Dynamic functional reorganization of the motor execution network after stroke

Dynamic functional reorganization of the motor execution network after stroke
of 15
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
  BRAIN A JOURNAL OF NEUROLOGY Dynamic functional reorganization of the motor execution network after stroke Liang Wang, 1, * Chunshui Yu, 2,3, * Hai Chen, 4 Wen Qin, 3 Yong He, 1 Fengmei Fan, 1 Yujin Zhang, 1 Moli Wang, 4 Kuncheng Li, 3 Yufeng Zang, 1 Todd S. Woodward 5 and Chaozhe Zhu 1 1 State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University, Beijing 100875, People’s Republic of China2 Department of Radiology, Tianjin Medical University General Hospital, Tianjin 300052, People’s Republic of China3 Department of Radiology, Xuanwu Hospital of Capital Medical University, Beijing 100053, People’s Republic of China4 Department of Neurology, Xuanwu Hospital of Capital Medical University, Beijing 100053, People’s Republic of China5 Department of Psychiatry, University of British Columbia, Vancouver, BC V6T 2A1, Canada*These authors contributed equally to this work.Correspondence to: Chaozhe Zhu, PhD,State Key Laboratory of Cognitive Neuroscience and Learning,Beijing Normal University,Beijing 100875,People’s Republic of ChinaE-mail: Numerous studies argue that cortical reorganization may contribute to the restoration of motor function following stroke.However, the evolution of changes during the post-stroke reorganization has been little studied. This study sought to identifydynamic changes in the functional organization, particularly topological characteristics, of the motor execution network duringthe stroke recovery process. Ten patients (nine male and one female) with subcortical infarctions were assessed by neurologicalexamination and scanned with resting-state functional magnetic resonance imaging across five consecutive time points in asingle year. The motor execution network of each subject was constructed using a functional connectivity matrix between21 brain regions and subsequently analysed using graph theoretical approaches. Dynamic changes in topological configurationof the network during the process of recovery were evaluated by a mixed model. We found that the motor execution networkgradually shifted towards a random mode during the recovery process, which suggests that a less optimized reorganization isinvolved in regaining function in the affected limbs. Significantly increased regional centralities within the network wereobserved in the ipsilesional primary motor area and contralesional cerebellum, whereas the ipsilesional cerebellum showeddecreased regional centrality. Functional connectivity to these brain regions demonstrated consistent alterations over time.Notably, these measures correlated with different clinical variables, which provided support that the findings may reflect theadaptive reorganization of the motor execution network in stroke patients. In conclusion, the study expands our understandingof the spectrum of changes occurring in the brain after stroke and provides a new avenue for investigating lesion-inducednetwork plasticity. Keywords:  stroke; network; small-world; connectivity; functional magnetic resonance imaging doi:10.1093/brain/awq043 Brain 2010: 133; 1224–1238  |  1224 Received August 16, 2009. Revised December 30, 2009. Accepted February 1, 2010  The Author (2010). Published by Oxford University Press on behalf of the Guarantors of Brain. All rights reserved.For Permissions, please email:  b  y  g u e s  t   onD  e c  em b  er 1  9  ,2  0 1 4 D  ownl   o a d  e d f  r  om   Introduction Motor deficit is the most prominent symptom in ischaemic stroke,and spontaneous recovery of motor function has been observedduring the first several months after stroke onset (Duncan  et al .,2000). This recovery has been commonly attributed to corticalreorganization, which has been confirmed by the findings fromfunctional neuroimaging studies, including the increased recruit-ment of contralesional motor areas (Johansen-Berg  et al ., 2002;Small  et al ., 2002; Ward  et al ., 2003; Lotze  et al ., 2006; Calautti et al ., 2007), increased activity in non-primary motor areas(Chollet  et al ., 1991; Weiller   et al ., 1992; Tombari  et al ., 2004),and the focalization of ipsilesional sensorimotor areas (Feydy  et al .,2002; Jaillard  et al ., 