Article

Parallel Interdigitated Distributed Networks within

the Individual Estimated by Intrinsic Functional
Connectivity
Highlights Authors
d Within-individual characterization of brain networks reveals Rodrigo M. Braga, Randy L. Buckner
new spatial details
Correspondence
d Group-defined networks fractionate into distinct parallel
[email protected] (R.M.B.),
networks in individuals [email protected] (R.L.B.)

d Parallel networks possess closely juxtaposed regions in

In Brief
numerous cortical zones
Braga and Buckner examine the detailed
d Networks share a conserved motif that may be organized organization of brain networks within
along a macroscale gradient individual people. They discovered that
multiple closely juxtaposed cortical
regions form parallel distributed
networks. Separate large-scale networks
may emerge from a common organizing
principle.

Braga & Buckner, 2017, Neuron 95, 457–471

July 19, 2017 ª 2017 The Authors. Published by Elsevier Inc.
http://dx.doi.org/10.1016/j.neuron.2017.06.038
Neuron

Article

Parallel Interdigitated Distributed Networks

within the Individual Estimated
by Intrinsic Functional Connectivity
Rodrigo M. Braga1,2,3,5,* and Randy L. Buckner1,3,4,*
1Department of Psychology, Center for Brain Science, Harvard University, Cambridge, MA 02138, USA
2The Computational, Cognitive & Clinical Neuroimaging Laboratory, Hammersmith Hospital Campus, Imperial College London,
London W12 0NN, UK
3Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA 02129, USA
4Department of Psychiatry, Massachusetts General Hospital, Charlestown, MA 02129, USA
5Lead Contact

*Correspondence: [email protected] (R.M.B.), [email protected] (R.L.B.)

http://dx.doi.org/10.1016/j.neuron.2017.06.038

SUMMARY et al., 1977) and anterograde (Cowan et al., 1972) tracers. For
example, by charting the laminar pattern of anatomical projec-
Certain organizational features of brain networks tions, a hierarchical pathway emerged from areas involved in
present in the individual are lost when central ten- visual perception through to areas enabling motor actions
dencies are examined in the group. Here we investi- (Maunsell and van Essen, 1983; Ungerleider and Desimone,
gated the detailed network organization of four 1986; Andersen et al., 1990; Boussaoud et al., 1990; see Shadlen
individuals each scanned 24 times using MRI. We and Newsome, 2001 for discussion). This canonical distributed
network comprises primary and secondary visual areas, parietal
discovered that the distributed network known as
association areas, and motor areas. These interconnected areas
the default network is comprised of two separate
form a partially modular distributed network that interacts with,
networks possessing adjacent regions in eight or but is anatomically distinguishable from, other processing hierar-
more cortical zones. A distinction between the net- chies (Felleman and Van Essen, 1991; Ungerleider and Desi-
works is that one is coupled to the hippocampal for- mone, 1986; Van Essen et al., 1992; see also Markov et al.,
mation while the other is not. Further exploration 2014). Notably, this canonical network involves areas distributed
revealed that these two networks were juxtaposed across temporal, parietal, and frontal cortices. As will be illus-
with additional networks that themselves fractionate trated below, this distributed pattern is a general motif that is
group-defined networks. The collective networks apparent across multiple large-scale networks (see Goldman-
display a repeating spatial progression in multiple Rakic, 1988).
cortical zones, suggesting that they are embedded Human neuroimaging studies are particularly useful for char-
acterizing distributed networks because they survey the whole
within a broad macroscale gradient. Regions con-
brain at once. Corbetta and Shulman (2002) highlighted a
tributing to the newly defined networks are spatially
network currently known as the ‘‘dorsal attention network’’
variable across individuals and adjacent to distinct (dATN) in the human neuroimaging literature, which is likely ho-
networks, raising issues for network estimation in mologous to later-stage components of the sensory-motor
group-averaged data and applied endeavors, hierarchy described above in the macaque (Vincent et al.,
including targeted neuromodulation. 2007; Patel et al., 2015). Network analysis based on intrinsic
functional connectivity (FC) consistently reveals the dATN (e.g.,
Fox et al., 2006; Vincent et al., 2008; see also Beckmann et al.,
INTRODUCTION 2005). Detailed analysis in relation to retinotopic areas recapitu-
lates, to first approximation, the full hierarchical pathway
The cerebral cortex possesses a complex tapestry of networks described in the macaque (Yeo et al., 2011). Thus, while there
that interact and compete in the service of information process- are caveats to interpreting networks observed by FC (Buckner
ing. Building on a history of assigning specialized functions to et al., 2013; Murphy et al., 2013; Smith et al., 2013; Power
brain regions, early seminal work by Norman Geschwind, Marsel et al., 2014), the results can generate hypotheses about the orga-
Mesulam, and others proposed ideas about how distributed re- nization of multiple large-scale networks that populate human
gions might interact to perform high-level tasks (e.g., Gesch- association cortex.
wind, 1965a, 1965b; Mesulam, 1981, 1990). A leap in progress Group-based studies using FC suggest that association cortex
occurred when networks began to be conceptualized within comprises about five major distributed networks (e.g., Yeo et al.,
the framework of anatomical connectivity patterns in the ma- 2011; Power et al., 2011; Doucet et al., 2011). These networks
caque, following the availability of both retrograde (Mesulam have sufficiently modular properties to be consistently identified

Neuron 95, 457–471, July 19, 2017 ª 2017 The Authors. Published by Elsevier Inc. 457
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
as isolated networks, even when constraints are relaxed to Power et al. (2011) noted at least two network subcomponents
emphasize interactions between networks (Yeo et al., 2014). (‘‘subgraphs’’ within their framework) that were not captured in
Among these, the ‘‘default network’’ (DN) has been extensively the major network descriptions. In detailed analysis of the DN,
studied. The DN is anatomically separate from the dATN and is there have been several proposed schemes to delineate subnet-
estimated to have expanded in hominin evolution (Buckner and works (e.g., Andrews-Hanna et al., 2010, 2014; Leech et al., 2011,
Krienen, 2013; Margulies et al., 2016; see also Hill et al., 2010). 2012; Braga and Leech, 2015). We do not attempt to integrate
Regions across the DN are correlated to a level akin to the local these various findings here but point out that analyses of group
sensory and motor networks despite being widely distributed data have always possessed features that are not fully accommo-
(Power et al., 2011; see also Greicius et al., 2003). Tracer studies dated by assuming a small number of major networks.
suggest that the DN discovered in the human may have an A second open question is whether there are global properties
anatomically connected homolog in the macaque (Buckner of organization that span across the networks. Margulies et al.
et al., 2008; Binder et al., 2009; see also Margulies et al., 2009; (2016) recently proposed a macroscale organization of networks
Buckner and Krienen, 2013). Several other networks, near the that moves outward from sensory-motor networks on one end
dATN and DN, each have their own distributed organization to the DN on the other end. This echoes an idea advanced by
(Yeo et al., 2011; Power et al., 2011; Doucet et al., 2011). Mesulam (1998) that the cortex exhibits a hierarchical organiza-
Several principles emerge through examining relations among tion progressing from unimodal areas to integrative transmodal
multiple large-scale networks. First, the distributed networks all areas (see also Jones and Powell, 1970). The repeating motif
have an organization that is similar to the dATN and DN, with and spatial juxtaposition of multiple networks suggest that the
each network possessing frontal, temporal, posterior parietal, macroscale organization of the distributed networks might be
frontal midline, and posterior midline components. Depending partially explained by developmental anchors or gradients
on how the network is estimated, certain components can be (Buckner and Krienen, 2013; Margulies et al., 2016; Krienen
underemphasized, e.g., the dATN is sometimes described and Buckner, 2017). Detailed analysis of cortical network archi-
without a frontal midline component, especially when ‘‘winner- tecture may provide evidence for or against macroscale gradi-
take-all’’ network assignments are used (e.g., Yeo et al., 2011). ents that span networks.
Targeted analyses often reveal there is a midline component To make progress on these questions, we need to move from
(e.g., Figure 5 in Fox et al., 2006; Figure 32 in Yeo et al., 2011). group-level description to the finer spatial scale that is acces-
The distributed networks follow a general motif that is roughly sible when studying organization within the individual. The
conserved even though each network contains spatially distinct majority of the literature on network architecture is based on
regions. Power et al. (2011) further noted that the networks have averaging data from groups of spatially normalized individuals.
similar spatial arrangements at their interfaces in multiple zones Group averaging was necessary in positron emission tomogra-
of cortex (e.g., temporal, parietal, frontal). That is, if a network lies phy (PET) studies as radiation places severe restrictions on
side by side with another network in parietal cortex, it is also repeat imaging. A major technical breakthrough came with the
likely to do so in frontal cortex. This echoes features proposed ability to average brain volumes across individuals to boost
by Goldman-Rakic (1988) (her Figure 4) and Mesulam (1981) signal to noise (e.g., Fox et al., 1985; Evans et al., 1992; Friston
(his Figure 4) based on anatomical data. Second, while the broad et al., 1995). However, group-averaging approaches are limited
architecture follows a general motif, there are differences that due to blurring over anatomical and functional variability (Stein-
distinguish the networks. For example, while the dATN is metz and Seitz, 1991; Silbersweig et al., 1993).
coupled to visual and motor regions, the DN is instead coupled Anatomical variability refers to the complex geometry of the
to the hippocampal formation, perhaps reflecting a mnemonic brain’s gross structural morphology (e.g., sulci and gyri) that
functional anchor (Greicius et al., 2004; Vincent et al., 2006; must be normalized across subjects. Surface-based normaliza-
Buckner et al., 2008; Andrews-Hanna et al., 2010). tion has improved this technical hurdle for the cerebral cortex
The discovery that cerebral association cortex possesses mul- (Fischl et al., 1999; Van Essen, 2005; Robinson et al., 2014),
tiple distributed networks is a major milestone for the field. How- but misregistration is still a challenge. Functional variability refers
ever, reliance on group-averaged estimates raises questions. A to the spatial arrangement of functional zones on the cortical sur-
first open question concerns whether the present convergence face. Functional organization is likely derived from microscopic
on major networks reflects the correct level of description. Simi- anatomical features, such as areal organization or anatomical
larities between human neuroimaging and monkey anatomical subdomains that possess distinct connections. The point here
findings suggest that the major networks described to date cap- is that these local organizational properties are not fully ac-
ture true organizational features. Nonetheless, several reports counted for by gross anatomical differences between individ-
note distinctions that fractionate the larger major networks, either uals. Histological data have illustrated that the border locations,
locally within a region or in complex ways across distributed re- size, and shape of architectonic areas vary on the surface be-
gions. For example, Smith et al. (2013) presented a clustering tween individuals (e.g., Rademacher et al., 1993; Rajkowska
analysis that illustrated both high-level structure that recapitu- and Goldman-Rakic, 1995; Amunts et al., 1999, 2000; Caspers
lated the major networks and a substructure that included spatial et al., 2006; Fischl et al., 2008; Henssen et al., 2016). Thus,
distinctions between nearby regions of cortex (see also Smith even if individuals could be brought into complete anatomical
et al., 2012). In the study by Yeo et al. (2011), they showed that alignment, spatial blurring across architectonically defined areas
the major networks broke down further when more fine-grained would persist. An additional consideration is that cortical organi-
network structure was examined (their 17-network parcellation). zation may possess complex topography that has side-by-side

