NeuroImage 235 (2021) 118027

Contents lists available at ScienceDirect

NeuroImage
journal homepage: www.elsevier.com/locate/neuroimage

Review

Understanding neural flexibility from a multifaceted definition

Dazhi Yin a,∗, Marcus Kaiser b,c,d,∗∗
a
Key Laboratory of Brain Functional Genomics (Ministry of Education and Shanghai), School of Psychology and Cognitive Science, East China Normal University,
Shanghai 200062, China
b
School of Computing, Newcastle University, Newcastle upon Tyne NE4 5TG, UK
c
School of Medicine, University of Nottingham, Nottingham NG7 2UH, UK
d
Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China

a r t i c l e i n f o a b s t r a c t

Keywords: Flexibility is a hallmark of human intelligence. Emerging studies have proposed several flexibility measurements
Brain dynamics at the level of individual regions, to produce a brain map of neural flexibility. However, flexibility is usually
Brain function inferred from separate components of brain activity (i.e., intrinsic/task-evoked), and different definitions are
Flexible regions
used. Moreover, recent studies have argued that neural processing may be more than a task-driven and intrinsic
Hubs
dichotomy. Therefore, the understanding to neural flexibility is still incomplete. To address this issue, we propose
Probabilistic view
Neural flexibility a multifaceted definition of neural flexibility according to three key features: broad cognitive engagement, dis-
tributed connectivity, and adaptive connectome dynamics. For these three features, we first review the advances
in computational approaches, their functional relevance, and their potential pitfalls. We then suggest a set of
metrics that can help us assign a flexibility rating to each region. Subsequently, we present an emergent proba-
bilistic view for further understanding the functional operation of individual regions in the unified framework of
intrinsic and task-driven states. Finally, we highlight several areas related to the multifaceted definition of neural
flexibility for future research. This review not only strengthens our understanding of flexible human brain, but
also suggests that the measure of neural flexibility could bridge the gap between understanding intrinsic and
task-driven brain function dynamics.

1. Introduction ies have described the functional flexibility of brain regions based on
how they are engaged in supporting dynamic reorganization of network
A fundamental question in cognitive neuroscience is how the func- structure during cognitive performance (Bassett et al., 2011; Cole et al.,
tional organization of the brain supports flexible and adaptive human 2013; Braun et al., 2015; Spielberg et al., 2015). In contrast, other neu-
behavior (Dehaene et al., 1998; Miller and Cohen, 2001). As functional roimaging studies emphasize that the function of a region is embodied
units, brain areas have been extensively investigated to explain their in its ability to intrinsically and adaptively reshape network configu-
functional roles in cognition and behavior (Misic and Sporns, 2016; rations. In other words, the way in which the intrinsic functional con-
Genon et al., 2018). For example, brain areas are traditionally seen nectivity of a region changes over time may be a characteristic feature
as having specific functional and network profiles (e.g., visual regions of its flexibility (Chen et al., 2016; Yin et al., 2016; Zhang et al., 2016;
or hub nodes). Moreover, the dysfunction of specific regions, including Pedersen et al., 2018; Yin et al., 2020). However, recent studies have ar-
their local activities and internal/external connections, has been demon- gued that neural processing may be more than a task-driven and intrin-
strated to be a biomarker for psychiatric disorders (Li et al., 2020). sic dichotomy, instead reflecting a single functional process (Bolt et al.,
In particular, emerging studies have proposed several flexibility met- 2018; Huk et al., 2018; Uddin, 2020b). Therefore, understanding neu-
rics at the level of individual regions, to produce a brain map of neural ral flexibility requires to link the intrinsic and task-driven brain function
flexibility. Specifically, brain regions that respond selectively to specific dynamics.
kinds of information content are described as functionally specialized, Regardless of task-driven or intrinsic brain dynamics, it should
whereas those that respond to a broad range of task demands are consid- also be noted that anatomical connectivity is the basis of localized
ered functionally diverse (Anderson et al., 2013) or flexible (Yeo et al., brain functions (Bressler and Kelso, 2001; Passingham et al., 2002;
2015). From dynamical frameworks of brain connectivity, some stud- Mars et al., 2018). A critical issue in neuroscience is how the fixed
brain structure gives rise to the dynamics of brain function. Accord-

∗
Correspondence author.
∗∗
Corresponding author at: School of Computing, Newcastle University, Newcastle upon Tyne NE4 5TG, UK.
E-mail addresses: [email protected] (D. Yin), [email protected] (M. Kaiser).

https://doi.org/10.1016/j.neuroimage.2021.118027

Available online 6 April 2021

1053-8119/© 2021 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

Fig. 1. The three key features for a multifaceted definition of neural flexibility: broad cognitive engagement, distributed connectivity, and adaptive connectome
dynamics.

ingly, the fast-growing field of network neuroscience has recently em- states, and both activity and connectivity can be static and dynamic. This
phasized the characterization of dynamical brain organization based operational definition could reconcile different computational methods
on network science tools, computational modeling, and statistical and brain states (i.e., intrinsic/task-driven). The current definition fo-
physics (Deco et al., 2015; Gonzalez-Castillo et al., 2015; Avena- cuses specifically on individual regions not because it is the only way
Koenigsberger et al., 2017; Bassett and Sporns, 2017; Breakspear, 2017; to describe neural flexibility, but because multiple studies with com-
Kaiser, 2017; Thompson et al., 2017; Bassett et al., 2018). Within the mon approaches have been conducted at the level of individual regions.
neuroanatomical network (structural connectivity), the nonlinear dy- In addition, defining neural flexibility at the level of individual regions
namics of neurons and neuronal populations result in patterns of statis- allows to show its heterogeneous profile across whole brain.
tical dependencies (functional connectivity) and causal interactions (ef- In the second section, we mainly discuss the neural flexibility in
fective connectivity), defining three major modalities of complex neural terms of the three features based on brain activation, connectivity pat-
systems (Sporns et al., 2004; Kaiser, 2020). A full understanding of the terns, and connectome dynamics. Specifically, we first review the ad-
neural flexibility should encompass information about structure, func- vances in computational approaches, and highlight their functional rel-
tion, and dynamics. evance and potential pitfalls. We then suggest a set of metrics that can
Considering that flexibility is usually inferred from separate com- help us assign a flexibility rating to each region and discuss the rela-
ponents of brain activity (i.e., intrinsic/task-evoked) and different def- tionship of our current hypotheses on the flexibility of individual re-
initions are used, the understanding to neural flexibility is still incom- gions with other neural processing theories. In the third section, we
plete. For example, answers to the following questions are unclear: Is present an emergent probabilistic view for further understanding the
neural flexibility task-driven or intrinsic, and how do they correspond? functional operation of individual regions in the unified framework of
To what extent are highly flexible regions members of the putative func- intrinsic and task-driven states. We begin with a discussion of brain-wide
tional or structural hubs? In addition, what is the role of regional flex- functional connectivity patterns in the shift between intrinsic and task-
ibility within the brain as a whole? In this review, we therefore pro- evoked states. This is followed by a discussion of the probabilistic views
pose a multifaceted definition of neural flexibility according to the three of regional activity and regional functional connectivity/co-activation
key features: broad cognitive engagement, distributed connectivity, and in the shift between intrinsic and task-driven states. After that, we
adaptive connectome dynamics (Fig. 1). In a formulated form, Flexibil- present a summarized probabilistic view to unify the intrinsic and task-
ity = Variability (activity profiles) ∗ Variability (connectivity profiles), evoked brain activities at the level of individual regions and also sug-
where the flexibility of a region is embodied by the degree its activity gest some potential issues to test. We highlight several areas related to
and connectivity profiles change during both resting and different task the multifaceted definition of neural flexibility for future research in

