doi: /brain/aws048 Brain 2012: 135;

(1)

BRAIN

A JOURNAL OF NEUROLOGY

Graph analysis of functional brain networks

for cognitive control of action in traumatic

brain injury

Karen Caeyenberghs,

1

Alexander Leemans,

2

Marcus H. Heitger,

1

Inge Leunissen,

1

Thijs Dhollander,

3

Stefan Sunaert,

4

Patrick Dupont

5

and Stephan P. Swinnen

1

1 Movement Control and Neuroplasticity Research Group, Department of Kinesiology, Biomedical Sciences Group, 3000 Leuven, Belgium 2 Image Sciences Institute, University Medical Centre Utrecht, 3564 CX Utrecht, The Netherlands

3 ESAT-PSI, Centre for the Processing of Speech and Images, Department of Electrical Engineering (ESAT), Science, Engineering and Technology Group, 3000 Leuven, Belgium

4 Radiology, Department of Imaging and Pathology, Biomedical Sciences Group, 3000 Leuven, Belgium

5 Laboratory for Cognitive Neurology, Research Group Experimental Neurology, Department of Neurosciences, Biomedical Sciences Group, 3000 Leuven

Correspondence to: Karen Caeyenberghs, Laboratory of Motor Control,

Research Centre for Motor Control and Neuroplasticity, Group Biomedical Sciences, K.U.Leuven,

Tervuursevest 101, B-3001 Heverlee, Belgium

E-mail: [email protected]

Patients with traumatic brain injury show clear impairments in behavioural flexibility and inhibition that often persist beyond the

time of injury, affecting independent living and psychosocial functioning. Functional magnetic resonance imaging studies have

shown that patients with traumatic brain injury typically show increased and more broadly dispersed frontal and parietal activity

during performance of cognitive control tasks. We constructed binary and weighted functional networks and calculated their

topological properties using a graph theoretical approach. Twenty-three adults with traumatic brain injury and 26 age-matched

controls were instructed to switch between coordination modes while making spatially and temporally coupled circular motions

with joysticks during event-related functional magnetic resonance imaging. Results demonstrated that switching performance

was significantly lower in patients with traumatic brain injury compared with control subjects. Furthermore, although brain

networks of both groups exhibited economical small-world topology, altered functional connectivity was demonstrated in

patients with traumatic brain injury. In particular, compared with controls, patients with traumatic brain injury showed increased

connectivity degree and strength, and higher values of local efficiency, suggesting adaptive mechanisms in this group. Finally,

the degree of increased connectivity was significantly correlated with poorer switching task performance and more severe brain

injury. We conclude that analysing the functional brain network connectivity provides new insights into understanding cognitive

control changes following brain injury.

Keywords:

traumatic brain injury; cognitive control; functional MRI; task switching; functional network

Abbreviations:

TBI = traumatic brain injury

Received August 1, 2011. Revised January 4, 2012. Accepted January 5, 2012. Advance Access publication March 16, 2012

by guest on October 19, 2015

http://brain.oxfordjournals.org/

(2)

Introduction

Many patients with traumatic brain injury (TBI) are faced with per-sistent cognitive deficits, including decreased mental flexibility, trou-ble with shifting sets, impaired attention, poor planning and organization, impaired judgement, deficits in verbal fluency, reduced working memory and increased impulsivity (Levin and Kraus, 1994; Miller, 2000; Godefroy, 2003), affecting independent living and psy-chosocial functioning (Garthet al., 1997). Hence, the burden on public health and social care is substantial (Thurmanet al., 1999). The increased use of neuroimaging techniques has substantially enhanced our understanding of the underlying pathophysiology of these persistent impairments.

A number of recent diffusion tensor imaging studies in adults with TBI have demonstrated reduced white matter microstructural integrity, particularly in the cingulum fibre bundles, the corpus callosum, inferior fronto-occipital fasciculus and superior longitu-dinal fasciculus. In some cases, significant correlations between diffusion tensor imaging metrics and cognitive function have been identified, such that degree of white matter pathology pre-dicts cognitive deficits (Salmond et al., 2006; Krauset al., 2007; Kennedy et al., 2009; Kinnunen et al., 2010). Moreover, a voxel-based morphometry study has provided evidence for abnor-mal grey matter density decreases in frontal and temporal cortices, cingulate gyrus, subcortical grey matter and the cerebellum in pa-tients with chronic TBI (Gale et al., 2005). This decreased grey matter concentration correlated with lower scores on tests of at-tention. Convergent evidence across various cognitive tasks also indicates abnormal functional MRI activation of frontal and par-ietal cortical regions (Perlstein et al., 2004; Scheibelet al., 2007; Newsome et al., 2008; Kimet al., 2009).

In the present study, we decided not to focus on these regional functional changes in specific areas but instead concentrate on the capacity of information flow within and between regions, which is equally crucial for behaviour. Indeed, TBI can be characterized as a ‘disconnection syndrome’ (Guye et al., 2010). TBI involves pro-gressive biochemical and structural changes, including degradation of the myelin sheath and further degradation of the axonal mem-brane (Ewing-Cobbset al., 2008; Sidaroset al., 2008). It is there-fore important to examine the characteristics of the brain network. Although many studies have elucidated the pathophysiology of cognitive recovery after TBI by focusing on local changes in brain structure and function, little is known about the insult-induced alterations in the interactions among the nodes of task-related functional networks in patients with TBI. Within this perspective, the concept of functional connectivity has emerged, referring to the statistical association between spatially distributed brain acti-vation time series (Lee et al., 2006). Graph theory is a powerful technique to map these relationships between spatially remote neurophysiological events in the brain. The area of graph theory is an established mathematical field and has proven a very effect-ive and informateffect-ive way to explore brain function and human behaviour (Bullmore and Sporns, 2009; Nakamura et al., 2009; Stam, 2010).

Graph theoretical measures have been applied previously to analyse functional brain connectivity in the context of TBI

(Nakamura et al., 2009; Castellanos et al., 2010, 2011). However, although graph theory offers a broad selection of meas-ures to examine and quantify relationships between activations in different brain regions, previous studies have frequently focused on only a few measures in each case, such as clustering coeffi-cient, network path length, small worldness or connectivity degree. Here, we applied a number of graph theory measures to task-related functional MRI data.

Furthermore, previous functional connectivity studies in TBI have primarily been limited to task-free (resting state) EEG or magnetoencephalogram recordings, identifying alterations in net-work architecture (Cao and Slobounov, 2010; Tsirkaet al., 2011). Although EEG and magnetoencephalogram signals supply a dis-tinctly high temporal resolution, their limited spatial resolution hampers the detailed study of activity within distributed task-related (sub)cortical brain regions and their associated inter-actions. As a whole, these studies have mainly investigated abnor-mal networks based on the spontaneous activity in the resting brain. However, task-related functional connectivity is a useful tool for investigating a more specific interdependence between brain regions.

Moreover, previous studies commonly used methods such as coherence or synchronization likelihood rather than partial correl-ations to quantify the association in activation between brain re-gions, thereby not addressing the confounding effects of indirect connections between network nodes (Castellanos et al., 2011). Also a template-based a priori parcellation of the brain was used to define the network nodes, thereby including brain regions with little or no task-related activation (Nakamuraet al., 2009). A recent study by Smith et al. (2011) provides strong indications that, in addition to using partial correlations to quantify unique functional associations between nodes, a data-driven approach by defining networks based only on areas showing clear task-related activation is preferable to template-based approaches in order to minimize confounds and obtain a better picture on functional con-nectivity within active neural networks. Therefore, we combined the partial correlation approach with a data-driven network def-inition, aiming to optimize the reliability of the results.

The present study was based on three main hypotheses: (i) Based on previous studies of neural network functionality in

TBI populations in a non-motor context (Nakamura et al., 2009; Castellanos et al., 2010, 2011), we expected an in-crease in functional connectivity in patients with TBI within the motor switching network. Behavioural studies have shown that switching between tasks and inhibition of re-sponses are diminished in patients with TBI, reflecting im-paired cognitive control of action (Mecklinger et al., 1999; Amos, 2002; Azouviet al., 2004; Levinet al., 2004; Larson

et al., 2006; Perlsteinet al., 2006). Here, we studied impair-ments in task switching during bimanual coordination (Kelso, 1995). This form of switching temporarily dissociates the limbs from sharing an ongoing movement plan, requiring inhibition of a part of the existing plan and facilitation of an alternative plan (Wenderothet al., 2009; Coxonet al., 2010).

