CC BY-NC-ND
0Accepted Manuscript
PII: DOI:
Reference:
Graph Analysis of EEG Resting State Functional Networks in Dyslexic Readers
G. Fraga González, M.J.W. Van der Molen, G. Zaric, M. Bonte, J. Tijms, L. Blomert, C.J. Stam, M.W. Van der Molen
S1388-2457(16)30453-9 http://dx.doi.Org/10.1016/j.clinph.2016.06.023 CLINPH 2007888
To appear in:
Clinical Neurophysiology
Accepted Date:
8 June 2016
Please cite this article as: Fraga González, G., Van der Molen, M.J.W., Zaric, G., Bonte, M., Tijms, J., Blomert, L., Stam, C.J., Van der Molen, M.W., Graph Analysis of EEG Resting State Functional Networks in Dyslexic Readers, Clinical Neurophysiology (2016), doi: http://dx.doi.org/10.1016/j.clinph.2016.06.023
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.
1 Graph Analysis of EEG Resting State Functional Networks in
2 Dyslexic Readers
3 Fraga González, G. -1 , Van der Molen, M.J.W.Z'J, Zaric, G. Bonte, M. Tijms, J.
der Molen, M.J.W.2,3, Zaric, G. 4, Bonte, A T 4 4 Blomert, L. 4,t, Stam, C.J. 6, Van der Molen, M.W. 1,7
6 1 Department of Psychology, University of Amsterdam, The Netherlands
7 2 Leiden University, Institute of Psychology, The Netherlands
rlands rlands
8 Leiden Institute for Brain and Cognition, Leiden University, The Netherlands
9 4 Department of Cognitive Neuroscience, Faculty of Psychology and Neuroscience,
10 University of Maastricht, The Netherlands
11 5 IWAL Institute, Amsterdam, The Netherlands
12 6 Department of Clinical Neuropsychology and MEG Center, Neuroscience Campus Amsterdam, VU
13 University Medical Center, Amsterdam, The Netherlands
14 7 Amsterdam Brain and Cognition, University of Amsterdam, The Netherlands
20 * Correspondence:
21 Gorka Fraga Gonzalez (Ph.D.)
22 Department of Psychology ,University of Amsterdam, Nieuwe Achtergracht 129B, 1018 WS
23 Amsterdam, The Netherlands
24 E-mail: G.FragaGonzalez@uva.nl
29 1 Leo Blomert passed away on November 25, 2012.
u: u.rrai
19 f Lec
Highlights
Organization of brain networks in dyslexies and typically-reading controls. Minimum spanning tree (MST) graphs were derived from connectivity matrices. Graph metrics in the theta-band showed less integrated network configuration in dyslexics.
40 Objective
Abstract
41 Neuroimaging research suggested a mixed pattern of functional connectivity abnormalities in
42 developmental dyslexia. We examined differences in the topological properties of functional
43 networks between 29 dyslexics and 15 typically reading controls in 3rd grade using graph
44 analysis. Graph metrics characterize brain networks in terms of integration and segregation.
45 Method
46 We used EEG resting-state data and calculated weighted connectivity matrices for multiple
47 frequency bands using the phase lag index (PLI). From the connectivity matrices we derived
48 minimum spanning tree (MST) graphs representing the sub-networks with maximum connectivity. Statistical analyses were performed on graph-derived metrics as well as on the
50 averaged PLI connectivity values.
51 Results
52 We found group differences in the theta band for two graph metrics suggesting reduced
53 network integration and communication between network nodes in dyslexies compared to
54 controls.
55 Conclusion
56 Collectively, our findings point to a less efficient network configuration in dyslexics relative
57 to the more proficient configuration in the control group.
58 Significance
59 Graph metrics relate to the intrinsic organization of functional brain networks. These metrics
60 provide additional insights on the cognitive deficits underlying dyslexia and, thus, may
61 advance our knowledge on reading development. Our findings add to the growing body
62 literature suggesting compromised networks rather than specific dysfunctional brain regions
63 in dyslexia.
66 (EEG), ^ ^ gKph ^
67 minimum spanning tree, dyslexia.
70 Introduction
71 Reading involves integrated functioning of complex brain networks. Distinct brain
72 systems, mostly in the left hemisphere, have been proposed to specialize during reading
73 acquisition (see a review in Schlaggar and McCandliss, 2007). Studies in developmental
74 dyslexia revealed various disturbances of the brain networks implicated in reading. Studies
75 using diffusion tensor imaging (DTI) to examine white matter properties of the main
76 pathways that constitute the anatomical basis of the network reported reduced connectivity in
77 dyslexia (for a review and meta-analysis Vandermosten et al., 2012). Similarly, a score of
78 functional magnetic resonance imaging (fMRI) studies in adults reported reduced
79 connectivity of the reading network (e.g., Pugh et al., 2000; Quaglino et al., 2008; Schurz et
80 al., 2014; Shaywitz et al., 2003; Stanberry et al., 2006; Van der Mark et al., 2011; but see
81 Richards and Berninger, 2008) and other connectivity disturbances (Finn et al., 2014; Wolf et
82 al., 2010). A recent MRI study examining the topological organization in Chinese dyslexic
83 children revealed a less integrated network organization relative to typically reading controls,
84 characterized by increased local processing and less long-range communication (Liu et al.,
85 2015).
86 The neural network studies highlight the interactive nature of brain systems implicated in
87 reading and underscore the relevance of connectivity to the study of dyslexia. Although
88 specific brain regions have been successfully linked to literacy acquisition, the understanding
89 of a highly complex cognitive function such as reading may also require a more integrative
90 and
holistic view of brain function, which can be conceptualized as a complex network
91 (Bullmore and Sporns, 2009). In relation to this, previous neuroimaging research showed that
92 examining the dynamics of spontaneous (task independent) activity in the brain provide us
93 with meaningful information about how different brain areas communicate (van den Heuvel
94 and Hulshoff Pol, 2010) and the underlying architecture of functional brain networks
95 (Gusnard and Raichle, 2001).
96 The goal of the current study was to examine global functional network connectivity and
97 organization in developmental dyslexia using the electroencephalogram (EEG) in resting-
98 state data. Previous fMRI studies on resting-state data revealed relations between resting-
99 state functional connectivity across the reading network with reading abilities in children and
100 adults (Koyama et al., 2011, 2013; Schurz et al., 2014; Zhang et al., 2014). Other studies
101 linked the strength of resting-state connectivity between the visual word recognition areas
102 and the dorsal attention network to age and reading skills (Vogel et al., 2012; Vogel et al.,
103 2014). The latter studies attest to the utility of resting-state data to characterize the functional
104 reading network (Hampson et al., 2006; Koyama et al., 2010).
105 In the current study we used graph analysis, which allows for modeling the organization of
106 resting-state whole-brain functional connectivity networks during development (Stam, 2014).
107 A 'graph' refers to an abstract representation of a network, consisting of a set of nodes
108 (vertices) and connections between them (edges). Various graph measures allow for
109 characterizing graph topologies in terms of the efficiency of information transfer and an
110 optimal balance between 'segregation' and 'integration' (see reviews in Bullmore and
111 Sporns, 2009; Bullmore and Sporns, 2012). Thus, a 'small-world' network topology,
112 characterized by a high clustering (related to high local connectedness and robustness) and a
113 short path length (related to high global efficiency) has been proposed as a plausible
114 configuration of highly efficient brain networks (Bassett and Bullmore, 2006). This topology
115 combines features from ordered or regular networks (high clustering) and random networks
116 (short path length).
117 A recent development in graph theory refers to minimum spanning tree (MST) analysis
118 (Stam et al., 2014). A tree is a loop-less sub-graph derived from a weighted connectivity
119 matrix, with a fixed number of nodes and edges; in the MST, the presence of a link is not
120 defined by a given threshold in the connectivity weights (see Methods). The advantage of
121 MST analysis over conventional graph procedures is that it minimizes bias when performing
122 direct comparisons between groups and experimental conditions (Tewarie et al., 2015). There
123 are two extreme tree topologies; path- and star-like configurations. Path-like confi gurations
124 consist of nodes that are all linked to two other nodes with the exception of the nodes at
125 either end of the path. Nodes with only one link in a tree are referred to as 'leaf' nodes (or
126 leaves) and the number of those nodes in a tree is the leaf number. Thus a path has a leaf
127 number of two. In contrast, star-like configurations consist of a central node connected to all
128 other nodes with only one link. Thus, a star consisting of N nodes has a leaf number of N-1.