2
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

the fourth section of the paper and present our conclusions in the fifth facilitate flexible and adaptive behavior. It should be noted that the
section. functional fingerprints of brain regions in these two systems are only
weakly distinct, and the regions in both systems have a diverse func-
2. Understanding neural flexibility from task-driven and intrinsic tional repertoire (Anderson et al., 2013). In addition, some of these cog-
states nitive control regions, such as the lateral frontoparietal and midcingulo-
insular areas, also support cognitive flexibility because they frequently
Considering that flexibility of a region is usually inferred by differ- respond to experimental paradigms that require task switching or set-
ent methods and from separate components of brain activity (i.e., task- shifting (Dajani and Uddin, 2015). Moreover, the activity of these re-
evoked/intrinsic), we review the advances in computational approaches gions is disrupted in various disorders such as autism spectrum disorder,
to describing the three proposed key features for a flexible region in this where the individual often shows stereotyped behavior (inflexibility)
section. We further discuss the extent to which regions with high flexibil- (Uddin, 2020a).
ity are putative functional and structural hubs, and how flexible regions While task-evoked activation approaches can describe the flexibil-
correspond to nodes identified from task-driven and intrinsic states. ity of a region, inferences about a specific functional role may be con-
strained by the chosen tasks. The complete functional repertoire of a
2.1. Task activation approaches region may be out of reach using a limited range of laboratory tasks
given the immense range of human behaviors that occur in real-life.
It is relatively easy to understand functionally specialized regions, However, a sufficiently meaningful characterization of a region could
such as visual and motor areas, which mainly process domain-specific be obtained via the use of naturalistic stimuli or a well thought set of
information. In contrast, functionally flexible regions are more complex, tasks, which should involve various cognitive domains and processes
which may reflect the need to follow diverse task demands. Hence, the (Huk et al., 2018). Additionally, as areas do not work in isolation, con-
profile of brain activation across different cognitive components may nectivity features are also needed to represent the function of brain re-
allow us to describe regional flexibility. gions.
The meta-analysis of aggregated brain activation is a promising ap-
proach for describing the flexibility of a region. For example, a pioneer-
ing study applied a clustering method to brain activation data, finding
2.2. Static functional and structural connectivity
that a substantial portion of the prefrontal cortex (PFC), including the
dorsolateral PFC, ventrolateral PFC, and dorsal anterior cingulate cor-
We now discuss the flexibility of brain regions as inferred from con-
tex (dACC), was recruited across a broad range of cognitive demands
nectivity patterns and graph theory. An important concept for describ-
(Duncan and Owen, 2000). This finding demonstrated that specific pre-
ing the functional role of network nodes is the extent to which they
frontal areas are functionally diverse. In contrast, Passingham et al. pro-
show a hub organization. Reflecting their importance, hubs can be deter-
posed the idea of a functional “fingerprint,” with a brain area’s function
mined through the number of connections of a node (i.e., node degree or
being based on a small set of “dimensions” (i.e., responsive properties)
strength for a weighted network), and through other centrality measures
(Passingham et al., 2002). Following this idea, Anderson et al. quanti-
such as betweenness or participation coefficient (Oldham et al., 2019).
fied functional diversity across brain regions based on large databases of
Hubs can be classified as provincial (intra-cluster) hubs that link ver-
functional brain imaging (Anderson et al., 2013). Brain activations were
tices primarily within a single cluster, or connector (inter-cluster) hubs
classified in terms of task domains, such as vision, attention, and lan-
that link multiple clusters to one another (Guimera and Amaral, 2005). A
guage, and a region’s functional fingerprint was then described with a
third type, bridge hubs, connect multiple modules (Fornito et al., 2015).
dimensional vector (Fig. 2A). The Shannon entropy of the functional fin-
Notably, identifying brain hubs depends in part on their methodologies,
gerprint was further defined as functional diversity. Regions with high
such as connectivity attributes (e.g., functional or structural), graph met-
diversity were identified in the lateral PFC, dorsomedial PFC, lateral
rics (e.g., degree or participant coefficient), and temporal scales (e.g.,
parietal cortex, anterior insula, thalamus, and putamen. Similar ideas
slow or fast) (van den Heuvel and Sporns, 2013b; de Pasquale et al.,
have been adopted to describe the functional diversity of each zone in
2018). Therefore, it is important to elucidate which putative hubs could
the human striatum (Pauli et al., 2016) and thalamus (Hwang et al.,
be interpreted as functionally flexible regions.
2017).
Unlike the traditional meta-analysis requiring prior classification of
task domains, Yeo et al. developed a formal mathematical model allow-
ing identification of cognitive components shared across tasks without 2.2.1. Static functional connectivity during task
predefining task membership, and enabling the derivation of quanti- Using graph theory, functional hubs can be probed during the per-
tative maps of functional specialization and flexibility across the cere- formance of specific tasks or cognitive processes. However, these hubs
bral cortex (Fig. 2B) (Yeo et al., 2015). The authors found that func- are task-dependent and often vary between different tasks or conditions.
tionally flexible regions participating in multiple cognitive components For example, an increase in the number of provincial hubs was found
were mainly located in the anterior insula/frontal operculum, intrapari- during a simple motor task (i.e., segregation increased) compared to
etal sulcus (IPS), dorsomedial PFC, and lateral PFC. Based on different a rest state. In contrast, there was an increase in the number of con-
analytical strategies, the flexible or diverse regions were consistently nector hubs during a working memory task (i.e., integration increased)
identified in the frontal and parietal lobes. (Cohen and D’Esposito, 2016). Additionally, when higher inhibitory
These functionally flexible regions are well recognized as part of a control is needed, the right inferior frontal sulcus and right anterior
top-down cognitive control system providing the signals that configure insula become more centralized network hubs (high nodal strength),
moment-to-moment information processing (Miller and Cohen, 2001; whereas the dorsal anterior cingulate becomes more interconnected
Corbetta and Shulman, 2002; Cole and Schneider, 2007). Considering within its local network (Spielberg et al., 2015). Although these stud-
the different timings needed in task control, two distinct control sys- ies suggest an adaptive reconfiguration of brain networks in response to
tems have been proposed: a frontoparietal system that consists of the different cognitive demands, the flexibility of brain regions is difficult
dorsolateral PFC and IPS mainly carrying start-cue and error-related ac- to deduce from the static, task-specific functional network. Instead, the
tivity (adaptive control), and a cingulo-opercular system that includes intrinsic functional connectivity pattern as measured during the rest-
the dACC/dorsomedial PFC and anterior insula/frontal operculum and ing state appears to constitute a faithful representation of its repertoire
contributes to the stable maintenance of task sets (Dosenbach et al., of evoked, extrinsic functional interactions (Vaidya and Gordon, 2013;
2008). The two control systems work with dissociable mechanisms to Ponce-Alvarez et al., 2015).

3
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

Fig. 2. Brain activation approaches to describing functional flexibility of brain regions. (A) The functional diversity of brain regions was defined as the Shannon
entropy of their functional fingerprints. The functional fingerprints of two representative brain regions are shown (left panel): dorsal anterior cingulate cortex (dACC)
and right anterior insula (AI). Regions with high diversity were primarily identified in the lateral prefrontal cortex (PFC), dorsomedial PFC, lateral parietal cortex,
AI, thalamus, and putamen (right panel, hot color denotes the regions with high functional diversity) (Anderson et al., 2013). (B) A hierarchical Bayesian model
was used to derive quantitative maps of functional specialization and flexibility across the cerebral cortex, in which the estimated model parameters include the
probability that a task would recruit a cognitive component for an activation focus, and the probability that a cognitive component would activate a brain voxel
for an activation focus (left panel). Functionally flexible regions were mainly located in the anterior insula/frontal operculum (aIns/oper), intraparietal sulcus (IPS),
dorsomedial PFC, and lateral PFC (right panel, hot color denotes the regions with high functional flexibility) (Yeo et al., 2015).

2.2.2. Static functional connectivity during rest tention, salience, and frontoparietal networks. These findings suggest
Although functional hubs identified from the resting state may im- that the true integrative core of a network is the diverse rather than
ply a more general role, the interpretation differs substantially with the rich club. It is also possible that the function of the rich club is to
different metrics. For example, based on degree/strength, regions in maintain stability in the entire network at a slow time scale, while the
the default mode network have been identified as functional hubs diverse club acts on a faster time scale to ensure flexibility (Gollo et al.,
from both resting-state functional MRI (fMRI) (Buckner et al., 2009; 2015). In addition, individual differences in the diversity of connector
Cole et al., 2010; Tomasi and Volkow, 2010; Zuo et al., 2012; Liang et al., hub connectivity are predictive of cognitive performance across a range
2013) and magnetoencephalography (MEG) (de Pasquale et al., 2012; of different tasks (Bertolero et al., 2018). Therefore, the participation
Maldjian et al., 2014; Garces et al., 2016). However, Power et al. ar- coefficient may be a promising measure for describing the functional
gued that degree-based hubs might merely be indicative of member- flexibility of brain regions.
ship within large brain systems, rather than playing an important role However, the calculation of the participation coefficient requires
in information processing (Power et al., 2013). Instead of degree, they dividing the brain network into different communities. Most studies
recommend defining hubs by the participation coefficient, which is a have used the modularity maximization approach to detect communi-
measure of how evenly distributed a node’s connectivity is across com- ties (Newman, 2006), with the assumption that each node (brain re-
munities. gion) belongs to a single community. On the other hand, many real-
The use of the participation coefficient suggests that the functional world networks, including biological and social networks, are charac-
hubs are mainly located in the salience and frontoparietal networks, terized by interwoven sets of overlapping communities (Gavin et al.,
rather than in the default mode network as previously cited. Moreover, 2002; Palla et al., 2005) where a node can belong to more than one
the hubs assessed by participation coefficient were in regions where community. For overlapping communities, the bridge hubs mentioned
brain activity is associated with several different cognitive functions above are naturally defined (Nepusz et al., 2008). The intuitive im-
(Bertolero et al., 2015). Bertolero et al. recently further defined two portance of a bridge is its potential to spread signals across multi-
dissociable clubs: rich (high degree) and diverse (high participation co- ple communities, thereby playing a pivotal role in distributed infor-
efficient) (Fig. 3A) (Bertolero et al., 2017). Rich club members were pri- mation processing. The overlapping community structure thus offers
marily located in visual, dorsal attention, and default mode networks, an alternative way to understand the functional flexibility of nodes in
while diverse club members were largely located in ventral/dorsal at- a network.

