J Neurophysiol 105: 2753–2763, 2011.
First published March 23, 2011; doi:10.1152/jn.00895.2010.
Brain activity at rest: a multiscale hierarchical functional organization
Gaëlle Doucet,1,2,3 Mikaël Naveau,1,2,3,4 Laurent Petit,1,2,3 Nicolas Delcroix,5 Laure Zago,1,2,3
Fabrice Crivello,1,2,3 Gaël Jobard,1,2,3,4 Nathalie Tzourio-Mazoyer,1,2,3 Bernard Mazoyer,1,2,3,4,6,7
Emmanuel Mellet,1,2,3,4 and Marc Joliot1,2,3
1
Univ. de Bordeaux, 2CNRS, 3CEA, Bordeaux; 4Université de Caen, 5GIP Cyceron, Caen; 6Institut Universitaire de France,
and 7CHU, Caen, France
Submitted 18 October 2010; accepted in final form 23 March 2011
conscious resting state; fMRI; functional connectivity; independent
component analysis; networks
studies have investigated the functional organization of the brain at rest (see
reviews of Bressler and Menon 2010; Fox and Raichle 2007;
Raichle 2010). All of these studies reported spontaneous,
structured neural activity, although by definition, no goaldirected mental activity and minimal perceptive input were
present. In short, two views have emerged regarding the
functional organization of the brain at rest: some authors have
emphasized two anticorrelated functional systems (Fox et al.
2005; Fransson 2005; Golland et al. 2008), whereas others
have suggested the existence of several modular systems on a
more local scale (He et al. 2009).
Pioneering the former view, Fox et al. (2005) investigated
the functional connectivity of predefined regions. They highlighted two large-scale networks in which activity was negatively correlated during rest: the so-called brain default-mode
network (DMN) (Raichle et al. 2001) and the network rou-
AN INCREASING NUMBER OF NEUROIMAGING
Address for reprint requests and other correspondence: M. Joliot, GIN,
UMR5296 CNRS CEA, Université Bordeaux Segalen, Case 71, 146 rue Léo
Saignat, 33076 Bordeaux cedex, France (e-mail: [email protected]).
www.jn.org
tinely activated during goal-directed attention activity (Corbetta and Shulman 2002). More recently, Golland et al. (2008)
generalized this partition to the whole brain using an unsupervised approach. They revealed an intrinsic system exhibiting
an inner-oriented mental activity and overlapping the DMN
and an extrinsic system driven by external inputs and activated
during sensory stimulations, including the attentional network
and the sensory-motor cortex.
In the second approach, which revealed a modular organization, authors investigated functional brain organization
through “anatomically and/or functionally associated components that perform specific biological functions” (He et al.
2009). These modules corresponded to sets of regions showing
highly coherent, spontaneous activity and included networks
involved in goal-directed activity as well as the DMN (Damoiseaux et al. 2006; He et al. 2009).
The two approaches are not mutually exclusive but apprehend the functional organization of the brain at rest, either from
a global (systems) or a local (modules) perspective. Nonetheless, previous works describing a global bipartition of the brain
resting-state organization did not investigate the local components of each partition. On the other hand, works that established a modular organization did not comment on the potential
desynchronization between modules, leading to superstructures
that could correspond to intrinsic and extrinsic systems. This
gap leaves open the questions of how the subcomponents of
each system interact and how this interaction allows the two
systems to balance their respective activities. Analysis of
temporal correlations evidenced between modules during the
resting state could shed light on this issue. Positive correlations
between networks observed at rest could indeed reflect rapid
information exchange between neuronal assemblies, whereas
negative correlations between modules or systems could demonstrate work in competition or achieving unrelated goals (Fox
et al. 2005).
In the present work, we combined individual independent
component analysis (ICA) with component classification
across subjects to build bridges between the system and module scales.
MATERIALS AND METHODS
Participants
One-hundred eighty healthy, young adults (89 males, 91 righthanded), aged 18 –57 years (26.0 ⫾ 6.6 years, mean ⫾ SD), participated in this study. Handedness was defined according to each
individual’s writing hand. The mean educational level of the participants was 15 ⫾ 3 years, indicating an average of 3 years at a
0022-3077/11 Copyright © 2011 the American Physiological Society
2753
Downloaded from http://jn.physiology.org/ by 10.220.33.2 on June 15, 2017
Doucet G, Naveau M, Petit L, Delcroix N, Zago L, Crivello F,
Jobard G, Tzourio-Mazoyer N, Mazoyer B, Mellet E, Joliot M.
Brain activity at rest: a multiscale hierarchical functional organization.
J Neurophysiol 105: 2753–2763, 2011. First published March 23,
2011; doi:10.1152/jn.00895.2010.—Spontaneous brain activity was
mapped with functional MRI (fMRI) in a sample of 180 subjects
while in a conscious resting-state condition. With the use of independent component analysis (ICA) of each individual fMRI signal and
classification of the ICA-defined components across subjects, a set of
23 resting-state networks (RNs) was identified. Functional connectivity between each pair of RNs was assessed using temporal correlation
analyses in the 0.01- to 0.1-Hz frequency band, and the corresponding
set of correlation coefficients was used to obtain a hierarchical
clustering of the 23 RNs. At the highest hierarchical level, we found
two anticorrelated systems in charge of intrinsic and extrinsic processing, respectively. At a lower level, the intrinsic system appears to
be partitioned in three modules that subserve generation of spontaneous thoughts (M1a; default mode), inner maintenance and manipulation of information (M1b), and cognitive control and switching
activity (M1c), respectively. The extrinsic system was found to be
made of two distinct modules: one including primary somatosensory
and auditory areas and the dorsal attentional network (M2a) and the
other encompassing the visual areas (M2b). Functional connectivity
analyses revealed that M1b played a central role in the functioning of
the intrinsic system, whereas M1c seems to mediate exchange of
information between the intrinsic and extrinsic systems.
2754
FUNCTIONAL BRAIN ORGANIZATION AT REST
university. All participants gave their informed, written consent, and
the local ethical committee (CPP de Basse Normandie, France)
approved the study.
each z-map was applied a threshold set at pc ⫽ 0.5 (Beckmann and
Smith 2004).