458 Neuron 95, 457–471, July 19, 2017

juxtapositions and finely interleaved structure. Fine spatial each subject’s 24 scans, slice-based tSNR ranged between
arrangements may reflect the complex geometry of small 127.2 and 336.3 for S1, 264.0 and 461.2 for S2, 123.2 and
cortical domains (e.g., Gordon et al., 2017a; Glasser et al., 286.6 for S3, and 204.7 and 493.1 for S4. Absolute motion
2016), such as has been hypothesized for face patches (Moeller (i.e., the accumulated displacement in each run) ranged between
et al., 2008), or the fine-grained structure due to topographically 0.303 and 1.505 mm for S1, 0.209 and 1.181 mm for S2, 0.474
arranged projections such as is present for eccentricity in the and 1.651 mm for S3, and 0.267 and 1.765 mm for S4. Spatial
early visual system (e.g., Maunsell and van Essen, 1983). properties of tSNR and fALFF are affected by susceptibility arti-
Normalization across subjects may blur these spatial features. facts (e.g., Ojemann et al., 1997). Figure S1 displays voxel-based
Recent human neuroimaging studies focusing on individuals tSNR and fALFF maps projected onto the cortical surface. High
have noted fine spatial details that are lost or attenuated by voxel-based tSNR and fALFF were achieved across nearly the
group averaging. Fedorenko et al. (2012) demonstrated that a entire cortical mantle, including ventral PFC and portions of the
language-preferential region in prefrontal cortex (PFC) appears anterior and ventral temporal lobe.
‘‘as an island’’ between regions showing domain-general
response properties. The exact positioning of the island moves Distinct Distributed Networks Fractionate the Canonical
from person to person. Michalka et al. (2015) characterized inter- Default Network within the Individual
digitated regions of PFC belonging to separate auditory and The goal of the present study was to identify spatially detailed
visual networks whose exact positions vary considerably across features of network organization in the individual. For each
subjects (Figure S1 in Michalka et al., 2015). Laumann et al. participant, half the data were used for discovery (n = 12) of
(2015) recently highlighted a network implication of closely juxta- network features that were later tested in the remaining indepen-
posed small regions. Two distinct networks were found in an in- dently collected sessions (n = 12). The discovery datasets were
dividual using nearby seed regions in left PFC; however, when analyzed blind to the hypothesis-testing datasets. An interactive
the same two regions were applied to group-averaged data, a seed selection and FC map viewing platform was established us-
single distributed network emerged (Figure 7 in Laumann et al., ing a bespoke high-resolution cortical mesh so as to minimize
2015). In an analysis motivated by architectonic subdivisions of spatial blurring during interpolation and allow networks to be
nearby prefrontal areas BA 44 and BA 45, Jakobsen et al. defined with high precision by selecting individual vertices.
(2016) (their Figure 11) observed a similar separation of networks In the discovery data, the observation was made in one sub-
in the individual. Glasser et al. (2016) noted that whereas in the ject that two similar, but distinct, networks could be observed
majority of cases a single contiguous region of lateral PFC was from two seed vertices selected from adjacent regions of left
found to belong to a left-lateralized language network (Figure S8 lateral PFC (Figure 1). Both networks were distributed across
in Glasser et al., 2016), a minority of individuals displayed two inferior parietal, lateral temporal, medial PFC, and posteromedial
‘‘split’’ non-contiguous regions. In a comprehensive analysis of cortices and resembled the canonical DN but with distinct
individual-specific network features, Gordon et al. (2017b) iden- nodes. The two networks had components that were closely
tified numerous reliable features that were not captured in group neighboring but separate in numerous cortical zones, including
estimates of network organization. These features form a distrib- temporal and ventral medial PFC, suggesting that the DN may
uted set of patches across the cortical mantle that were ‘‘too be comprised of distinct neighboring networks. We refer to these
infrequently present and/or spatially variable relative to their hypothesized networks as Network A and Network B (Figure 1).
size to emerge in group-average data.’’ Thus, analysis of individ- Investigating additional participants yielded a similar distinction.
ual brains reveals regional and network features that are under- An important difference between the networks was that
emphasized in group-averaged analyses. Network A showed correlation with a region in parahippocampal
Motivated by these findings, we conducted an extensive set of cortex (PHC), whereas no such evidence was found in Network B,
analyses focused on the individual. We discovered that the DN even when additional and more sensitive analyses were carried
could be reliably subdivided into parallel networks within the in- out focused on this region (Figure S2). The presence of correla-
dividual. Similar separations were made for other major net- tion with medial temporal lobe structures is anticipated (e.g.,
works. Regions of the separate networks lay side by side across Greicius et al., 2004; Vincent et al., 2006). However, that the
distributed zones of cortex and were sufficiently variable be- coupling was to one hypothesized network and not the other
tween individuals to obscure their existence in group-averaged was unexpected.
analyses. To make these observations, we analyzed data from The recurrence of the distributed pattern of regions across the
four individuals scanned repeatedly over 24 sessions. four participants provided strong evidence for two dissociable
networks (Figure 1).
RESULTS
Double Dissociation of the Two Networks within the
High Signal-to-Noise and Full-Brain Coverage Was Individual
Achieved in Each of Four Individuals The independent replication data were used to formally test the
The present study acquired extensive data over many functional hypothesis that Network A was dissociable from Network B.
MRI (fMRI) sessions in the same individuals. Two estimates of Using only the discovery datasets, a priori regions (single
data quality, slice-based temporal signal-to-noise ratio (tSNR) vertices) were selected that maximized the separation of the net-
and fractional amplitude of low-frequency fluctuations (fALFF; works from the two lateral PFC seed regions in the main regional
Zou et al., 2008), were calculated for each participant. Across zones of the cortex (temporal, inferior parietal, posteromedial,

Neuron 95, 457–471, July 19, 2017 459

Figure 1. Two Parallel Interdigitated Distributed Networks at or near the Canonical Default Network Are Revealed by Functional Connectivity
within Individuals
Each row illustrates functional connectivity (FC) maps from a single subject (S1–S4). Two networks were observed in each individual. Subject-specific seed
regions were placed in the left lateral PFC of the discovery dataset (white filled-in circle). The seed region labeled A* yielded Hypothesized Network A (left) and the
seed region labeled B* yielded Hypothesized Network B (right). Note that throughout the cortex, Networks A and B are adjacent to one another with slightly varied
positions from individual to individual. Hypothesized Network A includes posterior inferior parietal lobule, lateral temporal cortex, ventromedial PFC, retrosplenial/
ventral posteromedial cortex, and parahippocampal cortex. Hypothesized Network B includes the temporoparietal junction, lateral temporal cortex, an inferior
region of ventromedial PFC, a dorsal region of anteromedial PFC, and posterior cingulate cortex. Regions (hollow circles A and B) were selected to formally test
the distinction between the two networks in independent data. The surfaces are rotated by 19 along the y plane to better show the ventromedial PFC and
intraparietal sulcus. The same views are used in accompanying figures.

and medial PFC) and a region in PHC. These a priori regions were Networks A and B. The presence of interactions across all
targeted within each subject to locations where contiguous distributed zones of the cortex would provide strong evidence
vertices could be observed in Network A and Network B (Fig- that there was a network dissociation across the cortex.
ure 1). Using the independent replication dataset (n = 12 in Critically, the multiple tests across cortical zones and subjects
each subject), time courses were extracted from each subject’s targeted convergent evidence for network dissociation. Twenty
a priori regions. The r-to-z transformed Pearson’s product 2 3 2 ANOVAs were carried out across four subjects and five
moment correlation was computed between the two lateral cortical zones (temporal, inferior parietal, posteromedial, medial
PFC seed regions and each distributed test region (Figure 2). PFC, and PHC). The PHC region from Network A was paired
Two-way ANOVA was used to test the dissociation between with the posteromedial region from Network B. All 20 ANOVAs
the two PFC seed regions and the regions in each cortical were individually significant (p < 0.01) with most showing
zone. The critical test was whether there would be significant crossover interactions (Figure 2). One exception was the medial
interactions between seed and target regions belonging to PFC interaction that, while significant in all four subjects,

460 Neuron 95, 457–471, July 19, 2017

 Figure 2. Parallel Distributed Networks Are
Statistically Dissociated Using Indepen-
dent Data
Functional correlation strength was computed
between the two PFC seed regions yielding
Network A and Network B and the pairs of adja-
cent seed regions in lateral temporal (Temporal),
inferior parietal (Parietal), Posteromedial, and
Medial PFC cortices (regions shown in Figure 1).
This yielded a 2 3 2 contrast for each zone of
cortex (e.g., Networks A and B’s PFC regions
against Networks A and B’s Temporal regions). An
additional seed region in PHC was grouped with
Network B’s posteromedial region. Correlations
with Network A’s PFC region are shown in yellow
and Network B’s in red. Bars represent the mean
from the 12 sessions of the hypothesis-testing
dataset with SEM. All 20 2 3 2 ANOVA tests
were significant (**p < 0.01), with most showing a
cross-over interaction.

fractionate the canonical DN across

numerous cortical zones (Figure 3).

The Importance of Examining

Network Organization within the
Individual
The organization of the dissociated net-
works suggests why they might evade
group-averaged analyses. Spatial ‘‘hur-
dles’’ to group averaging have been pre-
viously reported (e.g., Fedorenko et al.,
2012’s Figure 1; Laumann et al., 2015’s
Figure 7). To quantify the effect of spatial
misalignment between individuals, we
took an approach using spatial yoking
between the individuals. For each individ-
ual, her spatially optimized regions from
the discovery dataset were applied to
her independent replication data, yielding
demonstrated a clear crossover in two subjects and minimal dif- an unbiased correlation matrix (Figure S5). Two clear clusters
ference for the most ventral region in two subjects. Another of strong within-network and minimal between-network
exception was the PHC interaction, which did not show a clear correlations were observed in each individual, with S1 and S4
crossover interaction in S1. Note that, in most cases, the cross- showing the strongest patterns. This illustrates that widely
over interactions are present even where regions are extremely distributed regions can show strong correlation with one
close to one another. another, while spatially adjacent regions can be embedded in
While the above analyses formally test the double dissocia- distinct correlated clusters. The same correlation matrices
tion, another source of evidence is that the spatial patterns showed minimal structure when regions from one person were
replicate within individuals across discovery and replication applied to another (e.g., Subject 1’s regions were used to
(hypothesis-testing) datasets (Figure S3). Additional analyses generate a matrix using Subject 2’s fMRI data), highlighting the
examined alternative methods for identifying these separate importance of respecting the exact spatial details present within
networks, including using data-driven clustering and estimation an individual.
of connectivity patterns in the volume in addition to the surface
(Figure S4). Visualization in the native volume is important Topography of Multiple Distinct Networks within the
because projection to the cortical surface can induce fraction- Individual
ations of single regions into multiple regions if they fall near The above analyses establish evidence for two distinct
sulcal boundaries. The combined results illustrate a robust neighboring networks that are likely components of what has
double dissociation between two distributed networks that been studied as the DN. This unexpected observation prompted

Neuron 95, 457–471, July 19, 2017 461

 Figure 3. Parallel Distributed Networks
Contain Juxtaposed Regions in Numerous
Cortical Zones
The two dissociated networks near the canonical
default network, Network A and Network B, are
shown for two subjects (S1 and S4) in a schematic
form on the same cortical surface representation.
The dashed boxes highlight nine cortical zones
where neighboring representations of the two
networks were found including: (1) dorsolateral
PFC, (2) inferior PFC, (3) lateral temporal cortex,
(4) inferior parietal lobule extending into the tem-
poroparietal junction, (5) posteromedial cortex,
(6) midcingulate cortex, (7) dorsomedial PFC,
(8) ventromedial PFC, and (9) anteromedial PFC.
Some zones, including the dorsal region along
the PFC (labeled 7), are subtle, but consistent, in
all subjects, suggesting that there exists small,
closely juxtaposed components of the two disso-
ciated networks.