(ii) Furthermore, our previous functional MRI study has reported negative correlations specifically in patients with

(3)

TBI between switching performance and levels of brain ac-tivation in the overacac-tivation areas (Leunissenet al., 2012). Hence, we examined if such correlations between blood oxygen level-dependent response and switching perform-ance are mirrored in similar associations between the graph theoretical network measures and switching perform-ance. To the best of our knowledge, the present study is the first to use a graph theoretical approach to examine changes in behavioural switching networks by measuring functional connectivity and network metrics in task-specific functional MRI data recorded from healthy volunteers and patients with TBI.

(iii) Finally, we expected a correlation between the degree of network connectivity and injury severity in patients with TBI, whereby the increased connectivity degree would be associated with more severe brain injury.

Materials and methods

Participants

Forty-nine adults participated in the study, including 26 control sub-jects (mean age = 25 years 3 months; SD = 3 years; 13 males and 13 females) and 23 patients with moderate to severe TBI (mean age = 24 years 5 months; SD = 5 years 6 months; 16 males and 7 females), who had sustained closed head trauma due to traffic accident or sport injury. Time since injury averaged 4 years 10 months prior to the study (SD = 2 years 7 months, range = 1 year 6 months 9 years 8 months). Injury severity of the patients with TBI was based on post-resuscitation Glasgow Coma Scale (Teasdale and Jennett, 1974) when available (mean Glasgow Coma Scale score = 7, SD = 3, range = 3–13; data available for eight patients), initial neuropathology,

or duration of loss of consciousness (530 min). The structural MRI

scans (see below) were inspected and classified by an experienced

neuroradiologist (S.S.) using the scheme of Adams et al. (1989),

which allows the identification of three grades of diffuse axonal injury (Table 1). In Grade 1 there is histological evidence of axonal injury in the white matter of the cerebral hemispheres, the corpus callosum, the brainstem and, less commonly, the cerebellum; in Grade 2 there is also a focal lesion in the corpus callosum; and in Grade 3 there is additionally a focal lesion in the dorsolateral quadrant or quadrants of the rostral brainstem. The median severity score of the patients with TBI was 2 (range = 0–2.5). Demographic and neuro-logical variables are provided in Table 1. The 26 control subjects had normal medical histories and MRI scans were rated as being ‘normal’ by a radiologist. All subjects were right-handed (laterality quotient: TBI: mean = 81, range = 22–100; control: mean = 92; range = 60– 100) as verified by the Edinburgh Handedness Inventory (Oldfield, 1971). General motor performance was assessed using both bilateral

items of the TEMPA (Desrosiers et al., 1995a) and the Purdue

Pegboard Test (Desrosiers et al., 1995b). The study was approved

by the local ethics committee for biomedical research. Informed con-sent was obtained from each patient or from the patient’s first degree relatives.

Behavioural testing

We used an established protocol applied previously in functional MRI

studies assessing young and elderly healthy subjects (Coxon et al.,

2010) and patients with TBI (Leunissenet al., 2012). In brief,

partici-pants were required to switch between coordination modes while making spatially and temporally coupled circular motions with joy-sticks. This task assesses cognitive control processes that are particu-larly affected in patients with TBI (Powell and Voeller, 2004; Cicerone et al., 2005; Reeset al., 2007). For each hand, the direction of circling could be either clockwise or counter-clockwise. An auditory metro-nome was used for pacing such that participants completed one circle per tone. Four possible bimanual movement patterns were

intro-duced: inward circles, outward circles, clockwise circles and

counter-clockwise circles. A constant visual display was shown com-prising two white circles, each with a curved white bar immediately above, and a central fixation cross on a black background. Instruction cues were conveyed by arrows, visible for 800 ms (Fig. 1). Green arrows were used as imperative cues. The initial pattern to adopt was indicated by two green arrows. A single green arrow pointing opposite to the actual movement direction indicated that the right hand must change direction (Switch). This resulted in a change of movement pattern from asymmetric to symmetric circling or from symmetric to asymmetric circling. A single green arrow pointing in the same direction as the actual movement indicated that no change was necessary for the right hand (Continue). In both cases, the left hand was to maintain moving in the already established direction. White arrows indicated the correct movement pattern 4 s after the imperative cue (Confirm). For participants, this served as confirmation that they were performing the correct pattern or it provided an oppor-tunity to correct their error.

Subjects performed five trial blocks, each 351 s in duration. A block comprised alternating movement (68 s) and rest (20 s) epochs to avoid fatigue. There were 14 switch trials, 14 continue trials, 4 reverse trials and 32 confirmation trials per block. Reverse trials were signalled by two green arrows, cueing a switch of both hands. They were included to change left hand direction within a movement epoch, but were of no interest for the analysis. One of the four possible movement pat-terns was used as an initial pattern for each movement epoch and the order was counterbalanced across runs. In total, there were 70 Switch trials of the right hand and 70 Continue trials.

MATLAB 7.7 (Mathworks) was used to analyse the kinematic data of the switching task. For each hand, a continuous estimate of angular

velocity was determined by!= d/dt with = arctan(x/y), where x

andy describe the mean corrected values for vertical and horizontal

joystick displacements, respectively. A second-order Butterworth

low-pass filter (cut-off 5 Hz) was applied to!. Two dependent variables

were of primary interest. First, for trials where the right hand changed direction, switch response time was determined as the interval

be-tween stimulus onset and the first zero crossing of!, indicating that

cycling direction had reversed. This measure is an indication of the speed with which an ongoing coordination pattern is suppressed and a different coordination pattern is adopted. Second, the percentage of contralateral disruptions was recorded when left-hand direction reversed until the correct pattern was displayed. This measure reflects to what extent the performer is able to perform the switch selectively with the right hand while the left hand continues its motion (i.e. div-ision of attention).

Functional MRI data acquisition and

pre-processing

A Siemens 3 T Magnetom Trio MRI scanner with standard head coil

was used for image acquisition. For all subjects, a high resolution T1

-weighted structural image was acquired using magnetization prepared

(4)

Table 1

Summary of demographic and injury characteristics for the TBI group

TBI Patient # Age (years, months)/gender Age at injury (years, months)

Cause of injury Acute scan within 24 h after injury, Lesion location/pathology

MRI scan at examination, Lesion location/ pathology

Grade

TBI 1 19,2/F 15,8 Traffic accident (L) PL/OL contusion, subdural haematoma (R) PL contusion, corpus callosum and OL shearing injuries, occipital horn lateral ventricle asymmetrically enlarged

2

TBI 2 27,6/F 25,2 Traffic accident TL contusion, (R) PL haemorrhage, (L) FL intraparenchymateus haemorrhagic contusion, subdural haematoma

Drain tract (R), (L) FL and TL contusion 2

TBI 3 22,9/F 21,3 Traffic accident (R) FL haemorrhage, (L) FL/TL and (L) PL and (R) orbito-frontal cortex contusion

Drain tract (R), haemosiderin deposits (R) PL and (R) orbito-frontal cortex

1 TBI 4 22,5/M 17,6 Traffic accident (L) FL shearing injuries, splenium and body

corpus callosum contusion

(R) FL contusion 1

TBI 5 28,1/M 18,6 Traffic accident FL contusion, (L) FL subdural haematoma, (L) TL and (R) PL haemorrhage

Drain tract (L), FL contusion 1.5 TBI 6 17,9/F 12,9 Traffic accident Contusion (location not specified in

available records)

– 0

TBI 7 16,1/F 8,2 Traffic accident (R) FL Subdural haematoma – 0

TBI 827,9/M 18,2 Traffic accident (R) OL/TL contusion Drain tract (L and R), haemosiderin deposits thalamus and (L) OL

1

TBI 9 27,2/M 20,9 Traffic accident (L) FL contusion, subarachnoidal bleeding, epi-dural haematoma, FL volume loss

Drain tract (L), wide extended FL and (R) OL/PL contusion, corpus callosum degeneration

2.5

TBI 10 34,6/M 28,9 Traffic accident (R) amygdale and basal ganglia and (R) PL haemorrhage, (L) FL inflammatory changes