4
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

Fig. 3. Measures of the diversity of a node’s connectivity pattern. (A) shows diverse and rich clubs identified by the measures of connector (higher participation
coefficient) and provincial (higher degree) hubs. The calculation of participation coefficient requires the division of the brain network into disjoint communities. The
diverse club is mainly involved in memory retrieval, ventral/dorsal attention, salience, and frontoparietal networks (Bertolero et al., 2017). (B) shows that diverse
regions are identified by the overlapping communities approach. The fractions of overlap are greater than 50% for frontoparietal (FPN), cingulo-opercular (CON),
salience (SAN), and subcortical (SUN) networks. Moreover, overlapping regions are significantly more flexible than non-overlapping ones, as shown by a larger Mi ,
the number of overlapping modules in which a node is involved (Lin et al., 2018).

A growing body of evidence from large-scale neuroimaging data community structure not only advances the understanding of function-
shows that both structural and functional brain networks overlap when ally flexible regions, but also provides a link between task-driven and
using analytic methods that allow community overlap. These methods intrinsic brain function.
include independent component analysis (Smith et al., 2012), latent The connectivity profile has created an important method of char-
Dirichlet allocation model (Yeo et al., 2014), maximal-clique based opti- acterizing the functional roles of brain regions. However, most graph
mization (Wu et al., 2011), mixed-membership algorithm (Najafi et al., metrics categorize hubs mainly by a node’s own connectivity profile,
2016), multi-objective evolutionary algorithm (Lin et al., 2018) and not taking into account how their neighbors’ connectivity profile might
non-negative matrix factorization (Aggarwal and Gupta, 2018). In also influence their role or importance. Accordingly, a statistical physics
resting-state functional brain networks, several association regions par- analysis method called k-shell decomposition has been used to assess
ticipate in multiple networks, whereas large portions of early sensory how central a node is in the network with respect to its neighbors with
and late motor cortices participate in single networks (Yeo et al., 2014). a higher k-value signifying a more central node belonging to a more con-
There is also a positive correlation between functional diversity (i.e., the nected neighborhood in the network (Lahav et al., 2016). For a given
extent to which a region activates tasks across domains) and member- network, the k-core is the maximal subgraph consisting of all nodes with
ship diversity (i.e., the extent to which a region has distributed member- at least k neighbors (Seidman, 1983; Arese Lucini et al., 2019). This net-
ship values). This indicates that regions that can be activated by a wide work core is considered an influential spreader, ensuring a more efficient
variety of tasks tend to participate in more communities when at rest spread of information (Kitsak et al., 2010).
(Najafi et al., 2016). Although densely overlapping regions were found In line with the idea that the connectivity profiles of a node’s neigh-
across all functional networks, the fractions of overlap were greater than bors might influence its role, Yin et al. recently presented a novel graph
50% for frontoparietal, cingulo-opercular, salience, and subcortical net- measure, the neighbor dispersion index. This index tests the hypothesis
works (Fig. 3B) (Lin et al., 2018). Moreover, the overlapping regions that the functional diversity of a brain region is embodied by the topo-
were significantly more flexible than non-overlapping regions when logical dissimilarity of its immediate neighbors in the large-scale func-
compared with a functional flexibility map. The concept of overlapping tional brain network (Yin et al., 2019a). The authors found that brain

5
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

regions in the frontoparietal and salience networks showed higher func- conditions (Fig. 4A) (Cole et al., 2013). The authors found that brain-
tional diversity compared to those in visual, auditory, and sensorimo- wide functional connectivity patterns of frontoparietal regions shifted
tor networks. Moreover, human fluid intelligence was associated with more than those of other networks following specific task demands.
the functional diversity of the dorsolateral PFC. To fully understand the Using the dynamical community detection technique, a high flexibil-
topological structure of a node, further research examining the various ity of frontal brain networks has been substantiated during working
topological layers that make up the network, such as the core-periphery memory processes. From these results, the flexibility of a region was
(Bassett et al., 2013) and minimum spanning tree (Stam et al., 2014) defined as the probability that a brain region changed its allegiance
architectures are required. to putative functional modules between any two consecutive time win-
dows (Braun et al., 2015). In addition, such brain flexibility is a poten-
2.2.3. Structural connectivity tial dynamic network marker related to learning (Bassett et al., 2011;
The structural connectivity fingerprint (Passingham et al., 2002) is a Bassett et al., 2015) and a genetic risk for schizophrenia (Braun et al.,
major determinant of an area’s function and is related to the functional 2016). However, an important question is whether task-driven flexibil-
fingerprint (Mars et al., 2018). For example, the structural connectivity ity is intrinsically encoded for the resting state. Moreover, though the
fingerprint may predict the pattern of individual functional activation definition is fine, one should be cautious to explain the neural flexi-
(Saygin et al., 2011; Osher et al., 2016). These empirical findings high- bility inferred from just one specific tasks, because it is probably task-
light a close coupling between the structural connectivity fingerprint dependent. In such a case, focusing on the change of flexibility between
and function. different cognitive processes or between different subject samples might
Based on diffusion MRI and various graph measures, structural hubs be more meaningful. To bridge the task-driven and intrinsic flexibility
have been identified in a set of relatively consistent regions. For exam- of brain organization will improve our understanding of brain function
ple, a large proportion of structural hubs were identified in the default dynamics.
mode, salience, and visual networks using degree (Nijhuis et al., 2013;
van den Heuvel and Sporns, 2013a), betweenness centrality (Gong et al., 2.3.2. Functional connectivity dynamics during rest
2009), participation index (Hagmann et al., 2008), and k-shell decom- In the absence of explicit tasks, the dynamical characteristics of brain
position (Hagmann et al., 2008; Lahav et al., 2016). Under the frame- regions are usually revealed by combining time-resolved and graph the-
work of overlapping community structure, the bridge nodes in structural oretical approaches. For example, resting-state fMRI and sliding-window
brain networks were mainly found in the temporal, visual, ventral pre- analysis identified transient degree-based hubs; the sensorimotor and
frontal cortices, and putamen (Wu et al., 2011). In addition to using sin- auditory areas showed the highest probability of being hubs over time
gle graph metrics, Wang et al. recently adopted a clustering method to (Liao et al., 2015). When imaged at higher temporal resolution with
further classify the structural hubs based on eight widely used network MEG, a set of network hubs (as based on degree and betweenness cen-
metrics (Wang et al., 2018). Although three categories of hubs (i.e., ag- trality) alternates over time in the resting state, mainly including the
gregated, distributed, and connector) were observed, most belonged to posterior cingulate cortex, supplementary motor area, and posterior IPS
default mode, visual, and sensorimotor networks. The accumulated evi- (de Pasquale et al., 2012; de Pasquale et al., 2016). These hubs are tran-
dence indicates that structural hubs largely overlap with the functional siently central and form a dynamic core that largely overlaps with previ-
hubs as identified by degree/strength, but are distinct in functionally ously reported rich clubs, though not with the diverse clubs (de Pasquale
flexible regions. et al., 2018). Therefore, the dynamic core is not sufficiently considered
In the static intrinsic functional network, connector hubs (identified to be functionally flexible, although it may play an important role in
by the participation coefficient) and bridge hubs (identified by overlap- information integration.
ping community structure) correspond well with the regions involved In contrast to investigating the dynamic core using traditional graph
in the performance of various cognitive tasks. The converging evidence theoretical methods, several recently established computational ap-
not only suggests that connector and bridge hubs facilitate flexibility in proaches have been used to characterize the intrinsic functional flexibil-
function, but also offers a link between intrinsic and task-driven brain ity of brain regions in the resting state. Specifically, Yin et al. proposed
functions. However, static task-specific network connectivity is not suf- a probabilistic framework in which the dynamic reconfiguration of in-
ficient for inferring the flexibility of brain regions. The structural hubs trinsic functional connectivity between brain regions was represented
likely provide a stable anatomical substrate for functional brain organi- as a probability distribution (Fig. 4B) (Yin et al., 2016). The Shannon
zation; however, they might not play a crucial role in the flexibility of entropy measurement of complexity was used to quantify regional flex-
brain function. ibility, characterizing the heterogeneous connectivity between a partic-
ular region and others over time. Notably, this measure is nonlinear and
requires setting a threshold for connectivity patterns at each window.
2.3. Functional connectivity dynamics Results indicated that regions showing high flexibility were mainly lo-
cated in the higher-order association cortex, such as the lateral PFC,
Task-based activation and static functional connectivity approaches lateral parietal cortex, lateral temporal lobule, and caudate, whereas
have provided opportunities to infer the functional repertoire of brain visual, auditory, and default mode areas exhibited low flexibility. This
regions. However, early studies paid little attention to the temporal intrinsic functional flexibility map has recently been revealed in non-
characteristics of functional organization. Over the past decade, a grow- human primates (Yin et al., 2019b), suggesting an evolutionary conser-
ing body of research has looked at the dynamical reorganization of brain vation of flexible brain organization. The flexibility of the right lateral
networks in the context of changing task demands (Davison et al., 2015; PFC has also exhibited an inverse U-shaped curve against the human life
Gonzalez-Castillo et al., 2015; Shine et al., 2016; Kucyi et al., 2018; span.
Shine et al., 2019) and resting state (Allen et al., 2014; Baker et al., 2014; Using linear correlation calculations without a threshold, Zhang and
de Pasquale et al., 2016; Vidaurre et al., 2017). Vidaurre et al. provided colleagues defined temporal variability as an index of nodal flexibil-
an insightful review on the topic (Vidaurre et al., 2018). Therefore, in ity, measuring the collective changes of a node’s functional connec-
the following section we focus on describing the flexibility of brain re- tivity profile at different time windows (Zhang et al., 2016). The au-
gions through time-varying functional connectivity. thors consistently found that the heteromodal association areas (e.g.,
frontoparietal and lateral temporal), limbic association areas, and cau-
2.3.1. Functional connectivity dynamics during tasks date showed high variability, while primary sensory cortices and default
An influential study has demonstrated the flexible hub theory mode regions showed low variability. Moreover, the range of variability
through a measure of global variable connectivity across a variety of task in controls was most likely to change for mental disorders. For exam-