For these analyses, all 24 data sessions

were combined to provide best estimate
maps for each participant. Seed regions
were placed in the left lateral PFC and
the frontal eye fields to identify the FPN
and dATN, respectively. For the FPN,
two seed regions were selected that
revealed networks that maximized the
following features: (1) the networks con-
tained strong representations in intra-
parietal, inferior temporal, and dorsome-
dial PFC, and (2) the networks occupied
neighboring, but distinct, regions in each
cortical location. For the dATN, similar
criteria were applied, but the distri-
bution instead included superior parie-
tal, occipitoparietal, and occipitotempo-
ral components.
As with the DN, both the FPN and
dATN were fractionated into two distinct,
parallel networks within the individual,
identifying a total of six networks with
close spatial arrangements (Figure 4).
The effects were clearest in S1 and S4
and suggestive in the other subjects. In
each case, the networks both resembled
the canonical group-average network but
inhabited separate subregions (Figure 5).
Important organizational details were
evident when all six networks were visu-
us to explore how these two networks relate to additional alized simultaneously. Figures 6 and 7 focus on the anterior
large-scale networks, the frontoparietal control network (FPN) midline and parietal and temporal lobes to illustrate two organi-
and the dATN. Two questions drove these analyses. First, are zational features. First, regions from each network are often
the other large-scale networks themselves fractionated and, if located in distinct locations in each cortical zone. The white lines
so, how are the newly detected networks organized? Second, in Figures 6 and 7 serve as landmarks to highlight non-overlap-
taken as a group, do the multiple large-scale networks possess ping features of the networks. Second, the networks display a
consistent spatial relations between networks? fine-scale interdigitation. For example, in the anterior midline,

462 Neuron 95, 457–471, July 19, 2017

 a broad ventral to dorsal progression is observed (Figure 6);
however, particularly for the DN, the representations from
Network B were positioned in between representations from
Network A. Similarly in the temporal lobe, the representation
from Network A is largely surrounded by representations from
Network B (Figure 7). In the parietal lobe, a broad posteroventral
to anterodorsal progression is observed across the networks,
with each of the six networks inhabiting a distinct region (Fig-
ure 6). The fractionated Networks A and B of the canonical
dATN showed three or more separate regions, as visualized
on the surface, which sequentially alternated along the intrapar-
ietal arc of the canonical dATN (Figures 4, 5, and 6).
Figures 8 and S6 show a diagrammatic representation of the
six different networks in two subjects to highlight the close,
parallel nature of their organization within frontal, parietal, and
temporal lobes. For an additional analysis, which should be
considered descriptive, a correlation matrix was constructed
(Figure S7) using regions chosen from all six networks (Fig-
ure S8). Data from all 24 sessions were used to construct each
matrix. Given that the regions were defined and tested on the
same data, the specific quantitative values of the within-network
correlations should not be interpreted; however, the between-
network correlations reflect an unbiased estimate of interactions
between networks. Of interest, certain networks showed hints of
interactions with other networks. For example, the FPN-A
showed slightly elevated correlation with DN-A; the FPN-B
showed slightly elevated correlation with dATN-A. These sug-
gestive interactions may be due to spatial blurring or to biologi-
cally meaningful factors and are presented purely for their ability
to generate future hypotheses.

DISCUSSION

The present study examined the organization of large-scale

distributed networks within the individual. We discovered that
the canonical DN fractionates into two parallel networks that
have juxtaposed regions throughout the cerebral cortex. Moti-
vated by this discovery, we examined the dATN and FPN and
found that each of these canonical networks also fractionates
into two parallel networks. The organization of the six identified
networks was charted and found to have a spatial progression
in multiple zones of cortex. These results are consistent with
the ideas that the large-scale networks (1) share a conserved
motif and (2) are embedded within a broad macroscale organiza-
tion. As a technical point, the present results underscore a need
to move from group-based analyses to examination of detailed
anatomy within the individual.

Canonical Networks Fractionate into Distinct Networks

within the Individual
The most pressing finding reported here is that three major net-
works (DN, FPN, and dATN) are each subdivided into parallel
spatially juxtaposed networks (Figures 1, 3, 4, and 5). Regions
Figure 4. Multiple Parallel Interdigitated Distributed Networks at or
near the Canonical Frontoparietal Control and Dorsal Attention Net-
works Estimated by Functional Connectivity within Individuals subjects (S1 and S4) are displayed. The canonical Frontoparietal Network
Best estimate maps (using all 24 sessions in each individual) of Networks A (middle) and Dorsal Attention Network (bottom) also each fractionate into two
and B that fractionate the default network are illustrated (top). Maps from two juxtaposed networks. Seed regions are illustrated by filled white circles.

Neuron 95, 457–471, July 19, 2017 463

Figure 5. Relationship of Parallel Interdigitated Networks to Canonical Networks from Group-Averaged Data
Each row illustrates how the networks identified in two individuals (S1 and S4) correspond to the well-characterized topography of group-derived networks. The
black border represents the outline of the canonical default, frontoparietal control, and dorsal attention networks (top to bottom) calculated using data from 1,000
subjects that were parcellated into seven networks (from Yeo et al., 2011). The correlation maps from each seed (white filled circle) are shown in color. Broadly, the
networks can be seen to occupy separate, closely juxtaposed regions that fall within the canonical network borders in most cases. Exceptions can also be found,
such as in the inferior frontal cortex in Default Network A and in the parietal lobe in Frontoparietal Control Network B, where the individual’s connectivity map
strays outside the group network borders.

of the separate networks lay side by side to one another by another’s (Figures 6 and 7). It is thus unsurprising that
across several cortical zones (Figure 3) and exhibited complex detailed within-subject analysis is needed to visualize the sepa-
topography, with one network’s region sometimes surrounded rate networks. We focused on the DN first to show that the

464 Neuron 95, 457–471, July 19, 2017

Figure 6. Detailed Anatomy of Six Distinct Networks: Parietal and Medial Prefrontal Cortices
The fine-scale interdigitation of the six identified networks is highlighted in regions where the macroscale organization is evident. White lines serve as landmarks
so that the relative position of each network can be appreciated across panels: Networks A and B of the Default Network (DN-A, DN-B), Frontoparietal Control
Network (FPN-A, FPN-B), and Dorsal Attention Network (dATN-A, dATN-B). In each row, FC maps from an individual are displayed for either medial frontal cortex
(top two rows) or lateral parietal cortex (bottom two rows).

within-subject network fractionation is reproducible across (Fig- 2011; Laird et al., 2011; Leech and Sharp, 2014; Andrews-Hanna
ure 1) and within individuals (Figure S3) and is statistically robust et al., 2014).
in an independent sample (Figure 2). We next characterized frac-
tionations of the FPN and the dATN (Figures 4, 5, 6, and 7). Relations to Prior Observations
These findings raise the prospect that the canonical networks There is growing consensus that idiosyncratic features exist
studied in group-averaged data consist of distinct functional net- within individuals that are not captured (or are attenuated) by
works that are blurred together by spatial averaging. The six net- examining group central tendencies (e.g., Fedorenko et al.,
works identified here appear to be fractionations of networks 2010, 2012; Mueller et al., 2013; Laumann et al., 2015; Glasser
identified in group studies (e.g., Yeo et al., 2011; Power et al., et al., 2016; Huth et al., 2016; Jakobsen et al., 2016; Gordon
2011; Doucet et al., 2011). This point is important to emphasize: et al., 2017a, 2017b). The DN was originally hypothesized based
it is not the case that more detailed analysis carved network on the distributed pattern of regions that increase activity in pas-
organization orthogonally to prior schemes, but rather fraction- sive relative to active non-self-referential tasks (Andreasen et al.,
ated the existing lower-resolution frameworks (Figure 5). It re- 1995; Shulman et al., 1997; Mazoyer et al., 2001; for review, see
mains to be determined whether these different networks Gusnard and Raichle, 2001). The DN was later estimated using
mediate separable functions across different task contexts. FC by placing a moderately sized seed region in the center of
Such a finding could help explain the heterogeneity in cognitive posteromedial cortex (often labeled posterior cingulate cortex
functions and clinical conditions ascribed to the canonical net- or ‘‘PCC’’) and plotting the correlation pattern (Greicius et al.,
works (e.g., Buckner et al., 2008; Spreng et al., 2009; Menon, 2003). One immediate feature of the resulting network that

Neuron 95, 457–471, July 19, 2017 465

 nodes in different zones of cortex. There is evidence for a dorsal
to ventral separation in the posterior midline (Margulies et al.,
2009; Leech et al., 2011; see also Vogt et al., 2006) that likely cor-
responds to the present DN-A and DN-B networks, although the
topography is not fully captured by a simple linear axis in all in-
dividuals (e.g., see S2 in Figure 1). Prior reports have also noted
that a subnetwork of the DN is coupled to the hippocampal for-
mation (e.g., Andrews-Hanna et al., 2010), consistent with
anatomical connectivity in the macaque (see Kahn et al., 2008
for review). The present hypothesized DN-A aligns well to the
‘‘hippocampal’’ subnetwork of Andrews-Hanna et al. (2010),
while DN-B does not. This separation also informs our under-
standing of the IPL.
While many analyses of the DN, especially those based on
increased response during passive tasks, reveal a large region
covering much of IPL, social tasks involving mentalizing
(theory-of-mind tasks) activate the more anterior temporo-
parietal junction (TPJ; Saxe and Kanwisher, 2003) while
tasks involving episodic remembering activate a caudal region
(e.g., Andrews-Hanna et al., 2014) near what might be a monkey
homolog of Opt (Pandya and Seltzer, 1982; for discussion, see
Yeo et al., 2011). Group-level FC analyses have noted that the
posterior portion of IPL is preferentially coupled to the PHC,
while the anterior portion is not (see Yeo et al., 2011’s Figure 30).
Furthermore, monkey anatomical tracing studies consistently
show that PHC projects to a circumscribed portion of area 7A
within Opt (e.g., Lavenex et al., 2002, Case M-2-90; Blatt et al.,
2003, Cases 1 and 5). These collective findings are consistent
with the distinction between DN-A, which is coupled to the
PHC and posterior IPL, and DN-B, which involves a more ante-
rior IPL region. What is novel in the present work is that this
distinction is now shown to be one spatial component of a
much broader separation of two parallel large-scale distributed
networks.
We know of no precedent for one feature of our results. The
ventral portion of the frontal midline is considered a projection
zone of limbic structures, including the amygdala (Ongu €r and
Price, 2000) and hippocampal formation (Rosene and Van Hoe-
sen, 1977). However, in the present study, the most ventral rep-
resentation of DN-B is inferior to DN-A. This is unexpected
because DN-B is differentiated from DN-A by its absence of
coupling to the PHC and retrosplenial cortex (Figure S2). This
Figure 7. Detailed Anatomy of Six Distinct Networks: Lateral Tem- may indicate that a subregion of ventromedial PFC is tied to a
poral Cortex
large-scale distributed network that is functionally separated
In a similar format to Figure 6, each column displays FC maps from an indi-
vidual to illustrate the fine-scale interdigitation in the lateral temporal cortex. from a direct limbic influence. The fine interdigitation shown in
Figure 6 reveals why group averaging will likely blur the two net-
works. In each individual, there are multiple regions for each
raises the possibility of further subdivision is its large size. For network that are interposed and whose positions spatially shift
example, the canonical group-averaged FC estimate of the DN between individuals. Anatomical tracing in the monkey will be
contains an extensive correlation pattern extending from the needed to substantiate whether there exists a ventral midline
dorsal extent of the frontal midline to ventromedial and orbito- region that is minimally connected to limbic structures.
medial PFC (e.g., Fox and Raichle, 2007; Buckner et al., 2008).
A number of group-based studies, including our own, have Parallel Large-Scale Distributed Networks Are an
sought to fractionate the DN into subnetworks (e.g., Margulies Organizing Principle of Association Cortex
et al., 2007, 2009; Buckner et al., 2008; Fransson and Marrelec, A notable feature of the identified networks is that each contains
2008; Andrews-Hanna et al., 2010, 2014; Leech et al., 2011, components in frontal, parietal, temporal, and frontal midline re-
2012). While the present fractionation is not contained within gions. This repeating pattern or motif has been discussed previ-
these past efforts, prior observations have noted juxtaposed ously in group data (e.g., Yeo et al., 2011; Power et al., 2011).

