:10.1177/1087054711427399. Barry RJ, Clarke AR, Hajos M, Dupuy FE, McCarthy R, Selikowitz M. EEG coherence and symptom profiles of children with Attention-Deficit/Hyperactivity Disorder. Clin Neurophysiol 2011;122:1327– 32. doi:10.1016/j.clinph.2011.01.007. Bazanova OM, Vernon D. Interpreting EEG alpha activity. Neurosci Biobehav Rev 2014;44:94–110. doi:10.1016/j.neubiorev.2013.05.007. Birot G, Spinelli L, Vulliémoz S, Mégevand P, Brunet D, Seeck M, et al. Head model and electrical source imaging: A study of 38 epileptic patients. NeuroImage Clin 2014;5:77–83. doi:10.1016/j.nicl.2014.06.005. Boersma M, Smit DJA, Boomsma DI, De Geus EJC, Delemarre-van de Waal HA, Stam CJ. Growing Trees in Child Brains: Graph Theoretical Analysis of Electroencephalography-Derived Minimum Spanning Tree in 5- and 7-Year-Old Children Reflects Brain Maturation. Brain Connect 2013;3:50– 60. doi:10.1089/brain.2012.0106. Boersma M, Smit DJA, De Bie HMA, Van Baal GCM, Boomsma DI, De Geus EJC, et al. Network analysis of resting state EEG in the developing young brain: Structure comes with maturation. Hum Brain Mapp 2011;32:413–25. doi:10.1002/hbm.21030. Buckner RL, Andrews-Hanna JR, Schacter DL. The brain’s default network: anatomy, function, and relevance to disease. Ann N Y Acad Sci 2008;1124:1–38. doi:10.1196/annals.1440.011.
T.W.P. Janssen - Neural network topology in ADHD
20
Bullmore E, Sporns O. The economy of brain network organization. Nat Rev Neurosci 2012;13:336– 49. doi:10.1038/nrn3214. Button KS, Ioannidis JP, Mokrysz C, Nosek BA, Flint J, Robinson ES, et al. Power failure: why small sample size undermines the reliability of neuroscience. Nat Rev Neurosci 2013;14:365–76. doi:10.1038/nrn3475. Cao M, Shu N, Cao Q, Wang Y, He Y. Imaging Functional and Structural Brain Connectomics in Attention-Deficit/Hyperactivity Disorder. Mol Neurobiol 2014;50:1111–23. doi:10.1007/s12035-0148685-x. Cao Q, Shu N, An L, Wang P, Sun L, Xia M-R, et al. Probabilistic Diffusion Tractography and Graph Theory Analysis Reveal Abnormal White Matter Structural Connectivity Networks in Drug-Naive Boys with Attention Deficit/Hyperactivity Disorder. J Neurosci 2013;33:10676–87. doi:10.1523/JNEUROSCI.4793-12.2013. Castellanos FX, Margulies DS, Kelly C, Uddin LQ, Ghaffari M, Kirsch A, et al. Cingulate-precuneus interactions: a new locus of dysfunction in adult attention-deficit/hyperactivity disorder. Biol Psychiatry 2008;63:332–7. doi:10.1016/j.biopsych.2007.06.025. Corp IBM. IBM SPSS Statistics for Windows 2015. Cortese S, Kelly C, Chabernaud C, Proal E, Di Martino A, Milham MP, et al. Toward systems neuroscience of ADHD: a meta-analysis of 55 fMRI studies. Am J Psychiatry 2012;169:1038–55. doi:10.1176/appi.ajp.2012.11101521. Craik FIM, Bialystok E. Cognition through the lifespan: Mechanisms of change. Trends Cogn Sci 2006;10:131–8. doi:10.1016/j.tics.2006.01.007. Fonov V, Evans A, McKinstry R, Almli C, Collins D. Unbiased nonlinear average age-appropriate brain templates from birth to adulthood. Neuroimage 2009;47:S102. doi:10.1016/S10538119(09)70884-5. Fonov V, Evans AC, Botteron K, Almli CR, McKinstry RC, Collins DL. Unbiased average ageappropriate atlases for pediatric studies. Neuroimage 2011;54:313–27.