(L) TL contusion 1

TBI 11 16,8/M 9,1 Traffic accident (L) TL and (L) FL punctiform and (R) mesen-cephalon contusion, (L) FL and (L) thalamus haemorrhagic injuries

Orbito-frontal cortex contusion, enlarged ventricles

1

TBI 12 33,8/M 27,9 Traffic accident Drain tract (R), thalamus injury, corpus cal-losum shearing injuries, (R) FL and (L) in-ferior FL and (R) OL contusion

2.5

TBI 13 26,9/F 23,9 Traffic accident FL injuries Drain tract (L), PL and OL/PL and FL and (R) TL shearing injuries, slightly enlarged ventricles

1

TBI 14 21,9/M 15,2 Traffic accident (L) FL contusion, (R) FL shearing injuries, bilat-eral haemosiderin deposits genu corpus cal-losum, diffuse axonal injuries

Superior and periventricular FL contusion, (R) orbito-frontal cortex and (R) OL shearing injuries

2.5

TBI 15 22,3/M 19,1 Traffic accident Contusion and DAI (location not specified in available records)

(L) thalamus and (L) TL and (L)

orbito-frontal cortex and (L) FL and (R) FL and central sulcus shearing injuries

2

TBI 16 31,7/M 29,6 Traffic accident (L) FL/TL haemorrhage and DAI, FL and TL/OL shearing injuries

TL and (R) orbito-frontal cortex and (R) in-ferior FL contusion, corpus callosum de-generation, asymmetric ventricles, (L) PL shearing injury

2

TBI 17 16,7/M 14,5 Fall Enlarged (R) lateral ventricle, (R) haematoma occipital horn lateral ventricle, hyperdensity (L) thalamus and PL/TL, (LH) shearing injuries

Drain tract (R), (L) corpus callosum and thalamus and (R) PL and (L) FL and (R) TL shearing injuries, occipital horn lateral ventricle asymmetrically enlarged

2

TBI 18 28,1/M 18,4 Traffic accident Hemosiderin deposits corpus callosum, DAI, ischaemic injury (L) occipital horn of lateral ventricle

Drain tract (R), (R) periventricular white matter FL and thalamus injuries, corpus callosum degeneration

1

TBI 19 27,9/M/ 24,9 Traffic accident (L) thalamus and (L) periventricular and corpus callosum and brainstem and TL shearing injuries

Drain tract, (L) thalamus and corpus callo-sum and (L) TL shearing injuries

2

TBI 20 30,9/M 28,3 Traffic accident Lesion and location not specified in available records

Drain tract (R), (L) inferior TL contusion, (L) anterior cingulus and (R) FL and central sulcus shearing injuries

2

TBI 21 24,1/M 21,8 Traffic accident (L) FL haematoma, FL parenchymal bleeding, subarachnoidal bleeding

Drain tract (R), orbito-frontal cortex and (L) cerebellum contusion

2 TBI 22 30,6/M 23,8 Traffic accident (R) intrathalamic haemorrhage, bilateral frontal

sheer haemorrhages, (R)

subarachnoid haemorrhage, (L) FL intracer-ebral haematoma

(L) inferior TL and wide extended FL contusion, thalamus injury

2.5

TBI 23 20,6/F 18,1 Traffic accident Diffuse axonal injuries, (L) FL/TL/PL subdural haematoma, FL contusion, injuries corpus callosum

FL and (R) PL contusion, orbito-frontal cortex shearing injuries, enlarged ventricles

2

F = female; FL = frontal lobe; L = left; M = male; OL = occipital lobe; PL = parietal lobe; R = right; TL = temporal lobe.

(5)

rapid gradient echo (MPRAGE; repetition time = 2300 ms, echo

time-= 2.98 ms, 111.1 mm voxels, field of view: 240256, 160

sa-gittal slices). Functional data (functional MRI) were acquired with a

descending gradient echo planar imaging (EPI) pulse sequence for T2

*-weighted images (repetition time = 3000 ms, echo time = 30 ms, flip

angle = 90_{, 50 oblique axial slices each 2.8 mm thick, inter-slice gap}

0.028 mm, in-plane resolution 2.52.5 mm, 8080 matrix).

In preparation for functional network analysis, the blood oxygen

level-dependent time series were passed through several

pre-processing steps using the SPM 5 software package (Wellcome Department of Imaging Neuroscience, University College, London) implemented in MATLAB 7.7 (Mathworks).

T1-weighted structural images were spatially normalized into

stand-ard Montreal Neurological Institute (MNI) space using the unified seg-mentation algorithm, which is less prone to normalization artefacts induced by changes in tissue signal (Ashburner and Friston, 2005;

Crinionet al., 2007). The unified segmentation algorithm is a

genera-tive model that combines tissue segmentation, bias correction and spatial normalization in the inversion of a single unified model (Ashburner and Friston, 2005). To further refine the unified

segmen-tation procedure for patients with clearly visible focal lesions (n= 9), a

lesion mask was manually drawn in MRIcron (Chris Rorden, University of South Carolina), smoothed with a Gaussian kernel of 6 mm full-width at half-maximum and used as a cost function mask. The

registration process simply ignores the information within the mask and prevents contribution of lesion voxels to the normalization pro-cedure. This combination of the unified segmentation procedure and cost function masking outperforms cost function masking alone

(Crinionet al., 2007).

The first three functional images of each subject’s dataset were

dis-carded to allow for T1equilibration. The remaining images were

spa-tially realigned to the first image in the time series, then corrected for differences in slice acquisition time by temporal interpolation to the middle slice (reference slice = 25). Functional images were spatially co-registered to the anatomical image, and normalized using the same transformation matrix applied to the anatomical image. Finally, the normalized functional images were smoothed with an isotropic 10 mm full-width at half-maximum Gaussian kernel.

The ensuing pre-processed functional MRI time series were analysed on a subject-by-subject basis using an event-related approach in the

context of the General Linear Model (Fristonet al., 1995). Events were

specified at the time of cue onset and modelled as delta functions convolved with the canonical haemodynamic response function and its temporal derivative. Four conditions were specified: correct Switch trials, correct Continue trials, Confirmation, and Rest. Switch condi-tions were parametrically modulated by response time. An additional regressor modelled events of no interest such as Initiation trials, Reversals, Switch trials with complete contralateral disruptions, and

Figure 1

Setup of motor switching task. (A) Subject position and custom MRI compatible joysticks are shown. Visual stimuli were

projected onto a screen that occluded vision below the shoulders and were viewed via a 45 _{mirror. Two circles and a central fixation}

cross were always visible. Imperative (green arrows) and confirmation cues (white arrows) were presented for 800 ms, separated by 4 s. Switching from symmetric to asymmetric (SW»ASYMM) or from asymmetric to symmetric (SW»SYMM) movement patterns required the right hand to change circling direction: clockwise (CW) or counter-clockwise (CCW). (B) Representative trace from one participant performing symmetric circles followed by a switch to asymmetric circles (LH = left hand; RH = right hand).Top: Left- and right-hand joystick displacement;bottom: radial velocity plotted against time (positive radial velocity = clockwise circles, negative radial velocity = counter-clockwise circles). The trace segments in dark grey highlight the right-hand switch and contralateral disruption

of the left hand. SwRT = switch response time. by guest on October 19, 2015

(6)

their subsequent correction on Confirmation trials. In addition, realign-ment parameters were included as covariates of no interest to correct for confounding effects of head movement. Translational motions did not exceed 1 voxel (TBI: mean = 0.55 mm, SD = 0.31; Controls: mean = 0.42, SD = 0.16). The intensity changes attributable to global signal noise due to changes in the magnetic field over time were ac-counted for by using the time series of the mean signal from the white

matter, CSF and out of brain voxels (Verhagen et al., 2008). Data

were filtered in the temporal domain using a high-pass cut-off of 1/ 128 Hz, and global differences in blood oxygen level-dependent signal were removed by scaling to the grand mean. For each subject, the

contrast of Switch4Continue was created (by applying weights to the

canonical haemodynamic response function for each condition). Both Switch and Continue conditions consisted of identical visual cues, but only the Switch trials required a change of movement plan.