466 Neuron 95, 457–471, July 19, 2017

 Figure 8. Diagrammatic Representation of
Six Parallel Distributed Networks within
One Individual
The central figure shows an illustration of the six
networks overlaid on the same cortical surface.
The top panel shows the lateral view, and the
lower panel shows the medial view. The different
colors correspond to the canonical network that
each network resembles (red, Default Network,
DN-A and DN-B; blue, Frontoparietal Network,
FPN-A and FPN-B; green, Dorsal Attention
Network, dATN-A and dATN-B). The names of the
networks are based on prior literature, recognizing
that the novel organization identified here may
lead to a reconsideration of the functional do-
mains. Data shown are from S4 (see also
Figure S6).

noted, DN-B does not couple to the hip-

pocampal formation, while DN-A does.
Similarly, dATN-B shows coupling to ret-
inotopic visual regions along the midline,
preferentially to the peripheral field repre-
sentation, while dATN-A does not (see
Figure S8). The presence of such differ-
ences may shed insight into the func-
tional role of the parallel networks and
their development. One possibility is
that there are broad constraints that
establish the same motif but, seeded by
competing inputs from limbic and sen-
sory systems, activity-dependent pro-
cesses differentiate the networks during
development.
A second important finding is the fine
spatial scale that differentiates neigh-
boring networks. The spatial interdigita-
tion of distinct regions was notable. In
the lateral temporal lobe, one network’s
region could be surrounded by another’s.
This fine-scale interdigitation has implica-
tions for interpreting prior network esti-
mations. For example, Mesulam (1981)
proposed a cortical network important
to spatial attention. Extensive findings
What is striking is how well the same distributed pattern ac- illustrate that certain areas near to the intraparietal sulcus form
counts for much of the newly fractionated networks. part of a sensory-motor hierarchy, sharing reciprocal projections
Goldman-Rakic (1988) suggested that this distributed motif with extrastriate visual cortex and the frontal eye fields (Maunsell
was a general organizing principle of association circuits. Based and van Essen, 1983; Ungerleider and Desimone, 1986; Ander-
on results from double-labeling tracer injections by Selemon and sen et al., 1990; Boussaoud et al., 1990). However, Mesulam’s
Goldman-Rakic (1985, 1988), she posited (1) that prefrontal and ideas drew on injections within the IPL area 7a in or near Opt
parietal areas are embedded within densely interconnected (e.g., Mesulam et al., 1977). This specific parietal association
distributed circuits that include midline and temporal areas and area is near to macaque area LIP but has a projection fingerprint
(2) that this same motif repeats across nearby zones forming that spares distant extrastriate areas while including cingulate
closely adjacent parallel networks. Our results are consistent and PHC (Andersen et al., 1990). Given the present results, it is
with her ideas. reasonable to suppose that past analyses of parietal association
One interesting finding is that there can be clear distinctions cortex may have lumped together injections in separate parallel,
even between networks with closely juxtaposed regions. As distributed networks.

Neuron 95, 457–471, July 19, 2017 467

Evidence for a Macroscale Organization of Association topography of these networks in subcortical and cerebellar
Cortex that Spans Networks structures.
An intriguing finding is revealed when the spatial relations across
all the networks are considered together (Figures 8 and S6). Conclusions
While the networks have complex interdigitated relationships, The present work reveals that there are parallel large-scale
there are also macroscale gradients that share the same general distributed networks that are spatially juxtaposed across the
progression in multiple zones of cortex. In parietal association cerebral cortex. The spatial scale of these networks is such
cortex, there is a caudal to rostral progression from DN-A that they become evident only when analyzed within the individ-
through to dATN-B (Figure 6). In temporal association cortex, ual. Discovery of the presence and description of details of these
there is a rostral to caudal progression through the same net- networks provide a foundation for future study of their functions.
works (Figure 7). The axis is imperfect, and it is sometimes un-
clear which network should be ordered before the other, but STAR+METHODS
the general ordering that repeats across distinct zones suggests
a broad macroscale organizing principle. Detailed methods are provided in the online version of this paper
Margulies et al. (2016) recently argued that association cortex and include the following:
possesses a macroscale gradient of networks from sensory-
motor networks on one end to the DN on the other. In agreement, d KEY RESOURCES TABLE
the possibility was recently raised that association networks d CONTACT FOR REAGENT AND RESOURCE SHARING
evolved from a prototypical distributed sensory-motor network d EXPERIMENTAL MODEL AND SUBJECT DETAILS
followed by a period of cortical expansion, which freed up zones B Participants
of association cortex from the constraints of primary sensory d METHOD DETAILS
input (Buckner and Krienen, 2013; Krienen and Buckner, 2017). B MRI Data Acquisition
The parallel and sequential nature of the presently defined net- B Data Preprocessing
works adds further support to these ideas. B Discovery of Networks within the Individual
B Hypothesis Testing to Dissociate Networks within the
Limitations and Technical Considerations Individual
An assumption behind our interpretation of the results is that FC B Effects of Misalignment between Individuals
across distributed regions is sufficiently constrained by direct B Additional Explorations Targeting Adjacent Networks
and polysynaptic anatomical circuits to provide insight into the or- within the Individual
ganization of distributed networks. Parallels with macaque anat- B Confirmation of Interdigitated Network Representa-
omy reinforce this assumption. However, details of the results tions in the Brain Volume
may be revised to the degree that factors beyond stable anatom- B Confirmation of Observed Parallel Networks Using
ical constraints contribute to the patterns. Exploration of monkey Data-Driven Network Parcellation
anatomy using multiple tracer injections from adjacent regions in B Additional Exploration of Network Organization along
the same animal would be a valuable complement to the present Regions of the Medial Temporal Lobe with Low
work. The anatomical origin of the fine spatial details are impor- Signal-To-Noise Ratio
tant to resolve because of their implications for clinical endeavors, B Test-Retest Reliability of Parallel Interdigitated
including presurgical planning and targeted neuromodulation. Networks
The present results suggest that targeted neuromodulation of d QUANTIFICATION AND STATISTICAL ANALYSIS
nearby cortical zones could have distinct effects because they
SUPPLEMENTAL INFORMATION
are embedded within anatomically separate networks. While
this general notion has been appreciated previously (e.g., Fox
Supplemental Information includes eight figures and can be found with this
et al., 2014), what is surprising is the fine spatial scale by which article online at http://dx.doi.org/10.1016/j.neuron.2017.06.038.
cortical zones participating in distinct networks are interdigitated.
Given that we were able to fractionate established large-scale AUTHOR CONTRIBUTIONS
networks by pushing the practical resolution of fMRI by targeting
the individual, it seems likely that our present estimates might R.M.B and R.L.B. designed the study, analyzed the data, interpreted the ex-
periments, and wrote the paper. R.M.B. collected the data.
also be fractionated further if higher resolution was achieved.
Despite efforts to minimize spatial blurring, we were unable to ACKNOWLEDGMENTS
confidently delineate all networks in all individuals. Two subjects,
S2 and S3, produced maps that were noticeably blurrier. This dif- We thank the Harvard Center for Brain Science neuroimaging core and FAS Di-
ference may be due to several factors, such as differences in vision of Research Computing. R. Mair and S. McMains assisted in optimizing
head motion, SNR, and misregistration. The reported summary acquisition. L. Farfel, M. Marotta, and R.M. Hutchison assisted in data acqui-
sition. R.M.H. and T.M. O’Keefe assisted in data preprocessing. R.M.B.
measures do not clearly indicate a cause. The fractionation of
was supported by Wellcome Trust grant 103980/Z/14/Z. This work was sup-
the DN into two parallel, distributed networks proved to be the
ported by Kent and Liz Dauten, NIH grant P50MH106435, and Shared Instru-
most robust finding observed clearly in all subjects. mentation Grant S10OD020039. The multi-band EPI sequence was gener-
The present study is also limited in that we studied only cere- ously provided by the Center for Magnetic Resonance Research (CMRR) at
bral cortical organization. Ongoing work is characterizing the University of Minnesota.