Classification of Individual ICs
Imaging Methods
Preprocessing of fMRI Data
Preprocessing was built on the basis of statistical parametric
mapping 5 subroutine (Wellcome Department of Neurology, London,
UK; www.fil.ion.ucl.ac.uk/spm). For each participant, anatomical
T1-weighted volume was segmented into three brain tissue classes
[gray matter (GM), white matter (WM), and cerebrospinal fluid
(CSF)] and spatially normalized using the stereotaxic space of the
Montreal Neurological Institute (MNI) template (Ashburner and Friston 2005). fMRI data were corrected for slice-timing differences and
motion (six parameters: three translations and three rotations) and
registered on the T2*-FFE volume. Combining the T2*-FFE with the
T1-weighted registration matrix and the previously computed T1weighted normalization matrix, fMRI data were then normalized to
the MNI stereotaxic space (2 ⫻ 2 ⫻ 2 mm3 cubic voxels; bounding
box: x ⫽ ⫺90 to 90 mm, y ⫽ ⫺126 to 91 mm, z ⫽ ⫺72 to 109 mm)
and spatially smoothed (Gaussian 6 mm full width at half maximum
filter). With the use of an in-house Matlab-based software, each voxel
of the image time series was motion corrected using time evolutions
from the six motion parameters and cleaned from contamination by
both WM and CSF signals using linear regression. For each individual, the temporal evolution of both WM and CSF was estimated in the
relevant brain tissue classes. Note that the global signal was not
removed from the data. Finally, fMRI data were temporally filtered
using a least-square linear-phase, finite impulse-response filter design
band-pass (low-cutoff frequency ⫽ 0.01 Hz ⫺ high-cutoff frequency
0.1 Hz).
ICA of Individual fMRI Data
Individual preprocessed functional data were analyzed using an
ICA (McKeown and Sejnowski 1998). Data were individually processed using Multivariate Exploratory Linear Optimized Decomposition into Independent Components software, version 3.05, included in
the FMRIB Software Library (http://www.fmrib.ox.ac.uk/fsl) (Smith
et al. 2004). The number of independent components (ICs) was
estimated by Laplace approximation (Minka 2000). A symmetric
approach of the FastICA algorithm (Hyvarinen 1999) was used for
computing ICAs. Each component was described by its spatial z-map
and a BOLD time series. With the use of a mixture model estimation,
J Neurophysiol • VOL
At the group level, we created sets of individual ICs. A set was
defined as a collection of ICs exhibiting similar spatial patterns in
different individuals. The set ownership of IC was assessed by an
unsupervised classification algorithm based on z-map spatial overlap.
For each pair of IC z-maps belonging to two individuals, a spatial
overlap parameter named ␥ was computed as the ratio of the number
of voxels of the two z-maps’ intersection to that of their union
␥(i, j) ⫽
ⱍIC 艚 IC ⱍ
ⱍIC 艛 IC ⱍ
i
j
i
j
where ICi (or ICj) is the number of significant voxels of individual IC
i (or j). ␥ was calculated for all pairs of ICs and all pairs of individuals.
When ␥ was observed above a cut-off value ␥c, the corresponding ICs
were said to be “overlapping”. To set up the ␥c cutoff, the distribution
of the maximal ␥ values with any other components was computed. It
showed a bimodal distribution, which could be adequately modeled
using two Gaussian distributions (see Supplemental Fig. S1). The first
Gaussian distribution included poorly overlapping ICs, whereas the
second distribution included highly overlapping ICs. The cutoff ␥c
was set at P ⫽ 0.0001 of the first distribution. This threshold excluded
subject-specific ICs from classification, as they did not have any
spatial overlap greater than ␥c with other ICs and selected the highest
overlapping components.
The classification algorithm proceeded in three steps (see Supplemental Fig. S2), consisting of 1) the creation of sets of overlapping
ICs (Fig. 1), 2) the iterative clustering of such sets (Fig. 1), and 3) the
iterative (re-) classification of misclassified or orphan ICs.
Step 1 implemented in an iterative manner a constraint of minimal
spatial variability between the spatial z-maps of a set, requesting that
all ICs of a given set be overlapping (Fig. 1). We went through the list
of pairs of ICs, starting from the highest overlap value and decreasing
up to the threshold ␥c ⫽ 0.2 and applied the following decisions.
● If none of both overlapping ICs belonged to an existing set, we
created a new set with them.
● If only one of these two ICs belonged to an existing set, and the
second one had an overlap higher than ␥c with all other ICs of this set
(overlapping criterion), we added it to this set. Additionally, a set was
constrained to include a maximum of one IC/subject.
● If ICs already belonged to different clusters, we did nothing.
This clustering method was a variation of the complete linkage
approach in hierarchical clustering (Ylipaavalniemi and Vigario
2008). Upon completion of Step 1, a voxelwise frequency map was
built for each set. Each voxel was assigned to the percentage of
participants contributing to the set at the voxel location. A threshold
of 0.5 was applied to each frequency map, resulting in set maps,
including voxels present in ⬎50% of the ICs of the set. For each pair
of thresholded frequency maps, a spatial overlap parameter named ␺
was computed as the ratio of the number of voxels of the two
frequency maps’ intersection to that of their union.
Step 2 performed the merging of sets based on the values of the spatial overlaps of their frequency maps (␺; Fig. 1). The procedure
described in Step 1 was applied to merge sets on the basis of the
overlap of their thresholded frequency maps. This step did not include
the criterion requesting that all of the spatial maps of the ICs of the set
showed a spatial overlap ⬎␥c with all others included in the set. This
procedure formed new sets of ICs. Frequency maps were computed
for each new set, and the procedure was iterated until the number of
sets remained constant.
Finally, Step 3 made corrections for misclassified ICs. Two types of
misclassified ICs were present and defined as: 1) an IC with “incorrect
assignment”, i.e., classified in one set but exhibiting its highest
105 • JUNE 2011 •
www.jn.org
Downloaded from http://jn.physiology.org/ by 10.220.33.2 on June 15, 2017
Imaging was performed on a Philips Achieva 3-Tesla MRI scanner.
Spontaneous brain activity was monitored using blood-oxygen leveldependent functional MRI (BOLD fMRI) while the participants performed a resting-state condition for 8 min [T2*-echo planar imaging,
sequence parameters: 240 volumes; repetition time (TR) ⫽ 2 s; echo
time (TE) ⫽ 35 ms; flip angle ⫽ 80°; 31 axial slices; 3.75 ⫻ 3.75 ⫻
3.75 mm3 isotropic voxel size]. Immediately before fMRI scanning,
participants were instructed to “keep their eyes closed, to relax, to
refrain from moving, to stay awake, and to let their thoughts come and
go”. Prior to the fMRI session, structural MR brain images were
acquired, including a high-resolution, three-dimensional T1-weighted
volume (sequence parameters: TR ⫽ 20 ms; TE ⫽ 4.6 ms; flip
angle ⫽ 10°; inversion time ⫽ 800 ms; turbo field echo factor ⫽ 65;
sense factor ⫽ 2; field of view ⫽ 256 ⫻ 256 ⫻ 180 mm; 1 ⫻ 1 ⫻ 1
mm3 isotropic voxel size), and T2*-weighted, multislice images were
also acquired [T2*-weighted fast-field echo (T2*-FFE), sequence
parameters: TR ⫽ 3,500 ms; TE ⫽ 35 ms; flip angle ⫽ 90°; sense
factor ⫽ 2; 70 axial slices; 2 ⫻ 2 ⫻ 2 mm3 isotropic voxel size]. No
cognitive training or task was realized by participants during the
imaging session.