129 Many different tree topologies are in between the two extreme configurations and they can be
130 characterized using a variety of metrics (review in Van Mieghem, 2014). We will apply the
131 tree measures that have been applied previously in EEG studies (see Methods section below).
132 The MST analysis has been successfully applied to EEG data from different populations.
133 A relatively early study indicated that patients with left vs. right temporal epilepsy could be
134 reliably discriminated in terms of large scale functional networks emerging just prior to the
135 onset of seizures (Lee et al., 2006). More recently, Fraschini et al., 2014 examined the effects
136 of vagal nerve stimulation in patients with pharmaco-resistant epilepsy. MST analysis yielded
137 a clear differentiation between responders vs. non-responders. Vagal nerve stimulation
138 shifted the network towards a more star-like network architecture in responders but not in
139 non-responders. Van Diessen et al., 2014 examined the effect of sleep deprivation on EEG
140 networks in children diagnosed with focal epilepsy. MST analysis revealed a shift to a more
141 path-like topology after sleep deprivation in children with focal epilepsy whereas a shift
142 towards a more star-like configuration was observed in controls. Vourkas et al., 2014
143 performed a MST analysis on the EEG recorded in children with mathematical difficulties
144 and typical controls during the performance of tasks with increasing difficulty. Although
145 group differences were absent in this study the MST parameters suggested a more centralized
146 and integrated network layout in the alpha bands of the EEG with increasing task demands.
147 Most relevant to the present study, Boersma et al., (2013) applied MST analysis to resting-
148 state EEG data of a large sample of 5- and 7-years old children. Developmental change was
149 observed for the EEG alpha band. More specifically, the MST analysis yielded increases in
150 diameter and eccentricity with advancing age while leaf number, degree and hierarchy
151 decreased. This pattern of results was interpreted to suggest a more integrated network
152 configuration in the 7- compared to the 5-years olds.
153 Collectively, the MST studies suggest this approach may provide a sensitive tool to assess
154 condition or group differences in network configuration. Previously, graph analysis of
155 magnetoencephalogaphic (MEG) data in dyslexic children and controls showed task-
156 dependent dysfunctional long- and short-range functional connectivity in the dyslexic
157 children (Vourkas et al., 2011). Another graph analysis from the same group of MEG data
158 obtained during rest revealed less organized network configuration in dyslexic children
159 (Dimitriadis et al., 2013). The current study will extend these findings by focusing on resting
160 state EEG data and by performing a MST analysis on these data. The use of resting state data
161 should indicate functional network differences between the groups that are not related to task-
162 related strategies and that are indicative of the underlying architecture of oscillatory EEG
163 activity. MST analysis goes beyond more conventional network analysis as (i) it allows an
164 unbiased network representation; (ii) it provides a comparison between groups/conditions
165 that is normalized; and (iii) it integrates features of small-worldness (clustering/ path length)
166 and scale-freeness (hubs) (e.g., Tewarie et al., 2015).
167 Methods
168 Participants
169 Twenty-nine third-grade dyslexic children (Mean age = 8.46; SD= 0.40) were recruited from
170 a nation-wide center for dyslexia in the Netherlands.1 All dyslexic children had a percentile
171 score of 10 or lower on a standard reading test and they participated in the EEG recordings
172 before starting their treatment program at the center. A group of 15 third-grade, control
173 children (8.75 ± 0.31 years old) was recruited from several primary schools attended by
174 children with the same socio-demographical background as the dyslexic group (see Table 1
175 for group characteristics). They had no history of reading difficulties and had a percentile
176 score of 25 or higher on standard reading tests (see below). All participants were native
177 Dutch speakers, received two and a half years of formal reading instruction in primary
178 education. Children with below average IQ (IQ < 85 on a non-verbal IQ-test), uncorrected
179 sight problems, hearing loss, diagnosis of ADHD or other neurological or cognitive
180 impairments were excluded. The study was approved by the Ethical Review Board of the
181 University and all parents or caretakers signed informed consent before the children
182 participated.
183 Behavio ral measurements
1 The current participants are part of a larger sample of 62 children taking part in the EEG recordings. From the original data set, resting-state data was not available for 3 participants due to complications during recording. Moreover, data from 6 participants were excluded due to excessive artifacts. In the remaining data (N= 53), the inspection of individual peak frequencies in the average spectra indicated that for the majority of participants the peak frequency fell within the low alpha (8-10 Hz) and high alpha (10-13 Hz) range (see Spectral Power section). We discarded data from children with a peak frequency equal or lower than 8 Hz as this might bias subsequent analysis in the lower frequency bands. A total of 9 subjects were excluded; 5 dyslexics (N=29) and 4 controls (N = 15). Demographic characteristics and reading scores of the complete sample are included in Supplementary Appendix A.
184 A series of tests was used to assess the reading skills of the participants (Fraga Gonzalez
185 et al., 2015). The children took the tests at their school.
186 Word reading skills were measured using a Dutch version of the One-minute test (Een-
187 Minuut-Test, EMT; Van den Bos et al., 1999), a time-limited test consisting of a list of 116
188 unrelated words of increasing difficulty. The number of correctly read words within 1 minute
189 serves as reading fluency score. Text reading fluency was assessed also using a test
190 consisting of a coherent text of increasing difficulty. The children were asked to read the
191 story out loud within 1 minute (Schoolvaardigheidstoets Technisch Lezen; de Vos, 2007). In
192 addition, the 3DM battery of tests (test reliability information available in Dyslexia
193 Differential Diagnosis; 3DM, Blomert and Vaessen, 2009) was individually administered.
194 The scores of the following 3DM subtests were used. Word Reading task: contains visually
195 presented high-frequency words, low-frequency words and pseudowords. Accuracy (%
196 correct) and fluency (correct words in 1 minute) were measured. Rapid automatized naming
197 (RAN): blocks of letters or numbers are presented and items have to be read as fast and
198 accurately as possible. Fluency is the time in seconds needed to name a screen of 15 items.
199 Letter-speech sound (LSS) association tasks: consist of identification and discrimination
200 tasks. In the identification task an aurally presented speech sound has to be matched to one
201 out of four visually presented letters. In the discrimination task the child has to judge whether
202 the speech sound and letter on the screen are congruent or incongruent. Computerized
203 Spelling: words are aurally presented and visually displayed on screen with missing letters.
204 Participants have to select the missing letter out of four alternatives. For the last two subtests,
205 accuracy (% correct) as well as response time (sec/item) is measured.
206 Finally, the RAVEN Coloured Progressive Matrices was used to obtain an estimate of
207 fluid IQ (RAVEN CPM; Raven et al., 1998) and the Child Behavior Checklist (CBCL) was
208 completed by the parents to exclude any additional behavioral problems (Achenbach et al.,
209 2008).
210 Procedure and equipment
211 EEG recordings were taken within a period of around 4 months and took place in a video-
212 controlled, dimly lit and air conditioned laboratory room. The participant and a lab assistant
213 were together at all times in the room while the experimenter controlling the recording was in
214 an adjacent room. The 2 minutes eyes-closed resting-state baseline was recorded at the
215 beginning of a longer experimental session (around 2 hours long, including visual and
216 audiovisual tasks). Children were instructed to keep their eyes closed and, when ready, to
217 make a button press to initiate the eyes-closed resting state EEG recording. Participants were
218 monitored at all times to ensure they complied to the instructions during the baseline
219 recording and that children did not show behavioral indications of drowsiness or sleep onset
220 during the recording.
221 EEG recording and signal processing
222 The EEG data were collected using a 64 channels Biosemi ActiveTwo system (Biosemi,
223 Amsterdam, Netherlands). EEG was recorded DC (low-pass: 5th order sync digital filter)
224 with a 1024 Hz sample rate. The Biosemi system uses two additional electrodes (Common
225 Mode Sense [CMS] and Driven Right Leg [DRL]) creating a feedback loop to replace the
226 conventional ground electrode (see www.biosemi.com/faq/cms&drl.htm for details). The
227 CMS electrode served as online reference. The 64 electrodes were distributed on the scalp
228 according to the 10-20 International system and applied using an elastic electrode cap
229 (Electro-cap International Inc.). Six external Flat-Type Active electrodes were used; four
230 electrodes for recording the vertical and horizontal electro-oculogram (EOG) and two
231 electrodes were placed at mastoids for off-line reference.
232 Continuous EEG data were imported in EEGLAB v.11.0.0.0b (Delorme and Makeig,
233 2004), an open source toolbox for Matlab (Mathworks, Inc.), using the averaged mastoids as
234 initial off-line reference. A two minutes long epoch was selected, time-locked to the button
235 press indicating the start of the eyes-closed resting-state recording.