468 Neuron 95, 457–471, July 19, 2017

Received: February 1, 2017 De Luca, M., Beckmann, C.F., De Stefano, N., Matthews, P.M., and Smith,
Revised: April 28, 2017 S.M. (2006). fMRI resting state networks define distinct modes of long-dis-
Accepted: June 23, 2017 tance interactions in the human brain. Neuroimage 29, 1359–1367.
Published: July 19, 2017 Doucet, G., Naveau, M., Petit, L., Delcroix, N., Zago, L., Crivello, F., Jobard, G.,
Tzourio-Mazoyer, N., Mazoyer, B., Mellet, E., and Joliot, M. (2011). Brain activ-
REFERENCES ity at rest: a multiscale hierarchical functional organization. J. Neurophysiol.
105, 2753–2763.
€rgel, U., Mohlberg, H., Uylings, H.B., and Zilles,
Amunts, K., Schleicher, A., Bu Evans, A.C., Marrett, S., Neelin, P., Collins, L., Worsley, K., Dai, W., Milot, S.,
K. (1999). Broca’s region revisited: cytoarchitecture and intersubject vari- Meyer, E., and Bub, D. (1992). Anatomical mapping of functional activation in
ability. J. Comp. Neurol. 412, 319–341. stereotactic coordinate space. Neuroimage 1, 43–53.
Amunts, K., Malikovic, A., Mohlberg, H., Schormann, T., and Zilles, K. (2000). Fedorenko, E., Hsieh, P.J., Nieto-Castañón, A., Whitfield-Gabrieli, S., and
Brodmann’s areas 17 and 18 brought into stereotaxic space-where and how Kanwisher, N. (2010). New method for fMRI investigations of language:
variable? Neuroimage 11, 66–84. defining ROIs functionally in individual subjects. J. Neurophysiol. 104,
Andersen, R.A., Asanuma, C., Essick, G., and Siegel, R.M. (1990). 1177–1194.
Corticocortical connections of anatomically and physiologically defined subdi- Fedorenko, E., Duncan, J., and Kanwisher, N. (2012). Language-selective and
visions within the inferior parietal lobule. J. Comp. Neurol. 296, 65–113. domain-general regions lie side by side within Broca’s area. Curr. Biol. 22,
Andreasen, N.C., O’Leary, D.S., Cizadlo, T., Arndt, S., Rezai, K., Watkins, G.L., 2059–2062.
Ponto, L.L., and Hichwa, R.D. (1995). Remembering the past: two facets of Felleman, D.J., and Van Essen, D.C. (1991). Distributed hierarchical process-
episodic memory explored with positron emission tomography. Am. J. ing in the primate cerebral cortex. Cereb. Cortex 1, 1–47.
Psychiatry 152, 1576–1585. Fischl, B. (2012). FreeSurfer. Neuroimage 62, 774–781.
Andrews-Hanna, J.R., Reidler, J.S., Sepulcre, J., Poulin, R., and Buckner, R.L. Fischl, B., Sereno, M.I., and Dale, A.M. (1999). Cortical surface-based analysis.
(2010). Functional-anatomic fractionation of the brain’s default network. II: inflation, flattening, and a surface-based coordinate system. Neuroimage 9,
Neuron 65, 550–562. 195–207.
Andrews-Hanna, J.R., Saxe, R., and Yarkoni, T. (2014). Contributions of Fischl, B., Rajendran, N., Busa, E., Augustinack, J., Hinds, O., Yeo, B.T.T.,
episodic retrieval and mentalizing to autobiographical thought: evidence Mohlberg, H., Amunts, K., and Zilles, K. (2008). Cortical folding patterns and
from functional neuroimaging, resting-state connectivity, and fMRI meta-ana- predicting cytoarchitecture. Cereb. Cortex 18, 1973–1980.
lyses. Neuroimage 91, 324–335.
Fox, M.D., and Raichle, M.E. (2007). Spontaneous fluctuations in brain activity
Beckmann, C.F., DeLuca, M., Devlin, J.T., and Smith, S.M. (2005). observed with functional magnetic resonance imaging. Nat. Rev. Neurosci. 8,
Investigations into resting-state connectivity using independent component 700–711.
analysis. Philos. Trans. R. Soc. Lond. B Biol. Sci. 360, 1001–1013.
Fox, P.T., Perlmutter, J.S., and Raichle, M.E. (1985). A stereotactic method of
Binder, J.R., Desai, R.H., Graves, W.W., and Conant, L.L. (2009). Where is the anatomical localization for positron emission tomography. J. Comput. Assist.
semantic system? A critical review and meta-analysis of 120 functional neuro- Tomogr. 9, 141–153.
imaging studies. Cereb. Cortex 19, 2767–2796.
Fox, M.D., Corbetta, M., Snyder, A.Z., Vincent, J.L., and Raichle, M.E. (2006).
Biswal, B., Yetkin, F.Z., Haughton, V.M., and Hyde, J.S. (1995). Functional Spontaneous neuronal activity distinguishes human dorsal and ventral atten-
connectivity in the motor cortex of resting human brain using echo-planar tion systems. Proc. Natl. Acad. Sci. USA 103, 10046–10051.
MRI. Magn. Reson. Med. 34, 537–541.
Fox, M.D., Buckner, R.L., Liu, H., Chakravarty, M.M., Lozano, A.M., and
Blatt, G.J., Pandya, D.N., and Rosene, D.L. (2003). Parcellation of cortical af- Pascual-Leone, A. (2014). Resting-state networks link invasive and noninva-
ferents to three distinct sectors in the parahippocampal gyrus of the rhesus sive brain stimulation across diverse psychiatric and neurological diseases.
monkey: an anatomical and neurophysiological study. J. Comp. Neurol. 466, Proc. Natl. Acad. Sci. USA 111, E4367–E4375.
161–179.
Fransson, P., and Marrelec, G. (2008). The precuneus/posterior cingulate cor-
Boussaoud, D., Ungerleider, L.G., and Desimone, R. (1990). Pathways for mo- tex plays a pivotal role in the default mode network: Evidence from a partial
tion analysis: cortical connections of the medial superior temporal and fundus correlation network analysis. Neuroimage 42, 1178–1184.
of the superior temporal visual areas in the macaque. J. Comp. Neurol. 296,
Friston, K.J., Ashburner, J., Frith, C.D., Poline, J.B., Heather, J.D., and
462–495.
Frackowiak, R.S.J. (1995). Spatial registration and normalization of images.
Braga, R.M., and Leech, R. (2015). Echoes of the brain: local-scale represen- Hum. Brain Mapp. 3, 165–189.
tation of whole-brain functional networks within transmodal cortex.
Geschwind, N. (1965a). Disconnexion syndromes in animals and man. I. Brain
Neuroscientist 21, 540–551.
88, 237–294.
Buckner, R.L., and Krienen, F.M. (2013). The evolution of distributed associa-
Geschwind, N. (1965b). Disconnexion syndromes in animals and man. II. Brain
tion networks in the human brain. Trends Cogn. Sci. 17, 648–665.
88, 585–644.
Buckner, R.L., Andrews-Hanna, J.R., and Schacter, D.L. (2008). The brain’s Glasser, M.F., Sotiropoulos, S.N., Wilson, J.A., Coalson, T.S., Fischl, B.,
default network: anatomy, function, and relevance to disease. Ann. N Y Andersson, J.L., Xu, J., Jbabdi, S., Webster, M., Polimeni, J.R., et al.; WU-
Acad. Sci. 1124, 1–38. Minn HCP Consortium (2013). The minimal preprocessing pipelines for the
Buckner, R.L., Krienen, F.M., and Yeo, B.T. (2013). Opportunities and limita- Human Connectome Project. Neuroimage 80, 105–124.
tions of intrinsic functional connectivity MRI. Nat. Neurosci. 16, 832–837. Glasser, M.F., Coalson, T.S., Robinson, E.C., Hacker, C.D., Harwell, J.,
Caspers, S., Geyer, S., Schleicher, A., Mohlberg, H., Amunts, K., and Zilles, K. Yacoub, E., Ugurbil, K., Andersson, J., Beckmann, C.F., Jenkinson, M., et al.
(2006). The human inferior parietal cortex: cytoarchitectonic parcellation and (2016). A multi-modal parcellation of human cerebral cortex. Nature 536,
interindividual variability. Neuroimage 33, 430–448. 171–178.
Corbetta, M., and Shulman, G.L. (2002). Control of goal-directed and stimulus- Goldman-Rakic, P.S. (1988). Topography of cognition: parallel distributed net-
driven attention in the brain. Nat. Rev. Neurosci. 3, 201–215. works in primate association cortex. Annu. Rev. Neurosci. 11, 137–156.
Cowan, W.M., Gottlieb, D.I., Hendrickson, A.E., Price, J.L., and Woolsey, T.A. Gordon, E.M., Laumann, T.O., Adeyemo, B., and Petersen, S.E. (2017a).
(1972). The autoradiographic demonstration of axonal connections in the cen- Individual variability of the system-level organization of the human brain.
tral nervous system. Brain Res. 37, 21–51. Cereb. Cortex 27, 386–399.

Neuron 95, 457–471, July 19, 2017 469

Gordon, E.M., Laumann, T.O., Adeyemo, B., Gilmore, A.W., Nelson, S.M., Leech, R., Braga, R., and Sharp, D.J. (2012). Echoes of the brain within the
Dosenbach, N.U., and Petersen, S.E. (2017b). Individual-specific features posterior cingulate cortex. J. Neurosci. 32, 215–222.
of brain systems identified with resting state functional correlations. Marcus, D.S., Harwell, J., Olsen, T., Hodge, M., Glasser, M.F., Prior, F.,
Neuroimage 146, 918–939. Jenkinson, M., Laumann, T., Curtiss, S.W., and Van Essen, D.C. (2011).
Greicius, M.D., Krasnow, B., Reiss, A.L., and Menon, V. (2003). Functional Informatics and data mining tools and strategies for the human connectome
connectivity in the resting brain: a network analysis of the default mode hy- project. Front. Neuroinform. 5, 4.
pothesis. Proc. Natl. Acad. Sci. USA 100, 253–258. Margulies, D.S., Kelly, A.M., Uddin, L.Q., Biswal, B.B., Castellanos, F.X., and
Greicius, M.D., Srivastava, G., Reiss, A.L., and Menon, V. (2004). Default- Milham, M.P. (2007). Mapping the functional connectivity of anterior cingulate
mode network activity distinguishes Alzheimer’s disease from healthy aging: cortex. Neuroimage 37, 579–588.
evidence from functional MRI. Proc. Natl. Acad. Sci. USA 101, 4637–4642.
Margulies, D.S., Vincent, J.L., Kelly, C., Lohmann, G., Uddin, L.Q., Biswal,
Greve, D.N., and Fischl, B. (2009). Accurate and robust brain image alignment B.B., Villringer, A., Castellanos, F.X., Milham, M.P., and Petrides, M. (2009).
using boundary-based registration. Neuroimage 48, 63–72. Precuneus shares intrinsic functional architecture in humans and monkeys.
Gusnard, D.A., and Raichle, M.E. (2001). Searching for a baseline: functional Proc. Natl. Acad. Sci. USA 106, 20069–20074.
imaging and the resting human brain. Nat. Rev. Neurosci. 2, 685–694. Margulies, D.S., Ghosh, S.S., Goulas, A., Falkiewicz, M., Huntenburg, J.M.,
Henssen, A., Zilles, K., Palomero-Gallagher, N., Schleicher, A., Mohlberg, Langs, G., Bezgin, G., Eickhoff, S.B., Castellanos, F.X., Petrides, M., et al.
H., Gerboga, F., Eickhoff, S.B., Bludau, S., and Amunts, K. (2016). (2016). Situating the default-mode network along a principal gradient of
Cytoarchitecture and probability maps of the human medial orbitofrontal macroscale cortical organization. Proc. Natl. Acad. Sci. USA 113,
cortex. Cortex 75, 87–112. 12574–12579.
Hill, J., Inder, T., Neil, J., Dierker, D., Harwell, J., and Van Essen, D. (2010). Markov, N.T., Vezoli, J., Chameau, P., Falchier, A., Quilodran, R., Huissoud, C.,
Similar patterns of cortical expansion during human development and evolu- Lamy, C., Misery, P., Giroud, P., Ullman, S., et al. (2014). Anatomy of hierarchy:
tion. Proc. Natl. Acad. Sci. USA 107, 13135–13140. feedforward and feedback pathways in macaque visual cortex. J. Comp.
Neurol. 522, 225–259.
Holmes, A.J., Hollinshead, M.O., O’Keefe, T.M., Petrov, V.I., Fariello, G.R.,
Wald, L.L., Fischl, B., Rosen, B.R., Mair, R.W., Roffman, J.L., et al. (2015). Maunsell, J.H., and van Essen, D.C. (1983). The connections of the middle
Brain Genomics Superstruct Project initial data release with structural, func- temporal visual area (MT) and their relationship to a cortical hierarchy in the
tional, and behavioral measures. Sci. Data 2, 150031. macaque monkey. J. Neurosci. 3, 2563–2586.
Huth, A.G., de Heer, W.A., Griffiths, T.L., Theunissen, F.E., and Gallant, J.L. Mazoyer, B., Zago, L., Mellet, E., Bricogne, S., Etard, O., Houdé, O., Crivello,
(2016). Natural speech reveals the semantic maps that tile human cerebral cor- F., Joliot, M., Petit, L., and Tzourio-Mazoyer, N. (2001). Cortical networks for
tex. Nature 532, 453–458. working memory and executive functions sustain the conscious resting state
Jakobsen, E., Liem, F., Klados, M.A., Bayrak, Sx., Petrides, M., and Margulies, in man. Brain Res. Bull. 54, 287–298.
D.S. (2016). Automated individual-level parcellation of Broca’s region based Mennes, M., Jenkinson, M., Valabregue, R., Buitelaar, J.K., Beckmann, C., and
on functional connectivity. Neuroimage. Published online September 30, Smith, S. (2014). Optimizing full-brain coverage in human brain MRI through
2016. http://dx.doi.org/10.1016/j.neuroimage.2016.09.069. population distributions of brain size. Neuroimage 98, 513–520.
Jenkinson, M., Bannister, P., Brady, M., and Smith, S. (2002). Improved opti- Menon, V. (2011). Large-scale brain networks and psychopathology: a unifying
mization for the robust and accurate linear registration and motion correction triple network model. Trends Cogn. Sci. 15, 483–506.
of brain images. Neuroimage 17, 825–841.
Mesulam, M.M. (1981). A cortical network for directed attention and unilateral
Jones, E.G., and Powell, T.P. (1970). An anatomical study of converging neglect. Ann. Neurol. 10, 309–325.
sensory pathways within the cerebral cortex of the monkey. Brain 93,
Mesulam, M.M. (1990). Large-scale neurocognitive networks and distributed
793–820.
processing for attention, language, and memory. Ann. Neurol. 28, 597–613.
Kahn, I., Andrews-Hanna, J.R., Vincent, J.L., Snyder, A.Z., and Buckner, R.L.
Mesulam, M.M. (1998). From sensation to cognition. Brain 121, 1013–1052.
(2008). Distinct cortical anatomy linked to subregions of the medial temporal
lobe revealed by intrinsic functional connectivity. J. Neurophysiol. 100, Mesulam, M.M., Van Hoesen, G.W., Pandya, D.N., and Geschwind, N. (1977).
129–139. Limbic and sensory connections of the inferior parietal lobule (area PG) in the
rhesus monkey: a study with a new method for horseradish peroxidase histo-
Krienen, F.M., and Buckner, R.L. (2017). Human association cortex:
chemistry. Brain Res. 136, 393–414.
expanded, untethered, neotenous, and plastic. In Evolution of Nervous
Systems 2e, Volume 4, J. Kaas, ed. (Elsevier), pp. 169–183. Michalka, S.W., Kong, L., Rosen, M.L., Shinn-Cunningham, B.G., and Somers,
D.C. (2015). Short-term memory for space and time flexibly recruit comple-
Laird, A.R., Fox, P.M., Eickhoff, S.B., Turner, J.A., Ray, K.L., McKay, D.R.,
mentary sensory-biased frontal lobe attention networks. Neuron 87, 882–892.
Glahn, D.C., Beckmann, C.F., Smith, S.M., and Fox, P.T. (2011). Behavioral in-
terpretations of intrinsic connectivity networks. J. Cogn. Neurosci. 23, Moeller, S., Freiwald, W.A., and Tsao, D.Y. (2008). Patches with links: a unified
4022–4037. system for processing faces in the macaque temporal lobe. Science 320,
1355–1359.
Laumann, T.O., Gordon, E.M., Adeyemo, B., Snyder, A.Z., Joo, S.J., Chen,
M.-Y., Gilmore, A.W., McDermott, K.B., Nelson, S.M., Dosenbach, N.U.F., Mueller, S., Wang, D., Fox, M.D., Yeo, B.T.T., Sepulcre, J., Sabuncu, M.R.,
et al. (2015). Functional system and areal organization of a highly sampled in- Shafee, R., Lu, J., and Liu, H. (2013). Individual variability in functional connec-
dividual human brain. Neuron 87, 657–670. tivity architecture of the human brain. Neuron 77, 586–595.
Lavenex, P., Suzuki, W.A., and Amaral, D.G. (2002). Perirhinal and parahippo- Mueller, S., Wang, D., Fox, M.D., Pan, R., Lu, J., Li, K., Sun, W., Buckner, R.L.,
campal cortices of the macaque monkey: projections to the neocortex. and Liu, H. (2015). Reliability correction for functional connectivity: theory and
J. Comp. Neurol. 447, 394–420. implementation. Hum. Brain Mapp. 36, 4664–4680.
Leech, R., and Sharp, D.J. (2014). The role of the posterior cingulate cortex in Murphy, K., Birn, R.M., and Bandettini, P.A. (2013). Resting-state fMRI con-
cognition and disease. Brain 137, 12–32. founds and cleanup. Neuroimage 80, 349–359.
Leech, R., Kamourieh, S., Beckmann, C.F., and Sharp, D.J. (2011). Ojemann, J.G., Akbudak, E., Snyder, A.Z., McKinstry, R.C., Raichle, M.E., and
Fractionating the default mode network: distinct contributions of the ventral Conturo, T.E. (1997). Anatomic localization and quantitative analysis of
and dorsal posterior cingulate cortex to cognitive control. J. Neurosci. 31, gradient refocused echo-planar fMRI susceptibility artifacts. Neuroimage 6,
3217–3224. 156–167.