FUNCTIONAL BRAIN ORGANIZATION AT REST
2755
overlap with ICs of another set or a number of overlaps (⬎␥c) higher
with another set than the one with which it belongs; 2) an IC with
“incorrect failure of assignment”, i.e., has not been assigned to a set,
although it is overlapping with at least one other IC classified in this
set.
a significant r value for that pair and declared that the RN pair was
significantly correlated at the group level if this proportion was
significantly larger than 0.5 (␹2 statistic, with a P value set at 0.05
corrected for the number of tested RN pairs).
Systems and Modules Analysis
Anatomical Characterization of Resting-State Networks
To extract local maxima for each set, z-maps of all ICs belonging
to the set were averaged and masked by the thresholded frequency
map. These maps provided a neuroanatomical description of restingstate networks (RNs) at play during the resting state. Local maxima of
each RN z-map were anatomically localized using automatic anatomical labeling (Tzourio-Mazoyer et al. 2002).
Temporal Correlations of RNs
The thresholded frequency map of each RN was used as a region of
interest. Thus for each individual and each RN, a BOLD fMRI time
series was computed by averaging the BOLD fMRI time series of all
voxels belonging to the RN-binarized frequency map. Then, for each
individual and each pair of RNs, we computed the Pearson’s linear
correlation coefficient r between the BOLD fMRI time series of two
RNs, declaring it significant when it was larger than a threshold set at
0.246 [which corresponds to a P value ⬍0.05 corrected for multiple
comparisons (number of RN pairs) and taking into account time-series
autocorrelations, i.e., using 78 degrees of freedom instead of 238]. For
each RN pair, we then looked at the proportion of subjects exhibiting
J Neurophysiol • VOL
The correlation coefficients of the individual RN pairs were averaged across the 180 participants. These averaged temporal correlation
coefficients were used to perform an average-linkage hierarchical
clustering of RNs (Johnson 1967). For each RNi/RNj pair, the averaged correlation coefficient value (ri,j) was first transformed into a
dissimilarity distance (di,j) using the following equation: di,j ⫽ (1 ⫺
ri,j)/2. Then, RNs were hierarchically aggregated according to a
minimal dissimilarity cluster distance (the distance between two
clusters is the average distance between all RN pairs with one RN in
each cluster). Clustering reliability was assessed with the approximately unbiased P value, computed as a result of multiscale bootstrap
resampling (Efron et al. 1996) using the “pvclust” function of the R
package (bootstrap was realized on 100,000 datasets including from
50% to 140% of the 180 participants’ sample data) (Suzuki and
Shimodaira 2006).
This hierarchical clustering algorithm provided three levels for
partitioning the set of RNs from systems (highest level), to modules
(intermediate level), down to RNs (lowest level). The threshold for the
modular partition was set to the lowest dissimilarity distance that
precluded the creation of a module, including only one RN. A spatial
map was generated for each system (or module) by combining the
105 • JUNE 2011 •
www.jn.org
Downloaded from http://jn.physiology.org/ by 10.220.33.2 on June 15, 2017
Fig. 1. Illustration of the 2 first steps of the classification algorithm of individual independent components (ICs). Step 1, left: Each set (A and B) is formed as
a group of ICs showing a spatial overlap superior to cut-off value (␥c) ⫽ 0.2 with all other ICs of the set (see histogram). Each set of ICs is associated to a
voxelwise “frequency map” coding the ratio of the number of contributing ICs at the voxel location divided by the number of ICs included in the set. Note that
in this example, 1 IC of the set labeled A has at least 1 spatial overlap inferior to ␥c with 1 IC of the set labeled B and thus precludes their aggregation in the
same set. ␥, Spatial overlap parameter. Step 2, right: The frequency map of each set is thresholded at 50%. The merging of sets of ICs occurs if the spatial overlap
(␺) between their respective thresholded frequency maps is superior to ␥c, and the sets were built of ICs from different individuals.
2756
FUNCTIONAL BRAIN ORGANIZATION AT REST
spatial maps of all RNs included in this system (or module). Finally,
a BOLD time series was computed for each system (or module), and
temporal correlation between systems (or modules) was computed and
tested for statistical significance in a way similar to that used for the
time series and temporal correlations of RNs.
We computed the averaged temporal correlations between systems
(or modules) among the 180 participants. An intramodular correlation
was also computed as the averaged temporal correlation computed
between all pairs of RNs of the module, among all participants.
RESULTS
RNs
J Neurophysiol • VOL
Systems and Modules Uncovered by Hierarchical Clustering
of RNs
Figure 3 displays the result of the hierarchical clustering of
the 23 RNs (left panel) based on the temporal correlation
matrix of pairs of RN’s BOLD fMRI signals (right panel). At
the root of the dendrogram, we observed a first partition in two
systems of networks (S1 and S2). As shown in Fig. 4 (left
panel), it is noteworthy that this partition segregated a system
of mainly sensory-oriented RNs (S2) from the other RNs (S1).
The partition of these systems provided five modules of
RNs. Three modules were issued from S1 (M1a, M1b, and
M1c) and two from the partition of S2 (M2a and M2b; Fig. 4,
right panel). Bootstrap analysis revealed that the clustering of
these five modules was extremely robust, found in ⬎99% of
the bootstrap samples. As expected, RNs within each module
were strongly synchronized together, with temporal correlations ranging from 0.44 ⫾ 0.23 within M1b to 0.68 ⫾ 0.16
within M2b (Table 1 and Fig. 3, right panel).
Description of Modules
M1a (Figs. 2 and 4, Supplemental Table S2). M1a included
most of the regions usually described as part of the DMN
(Mazoyer et al. 2001; Raichle et al. 2001).
The entire precuneus was included in M1a with a consistent
subdivision in each hemisphere: dorsal (RN1) and ventral
(RN2) parts along the parieto-occipital sulcus, a rostral part
along the marginal ramus (RN9), and a middle part (RN6).
Remaining on the medial wall, M1a bilaterally covered the
posterior cingulum (RN1, RN6), the anterior part of the calcarine sulcus (RN6), the medial orbito-frontal cortex, and anterior cingulum (RN2, RN6).