T.W.P. Janssen - Neural network topology in ADHD
21
doi:10.1016/j.neuroimage.2010.07.033. Fraschini M, Demuru M, Crobe A, Marrosu F, Stam C, Hillebrand A, et al. The effect of epoch length on estimated EEG functional connectivity and brain network organisation. J Neural Eng 2016;13:36015. doi:10.1088/1741-2560/13/3/036015. Frazier TW, Demaree H a, Youngstrom E a. Meta-analysis of intellectual and neuropsychological test performance in attention-deficit/hyperactivity disorder. Neuropsychology 2004;18:543–55. doi:10.1037/0894-4105.18.3.543. Gong G, He Y, Concha L, Lebel C, Gross DW, Evans AC, et al. Mapping Anatomical Connectivity Patterns of Human Cerebral Cortex Using In Vivo Diffusion Tensor Imaging Tractography. Cereb Cortex 2009;19:524–36. doi:10.1093/cercor/bhn102. Gratton G, Coles M, Donchin E. A new method for off-line removal of ocular artifact. Electroencephalogr Clin Neurophysiol 1983;55:468–84. Grave de Peralta Menendez R, Gonzalez Andino S, Lantz G, Michel CM, Landis T. Noninvasive localization of electromagnetic epileptic activity. I. Method descriptions and simulations. Brain Topogr 2001;14:131–7. Grave de Peralta Menendez R, Murray MM, Michel CM, Martuzzi R, Gonzalez Andino SL. Electrical neuroimaging based on biophysical constraints. Neuroimage 2004;21:527–39. doi:10.1016/j.neuroimage.2003.09.051. Hansen PC. Analysis of Discrete Ill-Posed Problems by Means of the L-Curve. SIAM Rev 1992;34:561–80. doi:10.1137/1034115. Hillebrand A, Tewarie P, van Dellen E, Yu M, Carbo EWS, Douw L, et al. Direction of information flow in large-scale resting-state networks is frequency-dependent. Proc Natl Acad Sci 2016;113:3867– 72. doi:10.1073/pnas.1515657113. Honey CJ, Kotter R, Breakspear M, Sporns O. Network structure of cerebral cortex shapes functional connectivity on multiple time scales. Pnas 2007;104:10240–5. doi:10.1073/pnas.0701519104. Jann K, Dierks T, Boesch C, Kottlow M, Strik W, Koenig T. BOLD correlates of EEG alpha phase-
T.W.P. Janssen - Neural network topology in ADHD
22
locking and the fMRI default mode network. Neuroimage 2009;45:903–16. doi:10.1016/j.neuroimage.2009.01.001. Janssen TWP, Geladé K, van Mourik R, Maras A, Oosterlaan J. An ERP source imaging study of the oddball task in children with Attention Deficit/Hyperactivity Disorder. Clin Neurophysiol 2016;127:1351–7. doi:10.1016/j.clinph.2015.10.051. Janssen TWP, Heslenfeld DJ, van Mourik R, Gelade K, Maras A, Oosterlaan J. Alterations in the Ventral Attention Network During the Stop-Signal Task in Children With ADHD: An Event-Related Potential Source Imaging Study. J Atten Disord 2015:Epub ahead of print; DOI: 10.1177/1087054715580847. doi:10.1177/1087054715580847. Konrad K, Eickhoff SB. Is the ADHD brain wired differently? A review on structural and functional connectivity in attention deficit hyperactivity disorder. Hum Brain Mapp 2010;31:904–16. doi:10.1002/hbm.21058. Kruskal JB. On the Shortest Spanning Subtree of a Graph and the Traveling Salesman Problem. Proc Am Math Soc 1956;7:48–50. Liu T, Chen Y, Lin P, Wang J. Small-World Brain Functional Networks in Children With AttentionDeficit/Hyperactivity Disorder Revealed by EEG Synchrony. Clin EEG Neurosci 2015;46:183–91. doi:10.1177/1550059414523959. Mantini D, Perrucci MG, Del Gratta C, Romani GL, Corbetta M. Electrophysiological signatures of resting state networks in the human brain. Proc Natl Acad Sci 2007;104:13170–5. doi:10.1073/pnas.0700668104. Michel CM, Murray MM, Lantz G, Gonzalez S, Spinelli L, Grave de Peralta Menendez R. EEG source imaging. Clin Neurophysiol 2004;115:2195–222. doi:10.1016/j.clinph.2004.06.001. Papo D, Zanin M, Martínez JH, Buldú JM. Beware of the Small-World Neuroscientist! Front Hum Neurosci 2016;10:1–4. doi:10.3389/fnhum.2016.00096. Pascual-Marqui RD, Sekihara K, Brandeis D, Michel CM. Imaging the electric neuronal generators of EEG/MEG. Electr. Neuroimaging, Cambridge, UK: Cambridge University Press; 2009, p. 49–77.