6
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

Fig. 4. Measures of nodal flexibility based on task-driven and intrinsic functional connectome dynamics. (A) shows a measure of global variable connectivity (GVC)
across different task conditions, indicating that frontoparietal regions (FPN) are flexible hubs (Cole et al., 2013). (B) shows an intrinsic flexibility map revealed by a
sliding-window method and a complexity measure (entropy). The regions showing high flexibility were consistently in the higher-order association cortex, such as
the lateral prefrontal cortex, lateral parietal cortex, and lateral temporal lobules, whereas visual, auditory, and default mode areas exhibited low flexibility (Yin et al.,
2016).

ple, compared to controls, regions of the default mode network demon- higher-order cognitive functions, including working memory, planning,
strated lower variability in patients with schizophrenia but higher vari- and reasoning. More recently, using a multilayer network approach,
ability in patients with autism/attention deficit hyperactivity disorder. the development of neural flexibility has been reveled in early infancy
In addition to brain disorders, verbal creativity in a healthy population (Yin et al., 2020). Results showed that the flexibility of higher-order
has been correlated with temporal variability of the intrinsic functional and motor regions increased significantly with age, while visual regions
connectivity patterns of the lateral PFC, the precuneus, and the parahip- exhibited a temporally stable pattern with low flexibility.
pocampal gyrus (Sun et al., 2019). From intrinsic brain function dynamics, these studies have revealed
In contrast, some studies have used more complicated methods to relatively consistent profiles of neural flexibility, even with distinct
describe the flexibility of a region. For instance, Chen et al. combined methods, and have also provided new insights into the neural basis of
a sliding-window analysis with a community detection technique to de- development, evolution, and neuropsychiatric disorders. Below, we dis-
fine two dynamical metrics: temporal flexibility (the tendency to inter- cuss the structural basis of neural flexibility.
act with brain regions outside its community) and the spatiotemporal
diversity coefficient (a measurement of the diversity of inter-community 2.3.3. Computational models of functional dynamics around a structural
interactions over time) (Chen et al., 2016). The authors observed that skeleton
regions in the salience, subcortical, and frontoparietal networks showed Considering that structure is the neural basis of function, it is impor-
high temporal flexibility and spatiotemporal diversity, whereas regions tant to understand how functional dynamics are generated from a struc-
in the visual, auditory, and motor networks showed low temporal flex- tural skeleton and thus support flexible brain function. In particular,
ibility and spatiotemporal diversity. Moreover, the temporal flexibility the dynamics of resting-state functional connectivity patterns have been
of a salience network can be a predictor of cognitive flexibility. Based considered a reflection of possible functional network configurations
on brain dynamics and a multilayer network model, Pedersen et al. re- around a stable anatomical skeleton (Deco et al., 2011). To link such
vealed that brain regions, particularly the default mode and frontopari- functional dynamics and structural constraints, several studies have fo-
etal regions, can transition between different network configurations cused on how different time-evolving functional networks emerge spon-
at a high rate (Pedersen et al., 2018). This network switching rate not taneously from a single structural network (Avena-Koenigsberger et al.,
only reflects a region’s flexibility, but also predicts the performance of 2017; Breakspear, 2017; Cabral et al., 2017). Empirical data has indi-