On the lateral view, M1a bilaterally encompassed the angular gyrus (RN2, RN6), extending into the inferior parietal gyrus
(RN1) and the middle occipital gyrus (RN6, RN9). Both
hippocampus and parahippocampal gyri (RN6) were also included bilaterally. M1a covered the posterior (RN9) and anterior (RN2) parts of the superior frontal gyri and to a lesser
extent, the anterior parts of the middle temporal (MT) gyri
(RN2), the left posterior part of both middle and inferior
temporal gyri (RN9), and the left supramarginal gyrus (RN9).
M1b (Figs. 2 and 4, Supplemental Table S3). In the frontal
cortex, M1b bilaterally encompassed the superior (RN10),
middle (RN3, RN7, RN10, RN11), and inferior frontal gyri.
All subsections of the inferior frontal gyrus were covered [the
opercular (RN7, RN14), triangular (RN3, RN7, RN10, RN11),
and orbital (RN3, RN10, RN11, RN14) sections, respectively].
In the parietal cortex, M1b bilaterally covered both superior
(RN7) and inferior parietal gyri (RN3, RN7, RN14), the
angular gyrus (RN10), and to a lesser extent, the left supramarginal gyrus (RN11). Note that M1b also bilaterally included the middle (RN10, RN11, RN14) and inferior (RN7,
RN10) temporal gyri.
On the medial wall, M1b bilaterally encompassed the medial
frontal cortex (RN3, RN10, RN11, RN14), extending to the
supplementary motor area (SMA; RN7 and RN11), as well as
the precuneus in its ventral part along the parieto-occipital
sulcus (RN10).
105 • JUNE 2011 •
www.jn.org
Downloaded from http://jn.physiology.org/ by 10.220.33.2 on June 15, 2017
The 180 individual analyses resulted in a total of 5,650 ICs
[an average of 31 components/subject; (min ⫺ max) ⫽ (24 ⫺
39)]. The distribution of the maximal ␥ values for each IC
showed a bimodal distribution, which could be adequately
modeled using two Gaussian distributions (see Supplemental
Fig. S1). The first Gaussian distribution peaking at ␥ ⫽ 0.10
included poorly overlapping ICs. This distribution also included a higher proportion of ICs localized in the WM and CSF
compared with those localized in the GM (70% of ICs for ␥ ⬍
0.10). The second distribution peaked at ␥ ⫽ 0.23, and 80% of
ICs above this threshold were localized in the GM. We set a
threshold ␥c ⫽ 0.2 so that remaining ICs have a very low
probability (P ⫽ 0.0001) to belong to the first distribution (see
Supplemental Fig. S1); this selection process led to select
2,438 ICs for the classification phase. The first step of the
classification retained 1,414 of the selected ICs (58%) allocated in 209 sets of overlapping ICs. The number of subjects
contributing to these sets ranged from 64 to 3 (median ⫽ 4).
Step 2 reduced the number of sets to 36 by merging sets on the
basis of their frequency maps. The number of subjects contributing to these sets ranged from 123 to 3 (median ⫽ 40). Upon
completion of Step 2, there were 290 ICs with incorrect
assignment and 894 ICs with incorrect failure of assignment.
Step 3 reclassified 271 of the 290 ICs with incorrect assignment
and classified 832 of the 894 ICs with incorrect failure of
assignment. Note that two sets were emptied by this procedure.
Overall, the classification algorithm successfully classified
2,227 of the 2,438 ICs (91%), uncovering 34 RNs; the number
of participants contributing to these RNs ranged from 155 to
three (median ⫽ 69).
Three RNs were discarded because they contained a large
fraction of CSF voxels; four were discarded because they were
localized in areas not fully covered by the MRI acquisition
field of view, namely the cerebellum (three) and the brain stem
(one); and one was discarded because it encompassed a lower
frontal zone with strong signal attenuation due to susceptibility
artefacts (Ojemann et al. 1997). In addition, three RNs with
corresponding sets, including ⬍5% of the subjects, were discarded, leaving 23 RNs for further analysis.
These RNs were numbered from 1 to 23, according to the
decreasing number of individual participants included in the
corresponding sets, which ranged from 155 participants for
RN1 to 16 participants for RN23 (see Supplemental Table S1).
Overall, the 23 RNs covered 68% of the cortical GM (Fig. 2),
and basal ganglia structures were present in only one network
(RN22). Most RNs were bilateral or weakly lateralized, except
RN14 (right-lateralized) and RN3 (left-lateralized) and to a
lesser extent, RN10 and RN11 (left-lateralized). A detailed
anatomical description of each RN is presented below and in
the supplemental materials (see Supplemental Tables S2–S6).
FUNCTIONAL BRAIN ORGANIZATION AT REST
2757
M1c (Figs. 2 and 4, Supplemental Table S4). M1c was first
composed of medial and lateral frontal regions. On the medial
wall, M1c encompassed the medial frontal cortex (RN18,
including the SMA) and both anterior and middle parts of the
cingulum (RN19, RN23). On the lateral surface, M1c bilaterally covered the lower half of the frontal cortex, namely the
anterior part of the middle frontal gyrus (RN18, RN19, RN23)
and the triangular (RN23) and orbital (RN18, RN23) parts of
the inferior frontal gyrus. It also included the anterior insular
cortex. Besides the frontal representation, M1c also bilaterally
encompassed the supramarginal gyri (RN23) and to a lesser
extent, the MT gyrus (RN23). Finally, M1c was the only
module to include subcortical structures, both putamen and
caudate nucleus, bilaterally (RN22).
J Neurophysiol • VOL
M2a (Figs. 2 and 4, Supplemental Table S5). M2a first
covered both primary (RN13) and secondary (RN5, RN16)
sensory and motor areas along the pre- and postcentral gyri,
respectively, at the locations representing the head (RN5) and
the upper (RN13) and lower (RN16) limbs (Stippich et al.
2002). M2a also bilaterally included the rolandic opercule
(RN4), extending into the insula (RN4, RN5, RN16).
In each hemisphere, this module covered the primary and
secondary auditory areas (RN20) within Heschl and superior
temporal gyri, respectively.
At the parietal level, M2a also bilaterally encompassed the
anterior part of both superior and inferior parietal lobules
(RN12). Both medial and lateral frontal eye fields (RN12), the
SMA (RN4), the temporo-parietal junction (RN4), and the
105 • JUNE 2011 •
www.jn.org
Downloaded from http://jn.physiology.org/ by 10.220.33.2 on June 15, 2017
Fig. 2. Spatial maps of 22 resting-state networks (RNs) projected on the 3-dimensional
surface of the Montreal Neurological Institute single-subject brain. Each RN is displayed in the module to which it belongs.
Note that RN22, which encompasses mainly
the basal ganglia, is not displayed.
2758
FUNCTIONAL BRAIN ORGANIZATION AT REST
MT/visual area (V5) area at the temporo-occipital junction
(RN12) were also bilaterally observed. Structures of the RN12
correspond to the so-called dorsal attentional network (Corbetta and Shulman 2002; Fox et al. 2005).