2005) and language areas (Saur   et al ., 2006).Moreover, the changes in functional and effective connectivity(Friston, 1994), such as increased coherence over the contra- lesional hemisphere (Gerloff  et al ., 2006), increased task-relatedcorticocortical coupling (Strens  et al ., 2004) and decreasedbidirectional coupling between ipsilesional supplementary motor area and primary motor area (Grefkes  et al ., 2008), also implythe existence of the functional reorganization. The cortical reorga-nization hypothesis is also supported by structural neuroimagingstudies, in which the increased cortical thickness in the ipsilesionalsensorimotor areas was found (Schaechter   et al ., 2006). In addi-tion, the white matter reorganization has been demonstrated bystudies finding increased integrity of whole brain white matter (Wang  et al ., 2006). Despite these advances in the motor-relatedreorganization literature, little is known about the dynamicchanges in the integrative ability of the whole motor networkassociated with revealed alterations of both local brain activityand functional and anatomical connectivity, which can enhanceour understanding of functional reorganization for the motor restoration following stroke.In recent years, graph theory has been introduced as a novelmethod or studying functional networks in the central nervoussystem (for a recent review, see Bullmore and Sporns, 2009).This approach, based on an elegant representation of nodes (ver-tices) and links (edges) between pairs of nodes, describes impor-tant properties of complex systems by quantifying topologies ofnetwork representations (Boccaletti  et al ., 2006). Nodes inlarge-scale brain networks usually represent anatomically definedbrain regions, while links represent functional or effective connec-tivity. Functional connectivity corresponds to magnitudes oftemporal correlations in activity (Friston  et al ., 1993) and mayoccur between pairs of anatomically unconnected regions.Depending on the measure, functional connectivity may reflectlinear or nonlinear interactions (Zhou  et al ., 2009), which can beestimated using many methods such as linear correlation (Horwitz et al ., 1998; Fox  et al ., 2005; Salvador   et al ., 2005), coherence(Sun  et al ., 2004), synchronization likelihood (Stam and van Dijk,2002), (constrained) principal (Friston  et al ., 1993; Woodward et al ., 2006) or independent component analysis (McKeown andSejnowski, 1998) and partial least squares (McIntosh  et al ., 1996).Effective connectivity represents direct or indirect influences thatone brain region exerts over another one (Friston, 1994), quanti- fied by various mathematical models, such as structural equationmodelling (McIntosh and Gonzalez-Lima, 1994), Granger causality(Roebroeck  et al ., 2005), multivariate autoregressive modelling(Harrison  et al ., 2003), dynamic causal modelling (Friston  et al .,2003) and Bayesian networks (Zheng and Rajapakse, 2006). Theabove-mentioned methods can really introduce measures thatdescribe the relationships between nodes. Based on these mea-sures, graph theoretical methods can build abundant models ofcomplex networks to characterize connection patterns within thebrain further from a perspective of topological organization. It hasbeen generally believed that functional segregation and integra-tion are two major organizational principles of the human brain.An optimal brain requires a balance between local specializationand global integration of brain functional activity (Tononi  et al .,1998). This is properly supported by graph indices [e.g. clusteringcoefficients (an index of functional segregation) and path length(an index of functional integration)] used in the analysis of func-tional brain networks (Bassett and Bullmore, 2006; Stam andReijneveld, 2007). The resultant coordinated patterns with highclustering coefficients and short path length, known as asmall-world network model (Watts and Strogatz, 1998), reflectthe need of the brain networks to satisfy the competitive demandsof local and global processing (Kaiser and Hilgetag, 2006). In addi-tion, graph theoretical methods also allow one to evaluate regionalcentrality in a graph using measures of centrality in contrast to theconnectivity methods mentioned above. So far, graph theoreticalapproaches have been applied to study development (Fair   et al .,2009; Supekar   