7
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

cated that the brain at wakefulness sequentially explores a rich reper- of regional brain dynamics contributing to flexible functional organiza-
toire of functional configurations, often including ones dissimilar to tion.
anatomical structure. In contrast, the more frequent functional connec- From a dynamical framework, flexible regions have been demon-
tivity patterns follow the structure of anatomical connectivity under strated in task-driven and resting states and in models of structural con-
anesthesia (Barttfeld et al., 2015). These observations suggest that dy- nectome dynamics. The convergent findings suggest that the flexibility
namics of intrinsic network connectivity may reflect a continuous stream of a region is intrinsic, not necessarily driven by extrinsic tasks, and
of ongoing cognitive processes and random fluctuations constrained by that this intrinsic flexibility may enable adaptive reconfigurations of
a fixed anatomical structure. Moreover, recent work (Medaglia et al., brain organization in response to changing cognitive demands. More-
2018) used a novel approach from graph signal processing to distill func- over, computational models can provide a mechanistic insight into how
tional brain signals into aligned and deviating components relative to a stable brain structure gives rise to a flexible functional organization.
the underlying anatomy, suggesting that anatomical networks provide However, there are also inconsistencies in the identified flexible regions,
organization to the flexible brain and facilitate greater mental agility. which may be biased by methodologies such as complexity. For exam-
In addition to the network-level relationships between functional dy- ple, the methods for defining flexibility are nonlinear for several au-
namics and fixed structure, computational models have been used to thors (Chen et al., 2016; Yin et al., 2016; Pedersen et al., 2018), but
unravel the mechanisms of structural hubs, contributing to the flex- linear for others (Cole et al., 2013; Zhang et al., 2016). In addition,
ible exploration of a rich functional repertoire. Under the Kuramoto the most of flexibility measurements at resting state are based on time-
model of structural connectivity, anatomical hubs play a leading role varying functional connectivity. Although dynamic functional connec-
in intermodular synchronization and integration in the human brain tivity at resting state has been widely conducted, debates remain exist
(Schmidt et al., 2015). A model of cascade spreading has been applied (Lurie et al., 2020). For example, to what extent do spontaneous tem-
to explore how the human structural connectome shapes the spread of poral fluctuations of functional connectivity reflect changes in neural
information (Misic et al., 2015). Results showed that structural hubs and network configurations, and what is the functional representation of the
a backbone of core pathways facilitated the early spreading of cascades. measured reconfigurations of functional connectivity? See (Kucyi et al.,
Moreover, polysensory association networks (i.e., the default mode, dor- 2018; Gonzalez-Castillo et al., 2019) for detailed discussions. Neverthe-
sal attention, and frontoparietal networks) were frequently the point of less, describing neural flexibility at intrinsic and task-evoked brain states
convergence for remote signals (high diversity), and initiated the largest simultaneously may facilitate understanding the cognitive and behav-
cascades regardless of where the other perturbation was seeded (high ioral relevance of time-varying fluctuations of intrinsic brain activity.
competitiveness) (Fig. 5A). These observations suggest a structural ba-
sis of the association networks for facilitating effective integration and 2.4. A set of metrics for assigning a flexibility rating to each region
dominating information-spreading dynamics.
The functional role of nodes can also be observed when network The research on the brain map of neural flexibility has attracted
dynamics are changed through external stimulation. Many models use great attention. However, different methods and brain states (i.e., task-
control theory to identify the driver nodes that determine the dynam- evoked/intrinsic) might give distinct flexibility profiles across cerebral
ics of the entire network (Liu et al., 2011). Such models were able to cortex. To reconcile these potential factors, we propose a multifaceted
predict roles of neurons of the roundworm Caenorhabditis elegans, pre- definition of neural flexibility. As an equation, Flexibility = Variability
dictions that were confirmed experimentally after the ablation of the (activity profiles) ∗ Variability (connectivity profiles), where the flexi-
neurons (Yan et al., 2017). In humans, control theory has been used to bility of a region is embodied by the degree its activity and connectivity
envision control hubs whose large-scale regional activity can move the profiles change during both resting and different task states, and both ac-
brain into new state trajectories that traverse diverse cognitive func- tivity and connectivity can be static and dynamic. Variability would best
tions. Gu et al. defined three kinds of control hubs for regional con- be defined as the complexity or dissimilarity of activity/connectivity
trollability, which are differentially located across cognitive systems profiles; therefore, flexibility has a positive value with the lower limit
(Gu et al., 2015). Specifically, average controllability hubs that can steer of zero if either activity, connectivity, or both do not vary.
the system into different states with little input energy are preferentially According to the multifaceted definition, we advise to consider a set
located in the default mode network. Modal controllability hubs that can of metrics in terms of the three key features: (1) broad cognitive engage-
push the brain into difficult-to-reach states requiring substantial energy ment: such as the diversity of functional fingerprint (Anderson et al.,
are predominately located in frontoparietal and cingulo-opercular net- 2013) or the flexibility (Yeo et al., 2015) of a region across a broad
works. Boundary controllability hubs that lie at the boundaries between range of cognitive demands; (2) distributed connectivity: such as the
network communities are distributed throughout all systems, with the membership diversity (overlapping communities) (Najafi et al., 2016),
two predominant systems being ventral and dorsal attention systems participant coefficient (disjoint communities) (Power et al., 2013), or
(Fig. 5B). neighbor dispersion index (Yin et al., 2019a) of a region for resting-state
Moreover, Bansal et al. investigated how regional brain stimula- functional brain networks; and (3) adaptive connectome dynamics: such
tion produces different patterns of synchronization across predefined as the variability of functional connectivity patterns of a region across
cognitive systems, detecting three dynamic states through the stimula- a broad range of cognitive demands (Cole et al., 2013); the temporal
tion of different brain regions: coherent, metastable, and chimera states variability (Zhang et al., 2016), switching rate (Pedersen et al., 2018),
(Bansal et al., 2019). Coherent states (global synchrony) are predomi- entropy (Yin et al., 2016), temporal flexibility (Chen et al., 2016), or al-
nantly produced by regional stimulation within subcortical and medial legiance (Yin et al., 2020) of a region for time-varying resting-state func-
default mode systems, reflecting that many of their constituent regions tional brain networks. Following the equation for flexibility, a combina-
are network hubs. Metastable states (absence of stable synchrony) fre- tion of different potential measurements from these three aspects will
quently occur after stimulation of nodes within cingulo-opercular and give a comprehensive flexibility rating to each region. In addition, com-
frontoparietal systems, possibly reflecting the fact that these systems are putational models, such as a chimera-based framework (Bansal et al.,
flexible and not constrained by their structural connectivity. Finally, 2019), can be used to test the structural basis of neural flexibility.
chimera states (with coexisting synchrony and desynchrony) emerge
upon the stimulation of nodes that are equally distributed across all 2.5. Relationship between our hypothesis describing the flexibility of
the cognitive systems, suggesting a complexity and flexibility in the individual regions and other neural processing theories
brain’s underlying architecture to support diverse processing require-
ments (Fig. 5C). These studies not only indicate that the structural net- Several theories and hypotheses have been proposed to describe how
work may shape neural communication, but also suggest a mechanism the brain is organized and operates functionally, but most focus on

8
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

Fig. 5. Computational models for structural connectome dynamics giving rise to flexible functional organization. (A) shows how two simultaneous perturbations
(indicated by arrows) develop into competing cascades. In a model of cascade spreading, the polysensory association networks (the default mode, dorsal attention,
and frontoparietal networks) were frequently the point of convergence for remote signals (high diversity), and initiated the largest cascades regardless of where the
other perturbation is seeded (high competitiveness) (Misic et al., 2015). (B) Control theory identifies control nodes whose large-scale regional activity can move the
brain into new trajectories that traverse diverse cognitive functions. A toy example (left panel) shows trajectories of state movement with varying amounts of energy
cost (indicated by the height of the purple bars). Hubs of average controllability are preferentially located in the default mode system; hubs of modal controllability
are predominantly located in cognitive control systems, including both the frontoparietal and cingulo-opercular systems; and hubs of boundary controllability are
distributed throughout all systems, with the two predominant systems being ventral and dorsal attention systems (right panel) (Gu et al., 2015). (C) Based on a
structural network, each brain region was computationally stimulated (left panel) and the spread of the stimulation measured through synchronization within the
brain network. Three dynamic states were observed through the stimulation of different brain regions (right panel): coherent states were produced predominantly by
regional stimulation within subcortical and medial default mode systems; metastable states frequently occurred after stimulation of nodes within cingulo-opercular
and frontoparietal systems; and chimera states emerged upon the stimulation of nodes that are equally distributed across all the cognitive systems (Bansal et al.,
2019).

whole-brain activity patterns rather than from the point-of-view of in- tiple functions (integration) (Tognoli and Kelso, 2014). The consistent
dividual regions. For instance, the neural reuse theory supposes that idea in our hypothesis is that a region exerts its potential functions by
different cognitive functions are supported by putting many of the same adaptively reshaping its connectivity/co-activation patterns. Although
neural circuits or co-activation patterns together in different arrange- neural reuse and coordination dynamics theories suggest how brain re-
ments (Anderson, 2010, 2014). Neural reuse, one of the network ap- gions are used and reused for various purposes in different cognitive
proaches to understanding the operation of the brain, emphasizes co- domains, the distinct roles of the individual regions are not well under-
operative interactions and a non-modular, many-to-many relationship stood.
between anatomical regions and complex cognitive functions and be- In contrast, the global neuronal workspace theory proposes two
haviors. Moreover, in response to adaptive cognition, the theory of coor- main computational spaces involved in brain processes: a unique global
dination dynamics proposes a metastable brain, where tendencies for re- workspace composed of distributed and heavily interconnected neurons
gions to express their specialized functions and autonomy (segregation) with long-range axons, and a set of specialized and modular perceptual,
coexist with their tendencies to couple and coordinate globally for mul- motor, memory, evaluative, and attentional processors (Dehaene et al.,