Construction of regions of interest

Regions of interest were defined based on the previously published event-related functional MRI studies of young adults, elderly and

pa-tients with TBI performing the switching task (Coxon et al., 2010;

Leunissenet al., 2012). Twenty-two 10-mm diameter spheres centred

on peak coordinates from the activation data (clusters) from both

studies (Coxon et al., 2010; Leunissen et al., 2012) were used to

define the switching-related areas. The areas constituting this network included the medial frontal cortex (supplementary motor area: pre-supplementary motor area and supplementary motor area-proper), anterior cingulate cortex, bilateral dorso-lateral prefrontal cortex (DLPFC), inferior frontal cortex (Brodmann area 44), basal ganglia (globus pallidus, putamen, subthalamic nucleus region), bilateral cere-bellum (lobule VI), right precuneus, left premotor cortex (dorsal and ventral), bilateral insula, and right superior and right inferior parietal

lobules. The majority of these regions have been identified as reflect-ing a ‘core’ network for the implementation of task sets (Dosenbach et al., 2006). Thus, the switching motor network in the current study included 22 brain areas and is listed in Table 2.

Construction of functional networks

We constructed a functional network for each subject using the fol-lowing multi-step procedure. First, we extracted the functional MRI time series. Second, we defined and calculated measures of connect-ivity. Next, a criterion was defined to select connections. Finally, graph measures were computed. These steps are described in detail in the following paragraphs.

The first step in all functional network analyses was to extract the blood oxygen level-dependent average time series for each region of

interest in each subject for the contrast Switch4Continue for the

runs. The average time series was taken as the first eigenvariate from the singular value decomposition of a matrix composed of each time series from all voxels within the node-sphere, the time series of all voxels of a region of interest was element-wise averaged, so as to obtain a single average time series for each region of interest. Within each region of interest, voxels which showed insufficient

acti-vation to exceed a threshold ofP50.001 (uncorrected) for the

con-trast Switch4Continue were not included in calculating the average

time series for each region of interest. Based on the average time series data, the network connectivity was then determined by calculat-ing undirected matrices of partial correlations between the regions of interest, quantifying the unique relationship between each pair of re-gions of interest. We used partial correlations to reduce indirect

dependencies by other brain regions (Hampson et al., 2002; Liu

et al., 2008). Amongst all methods of evaluating functional interde-pendencies between functional MRI time courses in different regions

Table 2

Cortical and subcortical regions defined as core switching network

Region name Hemisphere MNI coordinates

1 Anterior rostral cingulate cortex Right 8 18 35

2 Inferior frontal gyrus: pars opercularis (BA44) Left 48 5 25

3 Inferior frontal gyrus: pars opercularis (BA44) Right 53 10 20

4 Cerebellum (VI) Left 33 40 38

5 Cerebellum (VI) Right 35 53 30

6 Middle frontal gyrus: Dorsolateral prefrontal cortex Right 35 43 30

7 Middle frontal gyrus: Dorsolateral prefrontal cortex Left 38 35 28

8 Insula lobe Left 40 13 3

9 Insula lobe Right 45 18 0

10 Inferior parietal lobule Right 45 35 48

11 Pallidum Left 18 5 0

12 Pallidum Right 18 10 0

13 Precentral gyrus: ventral premotor cortex Left 53 3 40

14 Precuneus Right 13 60 55

15 Putamen Right 18 8 5

16 Putamen Left 18 8 5

17 Superior frontal gyrus: Dorsal premotor cortex Left 28 8 58

18 Medial superior frontal gyrus: preSMA

(Supplementary motor area) (BA6)

Right 0 10 53

19 Medial superior frontal gyrus: SMA proper (BA6) Left 3 10 55

20 Superior parietal lobule Right 40 45 60

21 Subthalamic nucleus Left 8 13 8

22 Subthalamic nucleus Right 10 13 10

BA = Brodmann area.

(7)

of interest, partial correlations have been found to provide the most

reliable approaches (Smithet al., 2011). The calculation of the matrix

of partial correlations makes use of the inverse of the sample covari-ance matrix. Specifically, the partial correlation matrix is a symmetric matrix in which each off-diagonal element is the correlation coefficient between a pair of variables after filtering out the contributions of all other variables included in the dataset. In the present study, therefore, the partial correlation between any pair of regions filters out the ef-fects of the remaining 20 brain regions. The procedure for obtaining partial correlation values here is consistent with other work (Salvador et al., 2005; Liuet al., 2008; Nakamuraet al., 2009).

Graph theoretical analysis

A binary graph (i.e. network), consisting of 22 nodes (brain regions) and undirected edges (functional connectivity) between nodes, was achieved by testing the null hypothesis that the partial correlation co-efficient was significantly different from zero between any region pairs. Two regions were considered functionally connected if their

par-tial correlation coefficient was significant atP= 0.0001, 0.0005, 0.001,

0.005 and 0.01, respectively, to ensure that the statistical results with regard to group effects between patients with TBI and controls were stable and reliable at different connection densities. It has been shown that manipulating the connection density in a network by varying the number of valid network connections can have a noticeable impact on

graph theoretical metrics (Van Wijket al., 2010). Hence, rather than

thresholding the connectivity matrices at only one arbitraryP

-thresh-old, we repeated the graph theoretical analyses across five threshold values. In each case, the thresholding of the connectivity matrices resulted in binary matrices where existing (valid) connections carried a value of 1 while the absence of a functional connection between network nodes was designated by a value of 0. Self-connections of nodes were not included in the analyses.

In addition to the binary matrices, we also calculated weighted matrices, where for each valid functional connection between a pair of nodes, the value of 1 in the binary matrix was replaced by the value of the partial correlation for this node-pair, with the partial correlation then representing a proxy measure for the weight/strength in connec-tion between this pair of nodes. This approach allowed us to examine connection strength between node-pairs and to calculate a mean con-nection strength for each node defined as the sum of its concon-nection weights (i.e. the sum of partial correlations for valid connections) with other nodes in the network. Subsequently, the mean connection strength of a network was then defined as the mean connection strength across all its network nodes.

We investigated the topological properties of the functional network at the global and regional (nodal) levels using the Brain Connectivity Toolbox (Rubinov and Sporns, 2010), quantifying the global network architecture in terms of the small-worldness, normalized clustering co-efficient, normalized path length, density, and global efficiency. We described the regional properties in terms of degree, strength, cluster-ing coefficient, path length and betweenness centrality. Based on the constructed functional network, we looked for significant differences in global and nodal properties between the patients with TBI and con-trols. We only provide brief, formal definitions of each of the network properties used in this study in the Supplementary material.

Statistical analysis

For the outcome measures of the switching task and the clinical tests of bimanual coordination (TEMPA, Purdue Pegboard Task),

two-sample t-tests were conducted for comparing the TBI with the

control group.

Statistical comparisons of all network measures (i.e. normalized clus-tering coefficient, normalized path length, degree, strength, density and local efficiency) between the two groups were performed by

using two-samplet-tests. Network measures were confirmed to have

a normal distribution using the Shapiro–Wilks Normality Test. If any change in the local topological properties was found between the two groups, we investigated the regions that showed significant differences in these topological properties. Bonferroni corrections for multiple

comparisons were made, hencePcorr50.002 was considered

signifi-cant following correction for the node-specific analyses regarding the 22 regions of interest. To determine whether the network topology in TBI was correlated with performance on the switching task, we calcu-lated Pearson’s correlation coefficients of the network parameters against Switch response time and percentage of contralateral disrup-tions. The grades of diffuse axonal injury were also used to assess relationships between the topological properties and severity score. To this end, we applied the Spearman’s rank correlation within the TBI group. Finally, to point out whether correlations between blood oxygen level-dependent response and switching performance are mir-rored in similar associations between the graph theoretical network measures and switching task performance on one hand and injury severity on the other hand, correlation analyses were conducted, in which the degree of two regions (pre-supplementary motor area and left inferior frontal gyrus) of the overactivation network was correlated with the behavioural parameters separately. Due to the number of correlations examined, only correlations below a statistical significance

level of P50.01 were considered significant. All statistical analyses

were performed with the Statistica software (StatSoft, Inc).

