Neuron

Article

Functional Network Organization of the Human Brain

Jonathan D. Power,1,* Alexander L. Cohen,1 Steven M. Nelson,2 Gagan S. Wig,1 Kelly Anne Barnes,1 Jessica A. Church,1
Alecia C. Vogel,1 Timothy O. Laumann,1 Fran M. Miezin,1,3 Bradley L. Schlaggar,1,3,4,5 and Steven E. Petersen1,2,3,5,6,7
1Department of Neurology
2Department of Psychology
3Department of Radiology
4Department of Pediatrics
5Department of Anatomy and Neurobiology
6Department of Neurosurgery
7Department of Biomedical Engineering

Washington University in Saint Louis, St. Louis, MO 63130, USA

*Correspondence: [email protected]
DOI 10.1016/j.neuron.2011.09.006

SUMMARY mental interest to neuroscientists because they offer the first

opportunity to comprehensively and noninvasively explore the
Real-world complex systems may be mathematically functional network structure of the human brain (Bullmore and
modeled as graphs, revealing properties of the Sporns, 2009).
system. Here we study graphs of functional brain Although a variety of methods may be used to study rs-fcMRI
organization in healthy adults using resting state data, one of the most powerful and flexible approaches is the
functional connectivity MRI. We propose two novel graph theoretic approach (Bullmore and Sporns, 2009; Rubinov
and Sporns, 2010). Within this framework, a complex system is
brain-wide graphs, one of 264 putative functional
formalized as a mathematical object consisting of a set of items
areas, the other a modification of voxelwise networks
and a set of pairwise relationships between the items. Items are
that eliminates potentially artificial short-distance called nodes, relationships are called ties, and collections of
relationships. These graphs contain many subgraphs these nodes with their ties are called graphs or networks. A short
in good agreement with known functional brain and incomplete list of established topics in graph theory includes
systems. Other subgraphs lack established func- quantifying hierarchy and substructure within a graph, identifying
tional identities; we suggest possible functional hubs and critical nodes, determining how easily traffic flows in
characteristics for these subgraphs. Further, graph different portions and at different scales of a network, and esti-
measures of the areal network indicate that the mating the controllability of a system (Liu et al., 2011; Newman,
default mode subgraph shares network properties 2010). Because graph theoretic analyses can model properties
with sensory and motor subgraphs: it is internally at the level of the entire graph, subgraphs, or individual nodes,
and because the brain itself is a complex network, graph theo-
integrated but isolated from other subgraphs, much
retic approaches are a natural and attractive choice for rs-fcMRI
like a ‘‘processing’’ system. The modified voxelwise
analysis.
graph also reveals spatial motifs in the patterning of A current obstacle to the graph-based study of functional
systems across the cortex. brain organization is that it very difficult to define the individual
nodes that make up a brain network. On first principles, treating
a graph as a model of a real system, if the nodes of the graph
INTRODUCTION do not accurately represent real items in the system, the graph
itself is a distorted model and graph theoretic properties will
Advances in neuroimaging that facilitate the study of brain rela- diverge from the true properties of the system (Butts, 2009;
tionships in humans have stimulated an enormous amount of Smith et al., 2011; Wig et al., 2011). The brain is a complex
scientific and medical interest in recent years (Biswal et al., network with macroscopic organization at the level of functional
1995; Bullmore and Sporns, 2009; Deco et al., 2011; Dosenbach areas and subcortical nuclei, but the number and locations
et al., 2010). Resting state functional connectivity MRI (rs-fcMRI), of these entities in humans is largely unknown. Standard
which measures spontaneous low-frequency fluctuations in approaches to forming whole-brain rs-fcMRI graphs often ignore
blood oxygen level dependent (BOLD) signal in subjects at this issue and define nodes as voxels (Buckner et al., 2009; Cole
rest, has attracted particular attention for its ability to measure et al., 2010; Fransson et al., 2011; Tomasi and Volkow, 2011;
correlations in neural activity (via BOLD signal) between distant van den Heuvel et al., 2008), large parcels from anatomically
brain regions. These correlations are of great interest to the based brain atlases (Hartman et al., 2011; He et al., 2009;
medical community because an increasing number of pathologic Meunier et al., 2009a; Spoormaker et al., 2010; Tian et al.,
conditions appear to be reflected in functional connectivity 2011), or random interpolations between voxels and parcels
between particular brain regions (Church et al., 2009; Seeley (Hayasaka and Laurienti, 2010; Meunier et al., 2009b). These
et al., 2009). At the same time, these correlations are of funda- approaches are not meant to correspond to macroscopic

Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc. 665
 Neuron
Functional Brain Networks

‘‘units’’ of brain organization, and thus there is no direct reason RESULTS

to believe that these approaches result in well-formed nodes
(Wig et al., 2011). Comparing Networks: Defining Four Brain-Wide
An overarching goal of this report is to, at least partially, over- Networks
come this obstacle. We have developed methods to define, as Two novel and two standard methods of graph definition were
best we can, a set of more appropriate nodes, and to define examined within a large cohort of healthy young adults (and in
a network based upon these nodes (and the ties between a matched replication cohort; see Table S1 available online).
them). We also propose a second novel brain network, based To reiterate, graphs are composed of a set of nodes and a set
on a modification of voxel-wise approaches, and examine of ties between nodes. Graphs were formed using the nodes
some of its properties in relation to the first graph. Before described below, and ties were defined using Pearson correla-
studying these graphs in detail, we are obliged to demonstrate tion coefficients between node rs-fcMRI timecourses. The cross
that they (1) display signs of accuracy, and (2) improve upon correlation matrix of a set of nodes thus defines a graph.
previous graph definitions. Because most graph theoretic techniques are developed (and
Our evaluation of rs-fcMRI brain graphs rests upon a simple are most meaningful) in sparse graphs (Newman, 2010), thresh-
and fundamental argument. Decades of PET and fMRI experi- olds were applied to the graphs to eliminate weak ties (such that
ments have defined functional systems as groups of brain correlations under the threshold were ignored). Because there is
regions that coactivate during certain types of task (e.g., the no ‘‘correct’’ threshold, all analyses were performed over a range
dorsal attention system, (Corbetta and Shulman, 2002; Corbetta of thresholds, typically beginning around 10% tie density (retain-
et al., 1995); here and elsewhere we replace common neurosci- ing the strongest 10% of correlations) and rising until the
entific usage of ‘‘network’’ with ‘‘system,’’ reserving the word networks became severely fragmented (see Supplemental
network for the graph theoretic sense, such that ‘‘dorsal attention Experimental Procedures).
network’’ becomes ‘‘dorsal attention system’’). A more recent The first novel graph (referred to as the areal graph) was
large literature indicates that rs-fcMRI signal is specifically and defined in accord with neurobiological principles. The brain is
highly correlated within these functional systems (e.g., within a complex network with a hierarchical spatial and functional
the visual system, default mode system, dorsal attention system, organization (in the cortex) at the level of neurons, local circuits,
ventral attention system, auditory system, motor system, etc.) columns, functional areas, and functional systems. Standard
(Biswal et al., 1995; Dosenbach et al., 2007; Fox et al., 2006; rs-fcMRI analyses use cubic voxels that are a few millimeters
Greicius et al., 2003; Lowe et al., 1998; Nelson et al., 2010a). on each side, and thus can potentially resolve brain relation-
There is a family of methods (subgraph detection) that is used ships at the level of areas. Centers of putative areas were
to break large networks into subnetworks of highly related identified using two independent methods operating on data
nodes (subgraphs), such that nodes within subgraphs are more sets that were not used in graph analyses (see Experimental
densely connected (here, correlated) to one another than to the Procedures). The first method was meta-analytic in nature (as
rest of the graph. We hypothesized that specific patterns of in Dosenbach et al., 2006), and explored a large fMRI data
high correlation within functional systems would be reflected set to identify voxels that were reliably and significantly modu-
as subgraphs within a brain-wide rs-fcMRI network. Thus, the lated when certain behaviors were demanded (e.g., button-
presence of subgraphs that correspond to functional systems pressing) or certain signal types were found (e.g., error-related
is an indication that a graph accurately models some features activity). The second method extended a recently developed
of brain organization, and the absence of such subgraphs raises technique of mapping cortical areas using rs-fcMRI to entire
suspicions that a graph may not be well-defined. cortical sheets (fc-Mapping) (Barnes et al., 2011; Cohen et al.,
With this hypothesis in mind, we open this report by studying 2008; Nelson et al., 2010a). The combination of these methods
the subgraph structures of four brain-wide graphs within a single yielded 264 putative areas spanning the cerebral cortex,
data set. As mentioned above, two novel graphs are studied: subcortical structures, and the cerebellum (see Experimental
a graph of putative functional areas (264 nodes), and a modifica- Procedures, Figure S1, and Table S1 for analysis details, and
tion of voxelwise networks that excludes short-distance correla- Table S2 for coordinates). Regions of interest (ROIs) were
tions (40,100 nodes). Two other standard graphs are used for modeled as 10 mm diameter spheres. Graphs were formed using
comparison: a graph of parcels from a popular brain atlas ROIs as nodes (n = 264) and ties terminating within 20 mm
(90 nodes), and a standard voxelwise graph (40,100 nodes). To of a source node center were set to zero to avoid possible
presage the results, subgraphs in the areal network are signifi- shared signal between nearby nodes. This procedure yielded
cantly more like functional systems than subgraphs in the atlas- graphs of putative functional areas in which each node repre-
based graph, and subgraphs in the modified voxelwise network sented, to the best of our capabilities, an element of brain
are more like functional systems than the standard voxelwise organization.
network. Additionally, despite great differences in network size The second novel graph that was examined was a modification
and definition, the areal and modified voxelwise subgraphs are of voxelwise networks in which all short-distance ties were
remarkably alike and contain many subgraphs corresponding excluded. This modification arose from several practical obser-
to known functional systems, bolstering confidence in their vations. First, nearby voxels share nonbiological signal (causing
accuracy. Given these findings, we perform a variety of further increased rs-fcMRI correlation), a result of unavoidable steps
analyses upon the novel graphs to learn more about functional in data processing (e.g., reslicing, blurring). Second, short-
brain organization, with some novel and interesting results. distance relationships are especially susceptible to spurious

