Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Save to My Library
Look up keyword
Like this
2Activity
0 of .
Results for:
No results containing your search query
P. 1
Adaptive Reconfiguration of Fractal Small-world Human Brain Functional Networks

Adaptive Reconfiguration of Fractal Small-world Human Brain Functional Networks

Ratings: (0)|Views: 17 |Likes:
Published by Taj Sangha

More info:

Published by: Taj Sangha on Jul 27, 2012
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

03/28/2014

pdf

text

original

 
Adaptive reconfiguration of fractal small-worldhuman brain functional networks
Danielle S. Bassett*
†‡
, Andreas Meyer-Lindenberg
†§
, Sophie Achard*, Thomas Duke
, and Edward Bullmore*
§
*Brain Mapping Unit, Department of Psychiatry, University of Cambridge, Addenbrooke’s Hospital, Cambridge CB2 2QQ, United Kingdom;
Unit for SystemsNeuroscience in Psychiatry, Genes, Cognition, and Psychosis Program, National Institute of Mental Health, National Institutes of Health, Bethesda, MD20892; and
Biological and Soft Systems, Department of Physics, Cavendish Laboratory, University of Cambridge, Cambridge CB3 0HE, United KingdomEdited by Marcus E. Raichle, Washington University School of Medicine, St. Louis, MO, and approved October 23, 2006 (received for review July 20, 2006)
Brain function depends on adaptive self-organization of large-scale neural assemblies, but little is known about quantitativenetwork parameters governing these processes in humans. Here,we describe the topology and synchronizability of frequency-specificbrainfunctionalnetworksusingwaveletdecompositionofmagnetoencephalographic time series, followed by constructionandanalysisofundirectedgraphs.Magnetoencephalographicdatawere acquired from 22 subjects, half of whom performed a finger-tapping task, whereas the other half were studied at rest. Wefound that brain functional networks were characterized by small-world properties at all six wavelet scales considered, correspond-ing approximately to classical
␦ 
(low and high),
␪ 
,
,
, and
␥ 
frequency bands. Global topological parameters (path length,clustering) were conserved across scales, most consistently in thefrequency range 2–37 Hz, implying a scale-invariant or fractalsmall-world organization. Dynamical analysis showed that net-works were located close to the threshold of order/disorder tran-sitioninallfrequencybands.Thehighest-frequency
␥ 
networkhadgreater synchronizability, greater clustering of connections, andshorter path length than networks in the scaling regime of (lower)frequencies. Behavioral state did not strongly influence globaltopology or synchronizability; however, motor task performancewas associated with emergence of long-range connections in both
and
␥ 
networks. Long-range connectivity, e.g., between frontaland parietal cortex, at high frequencies during a motor task mayfacilitate sensorimotor binding. Human brain functional networksdemonstrate a fractal small-world architecture that supports crit-ical dynamics and task-related spatial reconfiguration while pre-serving global topological parameters.
magnetoencephalography
͉
wavelet
͉
graph theory
͉
connectivity
͉
binding
C
oherent or correlated oscillation of large-scale, distributedneural networks is widely regarded as an important physi-ological substrate for motor, perceptual and cognitive represen-tations in the brain (1, 2). The topological description of brainnetworks promises quantitative insight into functionally relevantparameters because their topology strongly influences theirdynamic properties such as speed and specialization of informa-tion processing, learning, and robustness against pathologicalattack by disease (3).The topology of networks can range from entirely random tofully ordered (a lattice). In this spectrum, small-world topologyis characteristic of complex networks that demonstrate bothclustered or cliquish interconnectivity within groups of nodessharing many nearest neighbors in common (like regular lat-tices), and a short path length between any two nodes in thenetwork (like random graphs) (3). This is an attractive config-uration, in principle, for the anatomical and functional archi-tecture of the brain, because small-world networks are known tooptimize information transfer, increase the rate of learning, andsupport both segregated and distributed information processing(refs. 4–7; for review, see ref. 8). Small-world properties wererecently described empirically in neuroanatomical networks at amicroscopic level, e.g., the neuronal network of 
Caenorhabditis elegans
(5), and at the macroscopic level of interregional axonalconnectivity of the cat and macaque monkey cortices (9, 10).Neurophysiological networks inferred from patterns of corre-lated time-series activity in regions of monkey brains (11),regions of human brains (12, 13), and voxels of human functionalMRI data also demonstrate small-world topology (14, 15).However, the frequency dependence of small-world brain func-tional networks, and their sensitivity to different behavioralstates, has yet to be determined (16).To address the issue of frequency, we used a wavelet analysisofmagnetoencephalographic(MEG)signalsrecordedfromasetof 275 points overlying the scalp surface, to provide a time–frequency decomposition of human brain activity. This decom-position was followed by a correlation analysis in the waveletdomain to reveal which MEG signals represented similar phys-iological activity or were functionally connected, in each of six distinct wavelet scales or frequency intervals. Graph theory wasapplied to describe the topological properties of adjacencymatrices derived by a binary thresholding of the continuous wavelet correlation matrices. We used small-world metrics tocharacterize these undirected graphs representing brain func-tional networks and to compare network properties across therange of frequency intervals represented by scales of the wavelettransform. We then used these frequency-dependent topologicalanalyses to deduce emergent dynamical properties of the system;specifically, we derived an estimate of the synchronizability of each network.To address the second issue of state dependency, we extendedthe analysis to compare the properties of scale-specific networksderived from MEG data recorded with subjects either at rest orperforming a simple motor task (visually cued finger tapping). As well as considering the global topological and dynamicalparameters of resting and motor task-related networks, we alsocompared their spatial configurations. For example, we quanti-fied the physical distances between functionally connectednodes, and we mapped the spatial distributions of highly con-nected ‘‘hubs’’ and topologically pivotal nodes.Given the limited amount of prior data on small-world prop-erties of human brain functional networks measured by usingMEG (or electroencephalography) (12, 17), our hypotheticalpredictions were modest. We expected to find evidence for
Author contributions: D.S.B., A.M.-L., S.A., and E.B. designed research; D.S.B. performedresearch; A.M.-L., S.A., T.D., and E.B. contributed new reagents/analytic tools; D.S.B.analyzed data; and D.S.B., A.M.-L., and E.B. wrote the paper.The authors declare no conflict of interest.This article is a PNAS direct submission.Abbreviation: MEG, magnetoencephalographic.See Commentary on page 19219.
§
To whom correspondence may be addressed. E-mail: andreasm@mail.nih.gov oretb23@cam.ac.uk.This article contains supporting information online atwww.pnas.org/cgi/content/full/ 0606005103/DC1© 2006 by The National Academy of Sciences of the USA
19518–19523
͉
PNAS
͉
December 19, 2006
͉
vol. 103
͉
no. 51 www.pnas.org
͞
cgi
͞
doi
͞
10.1073
͞
pnas.0606005103
 