236 The 2 min EEG epoch was imported in Brain Vision Analyzer (Version 2.0.1.5528, ©
237 Brain Products) for further preprocessing. After importing, spline interpolation was applied to
238 channels with excessive artifacts. In the control group, interpolation was applied to data from
239 10 subjects (a maximum of 5 electrodes in one subject); in the dyslexic group interpolation
240 was applied to data from 8 subjects (a maximum of 5 electrodes in one subject). Data were
241 segmented in 30 epochs of 4 s (4096 sample points per epoch). The epochs were visually
242 inspected for eye blinks or muscle artifacts. For each subject 10 artifact-free epochs were
243 selected and exported to ASCII files.
244 The artifact-free epochs of 4 s were imported in Brainwave v0.9.117 (developed by C.S.;
245 freely available at http://home.kpn.nl/stam7883/brainwave.html) where data were re-
246 referenced to the average of all scalp channels before performing spectral power analysis,
247 functional connectivity and MST metrics.
248 For the analysis of connectivity strength (measured with phase lag index; see Functional
249 Connectivity section), besides mean connectivity, the following sub-averages were
250 calculated: frontal (including the electrodes Fp1, Fp2, AF3, AF4, AF7, AF8, F1, F2, F3, F4,
251 F5, F6, F7 and F8); central (including the electrodes FC1, FC2, FC3, FC4, FC5, FC6, C1, C2,
252 C3, C4, C5, C6, CP1, CP2, CP3, CP4, CP5 and CP6); temporal (including the electrodes
253 FT7, FT8, T7, T8, TP7 and TP8) and parietal-occipital (including the electrodes O1, O2,
254 PO3, PO4, PO7, PO8, P1, P2, P3, P4, P5, P6, P7, P8, P9 and P10). The mean connectivity
255 between the electrodes included in each sub-average was calculated. These sub-averages
256 were chosen to examine strength of both short-range and long-range connectivity across
257 broad cortical regions that previous studies have found relevant to reading and dyslexia. Note
258 that the graph measures, which are the main focus of the present analysis, were derived from
259 the complete connectivity matrix including the 64 scalp electrodes.
260 Spectral power
261 Spectral power was calculated for all EEG channels using Fast Fourier Transformation
262 (FFT) in Brainwave, with a frequency resolution of 1 / 4 s = 0.25 Hz. The relative power
263 values were calculated for the following frequency bands: delta (0.5-4 Hz), theta (4-8 Hz),
264 alpha (8-13 Hz), beta (13-30 Hz) and gamma (30-48 Hz). The broad alpha range was used
265 instead of the lower alpha (8-10 Hz) and upper alpha (10-13 Hz) as some participants showed
266 an average peak frequency within the upper alpha range. Power values were averaged over
267 epochs.
268 Functional connectivity
269 The Phase Lag Index (PLI) was used to calculate functional connectivity between all 64
270 electrodes for each frequency band and in each segment, separately. The PLI measures phase
271 synchronization based on the asymmetry of the distribution of instantaneous phase
272 differences between two signals, which is determined using the analytical signal based on the
273 Hilbert transformation (Stam et al., 2007). The PLI is less sensitive to common sources since
274 the zero-lag synchronization is removed from the analysis (Porz et al., 2014). Furthermore,
275 the PLI quantifies the relative phase distribution's asymmetry; that is, that the likelihood that
276 the phase difference Ap will be in the interval - n < Ap < 0 is different from the likelihood
277 that it will be in the interval 0 < Ap < n. This implies the presence of a consistent, nonzero
278 phase difference ('lag') between two time series. The distribution is expected to be
279 symmetric if there is no coupling or if the median phase difference is equal to or centers
280 around a value of 0 mod n. The PLI is obtained from time series of phase differences Ap (tk),
281 k=1...N by means of:
282 PLI = l<sign[sin(Ap(tk))]>l
283 Here sign is the signum function. The PLI ranges between 0 and 1. A PLI of zero
284 indicates either no coupling or coupling with a phase difference centered around 0 (mod n).
285 A PLI of 1 indicates perfect phase locking at a value of Ap different from 0 (mod n). The
286 stronger this nonzero phase locking is, the larger PLI will be.
287 Minimum Spanning Tree
288 The Minimum Spanning Tree (MST) sub-graph was calculated for each PLI matrix
289 derived per segment. A schematic of the analytic steps is shown in Fig. 1. The MST is a
290 unique sub-graph based on a weighted matrix that connects all nodes of the network but does
291 not contain circles or loops. The MST always contains m = N-1 links, where N is the number
292 of nodes. The MST was constructed by applying Kruskal's algorithm (Kruskal, 1956). This
293 algorithm orders the distance of all links in an ascending order followed by the construction
294 of the MST with the link of the shortest distance, and then adding the following shortest
295 dis
tance link until all nodes are connected in a loop-less sub-graph. If adding a new link
296 results in the formation of a cycle, this link is skipped. In the current case, we use a maximum
297 spanning tree, which is equivalent to an MST based upon 1-PLI, which represents the sub-
298 network with maximum connectivity.
299 ................................................................
300 Insert Figure 1 about here
301 ................................................................
302 MST metrics provides information about the topological properties of the tree
303 following tree measures were used in this study: Degree, leaf number, betweenness centrality
304 (BC), eccentricity, diameter, hierarchy (Th), and degree correlation (R). The measures are
305 summarized in Table 2 and examples of tree topologies with increasing leaf number are
306 presented in Fig. 2 (a detailed description in Stam et al., 2014). The degree of a node is its
307 number of connections (edges), and the leaf fraction (L) represents the number of nodes on
308 the tree with degree = 1. The leaf number has a lower bound of 2 and upper bound of N -1.
309 The leaf number presents an upper bound to the diameter of the MST, which is the largest
310 distance between any two nodes of the tree. The upper limit of the diameter is d = m- L + 2,
311 which implies that the largest possible diameter will decrease with the increasing leaf
312 number. Eccentricity of a node is defined as the longest distance between that node and any
313 other node and is low if this node is central in the tree. The BC of a node u is the number of
314 shortest paths between any pair of nodes i and j that are running through u, divided by the
315 total number of paths between i and j. The BC value ranges between 0 and 1 since it is a
316 fraction. The BC relates to the importance of a node within the network. The nodes with the
317 highest BC have the highest load. For instance, in a star-like tree, the central node has a BC
318 of 1 and it could be easily overloaded, while the leaf nodes have a BC of 0. Degree,
319 eccentricity and BC are different measures for relative nodal importance and may indicate the
320 critical nodes in a tree.
322 Insert Figure 2 about here
323 .................................................
324 For a tree topology to result in optimal network performance, it should conform to two
325 criteria. Firstly, efficient communication would require a small diameter. Secondly, the tre(
326 topology would require preventing overload of hub nodes by setting a maximal BC max for
327 any tree node. The balance between these two criteria is reflected by the tree hierarchy (Th)
328 measure (Boersma et al., 2013), which is defined as:
2 ^B C^ax
or is mu
329 To assure Th ranges between 0 and 1, the denominator is multiplied by 2. If L = 2 (line-
330 like topology), and m approaches infinity, then Th approaches 0. If L = m (star-like
331 topography), then Th approaches 0.5. For leaf numbers in between these extreme values, Th
332 has higher values.
333 Finally, the degree correlation is an index of whether the degree of a node is correlated
334 with the degree of its neighboring vertices to which it is connected. A positive degree
335 correlation indicates that the graph is assortative; if the degree correlation is negative the
336 graph is called disassortative. The degree correlations can be quantified by computing the
337 Pearson correlation coefficient of the degrees of pairs of nodes connected by an edge.
338 Interestingly, most social networks tend to be assortative, while most technological and
339 biological networks tend to be disassortative (Newman, 2003).
340 Statistical analysis
341 One-way ANOVAs were used for group comparisons in behavioral measures, relative
342 power, PLI averages and MST measures. As indicated above, PLI and MST were calculated
343 per segment and then averaged for each participant. Prior to analysis, the PLI and MST
344 measures were transformed to their natural logarithm, y=ln (x), to obtain normal
345 distributions. For the behavioral analyses, standardized scores were used instead of raw
346 scores, in order to assess the child's position within the distribution of a normative sample.
347 Due to reduced variance, no reliable norm scores were available for the accuracy measures of
348 the three subtasks of the 3DM word reading; thus raw scores were used for these measures.