9
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

1998). This theory supposes the coexistence of functionally integrated The intrinsic functional networks are essentially not static, but rather
and segregated subsystems. The integrated system (i.e., workspace neu- show a vital interplay of time and space. Deco et al. suggest critical-
rons) is spontaneously co-activated, forming variable spatiotemporal ac- ity as a principle of functional brain organization (Deco et al., 2013).
tivity patterns during the performance of effortful tasks. Similar to the Specifically, the intrinsic functional networks are continuously pulled
global workspace theory, the dynamic core theory suggests a unified towards multiple configurations, reflecting the range of available func-
neural process with a functional cluster emphasizing both its integration tional states. Moreover, the intrinsic functional network dynamics are
and constantly changing activity patterns (Tononi and Edelman, 1998). the gateway to the array of cognitive architectures that the brain has
Although both theories indicate a subset of interactive brain regions available, rather than the embodiment of specific cognitive or behav-
supporting adaptive functional integration, they do not encompass a ioral operations. Accordingly, Mantini et al. proposed that intrinsic brain
unique, invariant set of brain areas that can change in composition over networks represent a finite set of spatiotemporal basis functions from
time. Instead, our hypothesis mainly focuses on the operational char- which task-dependent networks are dynamically assembled and mod-
acteristics of individual regions. The consistency is that functional in- ulated during different behavioral states (Mantini et al., 2007). Addi-
tegration and specialization can be described through the flexibility of tionally, Sadaghiani and Kleinschmidt have proposed that spontaneous
brain regions in our hypothesis. Moreover, our hypothesis is suitable brain activity dynamics, representing an internal model, explore the
for describing both intrinsic and task-driven brain function dynamics. In space of possible brain states along trajectories that are similar to those
contrast, regions that are able to shift between intrinsic and task-driven induced by different task contexts (Sadaghiani and Kleinschmidt, 2013).
states are not highlighted in current workspace-centered theories. The In contrast, recent studies argued that brain function is beyond the sep-
subsequent section mainly focuses on understanding functional oper- aration of intrinsic and task-evoked components and should be repre-
ation of individual regions in the unified framework of intrinsic and sented as a single functional process (Bolt et al., 2018; Uddin, 2020b).
task-evoked states. The supporting evidence for this proposition is that intrinsic and task-
evoked activities show similar spatiotemporal dynamics, and the classic
3. An emergent probabilistic view for understanding functional linear superposition principle in support of the dichotomy of intrinsic
operation of individual regions in the unified framework of and task-evoked activity is challenged.
intrinsic and task-evoked states Considering the spatiotemporal similarity of brain-wide connectiv-
ity patterns, researchers have tried to understand the intrinsic and task-
With the development of cognitive neuroscience, brain function has evoked brain function in a unified framework. However, the empirical
been described from different perspectives, such as task-driven or in- models for linking intrinsic and task-evoked brain activities are still lack-
trinsic, regional or networked, and static or dynamic. However, a better ing at the whole-brain level. Instead, this modeling has emerged at the
understanding of the functional operation of the brain in a unified frame- level of specific regions, such as using a Bayesian model (Berkes et al.,
work is required. Importantly, strong behavioral and neurophysiological 2011).
evidence indicates that the brain works in a probabilistic manner, with
a representation of probabilistic distributions and performance of prob-
abilistic inference (Pouget et al., 2013; Meyniel et al., 2015; Hou et al., 3.2. Probabilistic view of regional activity in the transition between
2019). In neuroscience, probabilistic distributions are usually expressed intrinsic and task-driven states
by encoding probability distributions of neurons and the implementa-
tion of Bayesian decisions (Beck et al., 2008). However, most existing The function of individual regions has been explained from a prob-
probabilistic models are based on stimulus-evoked neural responses. abilistic framework (Pessoa, 2014). For example, the activation profile
Spontaneous brain activity has rarely been demonstrated in the context of a region across a wide range of cognitive processes may represent the
of probabilistic neural coding, although it is known to play a crucial probabilistic distribution of neural coding (Anderson et al., 2013). In
role in cortical computation (Ringach, 2009). Therefore, a probabilistic contrast, using Bayesian analysis, the engagement of a particular cog-
view linking spontaneous and task-evoked neural activities may offer a nitive process can be inferred from the activation of a particular brain
comprehensive understanding to functional operation of the brain. In region, called reverse inference (Poldrack, 2006). Moreover, a hierarchi-
the following sections, we first discuss brain-wide connectivity patterns cal Bayesian model enables the derivation of quantitative maps of func-
in the shift between intrinsic and task-evoked states, and then discuss tional specialization and flexibility across the cerebral cortex (Yeo et al.,
probabilistic views of regional activity and functional connectivity/co- 2015).
activation in the shift between intrinsic and task-evoked states. Importantly, similar patterns of regional activity are frequently ob-
served between intrinsic and task-evoked states (Sadaghiani and Klein-
3.1. Brain-wide connectivity patterns in the shift between intrinsic and schmidt, 2013; Bolt et al., 2018). For instance, evidence at the local level
task-evoked states showed that spontaneous brain activity encompasses a set of dynami-
cally switching cortical states (instantaneous activity patterns), several
Brain function is commonly thought to consist of separable intrinsic of which correspond closely to the functional maps induced by spe-
and task-driven components. However, at the whole-brain level, sub- cific stimuli (Kenet et al., 2003). Accordingly, Fiser et al. proposed that
stantial similarities in network structure are observed during rest and stimulus-evoked and spontaneous brain activity represent samples of a
during the performance of different cognitive tasks (Smith et al., 2009; posterior and a priori distributions, respectively (Fiser et al., 2010). In
Mennes et al., 2013; Cole et al., 2014; Krienen et al., 2014; Davison et al., the absence of external stimuli, sampling from the prior distribution
2015), although some differences were noticed. Furthermore, plausible can help drive the brain to states that are probable as valid inferences
inferences on functional operation shifting between intrinsic and task- once input arrives, thus playing a functional role of prediction or expec-
evoked states have also been suggested. For instance, one view indicates tation. Using a Bayesian model of sensory cortical processing, Berkes
that reconfigurable task-dependent functional network architecture is et al. related stimulus-evoked and spontaneous neural activities to pos-
shaped primarily by an intrinsic network architecture, and secondarily terior and prior distributions in an internal model, predicting that they
by evoked task-general and task-specific network changes (Cole et al., would match if the model is statistically optimal (Berkes et al., 2011).
2014; Krienen et al., 2014). This view actually relies on the hypothe- To test this prediction, the authors analyzed visual cortical activity of
sis that intrinsic network architecture is static and provides a structural awake ferrets during development, and demonstrated the progressive
constraint that acts as an attractor, pulling network configurations to- adaptation of internal models to the statistics of natural stimuli at the
wards a limited number of functional states (Fox and Raichle, 2007; neural level. These empirical studies suggest that intrinsic activity pat-
Krienen et al., 2014). terns of an area express its function in a probabilistic manner, and that

10
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

the stimulus-evoked activity patterns are sampled from a distribution of 3.4. An emergent probabilistic view for understanding functional operation
intrinsic activity patterns. of individual regions