M2b (Figs. 2 and 4 and Supplemental Table S6). M2b
bilaterally overlapped the majority of the occipital lobe. Along
the calcarine sulcus (RN17), M2b covered the primary V1 and
then encompassed the extrastriate visual cortex (Tootell et al.
1997), namely V2 and V3 (RN15), as well as V3a and V4
(RN8). On the medial occipital wall, M2b included the anterior
part of V1, the cuneus and lingual gyri (RN21).
Temporal Correlations
We will describe the significant correlations at the three
levels of clustering (systemic, modular, and RN) (Fig. 5).
At the highest hierarchical level, we found the two systems
S1 and S2 to be anticorrelated [r ⫽ ⫺0.23 ⫾ 0.27, mean ⫾ SD;
among the 180 participants, 71% of them (n ⫽ 127) showed a
significant anticorrelation (at P ⬍ 0.05) between both systems;
Fig. 4, left].
At the modular level (Fig. 5B), the pattern of negative
temporal correlations between modules was very specific:
significant negative correlations were exclusively observed
between M1a-b and M2a-b, with strongest anticorrelations
between M1b and M2a-b (Table 1). M1a was negatively
synchronized to a lesser extent with M2a.
A series of positive correlations also revealed a specific
pattern between modules (Fig. 5B, Table 1): M1a (DMN) was
synchronized with M1b (fronto-parieto-temporal networks),
which was itself synchronized with M1c (frontal/supramargin-
Fig. 4. Spatial maps of the 2 systems (S1, S2) and
the 5 modules, respectively. Within S1: M1a (yellow) included the default mode areas, M1b (pink)
included mainly parietal and frontal areas, M1c
(green) included mainly frontal areas, supramarginal gyri, and striatum; within S2: M2a (blue)
encompassed the parietal and temporal lobes, M2b
(cyan) encompassed the occipital lobe. All spatial
maps are overlaid onto lateral- and medial-inflated
surfaces of both hemispheres. Note that RN22,
encompassing the striatum (within M1c), is not
displayed.
J Neurophysiol • VOL
105 • JUNE 2011 •
www.jn.org
Downloaded from http://jn.physiology.org/ by 10.220.33.2 on June 15, 2017
Fig. 3. Hierarchical clustering analysis of the temporal correlations of the 23 RNs. The dendrogram of the analysis (left) was partitioned into 2 global systems (S1 and
S2) and 5 modules, 3 for S1 (M1a, M1b, M1c) and 2 for S2 (M2a, M2b). The 23 ⫻ 23 matrix, including the averaged temporal correlations of RNs, is on the right.
FUNCTIONAL BRAIN ORGANIZATION AT REST
2759
Table 1. Temporal correlations between spontaneous BOLD fMRI signal of modules
M1a
M1a
M1b
M1c
M2a
M2b
0.46 (0.11)
[0.20; 0.76]
0.24 (0.21)
[⫺0.32; 0.74]
0.44 (0.10)
[0.15; 0.76]
⫺0.12 (0.23)
[⫺0.59; 0.46]
0.41 (0.20)
[⫺0.14; 0.81]
0.50 (0.10)
[0.21; 0.78]
ⴚ0.28 (0.23)
[⫺0.71; 0.69]
ⴚ0.33 (0.24)
[⫺0.79; 0.36]
0.20 (0.23)
[⫺0.33; 0.78]
0.50 (0.13)
[0.22; 0.81]
0.19 (0.22)
[⫺0.40; 0.76]
ⴚ0.20 (0.24)
[⫺0.68; 0.42]
⫺0.12 (0.22)
[⫺0.61; 0.45]
0.33 (0.27)
[⫺0.33; 0.84]
0.68 (0.15)
[0.29; 0.95]
M1b
M1c
M2a
M2b
al/subcortical networks) and so on with M2a (sensory/motor/
attentional networks) and M2b (occipital networks).
At the RN level, significant temporal correlations represented 28% of all possible correlations between RNs, with 54
positive (Fig. 5C) and 17 negative correlations (Fig. 5D).
Concerning the pattern of negative correlations, the most
numerous significant correlations were between RNs of M1b
(fronto-parieto-temporal networks) and RNs of M2a (sensory/
motor/attentional networks; representing 60% of all significant
negative correlations). Two additional negative correlations
were revealed between RNs of M1b and M2b (RN3 and RN8,
respectively) and RNs of M1a and M2a (RN2 and RN4/12,
respectively). Finally, four additional negative correlations
were observed within the first system (S1): between two RNs
of M1a (RN2, core part of the DMN, and RN6) and two RNs
of M1c (RN18 and RN23).
As expected, only positive correlations were observed between RNs belonging to the same module. The series of
positive temporal correlations between RNs was consistent
with the correlation observed between the modules. In addition, one RN of M1a was positively synchronized with two
RNs included in the occipital lobe (M2b). Note that one
network (RN12) was also positively synchronized with one RN
of each subpart in S1 (RN9 of M1a, RN7 of M1b, and RN23
of M1c).
DISCUSSION
In the present work, spontaneous fluctuations of the BOLD
signal during the resting state were recorded in a large sample
of normal participants and analyzed using a fully unsupervised,
intersubject classifier. Our approach uncovered a hierarchical
functional parcellation of the brain into systems, modules, and
networks exhibiting complex, functional connectivity between
them. At the global level, we found two anticorrelated systems,
namely intrinsic (S1) and extrinsic (S2), associated with internal- and external-oriented processing, respectively (Golland et
al. 2008). Our study went further in partitioning these two
global systems into, respectively, three (M1a, M1b, M1c) and
two (M2a, M2b) distinct modules. We will first discuss the
partitioning at the coarsest level.
Two Anticorrelated Large-Scale Systems
The two large-scale and anticorrelated networks identified in
the present study appear to correspond with the so-called
J Neurophysiol • VOL
intrinsic and extrinsic systems previously reported by others
(Golland et al. 2007, 2008). Golland et al. (2007) defined the
extrinsic system as showing BOLD fluctuations locked with a
visual task, whereas the intrinsic system included all brain
areas whose activity was unrelated to that task. From a functional point of view, these extrinsic and intrinsic systems can
be considered as driven by sensory stimulation and by innerdriven mental activity, respectively. The S1 system reported
here, which includes the DMN, resembles the intrinsic system,
whereas our S2 system, which encompasses primary and secondary unimodal areas and the dorsal attention network,
closely matches the extrinsic system. However, our results
extend those from Golland et al. (2008) in showing that these
two systems can be defined without reference to any intercurrent, goal-directed task. In addition, we have demonstrated that
these two systems are temporally anticorrelated. Our findings
strengthen the proposal made by others that the brain is
basically organized into two anticorrelated systems. Fox et al.