349 Additionally, for the MST measures, Bonferroni correction for multiple comparisons was
350 applied to p values for each frequency band. Finally, to examine the relation between tree-
351 derived measures and reading, regression analysis was performed in dyslexics and controls
352 separately for the MST measures in which we found group differences and the main reading
353 scores.
354 The same set of analyses was performed on the data of a sub-sample of 15 randomly
355 selected dyslexics to evaluate whether different sample sizes dyslexic children (n=29) and
356 controls (n=15) had any effect in the group differences. The main pattern of results reported
357 below did not change and it is presented in Supplementary Appendix B.
is preser /
suits of the
358 Results
359 Behavior
360 The results of the ANOVAs for reading accuracy and speed measures are shown in Table
361 1. The tat le shows a deficit in dyslexics that is mainly manifested by substantial differences
362 in the reading fluency measures. The dyslexic group attained reasonably high levels of
363 accuracy, although significantly lower than those of the control group. With regard to the
364 letter-speech sound measures, only the fluency scores were sensitive to group differences.
365 Spectral power and functional connectivity
366 The power spectra averaged across all electrodes for each group are shown in Fig. 3.
367 Controls and dyslexics both showed prominent peak frequencies in the alpha band, which did
368 not differ between groups. The ANOVAs performed on the relative power values in each
369 frequency band revealed no significant differences (in the total average or regional sub-
370 averages) between groups. For each frequency band outliers and extreme values in relative
371 power were detected and excluded for the subsequent analyses of connectivity and graph
372 measures. Outliers and extreme values were defined based on 1.5 inter-quartile range steps.
373 Accordingly, for the theta band 2 subjects from the dyslexic group were excluded (N = 27).
374 For the alpha band 1 subject from the dyslexic group was excluded (N= 28). No outliers or
375 extreme values were detected in the delta, beta or gamma band.
376 ................................................................
377 Insert Figure 3 about here
378 ................................................................
379 The PLI total values and sub-averages were calculated for each frequency band. The
380 ANOVAs yielded no significant differences in functional connectivity (total network or sub-
381 networks) between groups (all p's >.05). The total PLI values for each frequency band are
382 presented in Tables 3 and 4.
383 MST analysis
384 MST analysis yielded significant between group effects in the theta band (see Table 3 and
385 Fig. 4). Leaf fraction, reflecting the integration of information within the network, was
386 significantly lower in dyslexics relative to typical readers, F (1, 40) = 10.24, p = .003, n =
387 0.20. The group effect on diameter, representing the efficiency of communication between
388 the nodes, was significant also, F (1, 40) = 4.27, p = .045, n2 = 0.10, indicating higher
389 diameter in dyslexics relative to controls. The group effect on eccentricity, relating to node
390 centrality, just fell short of significance, F (1, 40) = 3.47, p = .070, n2 = 0.08, suggesting a
391 trend for higher eccentricity in dyslexics compared to controls. These group differences are
392 displayed in Fig. 5. Collectively these results indicate a less integrated network organization
393 in dyslexic children compared to controls.
394 ...................................................
395 Insert Figure 4 and 5 about here
........
eccentri
397 For the alpha band, the group effects on diameter and eccentricity just failed to reach
398 significance, p = .080 and p = .098, respectively, suggesting trends for higher diameter and
399 eccentricity in dyslexics relative to controls2. Finally, for the gamma band, the ANOVA
400 revealed a somewhat higher hierarchy in controls relative to the dyslexic children but this
401 effect just failed to reach significance, F (1, 40) = 3.89, p = .055, n2 = 0.09. Group effects in
402 all other measures and frequency bands were not significant, ps > .124. Moreover, there were
403 no significant correlations between MST measures and reading performance.
405 The present study examined the topological characteristics of brain networks in dyslexics
406 and controls by applying MST analysis to eyes-closed resting state EEG. The present results
407 suggest that compared to controls, dyslexics may present differences in the way spontaneous
408 oscillato ry activity is o rganized . Our results showed a clear dissociation between PLI
2 In a control analysis including peak frequency as a covariate the main results in the theta band for leaf number remained significant (F = 9.94, p = .003, n2 = 0.20), while for the metric of diameter the group difference just fell short of significance (F = 4.07, p = .051, n2 = 0.09). Moreover, we performed additional analysis in the separate alpha 1 (8-10 Hz) and alpha 2 (10-13 Hz) frequency bands, similarly to other studies (e.g. Tewarie et al., 2014, Van Diessen et al., 2014). The group comparison in the alpha1 band revealed a significant effect in leaf number (F = 5.57, p = .023, n2 = 0.12), indicating a lower leaf number in dyslexics relative to control readers. No significant group effects were found in the other metrics. In the alpha2 band, we did not find significant group differences for any of the MST metrics analyzed.
404 Discussion
409 connectivity analyses vs. MST analyses of global network organization. That is, the PLI
410 analyses failed to reveal differences in connectivity strength between groups whereas the
411 MST analyses yielded between groups differences in network organization as revealed in the
412 theta band. The MST method should correct for potential bias in comparing networks (Stam
413 et al., 2014). Our pattern of findings presents another illustration of the differences between
414 connectivity vs. network analysis of EEG data (see also Stam and van Straaten, 2012). More
415 specifically, the MST analysis showed for dyslexic children a smaller leaf fraction indicating
416 less network integration compared to controls. In addition, there was a significant group
417 difference for diameter suggesting less communication between nodes of the network in
418 dyslexics compared to controls. In terms of the extreme tree topologies, the current pattern of
419 results suggests a more path-like configuration in dyslexic children and a more star-like
420 ,opology in ,ypically reading chlldren (see Flg. 4)-
421 The current group difference in network topology is indicative of a less integrated
422 network configuration in dyslexic children compared to controls (Olde Dubbelink et al.,
423 2014; Stam et al., 2014). This finding is in accordance with previous functional connectivity
424 studies suggesting a disrupted network structure and mixed patterns of connectivity
425 abnormalities in dyslexia (Frye et al., 2012; Koyama et al., 2013). A relevant consideration
426 when interpreting the current results is the relation between MST measures and more
427 conventional graph metrics pertaining to network models such as small-world and scale-free
428 networks. Tewari et al. (2015) examined this relation by performing an extensive and
429 sys tematic series of simulation studies. We observed for dyslexic children a lower leaf
430 fraction and a trend for higher diameter relative to controls. Tewarie et al., (2015) observed
431 that these two measures are strongly related to path length. MST leaf was negatively related
432 to path length. More specifically, MST leaf was low for trees derived from regular networks
433 and increased as these networks became more random. MST diameter, on the other hand, was
434 positively related to path length. That is, diameter increased as networks became more
435 regular. This finding is consistent with a recent study that examined the topology of structural
436 networks in Chinese dyslexics in which a longer path length was observed for dyslexics
437 relative to controls (Liu et al., 2015). Interestingly, in the Tewari et al. (2015) simulation
438 study, MST leaf fraction and diameter were also strongly related to the 'scale freeness' of the
439 network. In particular, leaf fraction increased from regular to random networks and it was
440 much larger for scale-free networks. Accordingly, the current results may also indicate a
441 deviation from scale-free topologies that is larger in dyslexics compared to controls. A scale-
442 free topology is indicative of the presence of highly connected hub nodes in the network
443 (Stam, 2014). In this regard, the current findings may suggest dysfunctional hub nodes in
444 dyslexia.
445 It should be emphasized that the main group differences in network organization were
446 found in the theta band. The present results are consistent with previous research on
447 functional and scaling aspects of oscillatory activity. Regarding general properties of the
448 brain as an oscillatory system, research suggests that slow oscillations such as theta recruit
449 large networks whereas higher frequencies are more confined to smaller networks (Buzsaki
450 and Draguhn, 2004). Further, it is proposed that synchronous activity of lower frequency
451 bands such as theta, mediate long range integration between processes involving several
452 cortical areas (von Stein and Sarnthein, 2000). More specifically, theta frequencies have been
453 related to long range interactions during top-down processes such as working memory
454 retention (von Stein and Sarnthein, 2000). In relation language-specific functions, it has been
455 suggested that synchronous theta activity may play an important role in speech processing
456 (Luo and Poeppel, 2007; Poeppel et al., 2008) and language comprehension (Bastiaansen et
457 al., 2008). Finally, the findings of current study support previous evidence suggesting
458 abnormalities in theta oscillations associated with reading difficulties (Arns and Peters, 2007;
459 Goswami, 2011; Klimesch, 1999; Marosi et al., 1995; Spironelli et al., 2008). Previous
460 reports of theta band abnormalities in dyslexics or poor readers relative to controls include
461 atypical lateralization in several reading tasks (Spironelli et al., 2008), increased power in
462 frontal and temporal regions at rest (Arns and Peters, 2007), higher coherence (Marosi et al.,
463 1995) and deficits in temporal sampling of phonological information (see review in
464 Goswami, 2011). The current results extend previous findings linking abnormalities in the
465 theta band to reading impaired groups by showing a less integrated network configuration in
466 dyslexics' theta spontaneous oscillatory activity.