small-world brain functional networks in MEG data recorded inboth behavioral states. Based on functional neuroimaging stud-ies of finger tapping and timing of motor performance, we alsoexpected to find task-related changes in the connectivity of motor, premotor, prefrontal, and lateral parietal cortical regions(18–26).
Results
Small-World Parameters of Scale-Specific Networks.
The small- world metric
␴ 
ranged between 1.7 and 2.0 (see Table 1 and Fig.1), but at most scales in both states
␴ 
Ϸ
1.9, indicating thatsmall-world topology was closely conserved over a wide range of frequencies. In resting networks, the scaling regime for
␴ 
was1.1–75 Hz; in motor task-related networks, the scaling regime was 2.2–75 Hz.The mean degree
k
Ϸ
13, the clustering coefficient
C
Ϸ
0.21,and the minimum path length
L
Ϸ
5.2 were also conserved overscales and states (Fig. 1). Although the scaling regimes for theseparameters did not include the highest or lowest frequencyintervals, they consistently extended from 2 to 37.5 Hz. The
␥ 
network was distinguished by higher degree, greater clustering,and shorter path length than the lower-frequency networks inthe scaling regime.
Degree Distributions of Scale-Specific Networks.
There was furtherevidence for scale-invariant topology when we compared thedegree distributions of the scale-specific networks in each be-havioral state (Fig. 1
F–H 
). For all networks, the degree distri-bution was best fit by an exponentially truncated power law, with very similar parameter values at all scales and states [seesupporting information (SI) Table 2for parameter values andSI Fig. 5for best fits].
Synchronizability of Scale-Specific Networks.
The synchronizabilityat all scales in both states was at or below the threshold of 0.01at which systems of various oscillators globally synchronize (seeFig. 1
 D
), suggesting that the brain networks are located dynam-ically on a critical point of the order/disorder transition (27).There was evidence of scale invariance in synchronizability, witha scaling regime consistently including the frequency range2–37.5 Hz. It was also notable that synchronizability in bothmotor and resting states was somewhat higher in the
␥ 
band,
Table 1. Global topological and dynamical properties of frequency-specific human brain functional networks
WaveletdecompositionlevelFrequencyrange, Hz Corr
␶ 
k L
␨ 
S (
ϫ
10
Ϫ
3