Results

Bimanual switching task performance

The kinematic results of the cognitive control task showed that the patient group took longer to implement pattern change via the right hand [t(47) = 4.35, P50.001; patients with TBI 76427 ms, controls 64529 ms], and resulted in more com-plete contralateral disruptions than the control subjects [t(47) = 2.53,P50.01; patients with TBI 6.295.96%, controls 2.933.03%]. Moreover, baseline motor differences between patients with TBI and controls were measured, using both bilateral items of the TEMPA and the Purdue Pegboard Test. The TBI group scored on average worse than the control group on the number of pegs in the bimanual condition of the Purdue Pegboard Test [t(41) = 4.17, P50.001; mean for controls: 13 pegs; mean for patients with TBI: 11 pegs], and on the summary score for the TEMPA bimanual tasks [t(41) = 5.56, P50.001; mean for controls: 9.89 s; mean for patients with TBI: 13.03 s], confirming the TBI group lower mean motor performance levels. However, no effect of these variables (as potential confounding factors) was found on the switch cost in our group analyses (all

P-values40.05). Therefore, we suggest that these switching problems in our patients with TBI rather reflect cognitive control deficits than merely being an expression of baseline motor differences.

(8)

Graph theory analyses at different

connectivity thresholds

The graph theory analyses were repeated across five connectivity threshold levels, i.e. for each subject, graph theory analyses were calculated based on connectivity matrices thresholded at

P= 0.0001, 0.0005, 0.001, 0.005 and 0.01 to define valid func-tional connections. This revealed that the statistical results of topo-logical changes between patients with TBI and controls were stable and essentially unaffected by manipulating the network density, reinforcing the robustness of the obtained findings (see Supplementary material). Overall, the results showed that the topological architecture of the functional networks was signifi-cantly altered in patients with TBI. These changes in patients with TBI included higher connectivity degree and strength, and higher values of network density and local efficiency. Similarly, the ‘small-world’ nature remained stable in both groups across all assessment points. Because the statistical results were equiva-lent across all network densities, we chose to report the data based on the P= 0.001 threshold (i.e. the threshold in the centre of the examined threshold range) as representative descrip-tion of TBI-induced changes in funcdescrip-tional connectivity in the as-sessed network.

Also, no significant effect of gender was found in our analyses (all P-values40.05). Therefore, the graph theory analyses were conducted without this variable as a potential confounding factor.

Overall topology of the functional

brain network

Both patients and healthy controls showed a small-world organ-ization of the functional brain network expressed by a41 (TBI

group: mean = 1.43, SD = 0.30; control group: mean = 1.42, SD = 0.26) and1 (TBI group: mean = 1.00, SD = 0.04; control group: mean = 1.01, SD = 0.04). Furthermore, the normalized clustering coefficient did not differ between the patients with TBI and healthy controls, nor did the overall normalized path length (P= 0.90 for , P= 0.74 for ). In other words, pa-tients with TBI displayed - and -values close to the values of the brain networks of healthy controls, suggesting an intact overall organization of the functional brain network in patients with TBI.

Between-group differences in functional

connectivity

Group comparisons revealed an increased level of connectivity degree [t(47) = 2.40, P50.05] in patients with TBI. Specifically, patients showed higher connectivity degree in left ventral pre-motor cortex (Pcorr50.002). Node specific values of connectivity

degree are presented in Fig. 2A and B. A significant effect for group was also seen on density, representing the total ‘wiring cost’ of the network. Density showed significantly higher values in patients with TBI (mean = 0.28, SD = 0.03) compared with con-trols (mean = 0.26, SD = 0.04) [t(47) = 2.40,P50.05].

A significant effect for group was also seen on the weighted connectivity strength, calculated as the sum of connection weights (i.e. the sum of partial correlations for valid connections) of a node with other nodes in the network. Connection strength showed significantly higher values in the TBI group (mean = 1.60, SD = 0.16) compared with controls (mean = 1.46, SD = 0.17) [t(47) = 3.11, P50.01]. Compared with controls, TBI children showed significantly higher connectivity strength in the left ventral premotor cortex (Pcorr50.002).

Figure 2

Group differences in connectivity degree. (A) Controls and (B) patients with TBI. Size of the regions of interest (spheres)

represents connectivity degree. The edge widths represent the strength of the partial correlations between nodes. Notably, the networks shown here were constructed by summing the partial correlation matrices of all subjects in each group (as shown by the colour bar). Yellow sphere indicating higher connectivity in left ventral premotor cortex (Pcorr50.002) in patients with TBI.

(9)

Identification of hubs

To identify the hub regions, we calculated the betweenness cen-trality for each node of each subject’s functional network. Then we calculated the mean betweenness centrality of each node by averaging across subjects for each group. For the TBI group, the identified hub nodes included right dorsolateral prefrontal cortex, right insula lobe and left dorsal premotor cortex (Fig. 3B). The hubs for the control group included only the right insula lobe (Fig. 3A).

Between-group differences in measures

of functional segregation

Results revealed higher values of local efficiency between the TBI and control group [t(47) = 2.29,P50.05].

Relationship between cognitive control

of action and network properties in

patients with traumatic brain injury

We considered only degree as network property in these analyses because it has been suggested to be the most important measure of network analysis and it has a straightforward neurobiological interpretation (Rubinov and Sporns, 2010): nodes with a high degree are functionally interacting with many other nodes in the network.

A significant positive correlation was found between mean con-nectivity degree of the network in the TBI group and percentage of contralateral disruptions (r= 0.42, P50.05, Fig. 4). In other words, increase in connectivity degree was associated with

poorer switching task performance (i.e. more contralateral disrup-tions). Percentage of contralateral disruptions is a measure of gen-eralized and undifferentiated responding, reflecting a lack of focal switching with irradiation of the response to the contralateral hand. No significant results were obtained with the other depend-ent variable, i.e. Switch response time. Moreover, no significant results were obtained within the control group. Similar results were obtained using different graph organizational measures, for ex-ample increased efficiency was associated with lower performance on the switching task.

In the primary functional MRI analysis, two regions (pre-supplementary motor area and left inferior frontal gyrus) of the overactivation network showed a significant correlation be-tween blood oxygen level-dependent activation and switching performance (Leunissen et al., 2012). The associations between switching performance and graph measures in these two areas revealed that percentage of contralateral disruptions was positively correlated to connectivity degree of left inferior frontal gyrus (r= 0.55, P50.01) (Fig. 5). No correlations with switching per-formance were found in the pre-supplementary motor area.

Relationship between injury severity

and network properties in patients

with traumatic brain injury

Regarding the injury severity (according to Adams et al., 1989), we observed a significant positive correlation between the sever-ity score and connectivsever-ity degree (r= 0.50, P50.01). Thus, the patients with more severe TBI showed a higher connectivity degree.

Figure 3

The hub regions distributed in the functional networks of the control (A) and TBI group (B). The hub nodes are shown in yellow

with node sizes indicating their betweenness values. For the TBI group, the identified hubs included right dorsolateral prefrontal cortex, right insula lobe, and left dorsal premotor cortex. The hub for the control group included only the right insula. Notably, the networks shown here were constructed by summing the partial correlation matrices of all subjects in each group (as shown by the colour bar). The edge widths represent the strength of the partial correlations between nodes.

(10)

Discussion

This is the first time that graph theoretical approaches reveal unique associations between network properties and behavioural functions in patients with TBI. Our results show altered functional connectivity in the networks of patients with TBI. In particular, compared with controls, patients with TBI show increased func-tional connectivity, and higher values of local efficiency, represent-ing adaptive mechanisms. Furthermore, the increased connectivity degree is significantly correlated with poorer switching task per-formance and more severe brain injury. It is clear from the present results that assessment of the functional brain networks provides new insights into the mechanism underlying cognitive control def-icits after brain injury.

Behavioural switching deficits

in patients with traumatic

brain injury

Patients with TBI demonstrated difficulties with cognitive control, presenting slower response times and more disruptions during the switching task. There is only limited previous evidence of impaired switching performance during bimanual coordination following TBI, despite the fact that slowing of motor functions is a frequent clinical outcome in these patients (Levinet al., 1990). More spe-cifically, the patients took longer to implement pattern change via the right hand and showed more left-hand reversals than the con-trol subjects. These results are consistent with previous studies using different cognitive task-switching paradigms in TBI (Larson

et al. 2006; Schroeter et al. 2007). These switching deficits in patients with TBI may be related to attentional problems when transferring between different motor tasks (Parket al., 2009). In other words, patients with TBI have difficulty shifting attention in order to perform on-line modification of motor patterns during the execution of complex movements.