666 Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc.
Neuron
Functional Brain Networks

Figure 1. Areal Subgraph Structure Is Highly

Similar across Cohorts and Subgraph Structure Is
Similar between Areal and Modified Voxelwise
Graphs
Top left: A spring embedded layout of the areal graph at
4% tie density visualizing the graph and the basis for
subgraphs.
Top right: For both cohorts, plots are shown of the areal
assignments into subgraphs (colors) at tie densities from
10% down to 2% in 1% steps. ROI ordering is identical,
and all subgraphs with fewer than four members are
colored white. The standard measure of subgraph
similarity, normalized mutual information, between node
assignments of the cohorts at identical tie densities
ranged from 0.86 to 0.92, indicating highly similar patterns
across cohorts (1 = identical assignments, 0 = no infor-
mation shared between assignments).
Bottom: subgraphs from three thresholds are shown
for the areal (spheres) and modified voxelwise graphs
(surfaces). Note the similarity of subgraph assignments
between networks, despite the great difference in network
size and cortical coverage, even in different subjects (main
versus replication cohorts). All areal subgraphs with fewer
than four members are colored white, and all modified
voxelwise subgraphs with fewer than 100 voxels are
colored white. Areal networks are shown at 10%, 3%, and
2% tie density (r > 0.16, 0.30, and 0.33), and modified
voxelwise networks are shown at 5%, 2%, and 0.5% tie
density (r > 0.16, 0.23, and 0.31).

shorten addresses of individual nodes). Other

algorithms were tested and yielded similar
results (Figure S2).
Figure 1 illustrates our methodology and high-
lights several important results. The first panel
augmentation by subject motion (Power et al., 2011). Third, as depicts the areal graph in a spring embedded layout and maps
will be seen shortly, voxelwise graphs are dominated at higher subgraphs onto nodes using colors, visibly demonstrating the
thresholds by short-distance relationships, which are logically basis for subgraphs. In spring embedded layouts, ties act as
partially artificial based on the above considerations. Modified springs to position nodes in space such that well-connected
voxelwise networks are presented in which all ties terminating groups of nodes are pulled together, providing an intuitive and
within 20 mm of a source node are excluded, though other informative picture of the graph. The second panel shows the
distances (e.g., 15 mm and 25 mm) were also tested, with similar subgraph assignments of the areal network in both cohorts
results (data not shown). over a range of thresholds (each chart consists of 9 columns of
The two standard methods of graph formation were parcel- 264 color entries). ROIs are ordered identically for both cohorts,
based and voxel-based. The parcel-based graph was formed and the patterns of subgraph assignment across cohorts are in
using the 90-parcel AAL atlas (Tzourio-Mazoyer et al., 2002), good agreement. The standard graph theoretic measure of
a popular method of graph formation. This atlas divides the similarity between two sets of node assignments is normalized
cortex and subcortical structures into parcels based upon mutual information (NMI), which measures how much informa-
anatomical landmarks. The voxel-based graph was defined tion one set of assignments provides about another set of
using all voxels within the AAL atlas (n = 40,100), and the modi- assignments. Values of 1 indicate identical assignments, and
fied voxelwise graph was also defined using these voxels. values of 0 indicate that no information is gained about the
second set of assignments by knowing the first. Between
Comparing Networks: Correspondence between cohorts, NMI ranges from 0.86 to 0.92 across thresholds, indi-
Subgraphs and Functional Systems cating very similar assignments.
Subgraphs were determined over a range of thresholds for each The subgraph charts contain subgraphs whose composition
graph using one of the best-performing subgraph detection remains quite constant over thresholds (e.g., the horizontal
algorithms currently available (Infomap) (Fortunato, 2010; Ros- bands of blue, red, or yellow) as well as subgraphs that are
vall and Bergstrom, 2008). This algorithm uses the map equation hierarchically refined as thresholds rise (e.g., cyan becoming
to minimize information theoretic descriptions of random walks cyan, orange, pink, and purple). These patterns can be seen on
on the graph (essentially assigning zip codes to subgraphs to brain surfaces (Figure 1, bottom) as relatively constant subgraph

Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc. 667
 Neuron
Functional Brain Networks

Figure 2. Many Modified Voxelwise Subgraphs Replicate across Cohorts and Even within Single Subjects
Select subgraphs from the modified voxelwise analysis are presented from a dorsal view for both cohorts and for an additional single subject. Cohort subgraphs
are taken from the 2% tie density analysis and subgraphs in the individual are taken from a 0.5% tie density analysis. The overall NMI between cohort assignments
at this threshold was 0.71, and NMI values between subgraphs from different cohorts are shown in the matrix to the right. Additional views of this data and
replications of subgraphs from other thresholds are found in Figure S3.

compositions for visual (blue), default (red), or fronto-parietal tend to merge, and subgraphs come to resemble a patchwork
(yellow) regions over thresholds, and as refinement of the large of local subgraphs across the cortex (see circled regions in
cyan subgraph into hand somatosensory-motor (cyan), face Figure S4).
somatosensory-motor (orange), auditory (pink), and cingulo- To more quantitatively assess subgraph correspondence to
opercular (purple) subgraphs. This bottom panel of Figure 1 plots functional systems, we used NMI to compare groups of coordi-
areal assignments (spheres) in the main cohort over the modified nates from functional systems with the subgraph identities of
voxelwise assignments (surfaces) in the replication cohort, the nodes nearest to the coordinates under each network defini-
demonstrating the similarity of subgraphs over thresholds tion. A one-factor ANOVA of NMI demonstrates an effect of
across different cohorts and even across graph definitions. As graph (p < 10 7; see Figure S5). The AAL-based graph displays
Figure 2 shows, the modified voxelwise graphs also replicate the lowest correspondence (NMI = 0.37 ± 0.04, significantly
well across cohorts and even in single subjects. Fuller visualiza- lower than all other graphs) across thresholds, and the variable
tions of these data and replications of subgraphs from other structure of the voxelwise graph is reflected in NMI that ranges
thresholds are found in Figure S3. widely over thresholds (0.58–0.86), in contrast to the stable
We predicted that well-formed graphs would possess well- and high NMI found in the areal (0.72 ± 0.06) and modified
formed subgraphs corresponding to major functional systems voxelwise graphs (0.87 ± 0.15). Importantly, as thresholds rise,
of the brain. Figure 3 gives an overview of how well each network NMI between functional systems and subgraphs increases for
met this prediction. At left, PET and fMRI data defining major the modified voxelwise analysis, but decreases for the standard
functional systems are shown. The next three columns display voxelwise analysis.
subgraphs from a single threshold of analysis for each graph
(a high threshold, tailored to each graph). In the second column, Choosing Network Definitions for Further Analysis
areal and modified voxelwise assignments are shown simulta- The areal and modified voxelwise graphs best meet our predic-
neously because they are in such good agreement. The areal tions about the correspondence between functional systems
and modified voxelwise graphs contain subgraphs that corre- and subgraphs within brain-wide networks. The poorer corre-
spond to each of the functional systems, and these subgraphs spondence in the AAL-based and standard voxelwise graphs
contain most or all of the brain regions implicated in the func- likely results from coarse, nonfunctionally based nodes in the
tional systems, and sometimes also some extra brain regions. AAL-based graph, and the effects of millions of artificially high
In contrast, the AAL-based graph is incapable of representing short-range correlations between nearby voxels in the standard
most functional systems at this threshold (or any threshold; voxelwise graph. We turn now from our focus upon confirmatory
see Figure S4). The standard voxel-based graph represents findings to novel observations about functional brain organiza-
some functional systems well (e.g., the default mode system), tion that can be drawn from the areal and modified voxelwise
but others are only incompletely represented. Examination of graphs. We shall continue to focus on the network at the level
other thresholds of the standard voxelwise graph (Figure S4) of subgraphs. We begin by discussing the identities of
indicates that at low to moderate thresholds, reasonable subgraphs, then examine the relationships and properties of
subgraph representations of some functional systems are found, particular subgraphs, and end with observations about relation-
but that as thresholds rise, portions of functional systems ships between all subgraphs.