467 The current results showed also a between-group effect in the gamma band that just fell
468 short of significance. The dyslexic children showed a lower tree hierarchy than controls. It
469 should be noted, however, that the gamma band in scalp EEG recordings may be strongly
470 affected by muscle artifact (Whitham et al., 2007). Consequently, a previous study using
471 graph analysis excluded the higher frequency gamma band from analysis (Lee et al., 2010).
472 In this regard, we hesitate to interpret the current findings for the gamma band, the more so
473 because we are dealing with child data that are typically more affected by muscle artifact
474 compared to adult participants.
475 It should be noted that the current study defined a broad alpha band (see Spectral Power).
476 Previously, it has been suggested that lower and upper frequency alpha bands may be
477 involved in different cognitive processes (see a review of some of these issues in van Diessen
478 et al., 2015). In addition, individual peak frequency varies with age and state of wakefulness
479 (Klimesch, 1999). Given the current scope and focus on resting-state and graph metrics, we
480 opted for a broad alpha band definition to avoid biases from individual peak variability.
481 Future studies could systematically investigate these issues and provide a more detailed
482 description of the cognitive processes associated with the frequency bands in which network
483 metrics are calculated.
484 The MST metrics of the EEG obtained during rest did not relate to the reading measures
485 differentiating children with dyslexia from controls. Previously, Dimitriades et al., (2013) did
486 observe a positive relation between local efficiency of temporo-parietal networks in the beta
487 band of the resting-state EEG and word reading measures in children with reading difficulties
488 but not in typical controls. Similarly, Vourkas et al. (2011) reported significant correlations
489 between graph metrics and phonological decoding ability but this relation was obtained for
490 task-related EEG. In view of the limited studies available to date we are reluctant to interpret
491 the current absence of a relation between reading ability measures and MST metrics. Future
492 studies should examine the potential relations between these measures more systematically
493 by comparing both resting-state and task-related EEG measures.
494 There are a few limitations to the current study. First, the current study used a modestly
495 sized EEG montage (64 electrodes). Although MST metrics are not affected by connectivity
496 strength and network density, some measures are sensitive to network size. Thus, our results
497 should be replicated by using a high-density electrodes array, or preferably MEG source
498 space networks, to assess relative nodal importance in network performance. Secondly,
499 although PLI is more robust than other connectivity measures to methodological problems
500 such as volume conduction (Stam et al., 2007), it is yet unclear how interpolation may affect
501 connectivity measures. At some instances we had to resort to interpolation and the potential
502 effects of interpolation should be examined more systematically. The current assessment of
503 this issue suggests that, in view of the limited number of interpolations, it seems unlikely that
504 interpolation impacted the connectivity weights. Moreover, a control analysis including only
505 participants without interpolated data continued to show the between-groups effect on leaf
506 number. The results of the control analysis are reported in Supplementary Appendix C.
507 Further, in the current study, the PLI sub-averages differed in the number of electrodes and
508 we cannot exclude the possibility that between-group differences in signal-to noise ratio
509 might have affected our MST metrics. It should be noted, however, that the Tewarie et al.
510 (2015) simulation studies indicated that the MST metrics are quite robust to noise. Finally,
511 another potential limitation relates to the stability of our network measures. The current
512 selection of segments was constrained by a relatively short baseline recording (2 minutes)
513 and the presence of artifacts which are common in children EEG resting-state data. Because
514 of this, we could not perform additional analyses on the effects of various epoch length in our
515 graph and connectivity strength metrics. Importantly, however, these issues were
516 systematically examined in a recent study suggesting that MST metrics derived from PLI are
517 almost unaffected by epoch length and produce stable results also for short epochs (Fraschini
518 et al., 2016). Those results suggest an advantage of tree-derived metrics when comparing
519 results across studies in contrast with traditional graph metrics.
520 Conclusion
521 In conclusion, the global organization of functional brain networks may be compromised
522 in dyslexics. The current MST analysis indicated a more path-like topology in dyslexics
523 compared to controls for the EEG theta band. This finding suggests a less integrated network
524 configuration in dyslexia. More specifically, the current results might indicate less efficient
525 long
-range connections in dyslexics, which would be in line with evidence suggesting
526 disrupted connectivity between the distant cortical areas of the reading network (Sandak et
527 al., 2004). The current findings also extend previous evidence suggesting abnormalities in
528 connectivity across multiple brain networks beyond the reading network in dyslexics (Finn et
529 al., 2014). The notion that dyslexics may present differences in widespread topology of brain
530 connectivity would be compatible with evidence and theoretical approaches suggesting
531 deficits in general sensory and attentional functions (e.g. visuospatial attention, visual
532 attention span, auditory processing, etc.). Importantly, MST metrics could help characterizing
533 the heterogeneity within dyslexia, as different underlying deficits may result in similar
534 reading impairments. Future studies employing MST analysis might want to adopt a
535 longitudinal perspective in examining the developmental trajectories of network organization
536 during reading acquisition. Furthermore, it would be of considerable interest to examine how
537 functional network organization is changed following reading intervention (e.g., Koyama et
538 al., 2013).
540 Conflict of interest
541 None of the authors have potential conflicts of interest to be disclosed.
543 Acknowledgements
544 We dedicate this paper to our co-author professor Leo Blomert. His contributions to the
545 initial stages of the project prior to his untimely death were significant. This project is part of
546 the research program "Fluent reading neurocognitively decomposed: The case of dyslexia -
547 HCMI 10-59" funded by the Netherlands Initiative Brain and Cognition, a part of the
548 Organization for Scientific Research (NWO) under grant number 056-14-015. The funders
549 had no role in study design, data collection and analysis, decision to publish, or preparation
550 of this manuscript.
553 References
554 Achenbach, T. M., Becker, A., Döpfner, M., Heiervang, E., Roessner, V., Steinhausen, H.C.,
555 & Rothenberger, A. (2008). Multicultural assessment of child and adolescent
556 psychopathology with ASEBA and SDQ instruments: research findings, applications,
557 and future directions. J Child Psychol Psyc, 49(3), 251-75. doi:10.1111/j.1469-
558 7610.2007.01867.x
559 Arns, M., & Peters, S. (2007). Different brain activation patterns in dyslexic children:
560 evidence from EEG power and coherence patterns for the double-deficit theory of
561 dyslexia. J Integr Neurosci, 6(1), 175-190. Retrieved from
562 http://www.worldscientific.com/doi/abs/10.1142/S0219635207001404
563 Bassett, D. S., & Bullmore, E. (2006). Small-world brain networks. Neuroscientist 12(6),
564 512-23. doi:10.1177/1073858406293182
565 Bastiaansen, M. C. M., Oostenveld, R., Jensen, O., & Hagoort, P. (2008). I see what you
566 mean: theta power increases are involved in the retrieval of lexical semantic
567 information. Brain Lang, 106(1), 15-28. doi:10.1016/j.bandl.2007.10.006
568 Blomert, L., & Vaessen, A. (2009). 3DM differential diagnostics for dyslexia: cognitive
569 analysis of reading and spelling. Amsterdam, the Netherlands: Boom Test.
570 Boersma, M., Smit, D. J., Boomsma, D. I., De Geus, E. J. C., Delemarre-van de Waal, H., &
571 Stam, C. J. (2013). Growing trees in child brains: graph theoretical analysis of
572 electroencephalography-derived minimum spanning tree in 5- and 7-year-old children
573 reflects brain maturation. Brain Connectivity, 3(1), 50-60. doi:10.1089/brain.2012.0106
574 Bullmore, E., & Sporns, O. (2009). Complex brain networks: graph theoretical analysis of
575 structural and functional systems. Nat Rev Neurosci, 10(3), 186-98.
576 doi: 10.1038/nrn2575
577 Bullmore, E., & Sporns, O. (2012). The economy of brain network organization. Nat Rev
578 Neurosci, 13(5), 336-49. doi:10.1038/nrn3214
579 Buzsáki, G., & Draguhn, A. (2004). Neuronal oscillations in cortical networks. Science, 304,
580 1926-1930. Retrieved from http://www.sciencemag.org/content/304/5679/1926.short
581 de Vos, T. (2007). Schoolvaardigheidstoets. Amsterdam, the Netherlands: Boom Test.