et al ., 2009), normal ageing (Achard and Bullmore,2007; Wu  et al ., 2007; Meunier   et al ., 2009) and neuropsychiatricdiseases (for a recent review, see Bassett and Bullmore, 2009).However, no study to date has used this model in an attemptto investigate the possible alterations in the brain functional net-works in stroke patients. Moreover, in previous studies the modelwas mainly used in cross-sectional studies. In the current study, alongitudinal design was employed to examine the changes in thenetwork topological pattern during stroke recovery.In this study, we focused on the motor execution network, dueto the importance of executive function in the process of strokerecovery (Wiese  et al ., 2005). We sought to investigate dynamicchanges in the topological patterns of the network during recoveryprocess. The main hypotheses were as follows:(i) Several recent studies have shown that the brain functionalnetworks shifted towards the topological pattern of randomnetworks in different types of brain pathology, such as braintumours (Bartolomei  et al ., 2006a), Alzheimer’s disease(Stam  et al ., 2009), schizophrenia (Micheloyannis  et al .,2006; Rubinov  et al ., 2009) and interictal recordings ofpatients with epilepsy pathological networks (Ponten  et al .,2007) and severe traumatic brain injury (Nakamura  et al .,2009). It is possible that network randomization may be afinal common pathway for different types of brain damage,resulting from a compensatory but non-optimized out-growth of new connections because of impaired normalconnection pathway. In the current study, we hypothesizedthat motor network randomization would be observedduring stroke recovery.Functional reorganization after stroke Brain 2010: 133; 1224–1238  |  1225  b  y  g u e s  t   onD  e c  em b  er 1  9  ,2  0 1 4 D  ownl   o a d  e d f  r  om   (ii) Recent longitudinal studies have showed progressiveimprovement in the ipsilesional primary sensorimotor cortex (Dijkhuizen  et al ., 2001; Feydy  et al ., 2002) andincreasing brain activity in controlesional cerebellum (Small et al ., 2002) after stroke; we hypothesized that graduallyincreased regional centralities and functional connectivityrelated to such regions in the network would be observedas time elapsed. Materials and methods Participants Ten right-handed patients (nine male and one female; mean age 48.3years; range 41–55 years) with left motor pathway subcortical strokewere enrolled from the inpatient services at the Xuanwu Hospital ofCapital Medical University (Beijing, China). All participants werefirst-onset stroke patients and showed motor deficits. None had ahistory of neurological or psychiatric disorders. Conventional magneticresonance images (MRI) did not find any abnormalities except for theinfarct lesion in each patient. A series of neurological examinationswere performed, including the Motricity Index, Modified RankinScale, the Barthel Index and the National Institutes of Health StrokeScale. The patients were scanned and clinically assessed at five timepoints, i.e. 1 week, 2 weeks, 1 month, 3 months and 1 year after stroke, as current literature suggests that the recovery process after stroke was assumed to consist of three phases (Saur   et al ., 2006). Theclinical characteristics of the stroke patients are summarized in Table 1.Nine age-matched healthy controls (mean age 48.1 years; range41–53 years) were recruited in a single run to identify thelesion-reduced functional reorganization in patients with stroke atthe early acute stage (about 2 weeks after stroke). In addition, tovalidate whether brain functional networks of controls exhibitedstable network topology, two groups of healthy subjects were scannedseparately in either a cross-sectional (36 subjects; mean age53.4 years; range 31–90 years) or longitudinal design (12 subjects;mean age 24.1 years; range 22–29 years), where time points weresplit into three 1-week intervals. The Ethics Committee of XuanwuHospital approved this experiment and each participant gave informedconsent. Table 1 Clinical and demographic data Patient number 1 2 3 4 5 6 7 8 9 10 Age (years) 42 48 53 52 52 51 43 50 55 41Gender M M M F M M M M M MLocalization of infarct IC IC IC IC IC IC IC IC IC ICCR CR CR CR CR CR CR CRBG BG BG BGPast medical history Nil HT Nil HT HT HT Nil HT HT DTHL DTThe number of scans 5 5 5 2 5 5 3 4 3 