668 Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc.
Neuron
Functional Brain Networks

Figure 3. Graph Definition Dictates Fidelity to Functional Brain Organization

At left, the task-defined locations of four established functional systems. The next three columns display, for the main cohort, the single subgraph that best
corresponds to each functional system under the four graph definitions. Circles are placed around small portions of subgraphs that might otherwise be over-
looked (there are small green regions within green circles). Data from a single threshold tailored to each graph are shown. The threshold was the next-to-highest
threshold that each graph can achieve before the graph becomes severely fragmented (defined by the giant component containing <50% of the nodes in the
graph). Tailored thresholds were 3% for the areal graph, 5% for the AAL-based graph, and 2% for both voxel-based graphs. Correspondence between these
functional systems and subgraphs is good for the areal and modified voxelwise graphs, intermediate for the voxelwise graph, and poor for the AAL-based graph.
Note especially the correspondence between areal (spheres) and modified voxelwise (surface) subgraphs, despite great differences in network size (n = 264
versus n = 40,100). See Figure S4 and Figure S5 for more comprehensive and quantitative presentations of subgraph assignments. Images in the left column are
modified from (Corbetta et al., 2008; Corbetta and Shulman, 2002; Dosenbach et al., 2007; Shulman et al., 1997).

Subgraph Identities ized to either side of the central sulcus (Penfield and Boldrey,
The identities of the red (default), yellow (fronto-parietal task 1937), a finding consistent with recent investigations of primary
control), green (dorsal attention), and teal (ventral attention) motor and somatosensory cortex in rodents (Matyas et al.,
subgraphs are already clear. The remaining major subgraphs 2010). The division into ventral and dorsal subgraphs roughly
are now considered. separates the face from the rest of the body, a distinction
Several subgraphs correspond to sensory and motor regions confirmed by button-pushing and verb generation meta-analysis
(Figure 4, left). A visual system (blue) was identified, spanning data (Figure S1). Similar dorsal/ventral distinctions have recently
most of occipital cortex, often including a small portion of been found (Yeo et al., 2011). Intriguingly, correlations between
superior parietal cortex and a portion of the postero-lateral meta-analytic face SSM (orange) and auditory (pink) ROIs are
thalamus (potentially lateral geniculate nucleus [LGN], see hori- higher than correlations between body SSM (cyan) and auditory
zontal sections). At moderate thresholds, somatosensory-motor ROIs (auditory-face r = 0.16, auditory-hand r = 0.05, p < 0.001,
(SSM) cortex (S1, M1, and some pre- and postcentral-gyrus significant in both cohorts). These differential correlations are
cortex) was divided into dorsal (cyan) and ventral (orange) unlikely to reflect only anatomical connectivity, but instead might
subgraphs. These subgraphs also included voxels in the parietal be related to the history of coactivation that these regions surely
operculum that likely correspond to the second somatosensory share as a function of oral/aural language. Thus, it appears that
area (S2) (Burton et al., 2008), as well as a portion of the thalamus somatosensory and motor cortex are functionally divided into
possibly corresponding to ventral posterior thalamus (VP). At a ventral facial representation and a dorsal representation of
high thresholds, an auditory subgraph (pink) emerged from the the rest of the body (called ‘‘hand’’ for brevity).
purple cingulo-opercular subgraph. Two cingulo-opercular subgraphs (black and purple, Fig-
Rather than a division between somatosensory and motor ure 4, middle) are identified, both encompassing regions in
regions, a division between dorsal and ventral SSM regions anterior cingulate/medial superior prefrontal cortex (aCC),
is found. Although motor and sensory function are typically anterior prefrontal cortex (aPFC), and the anterior insula (aI)
localized to the pre- and postcentral gyri, respectively, classic (with additional regions in inferior and middle frontal gyrus and
descriptions of stimulus-evoked responses and sensations in supramarginal gyrus at multiple thresholds). Two distributed
humans indicate that these processes are not exclusively local- functional systems have been ascribed to cingulo-opercular

Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc. 669
 Neuron
Functional Brain Networks

Figure 4. Subgraph Identities

Left: visual (blue), auditory (pink), and hand (cyan) and face (orange) sensory-somatomotor (SSM) subgraphs are shown for the areal network at 2% (spheres) and
the modified voxelwise network at 0.5% tie density (surface). The mean correlations in the main cohort between auditory processing (pink, MNI: 38 33 17) and
hand (cyan, 40 19 54) and face (orange, 49 11 35) regions are shown below. Auditory-face correlations are significantly higher than auditory-hand
correlations in both cohorts (p < 0.001, two-sample two-tail t test).
Bottom: slices from the 4% tie density modified voxelwise analysis, with labels on relevant thalamic nuclei (numbers are z coordinates).
Middle: two cingulo-opercular subgraphs shown from the 3% areal (spheres) and 2% tie density modified voxelwise analysis (surface). Middle, published ROIs
(cingulo-opercular task control [Dosenbach et al., 2007]; salience [Seeley et al., 2007]) or modified voxelwise subgraphs, with an overlaid heat map of on-cue
meta-analysis activation. On-cue activity localizes to the purple subgraph.
Bottom: very strong fc-Mapping gradients are displayed separating the black and purple subgraphs, indicating that they possess distinct rs-fcMRI signals.
Right: at top, three unknown subgraphs from the 0.5% tie density modified voxelwise analyses are shown. The salmon subgraph (gray in all other figures, here
salmon for contrast) is reproduced with a 2% areal subgraph overlaid as spheres, and the strongest activations from the memory retrieval meta-analysis are
shown below. The light blue subgraph is also reproduced and the coordinates of a putative functional system from Nelson et al. (2010a) are overlaid as tan
spheres.

cortex: a cingulo-opercular control system first described by subgraph more likely relates to a salience system, though the
Dosenbach et al. (2006) as the ‘‘core’’ of a task performance evidence for such assignments is provisional.
system, which is thought to instantiate and maintain set during At least three distributed subgraphs with previously unknown
task performance, and the salience system of Seeley et al. functional identities are also found (Figure 4, right). The first
(2007). Relative to the black subgraph, the purple subgraph subgraph (salmon in Figure 4, gray in Figure 1) includes parts
lies anterior and ventral in aCC, lateral in aPFC, and dorsal in of posterior cingulate, posterior medial parietal, and lateral
the aI. Three pieces of data hint at the identities of these parietal cortex. We are unaware of any earlier characterizations
subgraphs. First, the coordinates reported for the task control of this collection of brain regions as a coherent functional
network are dorsal to salience coordinates in the insula (Dosen- system, but we found that these regions display the strongest
bach et al., 2007; Seeley et al., 2007), although most other coor- activation in our memory retrieval meta-analysis. Another distrib-
dinates do not distinguish the competing functional systems. uted subgraph (light blue) is found in frontal, parietal, and
Second, on-cue activity localizes to the purple subgraph in the temporal cortex at higher thresholds of the modified voxelwise
aI, aCC, and aPFC (the task control system was defined over analysis. This set of regions is not a commonly described
a range of tasks by on-cue activity entering a task block, sus- functional system, but recent work (fMRI and rs-fcMRI) (Nelson
tained activity during a task block, and error-related activity). et al., 2010a) has indicated that a very similar set of regions
Finally, the fc-Mapping technique detects a strong border (tan spheres in Figure 4) interposed between fronto-parietal
between the black and purple subgraphs at many locations, indi- and default regions may be a functional system, also implicated
cating that rs-fcMRI signal differs strongly between these in memory retrieval. Another novel subgraph is shown in
subgraphs, consistent with prior reports (Nelson et al., 2010b). plum, with representation in fusiform cortex, the precuneus,
We suggest that the purple subgraph more closely represents lateral and medial posterior parietal cortex, and superior frontal
the cingulo-opercular task control system, whereas the black cortex.