T.W.P. Janssen - Neural network topology in ADHD
23
Pelham WE, Gnagy EM, Greenslade K, Milich R. Teacher ratings of DSM-III-R symptoms for the disruptive behaviour disorder. J Am Acad Child Adolesc Psychiatry 1992;31:210–8. Perrin F, Pernier J, Bertrand O, Echallier JF. Spherical splines for scalp potential and current density mapping. Electroencephalogr Clin Neurophysiol 1989;72:184–7. Petsche H, Kaplan S, von Stein A, Filz O. The possible meaning of the upper and lower alpha frequency ranges for cognitive and creative tasks. Int J Psychophysiol 1997;26:77–97. doi:10.1016/S0167-8760(97)00757-5. Shaw P, Eckstrand K, Sharp W, Blumenthal J, Lerch JP, Greenstein D, et al. Attentiondeficit/hyperactivity disorder is characterized by a delay in cortical maturation. Proc Natl Acad Sci 2007;104:19649–54. doi:10.1073/pnas.0707741104. Smit DJA, de Geus EJC, Boersma M, Boomsma DI, Stam CJ. Life-Span Development of Brain Network Integration Assessed with Phase Lag Index Connectivity and Minimum Spanning Tree Graphs. Brain Connect 2016;6:312–25. doi:10.1089/brain.2015.0359. Snyder SM, Hall JR. A meta-analysis of quantitative EEG power associated with attention-deficit hyperactivity disorder. J Clin Neurophysiol 2006;23:440–55. doi:10.1097/01.wnp.0000221363.12503.78. Sonuga-Barke EJS, Castellanos FX. Spontaneous attentional fluctuations in impaired states and pathological conditions: a neurobiological hypothesis. Neurosci Biobehav Rev 2007;31:977–86. doi:10.1016/j.neubiorev.2007.02.005. Sripada C, Kessler D, Fang Y, Welsh RC, Prem Kumar K, Angstadt M. Disrupted network architecture of the resting brain in attention-deficit/hyperactivity disorder. Hum Brain Mapp 2014;35:4693–705. doi:10.1002/hbm.22504. Stam CJ. Modern network science of neurological disorders. Nat Rev Neurosci 2014;15:683–95. doi:10.1038/nrn3801. Stam CJ, Nolte G, Daffertshofer A. Phase lag index: Assessment of functional connectivity from multi channel EEG and MEG with diminished bias from common sources. Hum Brain Mapp 2007;28:1178–
T.W.P. Janssen - Neural network topology in ADHD
24
93. doi:10.1002/hbm.20346. Stam CJ, Tewarie P, Van Dellen E, van Straaten ECW, Hillebrand A, Van Mieghem P. The trees and the forest: Characterization of complex brain networks with minimum spanning trees. Int J Psychophysiol 2014;92:129–38. doi:10.1016/j.ijpsycho.2014.04.001. Swanson JM, Schuck S, Mann M, Carlson C. Categorical and dimensional definitions and evaluations of symptoms of ADHD: The SNAP and SWAN Rating Scales. Univ California, Irvine 2006:Available at: http://www.adhd.net. Tewarie P, van Dellen E, Hillebrand A, Stam CJ. The minimum spanning tree: An unbiased method for brain network analysis. Neuroimage 2015;104:177–88. doi:10.1016/j.neuroimage.2014.10.015. Tijms BM, Wink AM, de Haan W, van der Flier WM, Stam CJ, Scheltens P, et al. Alzheimer’s disease: connecting findings from graph theoretical studies of brain networks. Neurobiol Aging 2013;34:2023–36. doi:10.1016/j.neurobiolaging.2013.02.020. Valera EM, Faraone S V., Murray KE, Seidman LJ. Meta-Analysis of Structural Imaging Findings in Attention-Deficit/Hyperactivity Disorder. Biol Psychiatry 2007;61:1361–9. doi:10.1016/j.biopsych.2006.06.011. van Diessen E, Numan T, van Dellen E, van der Kooi AW, Boersma M, Hofman D, et al. Opportunities and methodological challenges in EEG and MEG resting state functional brain network research. Clin Neurophysiol 2015;126:1468–81. doi:10.1016/j.clinph.2014.11.018. van Ewijk H, Heslenfeld DJ, Zwiers MP, Buitelaar JK, Oosterlaan J. Diffusion tensor imaging in attention deficit/hyperactivity disorder: a systematic review and meta-analysis. Neurosci Biobehav Rev 2012;36:1093–106. doi:10.1016/j.neubiorev.2012.01.003. van Wijk BCM, Stam CJ, Daffertshofer A. Comparing Brain Networks of Different Size and Connectivity Density Using Graph Theory. PLoS One 2010;5:e13701. doi:10.1371/journal.pone.0013701. Wang L, Zhu C, He Y, Zang Y, Cao Q, Zhang H, et al. Altered small-world brain functional networks in children with attention-deficit/hyperactivity disorder. Hum Brain Mapp 2009a;30:638–49.