(2005) and Fransson (2005) used a priori-defined seed regions
to identify two sets of regions with negatively correlated,
spontaneous signal fluctuations: one included the DMN and the
other, a fronto-parietal network overlapping the dorsal attention system. Our S1 and S2 systems included all areas belonging to the two sets reported by Fox et al. (2005). On the
contrary, sensory areas found in our S2 system were not
observed in the work of Fox et al. (2005), despite belonging to
the extrinsic network described by Golland et al. (2008). This
difference may be due to the use of an unsupervised method in
our study, with no a priori functional or anatomical limitations.
As a matter of fact, all previous works on the same topic have
used neuroanatomical priors, such as atlas-based region of
interest (e.g., He et al. 2009) or seed location, defined through
task-induced activation paradigms (e.g., Fox et al. 2005; Fransson 2005). Overall, the S1-S2 systems appear to reconcile and
extend previous descriptions of the brain-functional organization in two spatially distinct and anticorrelated, large-scale
brain networks.
Five Functionally Connected Modules
Our results also deepen the understanding of brain organization, as they reveal that these two systems can be segregated
into five distinct functional modules.
S1 (the intrinsic system) was split into three modules. The
M1a module included regions subtending the default-mode
105 • JUNE 2011 •
www.jn.org
Downloaded from http://jn.physiology.org/ by 10.220.33.2 on June 15, 2017
Each value is the average, SD, and range [minimum; maximum] of the 180 individual values. Bold indicates statistically significant correlations (P ⬍ 0.05
corrected). Off-diagonal values: temporal correlation between blood-oxygen level-dependent functional MRI (BOLD fMRI) signals of two different modules.
Diagonal values: temporal correlation between BOLD fMRI signals of two resting-state networks (RNs) averaged across all pairs of RNs belonging to the same
module.
2760
FUNCTIONAL BRAIN ORGANIZATION AT REST
processes. These regions were considered as supporting innerdirected processes such as spontaneous, self-related thoughts
(Vanhaudenhuyse et al. 2010). The second module (M1b) was
composed of a set of fronto-parieto-temporal networks involved in the storage and internal manipulation of verbal and
visual information (Curtis 2006; Curtis and D’Esposito 2003).
The final module (M1c) included insular, frontal, supramarginal, and subcortical areas and will be discussed according to
its relationships with other modules. S2 (the extrinsic system)
was subdivided: M2a was formed by primary and secondary
sensory, auditory, and motor cortices and by the dorsal attention network, whereas M2b included primary and secondary
visual areas.
We compared our modular partition with previous studies
based on an anatomically based regional atlas (He et al. 2009;
J Neurophysiol • VOL
Salvador et al. 2005). The Salvador et al. (2005) modular
partition followed essentially the lobar organization, thus only
the occipital module was common with our partition. Note that
comparison with this work is difficult, as they considered the
partial correlation between regions to build modular organization, and we considered the marginal correlation. In this sense
the work of He et al. (2009) was more comparable with ours.
We found a match for four modules out of their five identified
modules. On one side, discrepancies were observed on their
fifth module, which encompassed mainly basal ganglia, and
were poorly covered in our analysis. On the other side, they did
not describe an equivalent to our M1c module. We advocate
that this difference is likely related to a methodological choice.
The M1c module covered parts of three large areas, namely the
insula, supramarginal gyri, and anterior cingulate area included
105 • JUNE 2011 •
www.jn.org
Downloaded from http://jn.physiology.org/ by 10.220.33.2 on June 15, 2017
Fig. 5. Temporal correlations between the 2 systems (A), the 5 modules (B), and the 23 RNs (C and D), respectively. Black lines indicate statistically significant,
positive correlations; red lines indicate statistically significant, negative ones. The spatial maps of the systems/modules are overlaid onto lateral- and
medial-inflated surfaces of both hemispheres. The significance threshold was set at P ⬍ 0.05 corrected for the number of tests.
FUNCTIONAL BRAIN ORGANIZATION AT REST
J Neurophysiol • VOL
ported. Spreng et al. (2010) showed that this putative control
network was coactivated and synchronized with networks involved in both internally driven autobiographical and externally driven planning tasks. The parieto-frontal network described in these previous works included the M1b and M1c
modules described in the present study, indicating that this
control network is segregated in two highly synchronized
modules. Critically, we suggest that these two modules could
play a distinct role, since they are themselves synchronized
with distinct neural assemblies: M1c with the extrinsic system
via M2a (see above) and M1b with the default-mode module
M1a. More importantly, M1b exhibited negative temporal
correlations with both modules of the extrinsic system—a mark
of the competition between this module activity and that of the
extrinsic system as discussed above. We believe it is therefore
unlikely that M1b plays a role in the mediation between
intrinsic and extrinsic systems. Rather, we propose that this
role could be restricted to the module M1c, including the
salience network previously described.
Finally, we found that the two modules of the extrinsic
system, including sensory-motor areas, were positively correlated to each other. This fits with the claim that cortical areas
involved in sensory-input processing exhibit intense information exchange, sine qua non for a global and coherent representation of the external world (Mesulam 2008).
Global View
The aim of the present work was to explore the relationship
between the dichotomic and the modular organization of the
brain, which has resulted from various approaches to functional connectivity. Accordingly, we have shown that the
classical bipartition emerged from the interaction of five distinct modules. Figure 6 proposes a functional interpretation of
Fig. 6. Functional model of brain activity at rest. The functional relationships
are based on those observed between the modules (Fig. 5B). Red and gray
arrows indicate negative and positive correlations, respectively.
105 • JUNE 2011 •
www.jn.org
Downloaded from http://jn.physiology.org/ by 10.220.33.2 on June 15, 2017
in two modules, i.e., M1c/M2a, M1c/M1b, and M1c/M1a,
respectively. With the use of our approach, a large region can
express its functional variability and can belong to different
networks, in contrast to an anatomically based approach that
precludes such differentiation.
Regarding the temporal correlations between modules belonging to distinct systems, we reported the well-documented,
negative correlation between the default-mode (M1a) and the
sensory/motor/attentional (M2a) modules at rest (Fox et al.
2005; Fransson 2005; Tian et al. 2007). Such negative correlation has been interpreted as reflecting a balance between
inner thoughts and externally oriented activity. Our approach
revealed that the fronto-parieto-temporal (M1b) module was
the only one to be negatively synchronized with both modules
of the extrinsic system (Table 1). This latter finding leads us to
propose that the driving force of the intrinsic/extrinsic partition
at rest may rely more on this fronto-parieto-temporal (M1b)
module activity than that of the DMN proper (M1a), as suggested by others (Fox et al. 2005; Fransson 2005; Golland et al.