670 Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc.
Neuron
Functional Brain Networks

the second-order relationships among those other brain regions.

In other words, seed maps measure relationships in isolation,
whereas graphs capture these relationships and their context.
There is no necessary conflict in saying that a seed from dorsal
attention regions highlights broad swaths of cortex (the seed’s
voxelwise neighbors) and that graph-based analyses indicate
that some of these neighbors belong to other discrete sub-
graphs. Thus, the ‘‘task-positive system’’ seems to be com-
posed of at least three subgraphs, corresponding to distinct
attentional and task control systems.

The Default Mode System Is, from a Graph Theoretic

Perspective, Like Sensory and Motor Systems
Classic models of cognitive control posit that sensory informa-
tion is received, processed according to the demands of
a task, and an output is generated (Norman and Shallice,
1986). Processing at the input and output stages is thought to
Figure 5. The ‘‘Task Positive System’’ Consists of Multiple be relatively modular (not strictly in the graph theoretic sense),
Subgraphs, Including Dorsal Attention, Fronto-Parietal Task
whereas cognitive control mechanisms must flexibly adapt pro-
Control, and Cingulo-Opercular Task Control Systems
At left, the ‘‘task+ system’’ of Fox et al. (2005). At right, three subgraphs from
cessing to a wide range of task sets (Posner and Petersen, 1990).
the 0.5% tie density modified voxelwise analysis. The task+ system is On such an account, within a graph theoretic context, subgraphs
composed of at least three subgraphs, corresponding to the fronto-parietal thought to be responsible for task set or ‘‘control’’ ought to main-
task control, cingulo-opercular task control, and dorsal attention systems. tain a relatively diverse set of relationships, whereas sensory or
motor ‘‘processing’’ systems ought to have relatively compart-
mentalized sets of relationships.
We now shift from examining individual subgraphs to collec- The compartmentalization and diversity of relationships in
tions of subgraphs and their relationships to one another. graphs can be measured by two related, standard graph
measures: the local efficiency and participation coefficients of
The ‘‘Task-Positive System’’ Is Composed of Multiple nodes. Local efficiency is a measure of integration among the
Subgraphs Whereas the ‘‘Task-Negative System’’ neighbors of a node (the nodes a node has ties with): high local
Is Composed of a Single Subgraph efficiency means that a node is embedded within a richly con-
In an influential article, Fox et al. (2005) described a task-positive nected environment, and low local efficiency means that the
network that is broadly activated across tasks, and a task-nega- neighbors of the target node are sparsely connected to one
tive network that is broadly inactivated across tasks (Figure 5). another. The participation coefficient measures the extent to
Seed timecourses demonstrated that rs-fcMRI signal in one which a node connects to subgraphs other than its own. Low
network tended to rise as the signal in the other network fell, participation coefficients indicate that nodes are confined to
and the authors used seed correlation maps to suggest that interactions within their own subgraphs, whereas higher coeffi-
large portions of the brain are organized into two anticorrelated cients indicate that nodes connect to a variety of subgraphs.
networks. This framework is a useful heuristic, but the present Figure 6 plots subgraphs, local efficiency, and participation
results suggest a more complicated picture. coefficients for the areal graph over a range of thresholds.
The ‘‘task-negative system’’ corresponds predominantly to ‘‘Processing’’ systems ought to have high local efficiency and
a single subgraph (the default mode system), with possible low participation coefficients, reflected as hot colors in the
additional correspondence to the memory retrieval (salmon) middle panel and cool colors in the right panel of Figure 6. The
subgraph described above. The ‘‘task-positive system’’ is, visual (blue) and hand SSM (cyan) subgraphs meet this predic-
from a graph theoretic perspective, composed of at least three tion, as expected, and, intriguingly, so does the default mode
major subgraphs: the dorsal attention system (green), the system (red). The more diverse relationships of ‘‘control’’
fronto-parietal task control system (yellow), and the cingulo- systems, on the other hand, ought to be reflected in lower local
opercular task control system (purple). Because subgraphs are efficiencies and higher participation coefficients, seen as cooler
formed of nodes that are more related to one another than to colors in the middle panel and warmer colors in the right panel. In
the rest of the network, the rs-fcMRI timecourses of these comparison to ‘‘processing’’ systems, the fronto-parietal task
subgraphs must be distinct from one another. control (yellow) subgraph has significantly lower local efficiency
This highlights a fundamental difference between ‘‘resting and higher participation indices, as one would expect. ANOVA
state networks’’ defined by seed map analyses and the sub- and t tests confirm that these findings hold over a range of
graphs defined by graph-based approaches. Seed maps thresholds (see Figure 6).
measure only the relationships between a seed ROI and other These findings have several implications. Viewed from a graph
brain regions (usually voxels), whereas a graph of N nodes inte- theoretic perspective, sensory and motor systems and the
grates the information of N seed maps to capture not only the default mode system have similar levels of self-integration
relationships of a seed region to other brain regions, but also and self-containment. From the cognitive control perspective

Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc. 671
 Neuron
Functional Brain Networks

Figure 6. Default, Visual, and Somatosensory-Motor Systems Are Well-Integrated on Local Scales but Are Relatively Isolated in Relation to
Other Functional Systems
At top, the subgraphs, local efficiencies, and participation coefficients for all nodes in the areal network over a range of thresholds are shown. The local efficiency
of each node indicates the extent to which a node is embedded in a richly connected local environment. High (hot color) values indicate a richly connected local
environment. The participation coefficient of each node indicates the extent to which a node has ties to other subgraphs. Here, low (cool color) values indicate that
nodes are connected almost exclusively to members of their own subgraph. One-factor ANOVAs indicate a significant effect of subgraph at all thresholds for both
indices (all with p < 10 6), and post hoc t tests indicate that the cyan, blue, and red subgraphs have significantly higher local efficiencies and lower participation
coefficients at most or all thresholds than the yellow subgraph. Node assignments for a single threshold (4% tie density) are shown on a brain and in a spring
embedded layout, and the local efficiencies and participation coefficients of relevant subgraphs at this threshold are shown. Note that local efficiency is inde-
pendent of subgraph assignment, whereas participation coefficients depend upon subgraph assignment.

outlined above, these similarities would suggest that the default indeed have many ties, the isolated nature of the default mode
mode system acts more as a ‘‘processing system’’ than subgraph recasts the meaning of these nodes as hubs in the
a ‘‘control system’’ (in contrast with the fronto-parietal system). context of brain-wide rs-fcMRI networks.
Viewed from a perspective of temporal dynamics, the high simi-
larity of node relationships within SSM and visual systems and Functional Systems Are Arranged in Topological Motifs
the default mode system might indicate that these systems in across the Cortex
particular are relatively stationary, whereas other subgraphs One of the more striking features of the modified voxelwise anal-
such as task control systems might have more dynamic sets ysis is that subgraphs appear to be arranged in spatial motifs
of relationships. It should also be noted that several studies throughout the cortex. Figure 7 demonstrates the presence of
(Buckner et al., 2009; Cole et al., 2010) have implicated the motifs at a single threshold of the modified voxelwise analysis.
default mode system as the seat of the most prominent ‘‘hubs’’ For each subgraph, the distribution of its spatial interfaces
in rs-fcMRI brain graphs. Although default mode nodes may (defined as en face voxels) with other subgraphs is plotted,

672 Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc.
Neuron
Functional Brain Networks

Figure 7. Functional Systems Are Arranged into Topological Motifs across the Cortex
In charts, particular subgraphs at a single threshold are selected, the spatial boundaries of that subgraph are found, and the distribution of spatial interfaces (en
face voxels) to other subgraphs are calculated. The most frequent interfaces are plotted as percents of the total subgraph interface volume. Motifs are inferred by
finding instances where subgraphs interfacing with a subgraph are themselves very unlikely to interface. For instance, in the top chart, the light blue subgraph
interfaces most frequently with the yellow and red subgraphs, but red is only 3.6% of yellow’s interface, and yellow is only 2.6% of red’s interface. Below each
chart, plots of relevant subgraphs on brain surfaces visually demonstrate the repeated spatial patterns of subgraphs. Data from the modified voxelwise analysis at
1% tie density in the replication cohort are presented.