3.3. Probabilistic view of regional functional connectivity/co-activation in At the level of individual regions, the accumulative findings indicate
the transition between intrinsic and task-driven states that specific functional characteristics (e.g., diversity and flexibility) can
be consistently derived from intrinsic and task-evoked states in a prob-
Beyond probabilistic regional activity, the Bayesian approach entails abilistic framework. Moreover, empirical studies demonstrated that in-
an understanding of connectivity and, by implication, functional inte- trinsic activity patterns of a region express its function in a probabilistic
gration in the brain (Friston, 2012). In this case, brain regions may op- manner, and that the stimulus-evoked activity patterns are sampled from
erate in a probabilistic way through their functional connectivity. From a distribution of intrinsic activity patterns. In summarizing the emergent
the perspective of a static functional network, the connectivity profile evidence, we present a probabilistic view for understanding functional
of a node across different disjoint communities can be represented as a operation of individual regions in the unified framework of intrinsic and
probability-like distribution (Power et al., 2013). Recently, individual task-evoked states. In essence, the functional repertoire of each region
differences in the connectivity distribution of connector hubs have been in the brain can be represented by an intrinsic probabilistic distribution
demonstrated to be predictive of cognitive performance across differ- that embodies its constantly changing functional connectivity patterns
ent tasks (Bertolero et al., 2018). Similarly, the membership values of (or regional activity patterns) at rest, and where the specific cognitive
a node in overlapping community structures can also be expressed as a processes are supported by adaptively sampling connectivity configura-
probability-like distribution (Najafi et al., 2016). At the regional level, tions (or regional activity patterns) from the intrinsic probabilistic dis-
the diversity derived from these distributions is correlated with func- tribution (Fig. 6). In analogy to the concept of quantum mechanics, the
tional diversity as quantified by the extent to which a region activates for intrinsic function of a region is uncertain and expressed in a probabilis-
tasks that cross domains. The consistent functional characteristics (i.e., tic manner; its apparent function is detectable and deterministic only
diversity) of brain regions are inferred from intrinsic functional connec- when specific cognitive tasks or stimuli are performed.
tivity and task-evoked functional profiles in a probabilistic framework, Although this probabilistic view enables to explain the functional op-
suggesting a similar functional operation of a region between intrinsic eration of individual regions in the unified framework of intrinsic and
and task-driven states. In particular, recent work has revealed that the task-evoked states, empirical or modeling research are still needed. For
activation of a region during task performance is predicted and modeled example, Bayesian and other models dedicated to inferring task-specific
based on its intrinsic functional connectivity pattern (Cole et al., 2016; connectivity patterns from the probabilistic distribution of intrinsic
Tavor et al., 2016; Ito et al., 2017), see (Ito et al., 2020) for a review. functional connectivity need to be validated in future studies. Moreover,
This finding demonstrates that intrinsic functional connectivity plays a the probabilistic distribution of intrinsic functional connectivity reflects
role in shaping task-evoked activity flow among brain regions, further a priori information of an individual’s knowledge or experience and can
suggesting a mechanistic coupling between intrinsic and task-evoked be shaped by development, learning, and neuropsychiatric disorders
neural activities. (Edelman, 1993; Anderson, 2010; Tambini and Davachi, 2019). Models
However, the intrinsic functional connectivity pattern of a region as that use reshaped probabilistic distribution of intrinsic functional con-
mentioned in the above section is static. Recent studies have emphasized nectivity to predict task-specific connectivity patterns may show better
that dynamical connectivity patterns may provide more information. generalizability.
For example, the intrinsic functional connectivity pattern of a region is In addition, distinct from the traditional studies concerning
decomposed into different spontaneous co-activation maps with distinct structure-function coupling at the level of the entire brain, recent work
proportions or probabilities (Liu and Duyn, 2013; Karahanoglu and Van demonstrated that the relationships between structural and functional
De Ville, 2015). Yin et al. have also shown that the intrinsic functional connection profiles of individual regions varied markedly across the
connectivity pattern largely represents the connections that emerge over cerebral cortex (Vazquez-Rodriguez et al., 2019). Specifically, structure
time with high probability (Yin et al., 2016). This indicates that an ap- and function are closely aligned in unimodal cortex, but diverge in trans-
parent intrinsic functional connectivity pattern is likely a statistical av- modal cortex. So, it might also be necessary to perform the prediction
erage of dynamic brain co-activation or connectivity. between intrinsic and task-evoked brain states using different models at
Following the dynamical network framework, the probabilistic op- different regions.
eration of a region has been described in both task-evoked and intrinsic
states. For instance, the probability that a region changes its allegiance 4. Areas related to the multifaceted definition of neural flexibility
to modules between time windows while performing a task is defined as for future research
regional flexibility, and the probability can predict the amount of learn-
ing (Bassett et al., 2011) and executive function (Braun et al., 2015). In 4.1. Exploring neural flexibility with a multifaceted definition
contrast, an intrinsic probabilistic distribution reflecting the frequency
of connection between a region and the other brain regions across tem- Although neural flexibility has been widely investigated in develop-
poral windows at rest can also describe flexibility (Yin et al., 2016). ment, evolution, learning, and brain disorders, a systematic comparison
Moreover, the flexible regions identified from this distribution are con- between different methods and between brain states (i.e., intrinsic/task-
sistent with those identified by meta-analysis of aggregated brain activa- evoked) is still lacking. For example, both linear (Zhang et al., 2016)
tion. In addition, the flexibility of a region is consistently revealed by the and nonlinear (Yin et al., 2016) methods have been used. In particular,
variability of dynamic connectivity patterns at rest (Zhang et al., 2016) neural flexibility is usually studied based on either intrinsic (such as for
and across a wide range of different task demands (Cole et al., 2013). creativity (Sun et al., 2019)) or task-evoked (such as for skill learning
The convergent findings indicate that functional characteristics (i.e., (Bassett et al., 2011)) states. Few studies consider neural flexibility in
flexibility) of individual regions can be consistently derived from in- the unified framework of intrinsic and task-evoked brain dynamics. Ac-
trinsic and task-evoked dynamical connectivity in a probabilistic frame- cording to the currently proposed multifaceted definition, it is important
work. In other words, the functional repertoire of a region can be repre- to obtain a comprehensive brain map of neural flexibility. Simultane-
sented equally by its changing intrinsic functional connectivity patterns ously, this allows to identify a set of most sensitive and robust measure-
(prior probabilistic distribution) and task-evoked functional connectiv- ments to describe neural flexibility. Moreover, if effective metrics can
ity patterns across diverse tasks (posterior probabilistic distribution). be identified when only resting-state data is available, it will help map
The prior and posterior probabilistic distributions will match when myr- neural flexibility for the subjects (e.g., infant and severe patients) who
iad tasks are recruited. cannot perform experimental tasks.

11
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

Fig. 6. An emergent probabilistic view for understanding functional operation of individual regions in the unified framework of intrinsic and task-evoked states.
Specifically, the functional repertoire of each region in the brain can be represented by an intrinsic probabilistic distribution that embodies its constantly changing
functional connectivity patterns (or regional activity patterns) at rest. The specific cognitive processes are supported by adaptively sampling connectivity configura-
tions (or regional activity patterns) from the intrinsic probabilistic distribution. This intrinsic probabilistic distribution likely reflects an a priori constraint of brain
function, and may have been shaped by evolution, development, learning, and diseases.

Importantly, under rest and task conditions, emerging studies have posed an approach that includes an orthogonal weight modification al-
modeled the regional neural activity in terms of its trajectory in a mul- gorithm and a context-dependent processing module, enabling a descrip-
tidimensional space (Uddin, 2020b). This multidimensional space may tion of a neural network to progressively learn various mapping rules in
provide an opportunity for developing new measurements of neural flex- a context-dependent manner (Zeng et al., 2019). Moreover, Bouchacourt
ibility in the unified framework of intrinsic and task-evoked states. and Buschman presented a flexible model of working memory that main-
tains representations through random recurrent connections between
4.2. Exploring cognitive and pathophysiological mechanisms based on two layers of neurons: a structured ‘‘sensory’’ layer and a randomly con-
multifaceted-defined neural flexibility nected, unstructured layer (Fig. 7) (Bouchacourt and Buschman, 2019).
The introduction of random connections in this model allows a flexi-
Previous studies have demonstrated that specific brain regions play ble maintenance of any input and can explain the limited capacity of
a crucial role in both normal cognition and pathology. For instance, working memory. The functions of the two layers seem to correspond to
both fMRI (Schuck and Niv, 2019) and MEG (Wimmer et al., 2020) re- the flexible (e.g., frontoparietal network) and stable (sensory networks)
vealed that offline replay of activity patterns in the human hippocampus systems demonstrated in the brain. Moreover, the spontaneous brain
supports memory-mediated decisions. In contrast, combining striatal re- activity may provide the possibility for generating random connections.
gional signals and its connectivity were found to serve as neuroimaging Therefore, the neural flexibility of the large-scale brain organization,
biomarker for personalized schizophrenia identification (Li et al., 2020). particularly described from a multifaceted definition, could be a valu-
Fortunately, our proposed multifaceted definition of neural flexibility able framework for designing intelligent algorithms in the future.
inherently encompasses both internal properties and external connec-
tions of individual regions. Therefore, focusing on neural flexibility in a 5. Conclusions
combination of internal properties and external connections may pro-
vide new insights into understanding brain functional operation and The brain is a complex system with intricate spatiotemporal inter-
identifying biomarkers of neuropsychiatric disorders. Certainly, extract- actions between brain regions in both intrinsic and task-evoked states.
ing the internal and external features from complementary techniques Different methods may only capture specific functional characteristics
such as fMRI and MEG is necessary in some situations. of the brain. Although activation approaches can offer an intuitive un-
derstanding of the functional repertoire of an area, understanding of
4.3. Exploring neural flexibility-inspired intelligent algorithms its operational function may be lacking. In contrast, characterizing con-
nectivity patterns or connectome dynamics can provide mechanistic in-
Flexibility is a key hallmark of high-level intelligence. This abil- sights into an area’s function, but these insights often depend on the
ity enables both human and non-human primates to respond differ- methods used and the brain states involved. Moreover, the neural pro-
ently to identical stimuli under different contexts, for example, different cessing is usually studied based on separate components of brain activity
goals, environments, and internal states (Desimone and Duncan, 1995; (i.e., intrinsic/task-evoked). All these factors impede our understanding
Mante et al., 2013). However, most current artificial deep neural net- of brain function.
works (DNNs) exhibit stereotyped input-output mappings that are usu- To better understand neural flexibility, this review proposes a mul-
ally fixed once training is complete. To bridge the gap in the flexibil- tifaceted definition in terms of the three key features: broad cognitive
ity between current DNNs and primate brains, Zeng et al. recently pro- engagement, distributed connectivity, and adaptive connectome dynam-