582 Delorme, A., & Makeig, S. (2004). EEGLAB: an open source toolbox for analysis of single-
583 trial EEG dynamics including independent component analysis. J Neurosci Meth,
584 134(1), 9-21. doi:10.1016/j.j neumeth.2003.10.009
585 Dhar, M., Been, P., Minderaa, R., & Althaus, M. (2010). Reduced interhemispheric
586 coherence in dyslexic adults. Cortex, 46(6), 794-8. doi:10.1016/j.cortex.2009.09.006
587 Dimitriadis, S. I., Laskaris, N. A., Simos, P., Micheloyannis, S., Fletcher, J. M., Rezaie, R., &
588 Papanicolaou, A. C. (2013). Altered temporal correlations in resting-state connectivity
589 fluctuations in children with reading difficulties detected via MEG. NeuroImage, 83,
590 307-17. doi:10.1016/j.neuroimage.2013.06.036
591 Finn, E. S., Shen, X., Holahan, J. M., Scheinost, D., Lacadie, C., Papademetris, X., et al.
592 (2014). Disruption of functional networks in dyslexia: a whole-brain, data-driven
593 analysis of connectivity. Biol Psychiat, 76(5), 397-404.
594 doi:10.1016/j .biopsych. 2013.08.031
595 Fraga González, G., Zaric, G., Tijms, J., Bonte, M., Blomert, L., & van der Molen, M. W.
596 (2015). A randomized controlled trial on the beneficial effects of training letter-speech
597 sound integration on reading fluency in children with dyslexia. PloS one, 10(12),
598 e0143914.
599 Fraschini, M., Demuru, M., Puligheddu, M., Floridia, S., Polizzi, L., Maleci, A., et al. (2014).
600 The re-organization of functional brain networks in pharmaco-resistant epileptic patients
601 who respond to VNS. Neurosci Lett, 580, 153-7. doi:10.1016/j.neulet.2014.08.010
602 Fraschini, M., Demuru, M., Crobe, A., Marrosu, F., Stam, C. J., & Hillebrand, A. (2016). The
603 effect of epoch length on estimated EEG functional connectivity and brain network
604 organisation. J neural eng, 13(3), 036015.
605 Frye, R. E., Liederman, J., McGraw Fisher, J., & Wu, M.-H. (2012). Laterality of
606 temporoparietal causal connectivity during the prestimulus period correlates with
607 phonological decoding task performance in dyslexic and typical readers. Cereb Cortex,
608 22(8), 1923-34. doi:10.1093/cercor/bhr265
609 Goswami, U. (2011). A temporal sampling framework for developmental dyslexia. Trends
610 Cog Sci, 15(1), 3-10. doi:10.1016/j.tics.2010.10.001
611 Gusnard, D. A., & Raichle, M. E. (2001). Searching for a baseline: functional imaging and
612 the resting human brain. Nat Rev Neurosci, 2(10), 685-694.
613 Hampson, M., Tokoglu, F., Sun, Z., Schafer, R. J., Skudlarski, P., Gore, J. C., & Constable,
614 R. T. (2006). Connectivity-behavior analysis reveals that functional connectivity
615 between left BA39 and Broca's area varies with reading ability. Neurolmage, 31(2),
616 513-9. doi: 10.1016/j.neuroimage.2005.12.040
617 Klimesch, W. (1999). EEG alpha and theta oscillations reflect cognitive and memory
618 performance: a review and analysis. Brain Res Rev, 29(2-3), 169-95. Retrieved from
619 http://www.ncbi.nlm.nih.gov/pubmed/10209231
620 Koyama, M. S., Di Martino, A., Kelly, C., Jutagir, D. R., Sunshine, J., Schwartz, S. J., et al.,
621 (2013). Cortical signatures of dyslexia and remediation: an intrinsic functional
622 connectivity approach. PloS One, 8(2), e55454. doi:10.1371/journal.pone.0055454
623 Koyama, M. S., Di Martino, A., Zuo, X.-N., Kelly, C., Mennes, M., Jutagir, D. R., et al.,
624 (2011). Resting-state functional connectivity indexes reading competence in children
625 and adults. J Neurosci, 31(23), 8617-24. doi:10.1523/JNEUR0SCI.4865-10.2011
626 Koyama, M. S., Kelly, C., Shehzad, Z., Penesetti, D., Castellanos, F. X., & Milham, M. P.
627 (2010). Reading networks at rest. Cereb Cortex, 20(11), 2549-59.
628 doi: 10.1093/cercor/bhq005
629 Kruskal, J. B. (1956). On the shortest spanning subtree of a graph and the traveling salesman
630 problem. P Am Math Soc, 7(1), 48-50.
631 Lee, U., Kim, S., & Jung, K.-Y. (2006). Classification of epilepsy types through global
632 network analysis of scalp electroencephalograms. Phys Rev E, 73(4), 041920.
633 doi:10.1103/PhysRevE.73.041920
634 Lee, U., Oh, G., Kim, S., Noh, G., Choi, B., & Mashour, G. A. (2010). Brain Networks
635 Maintain a Scale-free Organization across Consciousness , Anesthesia , and Recovery.
636 Anesthesiology, 113, 1081-1091.
637 Liu, K., Shi, L., Chen, F., Waye, M. M. Y., Mok, V. C. T., Chu, W. C. W., & Wang, D.
638 (2015). Altered topological organization of brain structural network in Chinese children
639 with developmental dyslexia. Neurosci Lett, 589, 169-175.
640 doi:10.1016/j.neulet.2015.01.037
641 Luo, H., & Poeppel, D. (2007). Phase patterns of neuronal responses reliably discriminate
642 speech in human auditory cortex. Neuron, 54(6), 1001-10.
643 doi:10.1016/j.neuron.2007.06.004
644 Marosi, E., Harmony, T., & Becker, J. (1995). Electroencephalographic coherences
645 discriminate between children with different pedagogical evaluation. International J
646 Psychopysiol, 19, 23-32. Retrieved from
647 http://www.sciencedirect.com/science/article/pii/016787609400059N
648 Nagarajan, S., & Mahncke, H. (1999). Cortical auditory signal processing in poor readers. P
649 Natl Acad Sci USA, 96(11), 6483-6488. Retrieved from
650 http://www.pnas.org/content/96Z11/6483.short
651 Newman, M. E. J. (2003). Mixing patterns in networks. Phys Rev E Stat Nonlin Soft Matter
652 Phys, 67:026126.
653 Olde Dubbelink, K. T. E., Hillebrand, A., Stoffers, D., Deijen, J. B., Twisk, J. W. R., Stam,
654 C. J., & Berendse, H. W. (2014). Disrupted brain network topology in Parkinson's
655 disease: a longitudinal magnetoencephalography study. Brain, 137, 197-207.
656 doi: 10.1093/brain/awt316
657 Poeppel, D., Idsardi, W. J., & van Wassenhove, V. (2008). Speech perception at the interface
658 of neurobiology and linguistics. Philos Trans R Soc Lond B Biol Sci, 363(1493), 1071659 86. doi:10.1098/rstb.2007.2160
660 Porz, S., Kiel, M., & Lehnertz, K. (2014). Can spurious indications for phase synchronization
661 due to superimposed signals be avoided? Chaos, 24(3), 033112. Retrieved from
662 http://scholar.google.com/scholar?hl=en&btnG=Search&q=intitle:No+Title#0
663 Pugh, K. R., Mencl, W., Shaywitz, B., Shaywitz, S., Fulbright, R. K., Constable, R. T., et al.
664 (2000). The angular gyrus in developmental dyslexia: task-specific differences in
665 functional connectivity within posterior cortex. Psychol Sci, 11(1), 51-56. Retrieved
666 from http://pss.sagepub.com/content/! 1/1/51.short
667 Raven, J., Raven, J. C., & Court, J. H. (1998). Coloured progressive matrices. Oxford, UK:
668 Oxford Psychologists Press.
669 Sandak, R., Mencl, W., Frost, S. J., & Pugh, K. R. (2004). The Neurobiological Basis of
670 Skilled and Impaired Reading : Recent Findings and New Directions. Sci Stud Read,
671 8(3), 273-292.