5Scan time (day) 4 1 2 2 0 4 1 – 6 413 12 16 12 14 13 9 11 12 1332 35 34 – 30 27 – 33 31 29147 88 97 – 92 93 – 93 – 111354 301 350 – 369 411 300 432 – 375Motricity Index 33 0 14 14 141 14 28 – 37 0(0–200) 88 14 58 28 183 37 47 86 53 14130 19 88 – 198 47 – 138 91 33190 82 113 – 198 88 – 179 – 78190 95 113 – 198 116 130 183 – 83Modified Rankin Scale 5 5 5 5 5 5 5 – 5 5(0–5) 5 5 4 5 3 5 5 5 5 53 5 3 – 2 4 – 3 4 51 3 3 – 1 3 – 2 – 31 3 3 – 1 1 2 2 – 3Barthel Index 20 0 20 0 0 10 20 – 25 0(0–100) 55 25 60 25 85 25 30 15 25 1585 35 95 – 90 50 – 70 60 25100 80 95 – 100 75 – 100 – 60100 85 95 – 100 100 90 100 – 60National Institutes of Health Stroke Scale 10 14 8 11 5 10 7 – 8 15(0–15) 3 11 6 6 2 8 5 6 7 132 10 3 – 2 8 – 5 5 130 8 2 – 0 5 – 2 – 60 5 2 – 0 2 1 1 – 6 M = male; F = female; IC = internal capsule; CR = corona radiate; BG = basal ganglia; HT = hypertension; DT = diabetes; HL = hyperlipidaemia; ‘–’ = no functional MRI data. 1226  |  Brain 2010: 133; 1224–1238 L. Wang  et al.  b  y  g u e s  t   onD  e c  em b  er 1  9  ,2  0 1 4 D  ownl   o a d  e d f  r  om   Data acquisition All images were acquired on a Siemens Trio 3.0 Tesla MRI scanner (Siemens, Erlangen, Germany) at the Xuanwu Hospital of CapitalMedical University. The head of each participant was snugly fixedby foam pads to reduce head movements and scanner noise. All func-tional magnetic resonance imaging (fMRI) data of the whole brainfrom the top of the brain to the lower part of the medulla oblongatawere acquired using an echo-planar imaging sequence: 32 axial slices,thickness/gap=3/1mm, matrix=64  64, repetition time=2000ms,echo time=30ms, flip angle=90  , field of view=220mm  220mm.Structural images were obtained in a sagittal orientation employing amagnetization prepared rapid gradient echo sequence over the wholebrain: 176 slices, thickness/gap=1.0/0mm, matrix=256  224, repe-tition time=1600ms, echo time=2.6ms, flip angle=9  , field ofview=256mm  224mm. T 2 -weighted images were acquired using aturbo-spin-echo sequence: 20 axial slices, thickness/gap=5/6.5mm,matrix=512  416, repetition time=4140ms, echo time=92ms, flipangle=150  , field of view=187mm  230mm. During theecho-planar imaging data acquisition, subjects were instructed tokeep awake, relax with their eyes closed and remain motionless asmuch as possible. Each scan lasted for 6min and 180 image volumeswere obtained. For each patient, a different number of scans wereperformed after stroke. In total, 42 acquisitions (up to five scanningsessions per subject) were collected (Table 1). Preprocessing of functional MRI data For the dataset of each subject, the first 10 volumes were discarded toallow for magnetization equilibrium effects and the adaptation of thesubjects to the circumstances, leaving 170 volumes for further analysis.The resulting datasets were corrected for delay in slice acquisition andmotion using SPM5 ( software. Therealigned images were spatially normalized to the standard space ofthe Montreal Neurological Institute and smoothed (4mm isotropickernel). Finally, temporal filter (0.01–0.1Hz) was carried out basedon an ideal rectangle window filter. Regions of interest in the motor execution network In general, most stroke patients suffer from various degrees of motor deficit. The recovery from stroke is a complex process, which has beendemonstrated to be associated with functional reorganization acrossbrain areas (for a review, see Calautti and Baron, 2003). Recently, astudy has demonstrated functional reorganization of motor executionareas rather than motor preparation areas in post-stroke hemiparesis(Wiese  et al ., 2005). Therefore, in this study, we mainly focused onthe dynamic changes in the organization of the motor execution net-work controlling for the movement of the affected hand (right hand inthis study). We selected the regions of interest associated with themotor execution network from our previous work with a simplemotor task using the right hand (Jiang  et al ., 2004). The regions ofinterest included 24 regions, such as left primary motor cortex, bilat-eral dorsolateral and ventrolateral premotor cortex, bilateral superior parietal lobule, bilateral basal ganglia, bilateral thalamus, anterior infe-rior cerebellum, postcentral gyrus, dentate nucleus, fusiform gyrus,cuneus cortex and posterolateral cerebellum. Recent studies, however,reported that