Group differences in functional brain

networks

The human brain is a large, dynamic system with an optimal bal-ance between local specialization and global integration. Although small-world properties were present for both the normal control and TBI groups, the topological architecture of the functional net-works was significantly altered in patients with TBI.

First, an increase in connectivity degree and strength was con-sistently observed in patients with TBI, implying that their network connections were relatively denser than in controls. Connectivity degree increase has also been reported in studies of brain tumours (Bartolomei et al., 2006a,b; Bosma et al., 2008). Node-specific analyses showed increased degree and strength of functional con-nectivity in the left ventral premotor cortex, suggesting that this area may play an important role in the switching network in pa-tients with TBI compared with controls. The premotor cortex is composed of multiple areas, including the dorsal premotor area and ventral premotor area (Dum and Strick, 1991). The increase in connectivity degree was located within the ventral premotor area close to the M1 hand area. The ventral premotor area is densely interconnected with primary motor cortex and is involved in coordination (Sadato et al., 1997; Stephan et al., 1999; Debaere et al., 2001; Ehrsson et al., 2002). The strengthened connectivity degree in this region may reflect a functional com-pensation. However, a higher degree can also point to high levels of energetic cost indicative of an overcharged network that is unbalanced in the transmission versus energy consumption trade-off, as recently observed in a longitudinal study in patients with TBI using resting-state magnetoencephalogram recordings (Castellanos et al., 2011). Examination of the density values indeed revealed an increase in total ‘wiring cost’ of the network for the TBI group.

Second, our patients with TBI showed strong increases in local efficiency. The local efficiency is predominantly related to short-range connections between neighbouring nodes. The under-lying mechanisms of increased local efficiency of a network have

Figure 4

Plot indicating the relationship between percentage

of contralateral disruptions and connectivity degree of the whole network.

Figure 5

Plot indicating the relationship between percentage

of contralateral disruptions and connectivity degree of the left inferior frontal gyrus.

(11)

been widely discussed in various studies. For example, De Vico Fallani et al. (2007) reported that increased local efficiency in spinal cord injured patients could be attributable to functional re-organization. Latora and Marchiorio (2001) indicated that higher values of local efficiency suggest a larger level of internal organ-ization and fault tolerance in the face of external attack. Similarly, the higher local efficiency observed in TBI may imply that the network tends to form clusters of regions of interest which pre-serve efficient communication and this may be indicative of com-pensatory mechanisms.

The increased functional connectivity of the patients with TBI appears consistent with previous studies of brain injury. For ex-ample, Nakamura and colleagues (2009) showed comparable topological changes in six patients with severe TBI using weighted networks from resting state functional MRI. Specifically, they observed higher efficiency, lower path length, higher strength and a lower small-worldness in patients with TBI compared with control subjects. Similarly, analysis of weighted networks derived from resting-state magnetoencephalogram data revealed that pa-tients with TBI (n= 15) showed increased connectivity as shown by increased strength, decreased path length, increased efficiency and an increased clustering coefficient (Castellanoset al., 2011). Furthermore, several studies in patients with TBI have examined already how network connectivity progresses over time using a longitudinal design, showing connectivity increases shortly after traumatic injury followed by decreases during recovery (even restoring towards control values) (Nakamura et al., 2009; Castellanoset al., 2010, 2011). One piece of information missing from these previous studies is how the brain network metrics relate to cognitive and motor task performance to assess their associations with behaviour (Bullmore and Sporns, 2009). Here, we were able to correlate network parameters with switching be-haviour in patients with TBI for the first time.

Importance of specific brain regions

for network functionality

Functional network analysis estimated for TBI and controls re-vealed that both groups exhibit hubs. In particular, insular lobe acted as ‘hub’ in both groups, implying that removal of insula from the estimated patterns will cause a collapse of the whole functional network. The insula is situated within the lateral sulcus. This area is involved in discerning external from internal bodily signals (Platel et al., 1997; Bamiou et al., 2003; Thaut, 2003; Lewis et al., 2004; Heuninckx et al., 2005; Goble et al., 2011). Accordingly, the observed increased normalized between-ness in the insula may suggest that both groups relied heavily on the external visual signal to trigger coordination changes. The insula has also been identified as hub in a previous study using structural networks (Iturria-Medina et al., 2008). In the latter study, anatomical connection probabilities were estimated from diffusion tensor imaging techniques. A similar betweenness cen-trality analysis allowed identifying the most indispensable and crit-ical anatomcrit-ical areas: putamen, precuneus, insula, superior frontal and superior parietal region. These structural findings complement

our functional findings and support the notion of the insula as being one of the most vulnerable focal areas.

Two hub regions, the right dorsolateral prefrontal cortex and the left dorsal premotor cortex, presented as hubs in the TBI group only and support the increased penetration of cognition into action control. The dorsolateral prefrontal cortex receives visual, somatosensory and auditory information from the occipital, temporal and parietal cortices (Goldman-Rakic and Schwartz, 1982; Barbas and Pandya, 1989; Seltzer and Pandya, 1989; Pandya and Yeterian, 1990; Petrides and Pandya, 1999) and has preferential connections with the motor system structures (Miller and Cohen, 2001). Accordingly, the dorsolateral prefrontal cortex is hypothesized to play a central role in the cognitive control of motor behaviour (Miller and Cohen, 2001). The dorsolateral pre-frontal cortex is involved in replanning and generation of new movements, in task rehearsal, and in attention to action (Deiber

et al., 1991; Debaere et al., 2004; Puttemans et al., 2005). Therefore, assigning an increased importance to the dorsolateral prefrontal cortex in the functional network in patients with TBI may allude to less automatized movement generation. TBI most commonly disrupts frontal systems and therefore executive control processes (Hillaryet al., 2002). Moreover, Alstott and colleagues (2009) demonstrated that the target attack over the frontal lobes induces a severe disruption of the network.

The other hub region, the dorsal premotor cortex has been divided into a rostral and caudal part, i.e. the pre-dorsal premotor cortex and dorsal premotor cortex proper (Picard and Strick, 2001). In the present study, the region of interest was located within the caudal part of the dorsal premotor cortex. The dorsal premotor cortex proper, known to have direct projections to the spinal cord and primary motor cortex (Dum and Strick, 1991), is mainly involved in movement preparation and execution (Toni

et al., 1999; Picard and Strick, 2001; Thoenissen et al., 2002). Although no study has investigated structural or functional changes in the premotor cortex in patients with TBI, we speculate that the increased betweenness centrality of the dorsal premotor cortex in the functional network demonstrates the functional im-portance of higher control of bimanual coordination for the pa-tients with TBI. Overall, the emergence of dorsal premotor cortex and dorsolateral prefrontal cortex as hub regions in the TBI but not in the control group suggests that the patients with TBI relied on increased cognitive control and performance monitoring. This analysis of the hub-status of the brain regions in the switching network reveals new insights on the potential importance of cer-tain brain regions that go beyond mere identification of functional activation afforded by standard functional MRI.

Network properties and its relation to

cognitive control and traumatic brain

injury severity

To our knowledge, this is the first time that associations between cognitive control disabilities and functional reorganization in pa-tients with TBI from a network perspective have been reported. Increased connectivity degree was significantly correlated with percentage of contralateral disruptions, indicating that those

(12)

patients with TBI who showed higher connectivity degree dis-played lower switching task performance. Moreover, the patients’ severity score was also correlated with the connectivity degree, suggesting that more severe head injury was associated with a more profound change in network topology.

As discussed above, the increase in connectivity degree might have a hidden cost, to the extent that higher connectivity degree is partly non-functional, reflecting difficulties in recruiting specia-lized neural mechanisms and, as a result, may be associated with lower task performance (Hillary, 2008). Although an adaptation interpretation of higher connectivity degree in TBI adults is appeal-ing, it is clearly not the whole story. Subjects with TBI rely more on ‘cognitive reserve’ (as shown by the hub regions) and are thus more likely to reach a limit on the resources that can be brought to bear on task performance. We suggest that the better func-tional network cohesion in the TBI group may be an effect directly related to a poorer neurobiological substrate, i.e. structural discon-nection between brain areas. The increased functional connectivity in patients with TBI suggests a simplified functional system adjust-ing to a disrupted neurobiological substrate. If the integrity of the neurobiological system is disrupted, it is less versatile in coping with diversity in the activity levels among its network nodes. In other words, to manage decreased structural resources, the brain resorts to a simplified functional architecture characterized by increased functional coherence between network areas. As a result, this increased functional connectivity constrains the brain areas into synchronized activity. This may hamper behavioural flexibility, i.e. flexible coupling and decoupling between brain areas to produce goal-directed behaviour and adjustments to en-vironmental contingencies. This interpretation is consistent with previous research showing a reduced ability in patients with TBI to explore different functional states across brain regions (Hillary

et al., 2011; Medaglia et al., 2011). It would be intriguing in future studies to test directly how the structural network changes in TBI are associated with the alteration of functional brain net-works, by combining diffusion tensor imaging with functional MRI data.