)Resting1 37.575 0.18
Ϯ
0.02 0.50
Ϯ
0.05 16.3
Ϯ
5.1 4.5
Ϯ
0.5 0.23
Ϯ
0.02 1.9
Ϯ
0.2 61
Ϯ
14 9.7
Ϯ
1.92 18.837.5 0.26
Ϯ
0.02 0.74
Ϯ
0.04 12.6
Ϯ
3.1 5.2
Ϯ
0.5 0.21
Ϯ
0.02 1.9
Ϯ
0.1 70
Ϯ
41 8.2
Ϯ
2.33 9.418.8 0.30
Ϯ
0.03 0.81
Ϯ
0.03 12.4
Ϯ
1.8 5.4
Ϯ
0.4 0.20
Ϯ
0.01 1.9
Ϯ
0.2 100
Ϯ
72 6.3
Ϯ
2.74 4.79.4 0.30
Ϯ
0.03 0.82
Ϯ
0.03 12.3
Ϯ
2.0 5.4
Ϯ
0.4 0.21
Ϯ
0.01 1.9
Ϯ
0.2 106
Ϯ
75 6.4
Ϯ
3.35 2.34.7 0.30
Ϯ
0.02 0.81
Ϯ
0.02 12.5
Ϯ
2.0 5.2
Ϯ
0.4 0.21
Ϯ
0.01 2.0
Ϯ
0.1 118
Ϯ
71 7.6
Ϯ
2.96 1.12.3 0.33
Ϯ
0.05 0.83
Ϯ
0.02 13.7
Ϯ
3.3 5.1
Ϯ
0.4 0.23
Ϯ
0.02 1.9
Ϯ
0.1 137
Ϯ
62 6.0
Ϯ
2.4Tapping1 37.575 0.18
Ϯ
0.03 0.49
Ϯ
0.09 16.9
Ϯ
5.1 4.4
Ϯ
0.6 0.23
Ϯ
0.02 1.8
Ϯ
0.2 132
Ϯ
21 10.2
Ϯ
3.62 18.837.5 0.23
Ϯ
0.02 0.69
Ϯ
0.04 13.0
Ϯ
2.6 5.0
Ϯ
0.4 0.21
Ϯ
0.01 2.0
Ϯ
0.1 105
Ϯ
9 9.8
Ϯ
2.73 9.418.8 0.27
Ϯ
0.02 0.77
Ϯ
0.03 12.2
Ϯ
1.7 5.2
Ϯ
0.4 0.21
Ϯ
0.01 2.0
Ϯ
0.1 118
Ϯ
27 8.4
Ϯ
2.84 4.79.4 0.28
Ϯ
0.03 0.79
Ϯ
0.02 12.7
Ϯ
2.1 5.2
Ϯ
0.5 0.21
Ϯ
0.01 1.9
Ϯ
0.2 116
Ϯ
35 8.2
Ϯ
2.95 2.34.7 0.30
Ϯ
0.05 0.81
Ϯ
0.01 13.8
Ϯ
4.9 5.1
Ϯ
0.5 0.21
Ϯ
0.01 1.9
Ϯ
0.2 137
Ϯ
47 7.2
Ϯ
2.66 1.12.3 0.34
Ϯ
0.06 0.82
Ϯ
0.01 16.7
Ϯ
8.3 4.9
Ϯ
0.8 0.22
Ϯ
0.02 1.7
Ϯ
0.2 144
Ϯ
55 5.2
Ϯ
1.6
Corr, average correlation of the whole brain network before thresholding;
␶ 
, threshold applied to wavelet correlation matrices;
, average degree of thenetwork;
L
, average path length;
, average clustering;
␴ 
, small-world scalar value;
␨ 
, characteristic length scale in millimeters;
, synchronizability.
Fig. 1.
Scale-invariance of global topological and dynamical parameters of brain functional networks. Each image summarizes the group mean parametervalues over all wavelet scales and in both resting (red) and motor (blue) states; error bars represent 95% confidence interval; blue and red bars below the
axisindicatetheextentofthescalingregimeforeachparameterinbothresting(red)andmotor(blue)states.(
 A
)Averagepathlength,
L
.(
B
)Clustering,
.(
)Sigma,
␴ 
. (
D
) Synchronizability,
. (
) Characteristic length,
␨ 
(mm). (
H
) Parameters of an exponentially truncated power law degree distribution of the form
(
)
Ϸ
 A k 
Ϫ
1
e
k/k 
. (
) Coefficient,
A
. (
G
) Power law exponent,
. (
H
), Exponential cut-off degree,
.
Bassett
et al.
PNAS
͉
December 19, 2006
͉
vol. 103
͉
no. 51
͉
19519
     N     E     U     R     O     S     C     I     E     N     C     E     S     E     E     C     O     M     M     E     N     T     A     R     Y
 