and then these neighboring subgraphs are examined to see predominantly with red and yellow subgraphs, which are them-
whether they are themselves unlikely to interface (implying a selves miniscule portions of each others’ borders (red is 3.5%
3-step motif). For example, the light blue subgraph interfaces of yellow’s border, and yellow is 2.6% of red’s border), implying

Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc. 673
 Neuron
Functional Brain Networks

a yellow-light blue-red motif. Plots of relevant subgraphs on inform the degree to which the default mode system is seen as
brain surfaces visually confirm the presence of motifs. Three a processing type of system versus a control type of system.
instances of this motif are demonstrated, for the light blue, black Such a finding need not contradict the description of posterior
(salience), and green (dorsal attention) subgraphs. Other 3-step members of the default mode system as cortical hubs (Buckner
motifs are present but not shown (e.g., red-teal-purple), and et al., 2009), but it may alter the understanding of what it means
these motifs can be found up and down subgraph hierarchies to be a hub.
(i.e., thresholds).
A principal concern about such spatial motifs is that they are Integrating the Present Findings with Other Approaches
artifactual—that they arise as intermediate mixtures of adjacent to Whole-Brain rs-fcMRI Analysis
signals, particularly when averaging over subjects. Although Recent investigations into the structure of functional brain orga-
these concerns cannot be entirely excluded, several interposed nization using a variety of methods (Erhardt et al., 2010; Yeo
subgraphs (e.g., the green dorsal attention system or the teal et al., 2011) have found some similar (but not identical) sets of
ventral attention system) have firm and extensive experimental resting state networks as the subgraphs reported here. We
bases. If these are not considered artifactual, then other consider convergence across methods to be a key indicator of
subgraphs deserve similar consideration. the validity of findings. We find the graph theoretic framework
to be especially useful, because it is capable of describing
DISCUSSION the overall graph (no such measures are presented in this article,
but small-world measures are an example), portions of the
Task-Free Approaches Delineate Functional Systems system (e.g., subgraphs), or individual nodes of the system
across the Cortex (e.g., local efficiency) within a common framework.
At the onset of functional neuroimaging some 25 years ago, Our findings have substantial implications for past and future
investigators made educated guesses about the types of opera- graph-based analyses. By examining multiple network defini-
tions that the human brain must perform, and designed experi- tions within a single data set, we were able to show how network
mental paradigms to elicit such operations (Lueck et al., 1989; definition profoundly affects the properties of a network, and
Pardo et al., 1991; Petersen et al., 1988; Posner et al., 1988). therefore the conclusions one would draw about the brain. Our
Over time, evidence accumulated implicating collections of brain results demonstrate drawbacks in some previous approaches,
regions that were assumed to share the burden of some set of while offering new approaches that appear to more plausibly
cognitive operations, defining functional systems (Corbetta and represent brain organization.
Shulman, 2002; Dosenbach et al., 2006; Raichle et al., 2001). It is important to recognize that these new approaches to
Until the study of spontaneous BOLD activity, however, the graph definition are not equivalent or interchangeable. Note
association of regions within a functional system was to some that in this article we examine several graph theoretic properties
extent dependent upon sets of task paradigms. Task-based of the areal graph, but restrict our discussions of modified voxel-
approaches left functional systems open to an interpretation wise data to spatial observations. The areal graph is formed
that rather than being a fundamentally related group of brain using our best estimates of the functional ‘‘units’’ in the brain,
regions within a brain-wide context, a functional system thus and many properties of this network should be fairly direct reflec-
defined might be just a transient and task-specific association tions of functional brain organization. On the other hand, the
of brain regions. modified voxelwise graph is defined using volumetric elements
The subgraphs presented herein were derived in task-free (voxels), and this graph reflects volumetric properties of func-
data using methods with no prior information about node iden- tional organization. In this graph, most functional areas are prob-
tity. There is substantial agreement between aspects of para- ably represented by many voxels, and large functional areas (and
digm-driven functional system definition in neuroimaging, and functional systems) will dominate the graph structure regardless
paradigm-free subgraphs derived in task-free activity. Even if of their roles in information processing relative to smaller areas or
one were to object that the areal network included functional systems. This volume-based definition thus warps representa-
assumptions via meta-analytic localizers, the modified voxelwise tions of information processing, limiting the conclusions that
analysis, which returned very similar results, made no such can be drawn from this graph.
assumptions. In a brain-wide context, several functional systems
are distinguished from each other by spontaneous activity. Directions for Future Work
This task-free definition of brain functional organization can The analyses presented here suggest several avenues for future
inform perspectives on cognitive function. For example, dorsal inquiry. Within graphs that possess many subgraphs with strong
and lateral frontal cortex appears to be apportioned among correspondence to functional systems, we have detected addi-
a variety of distributed subgraphs, many of which correspond tional subgraphs with no such identity but with hints of shared
to functional systems with known characteristics (Figure 2). activity in certain contexts (e.g., memory retrieval activity in the
This organization does not appear consistent with accounts of salmon and light blue subgraphs). Unifying functional attributes
cognition that posit rostro-caudal gradients or hierarchies across among these subgraphs should be sought and tested. Our
frontal cortex (Badre and D’Esposito, 2009; O’Reilly, 2010). results demonstrate strong within-subgraph connectivity in
In a related manner, the finding of similar graph properties sensory, motor and default mode systems, especially in contrast
(relatively dense internal relationships and relatively few external to task control systems, suggesting that these systems may
relationships) in visual, SSM, and default mode systems may differ in the dynamics of their relationships with other subgraphs

674 Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc.
Neuron
Functional Brain Networks

over time. Our analyses only examined static pictures of graphs analyses demonstrate, the human cortex possesses a complex
obtained by summarizing activity over entire epochs into a single and dense topography of functional systems, underscoring the
correlation coefficient, and future work should explore if and need for ‘‘tedious anatomy’’ in neuroimaging studies (Devlin
how these relationships change over time. Perhaps the most and Poldrack, 2007).
obvious avenue for future work will lie in the comparison of graphs
across the lifespan and in disease. A recognized limitation within EXPERIMENTAL PROCEDURES
graph theoretic investigations of structural and functional
Subjects
brain networks is the current lack of validated parcellation strat-
Healthy young adults were recruited from the Washington University campus
egies (see Fornito et al., 2010; Wig et al., 2011; Zalesky et al.,
and the surrounding community. All subjects were native English speakers and
2010) for comprehensive discussions). We have derived and pre- right-handed. All subjects gave informed consent and were compensated for
sented a graph of 264 putative functional areas that displays a their participation.
plausible functional structure that should be sensitive to the orga-
nization of many functional systems. If the locations of functional Data Sets and Data Collection
areas do not greatly differ across populations (Barnes et al., This study utilized multiple data sets. The first and second data sets were used
for meta-analytic and fc-mapping analyses, respectively. The third data set
2011), this graph should be applicable to a wide variety of
was used for rs-fcMRI network analysis. The first (n > 300, detailed in Table
populations, such as clinical or developmental cohorts. S1) and second data sets (n = 40) were acquired on a Siemens 1.5 Tesla
MAGNETOM Vision MRI scanner (Erlangen, Germany) as described in Dosen-
Limitations bach et al. (2010). The third data set (n = 106: a 53 subject cohort, 52 subject
The present study should be considered a preliminary draft cohort, and an additional single subject) was acquired on a Siemens
of functional brain networks and has many limitations. The MAGNETOM Tim Trio 3.0T Scanner with a Siemens 12 channel Head Matrix
Coil (Erlangen, Germany) as described in Dosenbach et al. (2010). See Supple-
methods of locating putative functional areas may certainly
mental Experimental Procedures for acquisition details.
have overlooked, misplaced, or fabricated some areas. Addi-
tionally, the spherical ROIs used to model functional areas Data Processing
do not reflect the true shapes of functional areas. However, Functional images underwent standard fMRI preprocessing to reduce arti-
because subgraph structures in areal and modified voxelwise facts, register subjects to a target atlas, and resample the data on a 3 mm
networks were remarkably alike, this does not seem to have isotropic grid (Shulman et al., 2010). See Supplemental Experimental Proce-
crippled the endeavor. This study used a single signal (BOLD) dures for further details.
with known susceptibility artifacts in temporal and orbitofrontal
rs-fcMRI Processing
cortex. Accordingly, much remains to be discovered about For rs-fcMRI analyses, several additional preprocessing steps were utilized to
the organization of the ventral surface of the brain, as well as reduce spurious variance unlikely to reflect neuronal activity (Fox et al., 2009).
subcortical and cerebellar organization (see Buckner et al., These steps included: (1), a temporal band-pass filter (0.009 Hz < f < 0.08 Hz)
2011). One additional limitation inherent to fMRI is resolution: and spatial smoothing (6 mm full width at half maximum); (2), regression of six
voxels are 3 mm on each side, and partial voluming as well as parameters obtained by rigid body head motion correction; (3), regression of
the smoothing inherent in data processing limit the resolution the whole brain signal averaged across the whole brain; (4), regression of
ventricular signal averaged from ventricular ROIs; and (5), regression of white
that these studies can achieve. To offset these undesired effects,
matter signal averaged from white matter ROIs. The first derivatives of these
short-distance relationships were eliminated from areal and regressors were also regressed.
modified voxelwise analyses, and single subjects were exam-
ined. Future efforts that refine rs-fcMRI techniques and integrate Meta-Analytic ROI Definition
findings from other modalities, such as structural imaging, EEG, The first method of identifying putative functional areas searched a large fMRI
or MEG, will provide valuable additions and refinements to our data set acquired in a single scanner (data set 1) for brain regions that reliably
displayed significant activity when certain tasks were performed (e.g., button-
observations, both in terms of identifying the functional ‘‘units’’
pressing) or certain signal types (e.g., error-related activity) were expected
of the human brain and in more completely modeling functional (see Table S1). Meta-analyses identified 322 ROIs (10 mm diameter spheres,
brain networks in space and time. see Figure S1), which were reduced to a final collection of 151 nonoverlapping
meta-analytic ROIs. Full details of meta-analyses are available in Supple-
CONCLUSIONS mental Experimental Procedures.