470 Neuron 95, 457–471, July 19, 2017

Ongu €r, D., and Price, J.L. (2000). The organization of networks within the Temporally-independent functional modes of spontaneous brain activity.
orbital and medial prefrontal cortex of rats, monkeys and humans. Cereb. Proc. Natl. Acad. Sci. USA 109, 3131–3136.
Cortex 10, 206–219. Smith, S.M., Beckmann, C.F., Andersson, J., Auerbach, E.J., Bijsterbosch, J.,
Pandya, D.N., and Seltzer, B. (1982). Intrinsic connections and architectonics Douaud, G., Duff, E., Feinberg, D.A., Griffanti, L., Harms, M.P., et al.; WU-Minn
of posterior parietal cortex in the rhesus monkey. J. Comp. Neurol. 204, HCP Consortium (2013). Resting-state fMRI in the Human Connectome
196–210. Project. Neuroimage 80, 144–168.
Patel, G.H., Yang, D., Jamerson, E.C., Snyder, L.H., Corbetta, M., and Ferrera, Spreng, R.N., Mar, R.A., and Kim, A.S. (2009). The common neural basis of
V.P. (2015). Functional evolution of new and expanded attention networks in autobiographical memory, prospection, navigation, theory of mind, and the
humans. Proc. Natl. Acad. Sci. USA 112, 9454–9459. default mode: a quantitative meta-analysis. J. Cogn. Neurosci. 21, 489–510.
Power, J.D., Cohen, A.L., Nelson, S.M., Wig, G.S., Barnes, K.A., Church, J.A., Steinmetz, H., and Seitz, R.J. (1991). Functional anatomy of language process-
Vogel, A.C., Laumann, T.O., Miezin, F.M., Schlaggar, B.L., and Petersen, S.E. ing: neuroimaging and the problem of individual variability. Neuropsychologia
(2011). Functional network organization of the human brain. Neuron 72, 29, 1149–1161.
665–678. Torous, J., Kiang, M.V., Lorme, J., and Onnela, J.P. (2016). New tools for new
Power, J.D., Schlaggar, B.L., and Petersen, S.E. (2014). Studying brain orga- research in psychiatry: a scalable and customizable platform to empower data
nization via spontaneous fMRI signal. Neuron 84, 681–696. driven smartphone research. JMIR Ment Health 3, e16.
Rademacher, J., Caviness, V.S., Jr., Steinmetz, H., and Galaburda, A.M. Ungerleider, L.G., and Desimone, R. (1986). Cortical connections of visual area
(1993). Topographical variation of the human primary cortices: implications MT in the macaque. J. Comp. Neurol. 248, 190–222.
for neuroimaging, brain mapping, and neurobiology. Cereb. Cortex 3,
van der Kouwe, A.J., Benner, T., Fischl, B., Schmitt, F., Salat, D.H., Harder, M.,
313–329.
Sorensen, A.G., and Dale, A.M. (2005). On-line automatic slice positioning for
Rajkowska, G., and Goldman-Rakic, P.S. (1995). Cytoarchitectonic definition brain MR imaging. Neuroimage 27, 222–230.
of prefrontal areas in the normal human cortex: II. Variability in locations of
van der Kouwe, A.J., Benner, T., Salat, D.H., and Fischl, B. (2008). Brain
areas 9 and 46 and relationship to the Talairach Coordinate System. Cereb.
morphometry with multiecho MPRAGE. Neuroimage 40, 559–569.
Cortex 5, 323–337.
Van Dijk, K.R., Hedden, T., Venkataraman, A., Evans, K.C., Lazar, S.W., and
Robinson, E.C., Jbabdi, S., Glasser, M.F., Andersson, J., Burgess, G.C.,
Buckner, R.L. (2010). Intrinsic functional connectivity as a tool for human
Harms, M.P., Smith, S.M., Van Essen, D.C., and Jenkinson, M. (2014). MSM:
connectomics: theory, properties, and optimization. J. Neurophysiol. 103,
a new flexible framework for Multimodal Surface Matching. Neuroimage
297–321.
100, 414–426.
Van Essen, D.C. (2005). A Population-Average, Landmark- and Surface-based
Rosene, D.L., and Van Hoesen, G.W. (1977). Hippocampal efferents reach
(PALS) atlas of human cerebral cortex. Neuroimage 28, 635–662.
widespread areas of cerebral cortex and amygdala in the rhesus monkey.
Science 198, 315–317. Van Essen, D.C., Anderson, C.H., and Felleman, D.J. (1992). Information pro-
cessing in the primate visual system: an integrated systems perspective.
Saxe, R., and Kanwisher, N. (2003). People thinking about thinking people. The
Science 255, 419–423.
role of the temporo-parietal junction in ‘‘theory of mind’’. Neuroimage 19,
1835–1842. Vincent, J.L., Snyder, A.Z., Fox, M.D., Shannon, B.J., Andrews, J.R., Raichle,
M.E., and Buckner, R.L. (2006). Coherent spontaneous activity identifies a hip-
Selemon, L.D., and Goldman-Rakic, P.S. (1985). Longitudinal topography and
pocampal-parietal memory network. J. Neurophysiol. 96, 3517–3531.
interdigitation of corticostriatal projections in the rhesus monkey. J. Neurosci.
5, 776–794. Vincent, J.L., Patel, G.H., Fox, M.D., Snyder, A.Z., Baker, J.T., Van Essen,
D.C., Zempel, J.M., Snyder, L.H., Corbetta, M., and Raichle, M.E. (2007).
Selemon, L.D., and Goldman-Rakic, P.S. (1988). Common cortical and
Intrinsic functional architecture in the anaesthetized monkey brain. Nature
subcortical targets of the dorsolateral prefrontal and posterior parietal cortices
447, 83–86.
in the rhesus monkey: evidence for a distributed neural network subserving
spatially guided behavior. J. Neurosci. 8, 4049–4068. Vincent, J.L., Kahn, I., Snyder, A.Z., Raichle, M.E., and Buckner, R.L. (2008).
Evidence for a frontoparietal control system revealed by intrinsic functional
Setsompop, K., Gagoski, B.A., Polimeni, J.R., Witzel, T., Wedeen, V.J., and
connectivity. J. Neurophysiol. 100, 3328–3342.
Wald, L.L. (2012). Blipped-controlled aliasing in parallel imaging for simulta-
neous multislice echo planar imaging with reduced g-factor penalty. Magn. Vogt, B.A., Vogt, L., and Laureys, S. (2006). Cytology and functionally corre-
Reson. Med. 67, 1210–1224. lated circuits of human posterior cingulate areas. Neuroimage 29, 452–466.
Shadlen, M.N., and Newsome, W.T. (2001). Neural basis of a perceptual deci- Voyvodic, J.T., Glover, G.H., Greve, D., and Gadde, S.; FBIRN (2011).
sion in the parietal cortex (area LIP) of the rhesus monkey. J. Neurophysiol. 86, Automated real-time behavioral and physiological data acquisition and display
1916–1936. integrated with stimulus presentation for FMRI. Front. Neuroinform. 5, 27.
Shulman, G.L., Fiez, J.A., Corbetta, M., Buckner, R.L., Miezin, F.M., Raichle, Weiskopf, N., Hutton, C., Josephs, O., and Deichmann, R. (2006). Optimal EPI
M.E., and Petersen, S.E. (1997). Common blood flow changes across visual parameters for reduction of susceptibility-induced BOLD sensitivity losses: a
tasks: II. Decreases in cerebral cortex. J. Cogn. Neurosci 9, 648–663. whole-brain analysis at 3 T and 1.5 T. Neuroimage 33, 493–504.
Silbersweig, D.A., Stern, E., Frith, C.D., Cahill, C., Schnorr, L., Grootoonk, S., Yeo, B.T.T., Krienen, F.M., Sepulcre, J., Sabuncu, M.R., Lashkari, D.,
Spinks, T., Clark, J., Frackowiak, R., and Jones, T. (1993). Detection of thirty- Hollinshead, M., Roffman, J.L., Smoller, J.W., Zöllei, L., Polimeni, J.R., et al.
second cognitive activations in single subjects with positron emission tomog- (2011). The organization of the human cerebral cortex estimated by intrinsic
raphy: a new low-dose H2(15)O regional cerebral blood flow three-dimen- functional connectivity. J. Neurophysiol. 106, 1125–1165.
sional imaging technique. J. Cereb. Blood Flow Metab. 13, 617–629. Yeo, B.T.T., Krienen, F.M., Chee, M.W., and Buckner, R.L. (2014). Estimates of
Smith, S.M., Jenkinson, M., Woolrich, M.W., Beckmann, C.F., Behrens, T.E., segregation and overlap of functional connectivity networks in the human ce-
Johansen-Berg, H., Bannister, P.R., De Luca, M., Drobnjak, I., Flitney, D.E., rebral cortex. Neuroimage 88, 212–227.
et al. (2004). Advances in functional and structural MR image analysis and im- Zou, Q.H., Zhu, C.Z., Yang, Y., Zuo, X.N., Long, X.Y., Cao, Q.J., Wang, Y.F.,
plementation as FSL. Neuroimage 23 (Suppl 1 ), S208–S219. and Zang, Y.F. (2008). An improved approach to detection of amplitude of
Smith, S.M., Miller, K.L., Moeller, S., Xu, J., Auerbach, E.J., Woolrich, M.W., low-frequency fluctuation (ALFF) for resting-state fMRI: fractional ALFF.
Beckmann, C.F., Jenkinson, M., Andersson, J., Glasser, M.F., et al. (2012). J. Neurosci. Methods 172, 137–141.