brain activity in fusiform gyrus, cuneus cortex and pos-terolateral cerebellum were probably associated with visual represen-tation, motor imagery and instruction events (Allen  et al ., 1997;Hanakawa  et al ., 2008) rather than motor execution. Therefore,these five regions of interest were excluded from the current study.In addition, we made two modifications. First, we separated the sup-plementary motor area region of interest into two (left and right) inorder to study whether these performed different roles during therecovery process. Second, we added the right motor cortex into thestudied regions of interest since this region might play a pivotal role instroke recovery (Calautti and Baron, 2003). The srcinal motor cortexcoordinates were modified according to Fink  et al . (1997) and Ward et al .’s (2003) studies to locate accurately onto the motor hand area.Thus, a total of 21 regions of interest were obtained by creating10mm diameter spheres around the predefined coordinates(Table 2). In addition, to validate our results independently of theregions of interest selection, we applied the same analysis proceduresmentioned below to the motor-related and motor-imagery areasreported in Hanakawa  et al .’s (2008) study. Notably, from the meth-odological point of view, this study focused on the functional reorga-nization on the basis of the changes in topological patterns ofcoordinated networks, while many previous studies addressed thisissue using other approaches focusing on local features, such asbrain activity (for a review, see Calautti and Baron, 2003) and func-tional connectivity (Gerloff  et al ., 2006; Saur   et al ., 2006; Grefkes et al ., 2008). From a network perspective, the graph theoreticalapproaches employed in this study were interested in exploringdynamic changes in the topology of network organization duringstroke recovery, as opposed to comparing to the methods mentionedabove. Construction of brain functionalnetworks The time series of all voxels in each region of interest were extractedand averaged to obtain a representative time series. Using a multiplelinear regression model, spurious variance of the blood oxygen leveldependent signal unlikely reflecting neuronal activity was removedfrom the mean time series (the dependent variable) by regressingout signal attributable to the six parameters obtained by rigid-bodyhead motion correction (three for translation and three for rotation aspredictors). The residuals of this regression were then used to substi-tute for the raw mean time series of the corresponding regions.For each scan of every subject, we computed Pearson’s correlationcoefficients between the time series of all possible pairs of 21 regions,yielding one symmetric correlation matrix (i.e. functional connectivity).The network sparsity (i.e. connection density) was defined as thenumber of existing connections divided by all of their possible connec-tions (Achard and Bullmore, 2007; Wang  et al ., 2009), and used as athreshold measure to convert each correlation matrix into a graph. For a given sparsity, a data-specific correlation value can be determinedand separately used to threshold each correlation matrix. Only thoseabsolute correlation coefficients higher than the threshold value werereferred to as edge weights. We repeated the same procedure for allcorrelation matrices. To assure that the functional connectivity used inthis study reflected coupling between regions of interest, we per-formed statistical tests on the functional connectivity matrix con-structed from each participant in each session by using one-sample t  -tests ( P 5 0.01). The ratios of significant connections to all the pos-sible connections are represented in Supplementary Fig. S1. From thisfigure, we found that the minimum sparsity was slightly more than50%. Thus, the sparsity threshold of 0.5 was used to convert con-nectivity matrices into weighted networks (see supplementary mate-rials for the effect of different sparsity thresholds), which led to all Functional reorganization after stroke Brain 2010: 133; 1224–1238  |  1227  b  y  g u e s  t   onD  e c  em b  er 1  9  ,2  0 1 4 D  ownl   o a d  e d f  r  om   regions of interest included in the network (except 3 sessions of the42 scanning sessions including 19 regions). Graph theoretical approaches Small-world measures of a network (clustering