Our focused correlation analyses showed that increase in con-nectivity degree in the left inferior frontal gyrus was associated with poorer switching performance, i.e. more contralateral disrup-tions. The inferior frontal cortex is a critical node in the response inhibition network (Duann et al., 2009). It is part of the ventral attention system, which features in the detection of salient, behaviourally relevant targets (Corbetta and Shulman, 2002; Corbetta et al., 2008). For instance, in a stop signal task, the right inferior frontal cortex activates to a higher extent during stop (or no-go) trials compared with go trials, because the stop signal is both salient, less frequent and behaviourally relevant, and it demands a change of response. Leunissen and colleagues (2012) found that right inferior frontal cortex activation was common in both controls and patients with TBI during successful motor switching. More importantly, functional MRI overactivation in the left inferior frontal gyrus was observed only in the patients with TBI, reflecting stimulus driven attention and inhibition of the ongoing movement. We suggest that the left inferior frontal cortex, in addition to manifesting increases in intensity of activa-tion after brain injury, also shows changes in the nature and

strength of its functional interdependence. However, further stu-dies are required to confirm this suggestion.

There are some limitations to this study. First, changes of func-tional network topology in the switching network focused on 22 predefined regions of interest. This selection may have resulted in the omission of some switching-related regions or the inclusion of some regions that were no longer switching-related following TBI. However, a recent study by Smith et al. (2011) provides strong indications that an elaborate functional MRI-based functional protocol is preferable to template-based approaches in order to minimize confounds and obtain a better picture on functional con-nectivity within active neural networks. Second, Pearson’s correl-ations were employed to estimate the relcorrel-ationship between switching performance and connectivity degree of the brain re-gions. Correction for multiple comparisons was used, but future studies are needed to determine the optimal methodology to con-trol for multiple testing in a network setting. Alternative correction methods have been suggested (Zaleskyet al., 2010).

Even though our results should be interpreted with caution, to our knowledge, this is the first report providing new evidence for cognitive control deficits in patients with TBI from a network per-spective. Although small-world properties were present for both patient and control networks, the topological architecture of the functional networks was significantly altered in patients with TBI. Specifically, patients with TBI showed increased connectivity degree, increased density, and higher values of local efficiency, possibly representing an adaptive strategy. Furthermore, connect-ivity degree was significantly correlated with switching perform-ance and severity score. These brain-behaviour relations pave the way for topology-based brain network analyses that may ultim-ately serve as biomarkers to improve TBI diagnostics/prognosis and for follow-up of cognitive deficits.

Supplementary material

Supplementary material is available atBrainonline.

Funding

Support for this study was provided through a grant from the Research Program of the Research Foundation - Flanders (FWO) (Levenslijn G.0482.010 and G.A114.11) and Grant P6/29 from the Interuniversity Attraction Poles program of the Belgian federal government. Caeyenberghs K. is funded by a postdoctoral fellow-ship of the Research Foundation - Flanders (FWO).

References

Adams JH, Doyle D, Ford I, Gennarelli TA, Graham DI, McLellan DR. Diffuse axonal injury in head injury: definition, diagnosis and grading. Histopathology 1989; 15: 49–59.

Alstott J, Breakspear M, Hagmann P, Cammoun L, Sporns O. Modeling the impact of lesions in the human brain. PLoS Comput Biol 2009; 5: e1000408.

Amos A. Remediating deficits of switching attention in patients with acquired brain injury. Brain Inj 2002; 16: 407–13.

(13)

Ashburner J, Friston KJ. Unified segmentation. Neuroimage 2005; 26: 839–51.

Azouvi P, Couillet J, Leclercq M, Martin Y, Asloun S, Rousseaux M. Divided attention and mental effort after severe traumatic brain injury. Neuropsychologia 2004; 42: 1260–8.

Bamiou DE, Musiek FE, Luxon LM. The insula (Island of Reil) and its role in auditory processing. Literature review. Brain Res Brain Res Rev 2003; 42: 143–54.

Barbas H, Pandya DN. Architecture and intrinsic connections of the prefrontal cortex in the rhesus monkey. J Comp Neurol 1989; 286: 353–75.

Bartolomei F, Bosma I, Klein M, Baayen JC, Reijneveld JC, Postma TJ, et al. Disturbed functional connectivity in brain tumour patients: evalu-ation by graph analysis of synchronizevalu-ation matrices. Clin Neurophysiol 2006a; 117: 2039–49.

Bartolomei F, Bosma I, Klein M, Baayen JC, Reijneveld JC, Postma TJ, et al. How do brain tumors alter functional connectivity? A magne-toencephalography study. Ann Neurol 2006b; 59: 128–38.

Bosma I, Stam CJ, Douw L, Bartolomei F, Heimans JJ, van Dijk BW, et al. The influence of low-grade glioma on resting state oscillatory brain activity: a magnetoencephalography study. J Neurooncol 2008; 88: 77–85.

Bullmore E, Sporns O. Complex brain networks: graph theoretical ana-lysis of structural and functional systems. Nat Rev Neurosci 2009; 10: 186–98.

Cao C, Slobounov S. Alteration of cortical functional connectivity as a result of traumatic brain injury revealed by graph theory, ICA, and sLORETA analyses of EEG signals. IEEE Trans Neural Syst Rehabil Eng 2010; 18: 11–9.

Castellanos NP, Paul N, Ordonez VE, Demuynck O, Bajo R, Campo P, et al. Reorganization of functional connectivity as a correlate of cognitive recovery in acquired brain injury. Brain 2010; 133: 2365–81.

Castellanos NP, Leyva I, Buldu JM, Bajo R, Paul N, Cuesta P, et al. Principles of recovery from traumatic brain injury: reorganization of functional networks. Neuroimage 2011; 55: 1189–99.

Cicerone KD, Dahlberg C, Malec JF, Langenbahn DM, Felicetti T, Kneipp S, et al. Evidence-based cognitive rehabilitation: updated review of the literature from 1998 through 2002. Arch Phys Med Rehabil 2005; 86: 1681–92.

Corbetta M, Shulman GL. Control of goal-directed and stimulus-driven attention in the brain. Nat Rev Neurosci 2002; 3: 201–15.

Corbetta M, Patel G, Shulman GL. The reorienting system of the human brain: from environment to theory of mind. Neuron 2008; 58: 306–24.

Coxon JP, Goble DJ, Van IA, De VJ, Wenderoth N, Swinnen SP. Reduced basal ganglia function when elderly switch between coordinated movement patterns. Cereb Cortex 2010; 20: 2368–79.

Crinion J, Ashbumer J, Leff A, Brett M, Price C, Friston K. Spatial nor-malization of lesioned brains: Performance evaluation and impact on fMRI analyses. Neuroimage 2007; 37: 866–75.

De Vico FF, Astolfi L, Cincotti F, Mattia D, Tocci A, Marciani MG, et al. Extracting information from cortical connectivity patterns estimated from high resolution EEG recordings: a theoretical graph approach. Brain Topogr 2007; 19: 125–36.

Debaere F, Swinnen SP, Beatse E, Sunaert S, Van HP, Duysens J. Brain areas involved in interlimb coordination: a distributed network. Neuroimage 2001; 14: 947–58.

Debaere F, Wenderoth N, Sunaert S, Van HP, Swinnen SP. Changes in brain activation during the acquisition of a new bimanual coodination task. Neuropsychologia 2004; 42: 855–67.

Deiber MP, Passingham RE, Colebatch JG, Friston KJ, Nixon PD, Frackowiak RS. Cortical areas and the selection of movement: a study with positron emission tomography. Exp Brain Res 1991; 84: 393–402.