Neuron 95, 457–471, July 19, 2017 471

STAR+METHODS

KEY RESOURCES TABLE

REAGENT or RESOURCE SOURCE IDENTIFIER

Software and Algorithms
FreeSurfer Fischl, 2012 http://surfer.nmr.mgh.harvard.edu
MATLAB MathWorks www.mathworks.com
FSL Smith et al., 2004 https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/
Connectome Workbench Marcus et al., 2011 www.humanconnectome.org

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources should be directed to and will be fulfilled by the Lead Contact, Rodrigo M. Braga
([email protected] or [email protected]).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Participants
Four healthy female right-handed young adults (ages 21 to 26) were recruited from the greater Boston community for a study that
involved 24 separate MRI scanning sessions as well as extensive behavioral monitoring over a period of approximately 16 weeks.
None of the participants were local (Harvard) students or institutional employees, and all were paid for participation with milestone
payments at the end of the trial period (two weeks), after 12 MRI sessions, and after all 24 MRI sessions. Participants were screened
to exclude a history of neurological or psychiatric illness, or ongoing use of psychoactive medications. Seven participants were
enrolled and three chose not to continue within the first two weeks (which were described to all participants as a trial period).
Four participants were enrolled for the full 24-session extended study and all of these individuals completed all intended sessions.
The present paper concerns only the fMRI data, but enrolled individuals also participated in extensive behavioral testing and daily
monitoring of behavior via smart phones (Beiwe; Torous et al., 2016), sleep and activity monitoring via an Actigraph 2 wrist wearable
(Philips Respironics, Murrysville, PA, USA), as well as hearing tests at several times during the study (Model 2500 Microprocessor
Audiometer, AMBCO, Tustin, CA, USA). One subject required vision correction using MRI compatible glasses. Participants provided
written informed consent in accordance with the guidelines set by the Institutional Review Board of Harvard University.

METHOD DETAILS

MRI Data Acquisition

Data were collected on a 3T Siemens Prisma-fit MRI scanner (Siemens Healthcare, Erlangen, Germany) using the vendor’s 64-chan-
nel phased-array head-neck coil. Heads were immobilized with Siemens small foam head coil wedges. Each of the 24 MRI sessions
included one 7 m 2 s run of resting state data (passive fixation) to estimate intrinsic functional connectivity (Biswal et al., 1995) as well
as a number of other acquisitions (structural, ASL, and task-based functional runs). The resting state run was collected in the same
fixed order during every session near to the beginning of the session to optimize compliance. Participants were instructed to remain
still, stay awake and to fixate a centrally presented crosshair presented in black on a light gray background. The gray background
color was used instead of white to reduce glare, eye fatigue and discomfort. The position of the screen was adjusted at the beginning
of each session to ensure a comfortable, central viewing position to minimize muscle tension and head motion. Before each session
participants were encouraged to spend a few minutes finding a comfortable lying position that they could maintain for the entire ses-
sion. The scanner room lights were kept on to deter participants from becoming drowsy.
Eye closures and movements were monitored using the Eyelink 1000 Core Plus with Long-Range Mount. A video of the eye tracker
output was recorded in order to quantify compliance and arousal. Additional in-scanner physiological monitoring (Biopac Systems
Inc, Goleta, CA, USA) included a respirator belt around the chest to monitor breathing (Biopac, TSD221-MRI), electrodermal elec-
trodes (Biopac, EL508) attached to participants’ right sole to measure galvanic skin response, and a band pulse-oximeter (Biopac,
OXY-MRI-SENSOR) attached to participants’ right middle toe to measure oxygen saturation and pulse rate. See Voyvodic et al.
(2011) for details. The oximeter and electrodes were place on participants’ feet to keep their hands free to make responses using
button boxes during the in-scanner tasks. Immediately before each run, participants were asked to remain still for the entirety of

e1 Neuron 95, 457–471.e1–e5, July 19, 2017

the upcoming run. Following each run, participants were given feedback about noted movements and encouraged to stay still, in
order to establish an expectation that their compliance and movement were being carefully watched.
Functional imaging data were acquired using a multi-band gradient-echo echo-planar pulse sequence (Setsompop et al., 2012)
with acquisition parameters: TR 1000 ms; TE 32.6 ms; flip angle 64 ; 2.4 mm isotropic voxels; FOV 211 mm x 211 mm x 156 mm;
65 slices fully covering the cerebral cortex and cerebellum. Slice acquisition used interleaved simultaneous multi-slice 5x accelera-
tion. The sequence was a custom sequence generously provided by the Center for Magnetic Resonance Research (CMRR) at
University of Minnesota. Whole brain coverage and minimization of signal dropout due to magnetic susceptibility were achieved
by aligning slices to a plane 25 degrees from the anterior commissure-posterior commissure plane toward the coronal plane (Weis-
kopf et al., 2006; Mennes et al., 2014). This was implemented using an automated alignment procedure to ensure consistency across
sessions (van der Kouwe et al., 2005) and, in pilot acquisitions, was found to increase signal-to-noise in ventromedial prefrontal cor-
tex (PFC). A rapid T1-weighted structural image was also acquired in each session using a multi-echo MPRAGE three-dimensional
sequence (van der Kouwe et al., 2008) with acquisition parameters: TR 2200 ms; TE 1.57, 3.39, 5.21, 7.03 ms; flip angle 7 ; 1.2mm
isotropic voxels; 144 slices; FOV 230 mm x 230 mm x 173 mm, in-plane GRAPPA acceleration 4 (see Holmes et al., 2015 for empirical
results and discussion of comparability of this brief sequence to traditional longer acquisitions).

Data Preprocessing
Resting-state data were processed using methods similar to those previously described (Van Dijk et al., 2010; Yeo et al., 2011): (1) 12
initial volumes from each run were discarded to allow for T1-equillibration, (2) head motion was corrected using rigid body translation
and rotation (FSL, https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/; Jenkinson et al., 2002; Smith et al., 2004), (3) data were temporally low-pass
filtered at a threshold of 0.08 Hz, and (4) nuisance variables (6 motion parameters, mean whole-brain signal, mean ventricular signal,
mean deep cerebral white matter signal) and their temporal derivatives were regressed. Structural data were processed using the
FreeSurfer version 4.5.0 software package (http://surfer.nmr.mgh.harvard.edu; Fischl, 2012). For each anatomical image (one per
session), a surface mesh representation of the cortex was reconstructed and registered to a common spherical coordinate system
by aligning the major sulcal patterns to the FreeSurfer average template (Fischl et al., 1999). The preprocessed functional images
from each session were aligned to the cortical surface mesh reconstructed from that session’s anatomical image using bound-
ary-based registration (Greve and Fischl, 2009). Functional data were then propagated to the common spherical coordinate system
via sampling (trilinear interpolation) from the middle of the cortical ribbon in a single interpolation step.
Functional data were sampled to the fsaverage6 surface mesh (Fischl et al., 1999) containing 40,962 vertices per hemisphere, and
a 2mm full-width-at-half-maximum (FWHM) smoothing kernel was applied to the data in the surface space. A mesh resolution of
40,962 vertices was chosen to reduce blurring during the trilinear interpolation step and hence maximize the potential for observing
network distinctions, while keeping the computational burden manageable. A bespoke cortical surface template containing 40,962
vertices per hemisphere was produced using the Connectome Workbench’s command suite (Glasser et al., 2013). This was done so
that the functional connectivity analyses could be performed and visualized interactively within the Workbench’s flexible surface-
based visualization software, wb_view (Marcus et al., 2011). The bespoke template was created by combining the left and right
pial surfaces from the fsaverage6 freesurfer template into the CIFTI format using the Workbench commands. The pial surfaces
were then selectively inflated (smoothing cycles: 3, smoothing strength 0.7, smoothing-iterations 13, inflation factor 1.02) using
Workbench to allow visualization of the major cortical folds while maintaining the majority of the cortical surface visible.
Voxel-based tSNR maps were computed by taking the motion-corrected time series from each functional run and dividing the
mean signal at each voxel by its standard deviation over time. The tSNR maps were then averaged across functional runs within
the Discovery (n = 12), Replication (n = 12) and Full datasets (n = 24), and the resulting mean tSNR maps were projected to the cortical
surface for visualization using FreeSurfer (Figure S1). An additional metric of data quality, fractional Amplitude of Low Frequency
Fluctuations (fALFF), was also computed (Figure S1). fALFF maps were produced by normalizing the total power in the low
(0.01 – 0.08 Hz) frequency range by the total power across all frequencies (Zou et al., 2008).

Discovery of Networks within the Individual

For each participant, half of her data were used in a discovery manner to identify networks that would be later tested in the remaining
independently collected data sessions. The odd-numbered sessions (i.e., 1st, 3rd, 5th, etc) formed the discovery dataset (n = 12) while
the even-numbered sessions (i.e., 2nd, 4th, 6th, etc) were set aside as the hypothesis-testing dataset (next section). The discovery
datasets in each of the four participants were analyzed blind to the hypothesis-testing datasets.
For the discovery analysis, Pearson’s product moment correlations between the fMRI time series at each cortical surface vertex
were computed. This resulted in an 81,924 3 81,924 element cross-correlation matrix (40,962 vertices per hemisphere) for each of
the 12 fMRI runs from the discovery dataset. The matrices were then r-to-z transformed and averaged to yield a mean matrix with high
stability. These discovery matrices were used to explore network organization. The mean cross correlation matrices were assigned to
the bespoke cortical template so that individual seed vertices could be selected and their functional connectivity maps interactively
viewed using wb_view (Marcus et al., 2011). Individual vertices were selected from the general vicinity of expected locations of target
networks in PFC (as estimated from independent group-averaged data; n = 1000 from Yeo et al., 2011) and expectations from the
literature (e.g., Power et al., 2011; Yeo et al., 2011).

Neuron 95, 457–471.e1–e5, July 19, 2017 e2

 First, a seed vertex was selected from lateral PFC near regions which form part of the canonical default network, and the resulting
maps visualized. If the functional connectivity map revealed a network of distributed regions showing a robust correlation with the
seed region (z(r) z0.6), the seed vertex number was recorded for further analysis. If the map revealed no strong correlations, or a
diffuse network of correlations that were observably lower (z(r) z0.4), this was taken as evidence that the seed region was sampling
an area of signal dropout or containing a mixture of signals, respectively, and a different seed region was selected. The colorbar scale
was set between 0.2 and 0.6, using the JET256 palette in wb_view, so that the correlation structure could be adequately represented
and these subtle differences observed. After a robust network was detected and recorded, a different seed vertex was selected that
satisfied 4 criteria: the second seed region i) was within the lateral PFC, ii) was in the vicinity of the previously identified network’s
seed, iii) was in a region of the cortical surface that showed low correlation (z(r)<0.3) with the previous network’s seed region, and
iv) also showed robust correlation with a distributed set of regions. Thus the analysis converged on robust networks that had closely
neighboring representations within the lateral PFC.
The goal of this discovery procedure was not to confirm expectations from the group data, but to allow the individual participant
maps to be extensively interrogated, moving outward from properties expected from the group maps. Given prior analyses of within-
subject data (e.g., Figure 4 from De Luca et al., 2006; Figure 3 from Vincent et al., 2006; Figure 9 from Van Dijk et al., 2010), it was
unsurprising that many maps constructed within the individual participants resembled canonical networks discovered in group-
analyzed data. The targets of our exploration were network features that are not fully captured by group analyses (e.g., Laumann
et al., 2015; Gordon et al., 2017a, 2017b; see also Fedorenko et al., 2012).
As the results will demonstrate, seed regions placed in nearby regions of lateral PFC revealed two important features that led to the
regions selected for the hypothesis-testing phase of analysis. The first feature was that, like typical group-based analyses, the result-
ing distributed networks contained inferior parietal, temporal, medial prefrontal, and posteromedial cortical components near to what
has been described as the ‘default network’ (DN). Second, nearby seed regions yielded distributed networks that were closely
neighboring but separate throughout much of the distributed organization of the network. That is, two distinct networks were closely
interdigitated throughout the topography of the canonical DN suggesting the hypothesis that the DN may be comprised of multiple
neighboring networks. We refer to these hypothesized networks as Network A and Network B (Figure 1).