2008). At the functional level, this suggests a competition
between regions involved in storage of information (M1b) and
regions involved in low-level processing of sensory input (M2a
and M2b). Such competition has been previously evidenced in
task-induced activation studies. For instance, during mental
visual imagery tasks, numerous studies reported a deactivation
of visual areas, whereas a fronto-parietal network was activated (Mazard et al. 2002; Mellet et al. 2000). More recently,
Azulay et al. (2009) suggested a negative balance between
sensory cortex and lateral frontal and parietal cortex during a
verbal memory task. Our result strengthens the hypothesis that
even at rest, the sensory cortex must be isolated for an efficient
processing of internal representation (Azulay et al. 2009).
While correlations between S1 and S2 modules were essentially and logically negative, we also found a positive, temporal
correlation between a frontal/supramarginal/subcortical S1
module (M1c) and a sensory/motor/attentional S2 module
(M2a). This particular relationship appears to be mainly mediated by two M1c networks (RN18 and RN23). RN18, which
encompassed the inferior orbito-frontal, anterior insular, and
dorsal parts of the anterior cingulate cortices, corresponds to
the “salience” network, allowing a switch between the DMN
and the task-related networks (Bressler and Menon 2010).
These authors have further suggested that this network could
play a central role in “mediating dynamic interactions between
other large-scale brain networks involved in externally oriented
attention and internally oriented or self-related cognition”. Our
findings are in agreement with this statement, underscoring the
pivotal role of this network together with the additional networks that constituted the M1c. As a matter of fact, M1c was
the only module exhibiting positive correlations with modules
of both the extrinsic system (S2) and the intrinsic system (S1).
Although it belongs to the intrinsic system, M1c is in a position
to act as a go-between for externally and internally driven
cerebral activity.
Recent works have proposed that a parieto-frontal network
may mediate the balance between externally and internally
driven mental activity (Spreng et al. 2010; Vincent et al. 2008).
With the use of seed-based intrinsic connectivity on BOLD
data, Vincent et al. (2008) found a parieto-frontal network
distinct from both the dorsal-attentional network and the DMN;
however, temporal correlations between them were not re-
2761
2762
FUNCTIONAL BRAIN ORGANIZATION AT REST
ACKNOWLEDGMENTS
The authors are deeply indebted to Guy Perchey, Grégory Simon, and
Mathieu Vigneau for their help with data acquisition and analysis.
DISCLOSURES
No conflicts of interest, financial or otherwise, are declared by the author(s).
REFERENCES
Ashburner J, Friston KJ. Unified segmentation. Neuroimage 26: 839 – 851,
2005.
Azulay H, Striem E, Amedi A. Negative BOLD in sensory cortices during
verbal memory: a component in generating internal representations? Brain
Topogr 21: 221–231, 2009.
Beckmann CF, Smith SM. Probabilistic independent component analysis for
functional magnetic resonance imaging. IEEE Trans Med Imaging 23:
137–152, 2004.
Bressler SL, Menon V. Large-scale brain networks in cognition: emerging
methods and principles. Trends Cogn Sci 14: 277–290, 2010.
Buckner RL, Andrews-Hanna JR, Schacter DL. The brain’s default network: anatomy, function, and relevance to disease. Ann N Y Acad Sci 1124:
1–38, 2008.
J Neurophysiol • VOL
Christoff K, Gordon AM, Smallwood J, Smith R, Schooler JW. Experience
sampling during fMRI reveals default network and executive system contributions to mind wandering. Proc Natl Acad Sci USA 106: 8719 – 8724,
2009.
Corbetta M, Shulman GL. Control of goal-directed and stimulus-driven
attention in the brain. Nat Rev Neurosci 3: 201–215, 2002.
Curtis CE. Prefrontal and parietal contributions to spatial working memory.
Neuroscience 139: 173–180, 2006.
Curtis CE, D’Esposito M. Persistent activity in the prefrontal cortex during
working memory. Trends Cogn Sci 7: 415– 423, 2003.
Damoiseaux JS, Rombouts SA, Barkhof F, Scheltens P, Stam CJ, Smith
SM, Beckmann CF. Consistent resting-state networks across healthy subjects. Proc Natl Acad Sci USA 103: 13848 –13853, 2006.
Efron B, Halloran E, Holmes S. Bootstrap confidence levels for phylogenetic
trees. Proc Natl Acad Sci USA 93: 13429 –13434, 1996.
Fox MD, Raichle ME. Spontaneous fluctuations in brain activity observed
with functional magnetic resonance imaging. Nat Rev Neurosci 8: 700 –711,
2007.
Fox MD, Snyder AZ, Vincent JL, Corbetta M, Van Essen DC, Raichle
ME. The human brain is intrinsically organized into dynamic, anticorrelated
functional networks. Proc Natl Acad Sci USA 102: 9673–9678, 2005.
Fransson P. Spontaneous low-frequency BOLD signal fluctuations: an fMRI
investigation of the resting-state default mode of brain function hypothesis.
Hum Brain Mapp 26: 15–29, 2005.
Golland Y, Bentin S, Gelbard H, Benjamini Y, Heller R, Nir Y, Hasson U,
Malach R. Extrinsic and intrinsic systems in the posterior cortex of the
human brain revealed during natural sensory stimulation. Cereb Cortex 17:
766 –777, 2007.
Golland Y, Golland P, Bentin S, Malach R. Data-driven clustering reveals a
fundamental subdivision of the human cortex into two global systems.
Neuropsychologia 46: 540 –553, 2008.
He Y, Wang J, Wang L, Chen ZJ, Yan C, Yang H, Tang H, Zhu C, Gong
Q, Zang Y, Evans AC. Uncovering intrinsic modular organization of
spontaneous brain activity in humans. PLoS One 4: e5226, 2009.
Hyvarinen A. Fast and robust fixed-point algorithms for independent component analysis. IEEE Trans Neural Netw 10: 626 – 634, 1999.
Johnson SC. Hierarchical clustering schemes. Psychometrika 32: 241–254,
1967.
Mason MF, Norton MI, Van Horn JD, Wegner DM, Grafton ST, Macrae
CN. Wandering minds: the default network and stimulus-independent
thought. Science 315: 393–395, 2007.
Mazard A, Mazoyer B, Etard O, Tzourio-Mazoyer N, Kosslyn SM, Mellet
E. Impact of fMRI acoustic noise on the functional anatomy of visual mental
imagery. J Cogn Neurosci 14: 172–186, 2002.
Mazoyer B, Zago L, Mellet E, Bricogne S, Etard O, Houde O, Crivello F,
Joliot M, Petit L, Tzourio-Mazoyer N. Cortical networks for working
memory and executive functions sustain the conscious resting state in man.