We close with two broad points. First, there is a growing trend fc-Mapping ROI Definition
fc-Mapping techniques were applied to eyes-open fixation rs-fcMRI data from
to examine healthy and pathological brain activity in terms of
40 healthy young adults (data set 2: 27 M/13 F, average age = 26.4 years old,
networks (Bullmore and Sporns, 2009; Church et al., 2009; average RMS movement = 0.42 mm, average number of volumes = 432). See
Seeley et al., 2009). The sensitivity and specificity of such anal- Cohen et al. (2008) and Nelson et al. (2010a) for full conceptual and technical
yses is directly linked to the comprehensiveness and accuracy descriptions of fc-Mapping on cortical patches. Here, patches extending over
of the framework used to examine brain networks. The frame- the entire cortical surface (one per hemisphere) were used to define putative
work used in this report appears to be reasonably accurate, functional areas. This technique generated 254 ROIs across the cortex, which
were reduced to a final set of 193 nonoverlapping ROIs. See Supplemental
and is capable of describing networks as a whole, as subgraphs,
Experimental Procedures for further details.
or as individual nodes, making it a powerful tool for examining
functional relationships in the human brain. Second, the accu- Areal ROI Set Formation
racy of connectivity analyses depends upon the isolation of Meta-analytic ROIs and fc-Mapping ROIs were merged to form a maximally-
relevant or unique signals. As the areal and modified voxelwise spanning collection of ROIs. Meta-analytic ROIs were given preference, and

Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc. 675
 Neuron
Functional Brain Networks

nonoverlapping fc-Mapping ROIs were then added, resulting in 264 indepen- Subgraph Detection and Graph Analysis
dent ROIs. For a given network at a given threshold, the correlations below the threshold
were set to zero, and the resulting matrix was subjected to subgraph detection
Parcel-Based, Voxel-Based, and Modified Voxelwise algorithms. We utilized the Infomap algorithm, one of the best-performing
Network Formation algorithms on multiple benchmark networks (Fortunato, 2010; Lancichinetti
A 90-node parcel-based network was formed by using the 90-parcel auto- and Fortunato, 2009). Other algorithms were tried, with similar results.
mated anatomical labeling (AAL) atlas (Tzourio-Mazoyer et al., 2002) to assign Subgraph assignments were returned as numbers, which were then mapped
all voxels (n = 44,100) within the atlas into 90 parcels. An average timecourse onto nodes and ROIs as colors.
was formed for each parcel by averaging the timecourses of all nodes within Local efficiency was calculated after (Latora and Marchiori, 2001). Participa-
the parcel. A 44,100-node voxelwise network was defined from all voxels tion coefficients were calculated after (Guimerà et al., 2005). Binary networks
within the AAL atlas (Tzourio-Mazoyer et al., 2002). The modified voxelwise were used for calculations.
networks arose by masking out ties that terminated within 20 mm of the
source voxel. Distances of 15–25 mm were tested, with similar results across Computations and Visualizations
networks. Analyses were performed on all voxels in both hemispheres (n = MRI images were processed using in-house software. Network calculations
44,100), and also on all voxels within a single hemisphere (n = 22,050). Single were performed using MATLAB (The MathWorks, Natick, MA). The Infomap
hemisphere analyses were much less computationally demanding, permitting algorithm was provided by Rosvall and Bergstrom (2008). Network visualiza-
a wider range of analysis), and results between single- and dual-hemisphere tions were created using the Social Network Image Animator (SoNIA) software
analyses were similar. All figures except Figure 3 (both hemispheres were package (Bender-deMoll and McFarland, 2006). Brain surface visualizations
used for consistency with the voxelwise analysis and the rest of the literature were created using Caret software and the PALS surface (Van Essen, 2005;
in this figure) in the article portray single-hemisphere analyses. Van Essen et al., 2001).

Formation of Two Subject Cohorts for rs-fcMRI Network Analysis SUPPLEMENTAL INFORMATION
rs-fcMRI networks were studied in continuous eyes-open fixation data from
two cohorts (data set 3) of healthy young adults, matched for age, sex, move- Supplemental Information includes five figures, two tables, and Supplemental
ment and number of volumes in scans, as shown in Table S1. These subjects Experimental Procedures and can be found with this article online at
underwent a rigorous quality control process to correct for subject motion doi:10.1016/j.neuron.2011.09.006.
(Power et al., 2011). See Supplemental Experimental Procedures for details.
Reported numbers of volumes (time frames of rs-fcMRI data) and RMS are ACKNOWLEDGMENTS
for the final, usable, data (Table S1). Data cleaning for subject movement
during the scan removed 6% of the data from subjects (range 4%–8%), and We thank Nico Dosenbach, Thomas Pearce, Bradley Miller, and our reviewers
each cohort contained a mean of 350 frames of data per subject (range for their attentive reading of this manuscript. We thank Olaf Sporns and
215–501 frames). The single subject in Figure 2 had 1181 frames of data. Mika Rubinov for technical help with graph analysis, and Joe Dubis for help
with meta-analyses. This work was supported by NIH R21NS061144 (S.P.),
rs-fcMRI Graph Formation NIH R01NS32979 (S.P.), a McDonnell Foundation Collaborative Action
Given a collection of N ROIs (parcels, voxels, or putative areas), within each Award (S.P.), NIH R01HD057076 (B.L.S.), NIH F30NS062489 (A.L.C.), NIH
subject, timecourses are extracted for all ROIs and an N 3 N correlation matrix U54MH091657 (David Van Essen), and NSF IGERT DGE-0548890 (Kurt
is calculated. An average matrix is formed across all subjects in a cohort, and Thoroughman).
the diagonal is set to zero. This defines a weighted graph.
Typical graph analyses of weighted networks ignore negative ties and are Accepted: September 2, 2011
obliged to explore a range of thresholds to characterize the properties of Published: November 16, 2011
a network (Power et al., 2010; Rubinov and Sporns, 2010). Recent proposals
to incorporate negative weights into analyses of subgraph detection have REFERENCES
been made (Rubinov and Sporns, 2011; Traag and Bruggeman, 2009), but
here we follow the traditional approach. Many real-world networks have tie Badre, D., and D’Esposito, M. (2009). Is the rostro-caudal axis of the frontal
densities of a few percent or less (Newman, 2010), and the graph analytic lobe hierarchical? Nat. Rev. Neurosci. 10, 659–669.
techniques utilized here were developed upon such networks (Fortunato, Barnes, K.A., Nelson, S.M., Cohen, A.L., Power, J.D., Coalson, R.S., Miezin,
2010; Newman, 2010; Rosvall and Bergstrom, 2008). Accordingly, the anal- F.M., Vogel, A.C., Dubis, J.W., Church, J.A., Petersen, S.E., and Schlaggar,
yses presented here typically span a threshold range on the order of 10% B.L. (2011). Parcellation in left lateral parietal cortex is similar in adults and chil-
down to 1% tie density though the precise range depends upon the network dren. Cereb. Cortex., in press. Published online August 1, 2011. 10.1093/cer-
(for example, the AAL-based parcel network becomes severely fragmented cor/bhr189.
below 4% tie density and we do not present results from such thresholds). Bender-deMoll, S., and McFarland, D.A. (2006). The art and science of
In general, results are presented over a range of thresholds to give the reader dynamic network visualization. J. Soc. Struct. 7.
a sense of the dependence of a property upon thresholds, and no formal defi- Biswal, B., Yetkin, F.Z., Haughton, V.M., and Hyde, J.S. (1995). Functional
nition of threshold ranges is proposed, because it is essentially arbitrary. connectivity in the motor cortex of resting human brain using echo-planar
As noted in the text, short-range correlations can arise from shared patterns MRI. Magn. Reson. Med. 34, 537–541.
of local neuronal activity, but they can also arise from aspects of data process-
Buckner, R.L., Sepulcre, J., Talukdar, T., Krienen, F.M., Liu, H., Hedden, T.,
ing (e.g., reslicing, blurring), as well as motion-induced artifacts (Power et al.,
Andrews-Hanna, J.R., Sperling, R.A., and Johnson, K.A. (2009). Cortical
2011). Local correlations are thus combinations of neurobiological and artifac-
hubs revealed by intrinsic functional connectivity: mapping, assessment of
tual signal. To minimize the effects of questionable correlations on network
stability, and relation to Alzheimer’s disease. J. Neurosci. 29, 1860–1873.
structure, ties terminating within 20 mm of the source ROI are set to zero in
all areal network analyses and in the modified voxelwise analysis. Although Buckner, R.L., Krienen, F.M., Castellanos, A., Diaz, J.C., and Yeo, B.T.T.
this process does not completely remove the effect of reslicing and blurring (2011). The organization of the human cerebellum estimated by intrinsic func-
on correlations in the data (consider a voxel’s correlations to distant but adja- tional connectivity. J. Neurophysiol., in press. Published online July 27, 2011.
cent voxels), it removes a considerable portion of correlations of questionable 10.1152/jn.00339.2011.
origin. This procedure eliminated 635 (4.1%) of the 15,375 positive ties in the Bullmore, E., and Sporns, O. (2009). Complex brain networks: graph theoret-
areal network, and 15.3 million (4.2%) of 470 million ties in the single hemi- ical analysis of structural and functional systems. Nat. Rev. Neurosci. 10,
sphere voxelwise network. 186–198.