672 Schlaggar, B. L., & McCandliss, B. D. (2007). Development of neural systems for reading.
673 Ann Rev Neurosci, 30, 475-503. doi:10.1146/annurev.neuro.28.061604.135645
674 Schurz, M., Wimmer, H., Richlan, F., Ludersdorfer, P., Klackl, J., & Kronbichler, M. (2014).
675 Resting-State and Task-Based Functional Brain Connectivity in Developmental
676 Dyslexia. Cereb Cortex, 25 (10)., 3502.
677 Shaywitz, S., Shaywitz, B., Fulbright, R. K., Skudlarski, P., Mencl, W., Constable, R. T., et
678 al., (2003). Neural systems for compensation and persistence: young adult outcome of
679 childhood reading disability. Biol Psychiat, 54(1), 25-33. doi:10.1016/S0006-
680 3223(03)01836-X
681 Shiota, M., Koeda, T., & Takeshita, K. (2000). Cognitive and neurophysiological evaluation
682 of Japanese dyslexia. Brain Dev, 22(7), 421-426. Retrieved from
683 http://www.sciencedirect.com/science/article/pii/S0387760400001674
684 Spironelli, C., Penolazzi, B., & Angrilli, A. (2008). Dysfunctional hemispheric asymmetry of
685 theta and beta EEG activity during linguistic tasks in developmental dyslexia. Biol
686 Psychol, 77(2), 123-31. doi:10.1016/j.biopsycho.2007.09.009
687 Stam, C. J. (2014). Modern network science of neurological disorders. Nat Rev Neurosci,
688 i5(10), 683-695. doi:10.1038/nrn3801
689 Stam, C. J., Nolte, G., & Daffertshofer, A. (2007). Phase lag index: assessment of functional
690 connectivity from multi channel EEG and MEG with diminished bias from common
691 sources. Hum Brain Mapp, 28(11), 1178-93. doi:10.1002/hbm.20346
692 Stam, C. J., Tewarie, P., Van Dellen, E., van Straaten, E. C. W., Hillebrand, A., & Van
693 Mieghem, P. (2014). The trees and the forest: Characterization of complex brain
694 networks with minimum spanning trees. Int J Psychophysiol, 92(3), 129-38.
695 doi: 10.1016/j .ijpsycho.2014.04.001
696 Stam, C. J., & van Straaten, E. C. W. (2012). The organization of physiological brain
697 networks. Clin Neurophysiol, 123(6), 1067-87. doi:10.1016/j.clinph.2012.01.011
698 Stanberry, L. I., Richards, T. L., Berninger, V. W., Nandy, R. R., Aylward, E. H., Maravilla,
699 K. R., et al., (2006). Low-frequency signal changes reflect differences in functional
700 connectivity between good readers and dyslexics during continuous phoneme mapping.
701 Magn Reson Imaging, 24(3), 217-29. doi:10.1016/j.mri.2005.12.006
702 Tewarie, P., Van Dellen, E., Hillebrand, A., & Stam, C. J. (2015). The minimum spanning
703 tree: An unbiased method for brain network analysis. NeuroImage, 104, 177-188.
704 doi:10.1016/j.neuroimage.2014.10.015
705 van den Bos, K. P., Spelberg, L., Scheepsma, A. J. M., & De Vries, J. R. (1999). De Klepel:
706 Verantwoording, handleiding, Diagnostiek en behandeling. Lisse, the Netherlands:
707 Swets & Zeitlinger.
708 van den Heuvel, M. P., & Hulshoff Pol, H. E. (2010). Exploring the brain network: a review
709 on resting-state fMRI functional connectivity. Eur Neuropsychopharm, 20(8), 519-34.
710 doi:10.1016/j.euroneuro.2010.03.008
711 van der Mark, S., Klaver, P., Bucher, K., Maurer, U., Schulz, E., Brem, S., et al., (2011). The
712 left occipitotemporal system in reading: disruption of focal fMRI connectivity to left
713 inferior frontal and inferior parietal language areas in children with dyslexia.
714 NeuroImage, 54(3), 2426-36. doi:10.1016/j.neuroimage.2010.10.002
715 van Diessen, E., Otte, W. M., Braun, K. P., Stam, C. J., & Jansen, F. E. (2014). Does sleep
716 deprivation alter functional EEG networks in children with focal epilepsy? Front Syst
717 Neurosci, 8. doi:10.3389/fnsys.2014.00067
718 van Diessen, E., Numan, T., van Dellen, E., van der Kooi, A. W., Boersma, M., Hofman, D.
719 et al. (2015). Opportunities and methodological challenges in EEG and MEG resting
720 state functional brain network research. Clin Neurophysiol, 126(8), 1468-1481.
721 van Mieghem, P. (2014). Performance analysis of complex networks and systems. Cambridge
722 University Press.
723 Vandermosten, M., Boets, B., Wouters, J., & Ghesquiere, P. (2012). A qualitative and
724 quantitative review of diffusion tensor imaging studies in reading and dyslexia. Neurosci
725 Biobehav R, 36(6), 1532-52. doi:10.1016/j.neubiorev.2012.04.002
726 Vogel, A. C., Miezin, F. M., Petersen, S. E., & Schlaggar, B. L. (2012). The putative visual
727 word form area is functionally connected to the dorsal attention network. Cereb Cortex,
728 22(3), 537-49. doi:10.1093/cercor/bhr100
729 Vogel, A. C., Petersen, S. E., & Schlaggar, B. L. (2014). The VWFA: it's not just for words
730 anymore. Front Hum Neurosci, 8, 88. doi:10.3389/fnhum.2014.00088
731 von Stein, A., & Sarnthein, J. (2000). Different frequencies for different scales of cortical
732 integration: from local gamma to long range alpha/theta synchronization. Int J
733 Psychophysiol, 38(3), 301-313. Retrieved from
734 http://www. sciencedirect.com/science/article/pii/S 0167876000001720
735 Vourkas, M., Karakonstantaki, E., Simos, P., Tsirka, V., Antonakakis, M., Vamvoukas, M.,
736 et al. (2014). Simple and difficult mathematics in children: a minimum spanning tree
737 EEG network analysis. Neurosci Lett, 576, 28-33. doi:10.1016/j.neulet.2014.05.048
738 Vourkas, M., Micheloyannis, S., Simos, P., Rezaie, R., Fletcher, J. M., Cirino, P. T., &
739 Papanicolaou, A. C. (2011). Dynamic task-specific brain network connectivity in
740 children with severe reading difficulties. Neurosci Lett, 488(2), 123-8.
741 doi:10.1016/j.neulet.2010.11.013
742 Whitham, E. M., Pope, K. J., Fitzgibbon, S. P., Lewis, T., Clark, C. R., Loveless, S., et al.,
743 (2007). Scalp electrical recording during paralysis: quantitative evidence that EEG
744 frequencies above 20 Hz are contaminated by EMG. Clin Neurophysiol, 118(8), 1877745 88. doi:10.1016/j.clinph.2007.04.027
746 Wolf, R. C., Sambataro, F., Lohr, C., Steinbrink, C., Martin, C., & Vasic, N. (2010).
747 Functional brain network abnormalities during verbal working memory performance in
748 adolescents and young adults with dyslexia. Neuropsychologia, 48(1), 309-18.
749 doi:10.1016/j.neuropsychologia.2009.09.020
750 Zhang, M., Li, J., Chen, C., Xue, G., Lu, Z., Mei, L., et al., (2014). Resting-state functional
751 connectivity and reading abilities in first and second languages. NeuroImage, 84, 546752 53. doi:10.1016/j. neuroimage.2013.09.006
754 Figure Legends
755 Fig. 1. Schematic of the graph analysis. First, artifact-free epochs are filtered for each
756 frequency band (A). Secondly, the functional connectivity matrix based on Phase Lag Index
757 (PLI) is calculated for each frequency band and epoch (B). Finally, the Kruskal's algorithm is
758 applied to obtain the Minimum Spanning Tree (MST) matrix (C-left); the resulting loopless
759 graph is displayed on a scalp projection (C-middle) and as a tree (C-right). The tree view
760 shows the hierarchical structure of the graph starting from an arbitrary root node (in this case
761 FP1), the color map of the nodes from blue to red represents lower to higher betweenness
762 centrality. For illustrative purposes this figure shows the MST obtained from the PLI matrix
763 averaged across epochs and subjects of the control group (N=15).
764 Fig. 2. Examples of trees for increasing leaf number including the two extreme forms of
765 trees. All of them have 14 nodes (circles) and 13 edges (lines). On the left a line-like tree
766 with the lowest possible leaf number which is 2. The middle example shows a tree
767 configuration with eight leaf nodes. On the right, a star-like tree with the highest possible leaf
768 number which equals the number of edges.
769 Fig. 3. Power spectra averaged across 64 EEG scalp channels for the control and the
770 dyslexic group.