coefficient  C p , andshortest path length  L p ) were originally proposed by Watts andStrogatz (1998). Briefly, the  C p  is the average of the clustering coef-ficients over all nodes in a network, which quantifies the extent oflocal cliquishness or local efficiency of information transfer of a net-work. The  L p  of a network is the average minimum number of con-nections that link any two nodes of the network, which quantifies theability of parallel information propagation or global efficiency (Latoraand Marchiori, 2001) of a network. Most brain network studies todate have investigated the brain’s topological properties by analysingbinarized graphs in which every network edge has an equal weight of1. In this study, we characterized the dynamic changes in the coordi-nated pattern of motor execution networks by a weighted networkanalysis approach, which took into account of network edge strengthin terms of functional connectivity. Weighted clustering coefficient  For a weighted graph, the weighted clustering coefficient of a vertex  i is defined as (Barrat  et al ., 2004) C wi  ¼  1  s i ð k  i  1 Þ  ð  j , k  Þ w ij þ w ik  2  a ij a ik  a  jk  , where the normalizing factor   s i ð k  i  1 Þ  [  s i  is the strength of the vertexdefined as the sum of the weights  w ij (the correlation coefficientsbetween regions) of the connected edges:  s i  ¼   j w ij ] assures that0  C wi   1;  k  i  (generally called the node degree) is the number ofthe edges connected to the node  i ;  a ij  is the element of adjacencymatrix, which is 1 if there is a edge connecting the node  i  and node  j ,otherwise is 0. Thus, the weighted clustering coefficient of a weightednetwork with  N  nodes is defined as C w ¼  1 N  Ni ¼ N C wi  : Apart from the weighted clustering coefficient, we note that alter-native definitions have recently been proposed (Onnela  et al ., 2005;Stam  et al ., 2009). Weighted shortest path length The srcinal  L p  definition is problematic in graphs that include morethan one component. To avoid this situation,  L p  is measured here byusing an inverse of the harmonic mean of the minimum path length asproposed by Newman (2003). For a weighted graph, the weightedshortest path length is defined as L w ¼  N ð N  1 Þ  Ni ¼ 1  N j 6¼ i 1 = l wij where  l wij  ¼ min i ,  j ð sum ð d  ij ÞÞ  and d ij  ¼ 1 = w ij . Here, the shortestweighted path length  l wij  between any pair of node  i  and  j  inthe graph indicates the minimum value of the sum of transformedweights d ij  (i.e. functional distance) over all possible paths. Typically,regular networks are high  C w with large  L w but random networksare low  C w with small  L w . To correct for differences in the meanconnection weights across multiple scanning sessions and subjects,we computed the normalized  C w ( Gamma ¼ C w = C w rand ) and L w ( Lambda ¼ L w = L w rand ) by comparing  C w and  L w values with the Table 2 Regions of interest for the motor execution network ID Region Abbreviation Side MNI coordinate x y z  1 Superior cerebellum SCb R 16   59   212 Primary motor cortex M1 L   38   22 563 Primary motor cortex M1 R 38   22 564 Thalamus Th L   10   20 115 Superior parietal lobule SPL L   22   62 546 Supplementary motor area SMA L   5   4 577 Supplementary motor area SMA R 5   4 578 Dorsolateral premotor cortex PMd R 28   10 549 Ventrolateral premotor cortex PMv L   49   1 3810 Superior cerebellum SCb L   25   56   2111 Superior parietal lobule SPL R 16   66 5712 Dentate nucleus DN R 19   55   3913 Ventrolateral premotor cortex PMv R 53 0 2514 Anterior inferior cerebellum AICb L   22   45   4915 Anterior inferior cerebellum AICb R 16   45   4916 Postcentral gyrus PCG R 37   34 5317 Dorsolateral premotor cortex PMd L   22   13 5718 Basal ganglia BG R 22   2 1219 Basal ganglia BG L   25   14 820 Thalamus Th R 7   20 1121 Dentate nucleus DN L   28   55   43 Note that the regions are selected from a previous study (Jiang  et al ., 2004). We carefully examined the location of each region of interest with a 10mm diameter sphereand did not observe any overlap between each pair of regions by their Euclidean distance. MNI=Montreal Neurological Institute. 1228  |  Brain 2010: 133; 1224–1238 L. Wang  et al.  b  y  g u e s  t   onD  e c  em b  er 1  9  ,2  0 1 4 D  ownl   o a d  e d f  r  om 
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!