676 Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc.
Neuron
Functional Brain Networks

Burton, H., Sinclair, R.J., Wingert, J.R., and Dierker, D.L. (2008). Multiple pari- Guimerà, R., Mossa, S., Turtschi, A., and Amaral, L.A.N. (2005). The worldwide
etal operculum subdivisions in humans: tactile activation maps. Somatosens. air transportation network: Anomalous centrality, community structure, and
Mot. Res. 25, 149–162. cities’ global roles. Proc. Natl. Acad. Sci. USA 102, 7794–7799.
Butts, C.T. (2009). Revisiting the foundations of network analysis. Science 325, Hartman, D., Hlinka, J., Palus, M., Mantini, D., and Corbetta, M. (2011). The
414–416. role of nonlinearity in computing graph-theoretical properties of resting-state
Church, J.A., Fair, D.A., Dosenbach, N.U., Cohen, A.L., Miezin, F.M., Petersen, functional magnetic resonance imaging brain networks. Chaos 21, 013119.
S.E., and Schlaggar, B.L. (2009). Control networks in paediatric Tourette Hayasaka, S., and Laurienti, P.J. (2010). Comparison of characteristics
syndrome show immature and anomalous patterns of functional connectivity. between region-and voxel-based network analyses in resting-state fMRI
Brain 132, 225–238. data. Neuroimage 50, 499–508.
Cohen, A.L., Fair, D.A., Dosenbach, N.U.F., Miezin, F.M., Dierker, D., He, Y., Wang, J., Wang, L., Chen, Z.J., Yan, C., Yang, H., Tang, H., Zhu, C.,
Van Essen, D.C., Schlaggar, B.L., and Petersen, S.E. (2008). Defining func- Gong, Q., Zang, Y., and Evans, A.C. (2009). Uncovering intrinsic modular orga-
tional areas in individual human brains using resting functional connectivity nization of spontaneous brain activity in humans. PLoS ONE 4, e5226.
MRI. Neuroimage 41, 45–57.
Lancichinetti, A., and Fortunato, S. (2009). Community detection algorithms:
Cole, M.W., Pathak, S., and Schneider, W. (2010). Identifying the brain’s most a comparative analysis. Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 80,
globally connected regions. Neuroimage 49, 3132–3148. 056117.
Corbetta, M., and Shulman, G.L. (2002). Control of goal-directed and stimulus- Latora, V., and Marchiori, M. (2001). Efficient behavior of small-world
driven attention in the brain. Nat. Rev. Neurosci. 3, 201–215. networks. Phys. Rev. Lett. 87, 198701.
Corbetta, M., Shulman, G.L., Miezin, F.M., and Petersen, S.E. (1995). Superior Liu, Y.-Y., Slotine, J.-J., and Barabási, A.L. (2011). Controllability of complex
parietal cortex activation during spatial attention shifts and visual feature networks. Nature 473, 167–173.
conjunction. Science 270, 802–805.
Lowe, M.J., Mock, B.J., and Sorenson, J.A. (1998). Functional connectivity in
Corbetta, M., Patel, G., and Shulman, G.L. (2008). The reorienting system of single and multislice echoplanar imaging using resting-state fluctuations.
the human brain: from environment to theory of mind. Neuron 58, 306–324. Neuroimage 7, 119–132.
Deco, G., Jirsa, V.K., and McIntosh, A.R. (2011). Emerging concepts for the Lueck, C.J., Zeki, S., Friston, K.J., Deiber, M.P., Cope, P., Cunningham, V.J.,
dynamical organization of resting-state activity in the brain. Nat. Rev. Lammertsma, A.A., Kennard, C., and Frackowiak, R.S.J. (1989). The colour
Neurosci. 12, 43–56. centre in the cerebral cortex of man. Nature 340, 386–389.
Devlin, J.T., and Poldrack, R.A. (2007). In praise of tedious anatomy. Matyas, F., Sreenivasan, V., Marbach, F., Wacongne, C., Barsy, B., Mateo, C.,
Neuroimage 37, 1033–1041, discussion 1050–1058. Aronoff, R., and Petersen, C.C.H. (2010). Motor control by sensory cortex.
Dosenbach, N.U.F., Visscher, K.M., Palmer, E.D., Miezin, F.M., Wenger, K.K., Science 330, 1240–1243.
Kang, H.C., Burgund, E.D., Grimes, A.L., Schlaggar, B.L., and Petersen, S.E. Meunier, D., Achard, S., Morcom, A., and Bullmore, E. (2009a). Age-related
(2006). A core system for the implementation of task sets. Neuron 50, 799–812. changes in modular organization of human brain functional networks.
Dosenbach, N.U.F., Fair, D.A., Miezin, F.M., Cohen, A.L., Wenger, K.K., Neuroimage 44, 715–723.
Dosenbach, R.A.T., Fox, M.D., Snyder, A.Z., Vincent, J.L., Raichle, M.E.,
Meunier, D., Lambiotte, R., Fornito, A., Ersche, K.D., and Bullmore, E.T.
et al. (2007). Distinct brain networks for adaptive and stable task control in hu-
(2009b). Hierarchical modularity in human brain functional networks. Front.
mans. Proc. Natl. Acad. Sci. USA 104, 11073–11078.
Neuroinform. 3, 37.
Dosenbach, N.U., Nardos, B., Cohen, A.L., Fair, D.A., Power, J.D., Church,
Nelson, S.M., Cohen, A.L., Power, J.D., Wig, G.S., Miezin, F.M., Wheeler, M.E.,
J.A., Nelson, S.M., Wig, G.S., Vogel, A.C., Lessov-Schlaggar, C.N., et al.
Velanova, K., Donaldson, D.I., Phillips, J.S., Schlaggar, B.L., and Petersen,
(2010). Prediction of individual brain maturity using fMRI. Science 329,
S.E. (2010a). A parcellation scheme for human left lateral parietal cortex.
1358–1361.
Neuron 67, 156–170.
Erhardt, E.B., Rachakonda, S., Bedrick, E.J., Allen, E.A., Adali, T.l., and
Nelson, S.M., Dosenbach, N.U., Cohen, A.L., Wheeler, M.E., Schlaggar, B.L.,
Calhoun, V.D. (2010). Comparison of multi-subject ICA methods for analysis
and Petersen, S.E. (2010b). Role of the anterior insula in task-level control and
of fMRI data. Hum. Brain Mapp., in press. Published online December 15,
focal attention. Brain Struct. Funct. 214, 669–680.
2010. 10.1002/hbm.21170.
Newman, M.E.J. (2010). Networks: An Introduction (Oxford: Oxford University
Fornito, A., Zalesky, A., and Bullmore, E.T. (2010). Network scaling effects in
Press).
graph analytic studies of human resting-state FMRI data. Front. Syst.
Neurosci. 4, 22. Norman, D.A., and Shallice, T. (1986). Attention to action: willed and automatic
control of behavior. In Consciousness and Self-Regulation, R.J. Davidson,
Fortunato, S. (2010). Community detection in graphs. Phys. Rep. 486, 75–174.
G.E. Schwartz, and D. Shapiro, eds. (New York: Plenum Press), pp. 1–18.
Fox, M.D., Snyder, A.Z., Vincent, J.L., Corbetta, M., Van Essen, D.C., and
O’Reilly, R.C. (2010). The what and how of prefrontal cortical organization.
Raichle, M.E. (2005). The human brain is intrinsically organized into dynamic,
Trends Neurosci. 33, 355–361.
anticorrelated functional networks. Proc. Natl. Acad. Sci. USA 102, 9673–
9678. Pardo, J.V., Fox, P.T., and Raichle, M.E. (1991). Localization of a human
system for sustained attention by positron emission tomography. Nature
Fox, M.D., Corbetta, M., Snyder, A.Z., Vincent, J.L., and Raichle, M.E. (2006).
349, 61–64.
Spontaneous neuronal activity distinguishes human dorsal and ventral atten-
tion systems. Proc. Natl. Acad. Sci. USA 103, 10046–10051. Penfield, W., and Boldrey, E. (1937). Somatic motor and sensory representa-
Fox, M.D., Zhang, D., Snyder, A.Z., and Raichle, M.E. (2009). The global signal tion in the cerebral cortex of man as studied by electrical stimulation. Brain
and observed anticorrelated resting state brain networks. J. Neurophysiol. 60, 389–443.
101, 3270–3283. Petersen, S.E., Fox, P.T., Posner, M.I., Mintun, M., and Raichle, M.E. (1988).
Fransson, P., Aden, U., Blennow, M., and Lagercrantz, H. (2011). The func- Positron emission tomographic studies of the cortical anatomy of single-
tional architecture of the infant brain as revealed by resting-state fMRI. word processing. Nature 331, 585–589.
Cereb. Cortex 21, 145–154. Posner, M.I., and Petersen, S.E. (1990). The attention system of the human
Greicius, M.D., Krasnow, B., Reiss, A.L., and Menon, V. (2003). Functional brain. Annu. Rev. Neurosci. 13, 25–42.
connectivity in the resting brain: a network analysis of the default mode Posner, M.I., Petersen, S.E., Fox, P.T., and Raichle, M.E. (1988). Localization
hypothesis. Proc. Natl. Acad. Sci. USA 100, 253–258. of cognitive operations in the human brain. Science 240, 1627–1631.

Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc. 677
 Neuron
Functional Brain Networks

Power, J.D., Fair, D.A., Schlaggar, B.L., and Petersen, S.E. (2010). The devel- Spoormaker, V.I., Schröter, M.S., Gleiser, P.M., Andrade, K.C., Dresler, M.,
opment of human functional brain networks. Neuron 67, 735–748. Wehrle, R., Sämann, P.G., and Czisch, M. (2010). Development of a large-
Power, J.D., Barnes, K.A., Snyder, A.Z., Schlaggar, B.L., and Petersen, S.E. scale functional brain network during human non-rapid eye movement sleep.
(2011). Spurious but systematic correlations in functional connectivity MRI J. Neurosci. 30, 11379–11387.
networks arise from subject motion. Neuroimage, in press. Published online Tian, L., Wang, J., Yan, C., and He, Y. (2011). Hemisphere- and gender-related
October 14, 2011. 10.1016/j.neuroimage.2011.10.018. differences in small-world brain networks: a resting-state functional MRI study.
Raichle, M.E., MacLeod, A.M., Snyder, A.Z., Powers, W.J., Gusnard, D.A., and Neuroimage 54, 191–202.
Shulman, G.L. (2001). A default mode of brain function. Proc. Natl. Acad. Sci. Tomasi, D., and Volkow, N.D. (2011). Functional connectivity hubs in the
USA 98, 676–682. human brain. Neuroimage 57, 908–917.
Rosvall, M., and Bergstrom, C.T. (2008). Maps of random walks on complex Traag, V.A., and Bruggeman, J. (2009). Community detection in networks with
networks reveal community structure. Proc. Natl. Acad. Sci. USA 105, 1118– positive and negative links. Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 80,
1123. 036115.
Rubinov, M., and Sporns, O. (2010). Complex network measures of brain Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O.,
connectivity: uses and interpretations. Neuroimage 52, 1059–1069. Delcroix, N., Mazoyer, B., and Joliot, M. (2002). Automated anatomical
Rubinov, M., and Sporns, O. (2011). Weight-conserving characterization of labeling of activations in SPM using a macroscopic anatomical parcellation
complex functional brain networks. Neuroimage 56, 2068–2079. of the MNI MRI single-subject brain. Neuroimage 15, 273–289.
Seeley, W.W., Menon, V., Schatzberg, A.F., Keller, J., Glover, G.H., Kenna, H., van den Heuvel, M.P., Stam, C.J., Boersma, M., and Hulshoff Pol, H.E. (2008).
Reiss, A.L., and Greicius, M.D. (2007). Dissociable intrinsic connectivity Small-world and scale-free organization of voxel-based resting-state func-
networks for salience processing and executive control. J. Neurosci. 27, tional connectivity in the human brain. Neuroimage 43, 528–539.
2349–2356. Van Essen, D.C. (2005). A Population-Average, Landmark- and Surface-based
Seeley, W.W., Crawford, R.K., Zhou, J., Miller, B.L., and Greicius, M.D. (2009). (PALS) atlas of human cerebral cortex. Neuroimage 28, 635–662.
Neurodegenerative diseases target large-scale human brain networks. Neuron Van Essen, D.C., Drury, H.A., Dickson, J., Harwell, J., Hanlon, D., and
62, 42–52. Anderson, C.H. (2001). An integrated software suite for surface-based
Shulman, G.L., Fiez, J.A., Corbetta, M., Buckner, R.L., Miezin, F.M., Raichle, analyses of cerebral cortex. J. Am. Med. Inform. Assoc. 8, 443–459.
M.E., and Petersen, S.E. (1997). Common blood flow changes across visual Wig, G.S., Schlaggar, B.L., and Petersen, S.E. (2011). Concepts and principles
tasks: II. Decreases in cerebral cortex. J. Cogn. Neurosci. 9, 648–663. in the analysis of brain networks. Ann. N Y Acad. Sci. 1224, 126–146.
Shulman, G.L., Pope, D.L.W., Astafiev, S.V., McAvoy, M.P., Snyder, A.Z., and Yeo, B.T.T., Krienen, F.M., Sepulcre, J., Sabuncu, M.R., Lashkari, D.,
Corbetta, M. (2010). Right hemisphere dominance during spatial selective Hollinshead, M., Roffman, J.L., Smoller, J.W., Zollei, L., Polimeni, J.R., et al.
attention and target detection occurs outside the dorsal frontoparietal (2011). The organization of the human cerebral cortex estimated by functional
network. J. Neurosci. 30, 3640–3651. connectivity. J. Neurophysiol. 106, 1125–1165.
Smith, S.M., Miller, K.L., Salimi-Khorshidi, G., Webster, M., Beckmann, C.F., Zalesky, A., Fornito, A., Harding, I.H., Cocchi, L., Yücel, M., Pantelis, C., and
Nichols, T.E., Ramsey, J.D., and Woolrich, M.W. (2011). Network modelling Bullmore, E.T. (2010). Whole-brain anatomical networks: does the choice of
methods for FMRI. Neuroimage 54, 875–891. nodes matter? Neuroimage 50, 970–983.

678 Neuron 72, 665–678, November 17, 2011 ª2011 Elsevier Inc.