771 Fig. 4. MST matrices (left panels) and MST graph in scalp view (center panel) and tree
772 view (right panel) for the theta band for controls (above) and dyslexics (below). For 73 illustra
774 Fig. 5. (A) Group averages for leaf fraction, (B) eccentricity and diameter measures of the
773 illustrative purposes the MST algorithm was performed on the averaged PLI matrices.
775 MST. Open bars refer to controls and filled bars to dyslexics. * p < 0.05.
Tables
Table 1.
Sample characteristics and descriptive statistics showing reading accuracy and fluency scores.
Controls M(SD)
Sex ratio (m:f) Handedness (L:R)* Age
RAVEN - IQ test a
3DM Word reading - accuracy b High Frequency Low Frequency Pseudo Total [T]c
3DM Word reading - fluency [T] High Frequency Low Frequency Pseudo Total
One-Minute Test -fluency [SS]d Text Reading - fluency [T]**
3DM Spelling - accuracy[T] 3DM Spelling - fluency[T]
3DM Phoneme deletion - accuracy [T]**
Letter-speech sound associations [T]
L-SS identificacion - accuracy
iscrimination - accuracy L-SS identificacion - fluency L-SS discrimination - fluency**
ÖSS dis
LSS L-SS
3DM Naming speed scores[T] Letters numbers Total
15 6:9 2:10 8.75 (0.31) 6.70 (1.51)
Dyslexies M(SD)
p-value
29 16:13 2:27 8.96 (0.40) 7.11 (1.51)
0.07 0.02
99.28 (1.05) 93.10 (5.93) .000 0.27
98.32 (2.54) 86.31 (14.48) .003 0.19
88.70 (8.48) 73.33 (17.43) .003 0.20
51.40 (8.00) 33.72 (12.58) .000 0.37
54.27 (7.58) 31.38 (6.14) .000 0.74
56.80 (8.98) 32.07 (6.46) .000 0.72
54.93 (9.71) 30.93 (6.37) .000 0.70
55.93 (9.51) 31.00 (5.40) .000 0.75
12.07 (2.94) 3.97 (1.97) .000 0.74
55.27 (8.41) 33.21 (6.30) .000 0.70
51.73 (8.62) 36.21 (6.70) .000 0.51
54.33 (9.90) 36.55 (6.01) .000 0.57
53.73 (8.39) 39.61 (8.32) .000 0.41
46.87 (8.65) 43.34 (12.99) .350 0.02
50.80 (10.28) 44.43 (9.63) .050 0.09
51.53 (7.67) 41.79 (6.97) .000 0.30
51.73 (7.36) 45.46 (8.95) .025 0.12
50.93 (6.95) 36.57 (8.05) .000 0.45
52.73 (10.67) 36.21 (8.50) .000 0.43
50.80 (7.73) 35.54 (9.15) .000 0.42
a C scores (M = 5, SD = 2).b Raw scores. c T scores (M = 50, SD = 10). d SS scores (M = 10, SD= 3) *Data missing for 3 participants; Typical N = 12. ** Data missing for one participant; Dyslexics N = 28.
Table 2.
MST measures summary.
Degree
Leaf fraction
Diameter
Eccentricity
BC Betweenness Centrality k Kappa Th Tree Hierarchy
R Degree correlation
Number of nodes in MST Number of links in the MST Number of neighbors for a given node in the MST Fraction of nodes with degree = 1 (leafs) in the MST. Largest distance between any two nodes of the tree. Longest distance between a reference node and any other node
Fraction of all shortest paths that pass through a particular node
Measure of the broadness of the degree distribution (degree divergence)
A hierarchical metric that quantifies the trade-off between large scale integration in the MST and the overload of central nodes
Correlation between the degrees of a node and the degree of the neighboring vertices to which it is connected
Table 3.
PLI average and MST measures.
Typical (N = 15) M SD
Dyslexies (N =29 ) M SD
p value n 2
Delta PLI
Degree Leaf
Eccentricity Kappa Diameter BC
Degree Correlation Hierarchy
Thetaa PLI
Degree Leaf
Eccentricity Kappa Diameter BC
Degree Correlation Hierarchy
Alphab PLI
Degree Leaf
Eccentricity Kappa Diameter BC
Degree Correlation Hier
0.202 0.163 0.583 0.168 3.551 0.216 0.704 -0.325 0.418
0.176 0.152 0.584 0.169 3.415 0.216 0.701 -0.327 0.419
0.209 0.187
0.197 0.713 -0.345 0.441
(0.012) (0.022) (0.012) (0.010) (0.259) (0.013) (0.026) (0.030) (0.015)
(0.008) (0.011) (0.013) (0.011) (0.149) (0.015) (0.022) (0.037) (0.015)
(0.033) (0.028) (0.030) (0.013) (0.398) (0.018) (0.025) (0.028) (0.018)
0.207 0.160 0.576 0.170 3.501 0.219 0.698 -0.321 0.417
0.174 0.148 0.569 0.174 3.341
0.696 -0.319 0.412
0.200 0.185 0.609 0.160 3.892 0.206 0.712 -0.352 0.432
(0.013) (0.022) (0.020) (0.011) (0.270) (0.014) (0.033) (0.039) (0.023)
(0.0 (0.015)
(0.015) (0.008) (0.165) (0.011)
(0.023) (0.031) (0.016)
(0.037) (0.033) (0.027) (0.011) (0.466) (0.015) (0.032) (0.038) (0.024)
1.94 0.31 1.37 0.36 0.37 0.33 0.
00..2023
0.70 0.86
3.47 2.14 4.27 0.49 0.55 2.13
0.68 0.11 2.59 2.86 0.37 3.22 0.03 0.29 1.74
.171 .580 .248 .552 .543 .568 .577 .643 .865
.408 .359
.003* .070 .151
.489 .463 .152
.413 .744 .115
.098 .545 .080 .862 .593 .194
0.04 0.01 0.03 0.01 0.01 0.01 0.01 0.00 0.00
0.02 0.02
0.08 0.05
0.01 0.01 0.05
0.02 0.00 0.06 0.07 0.01 0.07 0.00 0.01 0.04
text represents significant results (p < 0.05 ); italic text represents results at trend level; Significant after Bonferroni correction at p = 0.059 (i.e., p < 0.006).
Notes. Bold tex
a Two outliers based on spectral power excluded; Dyslexies N = 27.b One outlier based on spectral power excluded; Dyslexies N = 28. MST, minimum spanning tree; PLI, phase lag index; BC, betweenness centrality.
Table 4.
PLI average and MST measures.
Typical (N = 15)
Dyslexics (N =29 )
M SD M SD F p value n 2
PLI 0.099 (0.006) 0.101 (0.010) 0.43 .514 0.01
Degree 0.160 (0.019) 0.162 (0.019) 0.09 .770 0.00
Leaf 0.582 (0.018) 0.580 (0.022) 0.07 .796 0.00
Eccentricity 0.168 (0.010) 0.168 (0.009) 0.00 .958 0.00
Kappa 3.516 (0.244) 3.542 (0.275) 0.08 .773 0.00
Diameter 0.216 (0.013) 0.216 (0.012) 0.06 .811 0.00
BC 0.691 (0.026) 0.702 (0.021) 2.28 .138 0.05
Degree Correlation -0.319 (0.027) -0.316 (0.040) 0.18 .675 0.00
Hierarchy 0.425 (0.015) 0.417 (0.020) 1.72 .197 0.04
PLI 0.092 (0.006) 0.091 (0.006) 0.04 .840 0.00
Degree 0.224 (0.041) 0.214 (0.061) 0.69 .410 0.02
Leaf 0.637 (0.034) 0.621 (0.038) 2.04 .160 0.05
Eccentricity 0.152 (0.011) 0.158 (0.015) 2.02 .162 0.05
Kappa 4.587 (0.875) 4.468 (1.381) 0.41 .528 0.01
Diameter 0.196 (0.014) 0.205 (0.019) 2.31 .136 0.05
BC 0.725 (0.027) 0.725 (0.032) 0.00 .942 0.00
Degree Correlation -0.366 %0.031) -0.356 (0.040) 0.89 .350 0.02
Hierarchy 0.443 (0.020) 0.431 (0.016) 3.89 .055 0.09
Notes. Bold text represents significant results ( p < 0.05 ); italic text represents results at trend level; MST, minimum spanning tree; PLI, phase lag index; BC, betweenness centrality.
Control (N = 15)
Dyslexies (N = 27)
PLI MST matrix MST graph