Brain Res Bull 54: 287–298, 2001.
McKeown MJ, Sejnowski TJ. Independent component analysis of fMRI data:
examining the assumptions. Hum Brain Mapp 6: 368 –372, 1998.
McKiernan KA, D’Angelo BR, Kaufman JN, Binder JR. Interrupting the
“stream of consciousness”: an fMRI investigation. Neuroimage 29: 1185–
1191, 2006.
Mellet E, Tzourio-Mazoyer N, Bricogne S, Mazoyer B, Kosslyn SM, Denis
M. Functional anatomy of high-resolution visual mental imagery. J Cogn
Neurosci 12: 98 –109, 2000.
Menon V, Uddin LQ. Saliency, switching, attention and control: a network
model of insula function. Brain Struct Funct 214: 655– 667, 2010.
Mesulam M. Representation, inference, and transcendent encoding in neurocognitive networks of the human brain. Ann Neurol 64: 367–378, 2008.
Minka T. Automatic choice of dimensionality for PCA. MIT Media Laboratory Perceptual Computing Section Technical Report, No. 514, 2000.
Ojemann JG, Akbudak E, Snyder AZ, McKinstry RC, Raichle ME,
Conturo TE. Anatomic localization and quantitative analysis of gradient
refocused echo-planar fMRI susceptibility artifacts. Neuroimage 6: 156 –
167, 1997.
Raichle ME. Two views of brain function. Trends Cogn Sci 14: 180 –190,
2010.
Raichle ME, MacLeod AM, Snyder AZ, Powers WJ, Gusnard DA, Shulman GL. A default mode of brain function. Proc Natl Acad Sci USA 98:
676 – 682, 2001.
Salvador R, Suckling J, Coleman MR, Pickard JD, Menon D, Bullmore E.
Neurophysiological architecture of functional magnetic resonance images of
human brain. Cereb Cortex 15: 1332–1342, 2005.
105 • JUNE 2011 •
www.jn.org
Downloaded from http://jn.physiology.org/ by 10.220.33.2 on June 15, 2017
this interaction, where each of the five modules would support
a distinct, functional role.
We propose that the first module (M1a) of the intrinsic system is
associated with “emergence of spontaneous thoughts”. More
specifically, spontaneous thoughts refer to any internally oriented thoughts occurring naturally at rest, including mental
images, reminiscence of past experiences based on episodic
memory, and making plans (Mazoyer et al. 2001). Emergence
of these spontaneous thoughts was associated with activity of
the regions of the M1a module according to the studies realized
by Mason et al. (2007) or McKiernan et al. (2006). In fact, they
showed that the frequency of spontaneous thoughts was positively correlated with the activity measured in the brain regions
associated to the DMN (see the review of Buckner et al. 2008).
We suggest that the second module (M1b) of the intrinsic
system supports working memory processes. Especially, these
lateral frontal-parietal regions were described as engaged during both active manipulation and maintenance of internal
information (see reviews of Curtis and D’Esposito 2003; Wager and Smith 2003), such as spontaneous thoughts at rest
(Christoff et al. 2009).
Regarding the third module (M1c), our results support that it
could play a crucial role in switching between exogenous- and
endogenous-oriented mental activities. Such a role has been
previously suggested using a different approach (see the review
of Menon and Uddin 2010).
Finally, we suggest that regions associated with manipulation of information may reflect their central role, dealing with
spontaneous, unconstrained, and goal-directed, inner-oriented
thoughts by being positively correlated with the other brain
areas of the intrinsic system. We also suggest that brain
networks engaged during goal-directed activity act as a unique
mediator that functionally interacts with regions associated
with both internal- and external-oriented activity at rest. A
broad implication of this general attempt to describe each
modular functional role and interactions is a call for a detailed
and systematic study of the global, functional organization of
the brain at rest but also during goal-directed, cognitive tasks
on the same populations.
FUNCTIONAL BRAIN ORGANIZATION AT REST
Smith SM, Jenkinson M, Woolrich MW, Beckmann CF, Behrens TE,
Johansen-Berg H, Bannister PR, De Luca M, Drobnjak I, Flitney DE,
Niazy RK, Saunders J, Vickers J, Zhang Y, De Stefano N, Brady JM,
Matthews PM. Advances in functional and structural MR image analysis
and implementation as FSL. Neuroimage 23, Suppl 1: S208 –S219, 2004.
Spreng RN, Stevens WD, Chamberlain JP, Gilmore AW, Schacter DL.
Default network activity, coupled with the frontoparietal control network,
supports goal-directed cognition. Neuroimage 53: 303–317, 2010.
Stippich C, Ochmann H, Sartor K. Somatotopic mapping of the human
primary sensorimotor cortex during motor imagery and motor execution
by functional magnetic resonance imaging. Neurosci Lett 331: 50 –54,
2002.
Suzuki R, Shimodaira H. Pvclust: an R package for assessing the uncertainty
in hierarchical clustering. Bioinformatics 22: 1540 –1542, 2006.
Tian L, Jiang T, Liu Y, Yu C, Wang K, Zhou Y, Song M, Li K. The
relationship within and between the extrinsic and intrinsic systems indicated
by resting state correlational patterns of sensory cortices. Neuroimage 36:
684 – 690, 2007.
2763
Tootell RB, Mendola JD, Hadjikhani NK, Ledden PJ, Liu AK, Reppas JB,
Sereno MI, Dale AM. Functional analysis of V3A and related areas in
human visual cortex. J Neurosci 17: 7060 –7078, 1997.
Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O,
Delcroix N, Mazoyer B, Joliot M. Automated anatomical labeling of
activations in SPM using a macroscopic anatomical parcellation of the MNI
MRI single-subject brain. Neuroimage 15: 273–289, 2002.
Vanhaudenhuyse A, Demertzi A, Schabus M, Noirhomme Q, Bredart S,
Boly M, Phillips C, Soddu A, Luxen A, Moonen G, Laureys S. Two
distinct neuronal networks mediate the awareness of environment and of
self. J Cogn Neurosci, 2010.
Vincent JL, Kahn I, Snyder AZ, Raichle ME, Buckner RL. Evidence for a
frontoparietal control system revealed by intrinsic functional connectivity. J
Neurophysiol 100: 3328 –3342, 2008.
Wager TD, Smith EE. Neuroimaging studies of working memory: a metaanalysis. Cogn Affect Behav Neurosci 3: 255–274, 2003.
Ylipaavalniemi J, Vigario R. Analyzing consistency of independent components: an fMRI illustration. Neuroimage 39: 169 –180, 2008.
Downloaded from http://jn.physiology.org/ by 10.220.33.2 on June 15, 2017
J Neurophysiol • VOL
105 • JUNE 2011 •
www.jn.org