Hypothesis Testing to Dissociate Networks within the Individual

The discovery phase of data analysis led to the hypothesis that two distinct interdigitated networks exist that each are within or near
to the canonical DN (Figure 1). The two networks were present in each individual participant. The goal of the hypothesis-testing phase
was to use independent data in each participant to support or refute the possibility of two dissociable networks. To conduct this anal-
ysis, two distinct neighboring lateral PFC seed regions were selected within each participant as well as pairs of regions throughout the
cortex based only on the discovery datasets that maximized the separation of the distributed networks. A priori regions (single
vertices) were selected in each of the main regional zones of the cortex (temporal, inferior parietal, posteromedial, and medial pre-
frontal). These a priori regions were targeted to locations where contiguous vertices could be observed in Network A and Network B
(Figure 1). A seed region was also selected in parahippocampal cortex (PHC) to quantify the representation of Network A in this re-
gion, given a prior literature linking the hippocampal formation to the DN (Greicius et al., 2004; Vincent et al., 2006; Kahn et al., 2008).
These regions were then statistically tested in the independent data to formally dissociate the two networks.
The critical test was whether there would be significant interactions between the two lateral PFC seed regions and Network A and
Network B regions in each of the zones of cortex in the independent hypothesis-testing datasets. The presence of significant inter-
actions would be evidence for regional dissociations. The presence of interactions across all distributed zones of the cortex would be
strong evidence that there was a complete network dissociation across the cortex, in essence establishing dissociable but adjacent
networks. For each of the 12 hypothesis-testing data sessions within each participant, values representing the r-to-z transformed
Pearson’s product moment correlation for each of the two lateral PFC seed regions were computed in relation to each of the distrib-
uted test regions. Statistical tests were performed as a two-way ANOVA. In each ANOVA, the independent (classification) variables
were the PFC seed location (A and B) and the a priori test regions (A and B) within one of the separate zones of cortex. The dependent
variable was the r-to-z transformed Pearson’s product moment correlation between each seed and test region. Each cell contained a
correlation measure from each of the 12 sessions in the hypothesis-testing dataset, producing a balanced 2 X 2 factorial design with
12 elements in each cell. This 2-factor analysis was repeated for each of the separate cortical zones (temporal, inferior parietal, post-
eromedial, and medial prefrontal). Statistical significance level was set at p < 0.01. Critically, the multiple tests across the network and
across subjects were not independent in the sense that they were targeting convergent evidence for network dissociation. Thus
multiple, repeated significant results in the ANOVA across regions and participants would provide a high level of certainty for disso-
ciation. Sporadic significance that occurred in 1 in 20 or 1 in 100 tests that showed no specific pattern might be indicative of false
positives. As the results will reveal, the data patterns and hypothesis-directed statistical tests were quite clear in the weight of their
evidence.

Effects of Misalignment between Individuals

The details of network organization that are revealed by our analyses within individuals suggest a level of spatial and anatomical
specificity that would likely be lost or underappreciated when central tendencies across participants are extracted in group-averaged
data or when details of anatomy in one participant are assumed to apply to another. That intuition can be appreciated visually by

e3 Neuron 95, 457–471.e1–e5, July 19, 2017

examining Figures 1, 3, 4, 5, 6, and 7, and also in results from other papers (e.g., Figure 7 from Laumann et al., 2015). To formally
explore this issue, we asked to what degree network structure present using individually-tailored regions is lost when one person’s
anatomy is assumed to apply to another person.
For these analyses, in each of the four participants we extracted the hypothesized cortical regions that were components of DN-A
and DN-B from the discovery datasets (lateral prefrontal, temporal, inferior parietal, posteromedial, and medial prefrontal, but not the
parahippocampal region; Figure 1), yielding 10 individually-tailored regions per subject. These regions were then used to construct a
10 3 10 matrix of correlations in the hypothesis-testing dataset for each participant. To address the question of whether misalign-
ment between individuals affects the results, we recalculated the matrices in each participant using the regions defined in the other
three participants (Figure S5). The goal of this analysis is to provide some level of visualization of what is maintained and what is lost
when misalignment is present across individuals. Our focus here is on the specific dissociation between DN-A and DN-B as an
example of an important feature of functional-anatomical organization (see Gordon et al., 2017a for a conceptually similar analysis
performed for algorithmically defined individually-tailored patches). As can be seen in Figure S5, the clustering of individually tailored
regions into the two distinct networks breaks down when seed region locations from a different subject are used. This result illustrates
that the spatial variation between individuals is enough to obscure the clear and reproducible network dissociation that is uncovered
within the individuals.

Additional Explorations Targeting Adjacent Networks within the Individual

The main analyses explored and tested for dissociation of two distinct neighboring networks that fractionate the canonical DN. This
was an unexpected result and encouraged further exploration to determine whether there were further network fractionations. For
these additional, post hoc explorations the full dataset for each participant was used to dissociate networks. Results are reported
for network properties that were evident in at least two separate participants.

Confirmation of Interdigitated Network Representations in the Brain Volume

Surface projection provides a convenient means to visualize cortical organization and the juxtaposition of networks on the continuous
surface. However, projection to the surface involves a spatially-nonlinear transformation that can project nearby voxels in the volume
to distant positions on the surface (such as occurs when folds near the crowns of separate gyri abut one another). It is thus important
to check that the observed interdigitation between the dissociated networks is present in both surface and volume representations
and is not the result of the cortical sampling procedure used. Moreover, volume-based analysis will be required for many applied
endeavors including presurgical planning and localization for neuromodulation (e.g., transcranial magnetic and direct-current
stimulation).
To explore these issues, we reproduced the findings of two dissociated networks linked to the default network in the volume. The
functional data from each subject were preprocessed as described for the surface-based pipeline, with the exceptions that smooth-
ing was performed in the volume at 2mm FWHM prior to regression of nuisance variables and bandpass filtering, and that data were
not projected to the surface. The first 8 resting state runs from each subject were concatenated in time and AFNI’s InstaCorr (R. Cox
and Z. Saad, 2010, International Conference on Resting-State Functional Brain Connectivity, conference; https://afni.nimh.nih.gov/
afni) software was used to interactively select seed voxels and view the resulting FC maps. Two seed voxels were selected in left
lateral PFC that maximized the separation between the two FC map representations in posteromedial cortex, namely a ventral
representation in retrosplenial cortex for DN-A, and the dorsal posterior cingulate cortex representation for DN-B. When the two
candidate seed regions were identified, the seed voxel locations were recorded and FC maps were produced in all 24 runs of the
data for each subject.
Pearson’s product moment correlations were calculated between the fMRI time series at each voxel and the seed voxel. The maps
were then r-to-z transformed and averaged to produce a single FC map for each of the two seed regions. The question was whether
the two parallel networks would retain the critical features observed in the cortical surface when viewed in the volume (Figure S4).
More broadly, while the surface-based visualization proved most effective for discovering organizational details, volume-based anal-
ysis is important to fully understand the underlying organizational features, including the possibilities that complex geometry of the
sulcal patterns and also non-neuronal structures such as vessels may contribute to estimated features of cortical organization.

Confirmation of Observed Parallel Networks Using Data-Driven Network Parcellation

The network dissociations reported were discovered and replicated using seed-based methods within-subjects that generalized
across subjects. If the network dissociations are robust, they should also be observable using multiple approaches. That is, the
path we initially employed to uncover them should not be the only way to reveal their presence and their specific spatial organization.
To investigate whether the fractionation of the default network generalized across approaches, we explored network organization
within the individual participants using data-driven clustering techniques.
We concatenated the surface-projected timeseries data from the Discovery dataset (n = 12), and used MATLAB’s kmeans function
(version R2015a; MathWorks, Natick, MA) to parcellate the vertex timeseries into clusters. As discussed extensively in our earlier
analysis of group-based network estimates (Yeo et al., 2011), clustering approaches yield multiple solutions to data parcellations
at many levels of clustering. Here we explored a relatively low dimensional fractionation (k = 12) that was able to yield the key disso-
ciated networks including many of their distributed subcomponents.

Neuron 95, 457–471.e1–e5, July 19, 2017 e4

Additional Exploration of Network Organization along Regions of the Medial Temporal Lobe with Low
Signal-To-Noise Ratio
As the results unfolded, the DN-A was found to possess a clear representation in the PHC, while no such representation was detected
for DN-B. As portions of the medial temporal lobe are susceptible to signal drop-out in fMRI (Ojemann et al., 1997), additional ana-
lyses were conducted to try to find evidence for a representation of DN-B in this region. Additional analyses were focused on the
subject that showed the most robust network dissociations (S4). First, the FC maps for DN-A and DN-B from the PFC seeds were
disattenuated by using the reliability of the functional connectivity maps across runs as an estimate of the signal dropout (Mueller
et al., 2015; Figure S2A). This revealed two representations belonging to DN-B in the inferior temporal lobe along zones of suscep-
tibility artifact, but in regions located well outside the PHC. Second, seed vertices were selected in a posterior to anterior progression
along the medial temporal lobe, and the resulting FC maps were observed (Figure S2B). This analysis also did not reveal evidence for
a representation of DN-B in the vicinity of the representation of DN-A. We also further explored whether other networks might be
detected along the inferior and anterior temporal lobe, once signal dropout was partially accounted for. We describe these additional
observations as potential avenues for further more detailed investigations as well as a reminder that, even with our numerous steps to
increase signal-to-noise through signal averaging and our use of small voxels and acceleration during acquisition, signal loss due to
susceptibility artifacts is still a problem in certain zones of cortex.

Test-Retest Reliability of Parallel Interdigitated Networks

The FC maps for DN-A and DN-B were produced from the Discovery and Replication datasets (Figure S3). Additionally, the mean
connectivity matrix from the Discovery (n = 12) and Replication (n = 12) datasets were correlated to show the consistency of the con-
nectivity patterns within each subject (S1: r = 0.88; S2: r = 0.95; S3: r = 0.87; S4: r = 0.94, all p < 0.001).

QUANTIFICATION AND STATISTICAL ANALYSIS

This study includes n = 4 participants, each of which were scanned over 24 fMRI sessions. Each participant’s imaging data were
divided into discovery (n = 12; odd-numbered runs) and replication (n = 12; even-numbered runs) samples to allow for data explo-
ration and statistical testing using independent data points. Functional connectivity between brain regions was calculated in MATLAB
(version 7.4; http://www.mathworks.com; MathWorks, Natick, MA) using Pearson’s product moment correlations which were r to z
transformed prior to averaging or comparison. Statistical tests were performed as a two-way ANOVA using MATLAB’s anova2 func-
tion (version R2015a). Statistical significance was set to p < 0.01. Network parcellation was performed using MATLAB’s kmeans func-
tion (version R2015a).

e5 Neuron 95, 457–471.e1–e5, July 19, 2017