T.W.P. Janssen - Neural network topology in ADHD
25
doi:10.1002/hbm.20530. Wang L, Zhu C, He Y, Zang Y, Cao Q, Zhang H, et al. Altered small-world brain functional networks in children with attention-deficit/hyperactivity disorder. Hum Brain Mapp 2009b;30:638–49. doi:10.1002/hbm.20530. Webb JT, Ferguson MA, Nielsen JA, Anderson JS. BOLD Granger causality reflects vascular anatomy. PLoS One 2013;8:e84279. doi:10.1371/journal.pone.0084279. Wechsler D. Wechsler intelligence scale for children - third edition. Psychological Corporation, San Antonio, TX. 1991.
T.W.P. Janssen - Neural network topology in ADHD
26
Legends
Figure 1. Minimum spanning tree (MST) characteristics. Note. Schematic illustration of different trees, ranging between two extremes, (A) a line-like configuration, and (C) a star-like configuration. The middle tree (B) is an intermediate between A and C, displaying characteristics of both extremes. Nodes are represented by circles, and edges are represented by connecting lines. Dark circles are leaf nodes and have a degree of 1. Node 3 has the highest degree in tree B, since it is connected to 3 other nodes. Node 4 has the highest Betweenness Centrality (BC); the highest number of shortest paths between any pair of nodes run through node 4. Diameter is the longest distance between any two nodes. Nodes 1 and 8 are most distant with a diameter of 6. Tree B is an adaptation based on Boersma et al. (2013).
Figure 2. Continuous and dichotomous relations between alpha MST indices and ADHD. Note. Left figure depicts correlations between mean alpha diameter and the Attention scale of the SWAN questionnaire as reported by parents for TD and ADHD separately. The Attention scale ranges from above average attention skills (-3), to average attention skills (0) and below average attention skills (3). Please note different correlation signs for TD (positive) and ADHD (negative) that cross around average attentional skills (0 at x-axis). A quadratic function best describes the relation between alpha diameter and attentional skills (black line), when considering attention as a continuous construct. Right figures show means and 95%CI for both MST indices in the alpha band for TD and ADHD.
Figure 3. Spectral power for theta, alpha and beta bands in source space for TD and ADHD. Note. Relative power (%) in theta (4-8Hz), alpha (8-13Hz) and beta (13-25Hz) frequency bands in source space for 78 nodes. F-values are color-coded and represent group differences, with lower bound significance at F(1,83)=6.950, p<.010
T.W.P. Janssen - Neural network topology in ADHD
27
Figure 4. Betweenness Centrality (BC) for alpha and beta bands in source space for TD and ADHD. Note. Betweenness Centrality (BC) in alpha (8-13Hz) and beta (13-25Hz) frequency bands in source space for 78 nodes. F-values are color-coded and represent group differences, with lower bound significance at F(1,83)=6.950, p<.010
Figure 5. Post-hoc tests for differences between TD and ADHD for BC in alpha1 and alpha2 bands. Note. Betweenness Centrality (BC) in alpha1 (8-10Hz) and alpha2 (10-13Hz) frequency bands in source space for 78 nodes. F-values are color-coded and represent group differences, with lower bound significance at F(1,83)=3.956, p<.050
T.W.P. Janssen - Neural network topology in ADHD
28
Tables
Table 1. Group characteristics and task performance. ADHD