 which reflects the distinctive topological properties (greaterdensity and clustering) of the
␥ 
network.
Spatial Configuration of Scale-Specific Networks.
The spatial distri-bution of network hubs was also broadly similar across scales andstates (see Fig. 2 andSI Fig. 5). SeeSI Fig. 6for average hub distributions across all scales in both states. However, there werestriking differences between scales and states in the physicaldistance between functionally connected network nodes (seeFig. 3).In the resting state, long-range functional connectivity be-tween brain regions was stronger at low frequencies. At higherfrequencies(
,
␥ 
),long-rangeconnectivitywasweaker,andmostof the edges in the graph represented high-density local con-nections (see Figs. 1
 E
and 4), shown by the increase in charac-teristic length scale of network edges
␨ 
, going from high to lowfrequency scales; and by the increasing number of connectorcompared with provincial nodes at low frequencies (seeSI Fig.7for a schematic andSI Fig. 8for distributions of provincial and connector hubs in both states and all frequency bands).In the finger-tapping state, long-range functional connectionsemerged more strongly at high frequencies (
,
␥ 
), shown by thesignificant motor task-related increases in characteristic lengthscale of edges in high-frequency motor networks. It is alsorepresentedbytheshiftfromresting-state
␥ 
networksdominatedby provincial hubs (predominantly connected to locally neigh-boring regions of bilateral occipital, parietal, and central cortex)to motor
␥ 
networks with a larger number of connector hubs inmedial premotor and bilateral prefrontal cortex. Some of thenew long-range connections engendered by task performance athigh frequencies link to topologically pivotal nodes in rightmedial premotor and prefrontal cortex with high betweennessscores (see Fig. 3; and seeSI Fig. 9for betweenness distributionsat all frequencies). This indicates that task performance isassociated with reconfiguration of high-frequency networks tofavor long-distance connections between prefrontal and premo-
Fig.2.
Self-similarityofspatialdistributionofhighlyconnectednetworknodesor‘‘hubs’’inthefrequencyrange2–38Hz(64).Eachcolumnshowsthesurfacedistribution of the degree of network nodes in frequency bands
to
: red represents nodes with high degree. The last column shows the spatial distributionofdegreeaveragedoverthesefourfrequencybands,whichemphasizesthesimilarityofspatialconfigurationsacrossscales.SeeSIFig.5forthehubdistributionsin both states at all frequency bands.
Fig.3.
State-relateddifferencesinspatialconfigurationofthehighestfrequency
␥ 
network.Thetoprowshowsthedegreedistributionandbetweennessscoresfor the resting state
␥ 
network; the middle row shows the same maps for the motor
␥ 
network; the bottom row shows the between-state differences in degreeand betweenness. It is clear that motor task performance is associated with emergence of greater connectivity in bilateral prefrontal and premotor nodes, andappearance of topologically pivotal nodes (with high betweenness scores) in medial premotor, right prefrontal, and parietal areas. SeeSI Fig. 7for thebetweenness distributions in both states at all frequency bands.
19520
͉
www.pnas.org
͞
cgi
͞
doi
͞
10.1073
͞
pnas.0606005103 Bassett
et al.

You're Reading a Free Preview

Download
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->