Desrosiers J, Hebert R, Bravo G, Dutil E. The Purdue Pegboard Test: normative data for people aged 60 and over. Disabil Rehabil 1995; 17: 217–24.

Desrosiers J, Hebert R, Bravo G, Dutil E. Upper extremity performance test for the elderly (TEMPA): normative data and correlates with sen-sorimotor parameters. Test d’Evaluation des Membres Superieurs de Personnes Agees. Arch Phys Med Rehabil 1995; 76: 1125–9. Dosenbach NU, Visscher KM, Palmer ED, Miezin FM, Wenger KK,

Kang HC, et al. A core system for the implementation of task sets. Neuron 2006; 50: 799–812.

Duann JR, Ide JS, Luo X, Li CS. Functional connectivity delineates distinct roles of the inferior frontal cortex and presupplementary motor area in stop signal inhibition. J Neurosci 2009; 29: 10171–9.

Dum RP, Strick PL. The origin of corticospinal projections from the pre-motor areas in the frontal lobe. J Neurosci 1991; 11: 667–89. Ehrsson HH, Kuhtz-Buschbeck JP, Forssberg H. Brain regions

con-trolling nonsynergistic versus synergistic movement of the digits: a functional magnetic resonance imaging study. J Neurosci 2002; 22: 5074–80.

Ewing-Cobbs L, Prasad MR, Swank P, Kramer L, Cox CS, Fletcher JM, et al. Arrested development and disrupted callosal microstructure fol-lowing pediatric traumatic brain injury: relation to neurobehavioral outcomes. Neuroimage 2008; 42: 1305–15.

Friston KJ, Holmes AP, Poline JB, Grasby PJ, Williams SC, Frackowiak RS, et al. Analysis of fMRI time-series revisited. Neuroimage 1995; 2: 45–53.

Gale SD, Baxter L, Roundy N, Johnson SC. Traumatic brain injury and grey matter concentration: a preliminary voxel based morphometry study. J Neurol Neurosurg Psychiatry 2005; 76: 984–8.

Garth J, Anderson V, Wrennall J. Executive functions following moderate to severe frontal lobe injury: impact of injury and age at injury. Pediatr Rehabil 1997; 1: 99–108.

Goble DJ, Coxon JP, Van IA, Geurts M, Van HW, Sunaert S, et al. The neural basis of central proprioceptive processing in older versus younger adults: An important sensory role for right putamen. Hum Brain Mapp 2011; 22.

Godefroy O. Frontal syndrome and disorders of executive functions. J Neurol 2003; 250: 1–6.

Goldman-Rakic PS, Schwartz ML. Interdigitation of contralateral and ip-silateral columnar projections to frontal association cortex in primates. Science 1982; 216: 755–57.

Guye M, Bettus G, Bartolomei F, Cozzone PJ. Graph theoretical analysis of structural and functional connectivity MRI in normal and patho-logical brain networks. Magn Resonan Materials Phys Biol Med 2010; 23: 409–21.

Hampson M, Peterson BS, Skudlarski P, Gatenby JC, Gore JC. Detection of functional connectivity using temporal correlations in MR images. Hum Brain Mapp 2002; 15: 247–62.

Heuninckx S, Wenderoth N, Debaere F, Peeters R, Swinnen SP. Neural basis of aging: the penetration of cognition into action control. J Neurosci 2005; 25: 6787–96.

Hillary FG, Steffener J, Biswal BB, Lange G, DeLuca J, Ashburner J. Functional magnetic resonance imaging technology and traumatic brain injury rehabilitation: guidelines for methodological and concep-tual pitfalls. J Head Trauma Rehabil 2002; 17: 411–30.

Hillary FG. Neuroimaging of working memory dysfunction and the di-lemma with brain reorganization hypotheses. J Int Neuropsychol Soc 2008; 14: 526–34.

Hillary FG. Determining the nature of prefrontal cortex recruitment after traumatic brain injury: a response to Turner. Front Syst Neurosci 2011; 5: 24.

Iturria-Medina Y, Sotero RC, Canales-Rodriguez EJ, Aleman-Gomez Y, Melie-Garcia L. Studying the human brain anatomical network via diffusion-weighted MRI and Graph Theory. Neuroimage 2008; 40: 1064–76.

Kelso JA, Fuchs A. Self-organizing dynamics of the human brain: critical instabilities and Sil’nikov chaos. Chaos 1995; 5: 64–9.

Kennedy MR, Wozniak JR, Muetzel RL, Mueller BA, Chiou HH, Pantekoek K, et al. White matter and neurocognitive changes in adults with chronic traumatic brain injury. J Int Neuropsychol Soc 2009; 15: 130–6.

(14)

Kim YH, Yoo WK, Ko MH, Park CH, Kim ST, Na DL. Plasticity of the attentional network after brain injury and cognitive rehabilitation. Neurorehabil Neural Repair 2009; 23: 468–77.

Kinnunen KM, Greenwood R, Powell JH, Leech R, Hawkins PC, Bonnelle V, et al. White matter damage and cognitive impairment after traumatic brain injury. Brain 2011; 134: 449–63.

Kraus MF, Susmaras T, Caughlin BP, Walker CJ, Sweeney JA, Little DM.

White matter integrity and cognition in chronic traumatic

brain injury: a diffusion tensor imaging study. Brain 2007; 130: 2508–519.

Larson MJ, Perlstein WM, Demery JA, Stigge-Kaufman DA. Cognitive

control impairments in traumatic brain injury. J Clin Exp

Neuropsychol 2006; 28: 968–86.

Latora V, Marchiori M. Efficient behavior of small-world networks. Phys Rev Lett 2001; 87: 198701.

Lee L, Friston K, Horwitz B. Large-scale neural models and dynamic causal modelling. Neuroimage 2006; 30: 1243–54.

Leunissen I, Coxon JP, Geurts M, Caeyenberghs K, Michiels K, Sunaert S, et al. Disturbed cortico-subcortical interactions during motor task switching in traumatic brain injury. Hum Brain Mapp 2012; 30. Levin H, Kraus MF. The frontal lobes and traumatic brain injury.

J Neuropsychiatry Clin Neurosci 1994; 6: 443–54.

Levin HS, Gary HE Jr, Eisenberg HM, Ruff RM, Barth JT, Kreutzer J, et al. Neurobehavioral outcome 1 year after severe head injury. Experience of the Traumatic Coma Data Bank. J Neurosurg 1990; 73: 699–709.

Levin HS, Hanten G, Zhang L, Swank PR, Hunter J. Selective impairment of inhibition after TBI in children. J Clin Exp Neuropsychol 2004; 26: 589–97.

Lewis PA, Wing AM, Pope PA, Praamstra P, Miall RC. Brain activity correlates differentially with increasing temporal complexity of rhythms during initialisation, synchronisation, and continuation phases of paced finger tapping. Neuropsychologia 2004; 42: 1301–12.

Liu Y, Liang M, Zhou Y, He Y, Hao Y, Song M, et al. Disrupted small-world networks in schizophrenia. Brain 2008; 131: 945–61. Mecklinger AD, von Cramon DY, Springer A, Matthes-von CG. Executive

control functions in task switching: evidence from brain injured pa-tients. J Clin Exp Neuropsychol 1999; 21: 606–19.

Medaglia JD, Chiou KS, Slocomb J, Fitzpatrick NM, Wardecker BM, Ramanathan D, et al. The Less BOLD, the Wiser: Support for the latent resource hypothesis after traumatic brain injury. Hum Brain Mapp 2011.

Miller EK. The prefrontal cortex and cognitive control. Nat Rev Neurosci 2000; 1: 59–65.

Miller EK, Cohen JD. An integrative theory of prefrontal cortex function. Annu Rev Neurosci 2001; 24: 167–202.

Nakamura T, Hillary FG, Biswal BB. Resting network plasticity following brain injury. PLoS One 2009; 4: e8220.

Newsome MR, Steinberg JL, Scheibel RS, Troyanskaya M, Chu Z, Hanten G, et al. Effects of traumatic brain injury on working memory-related brain activation in adolescents. Neuropsychology 2008; 22: 419–25.

Oldfield RC. The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia 1971; 9: 97–113.

Pandya DN, Yeterian EH. Prefrontal cortex in relation to other corti