12
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

Fig. 7. A flexible model of working memory with a structured sensory network and a random network. The introduction of random connections in this model allows
for a flexible maintenance of any input and can explain the limited capacity of working memory. (A) shows a model layout with a sensory network composed of eight
ring-like sub-networks. The inset shows center-surround connectivity within a sensory sub-network. The connections to the random network are randomly assigned
and balanced. (B) shows a raster plot of an example trial with eight sensory sub-networks (512 neurons each) randomly connected to the same random network
(1,024 neurons). Six sensory sub-networks (1–6) received a Gaussian input for 0.1 s during the ‘‘stimulus presentation’’ period (shaded blue region) (Bouchacourt and
Buschman, 2019).

ics. This operational definition could reconcile different computational Acknowledgements

methods and brain states (i.e., intrinsic/task-evoked). Furthermore, the
accumulative findings show consistent functional characteristics (e.g., Dazhi Yin was supported by National Key Research and Develop-
diversity and flexibility) of individual regions between intrinsic and ment Program of China (2017YFC1310400), National Natural Science
task-evoked states in a probabilistic framework. In summarizing the Foundation of China (31600869, 81527901); Marcus Kaiser was sup-
emergent evidence, we have presented a probabilistic view for under- ported by Wellcome Trust (102037), Engineering and Physical Sciences
standing the functional operation of individual regions in the unified Research Council (NS/A000026/1, EP/N031962/1), Medical Research
framework of intrinsic and task-evoked states. This work not only con- Council (MR/T004347/1), and the Guangci Professorship Program of
solidates our understanding of flexible brain organization, but also sug- Ruijin Hospital (Shanghai Jiao Tong Univ.).
gests that the measure of neural flexibility could bridge the gap between
understanding intrinsic and task-driven brain function. Further, explor- References
ing neural flexibility from a multifaceted definition may provide new
insights into cognitive flexibility, pathophysiology of neuropsychiatric Aggarwal, P., Gupta, A., 2018. Low rank and sparsity constrained method for identifying
overlapping functional brain networks. PLoS One 13, e0208068. doi:10.1371/jour-
disorders, and inspire future artificial intelligence research.
nal.pone.0208068.
Allen, E.A., Damaraju, E., Plis, S.M., Erhardt, E.B., Eichele, T., Calhoun, V.D., 2014. Track-
ing whole-brain connectivity dynamics in the resting state. Cereb. Cortex 24, 663–676.
doi:10.1093/cercor/bhs352.
Declaration of Competing Interest Anderson, M.L., 2010. Neural reuse: a fundamental organizational principle of the brain.
Behav. Brain Sci. 33, 245–266. doi:10.1017/S0140525X10000853.
Anderson, M.L., 2014. After Phrenology: Neural Reuse and the Interactive Brain. MIT
The authors have declared that no competing interests exist. Press.
Anderson, M.L., Kinnison, J., Pessoa, L., 2013. Describing functional di-
Data for reference versity of brain regions and brain networks. NeuroImage 73, 50–58.
doi:10.1016/j.neuroimage.2013.01.071.
Arese Lucini, F., Del Ferraro, G., Sigman, M., Makse, H.A., 2019. How the brain
This is a Review Article. All the data and sources mentioned in this transitions from conscious to subliminal perception. Neuroscience 411, 280–290.
paper were cited and stated in the corresponding positions. doi:10.1016/j.neuroscience.2019.03.047.

13
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

Avena-Koenigsberger, A., Misic, B., Sporns, O., 2017. Communication dynamics in com- Dajani, D.R., Uddin, L.Q., 2015. Demystifying cognitive flexibility: Implications
plex brain networks. Nat. Rev. Neurosci. 19, 17–33. doi:10.1038/nrn.2017.149. for clinical and developmental neuroscience. Trends Neurosci. 38, 571–578.
Baker, A.P., Brookes, M.J., Rezek, I.A., Smith, S.M., Behrens, T., Probert Smith, P.J., Wool- doi:10.1016/j.tins.2015.07.003.
rich, M., 2014. Fast transient networks in spontaneous human brain activity. Elife 3, Davison, E.N., Schlesinger, K.J., Bassett, D.S., Lynall, M.E., Miller, M.B., Grafton, S.T.,
e01867. doi:10.7554/eLife.01867. Carlson, J.M., 2015. Brain network adaptability across task states. PLoS Comput. Biol.
Bansal, K., Garcia, J.O., Tompson, S.H., Verstynen, T., Vettel, J.M., Muldoon, S.F., 11, e1004029. doi:10.1371/journal.pcbi.1004029.
2019. Cognitive chimera states in human brain networks. Sci. Adv. 5, eaau8535. de Pasquale, F., Corbetta, M., Betti, V., Della Penna, S., 2018. Cortical cores in network
doi:10.1126/sciadv.aau8535. dynamics. NeuroImage 180, 370–382. doi:10.1016/j.neuroimage.2017.09.063.
Barttfeld, P., Uhrig, L., Sitt, J.D., Sigman, M., Jarraya, B., Dehaene, S., 2015. Signature de Pasquale, F., Della Penna, S., Sporns, O., Romani, G.L., Corbetta, M., 2016. A dynamic
of consciousness in the dynamics of resting-state brain activity. Proc. Natl. Acad. Sci. core network and global efficiency in the resting human brain. Cereb. Cortex 26,
USA. 112, 887–892. doi:10.1073/pnas.1418031112. 4015–4033. doi:10.1093/cercor/bhv185.
Bassett, D.S., Sporns, O., 2017. Network neuroscience. Nat. Neurosci. 20, 353–364. de Pasquale, F., Della Penna, S., Snyder, A.Z., Marzetti, L., Pizzella, V., Romani, G.L.,
doi:10.1038/nn.4502. Corbetta, M., 2012. A cortical core for dynamic integration of functional net-
Bassett, D.S., Zurn, P., Gold, J.I., 2018. On the nature and use of models in network neu- works in the resting human brain. Neuron 74, 753–764. doi:10.1016/j.neuron.2012.
roscience. Nat. Rev. Neurosci. 19, 566–578. doi:10.1038/s41583-018-0038-8. 03.031.
Bassett, D.S., Yang, M., Wymbs, N.F., Grafton, S.T., 2015. Learning-induced autonomy of Deco, G., Jirsa, V.K., McIntosh, A.R., 2011. Emerging concepts for the dynamical or-
sensorimotor systems. Nat. Neurosci. 18, 744–751. doi:10.1038/nn.3993. ganization of resting-state activity in the brain. Nat. Rev. Neurosci. 12, 43–56.
Bassett, D.S., Wymbs, N.F., Porter, M.A., Mucha, P.J., Carlson, J.M., Grafton, S.T., 2011. doi:10.1038/nrn2961.
Dynamic reconfiguration of human brain networks during learning. Proc. Natl. Acad. Deco, G., Jirsa, V.K., McIntosh, A.R., 2013. Resting brains never rest: computa-
Sci. USA 108, 7641–7646. doi:10.1073/pnas.1018985108. tional insights into potential cognitive architectures. Trends Neurosci. 36, 268–274.
Bassett, D.S., Wymbs, N.F., Rombach, M.P., Porter, M.A., Mucha, P.J., Grafton, S.T., 2013. doi:10.1016/j.tins.2013.03.001.
Task-based core-periphery organization of human brain dynamics. PLoS Comput. Biol. Deco, G., Tononi, G., Boly, M., Kringelbach, M.L., 2015. Rethinking segregation and in-
9, e1003171. doi:10.1371/journal.pcbi.1003171. tegration: contributions of whole-brain modelling. Nat. Rev. Neurosci. 16, 430–439.
Beck, J.M., Ma, W.J., Kiani, R., Hanks, T., Churchland, A.K., Roitman, J., Shadlen, M.N., doi:10.1038/nrn3963.
Latham, P.E., Pouget, A., 2008. Probabilistic population codes for bayesian decision Dehaene, S., Kerszberg, M., Changeux, J.P., 1998. A neuronal model of a global
making. Neuron 60, 1142–1152. doi:10.1016/j.neuron.2008.09.021. workspace in effortful cognitive tasks. Proc. Natl. Acad. Sci. USA 95, 14529–14534.
Berkes, P., Orban, G., Lengyel, M., Fiser, J., 2011. Spontaneous cortical activity reveals doi:10.1073/pnas.95.24.14529.
hallmarks of an optimal internal model of the environment. Science 331, 83–87. Desimone, R., Duncan, J., 1995. Neural mechanisms of selective visual attention. Annu.
doi:10.1126/science.1195870. Rev. Neurosci. 18, 193–222. doi:10.1146/annurev.ne.18.030195.001205.
Bertolero, M.A., Yeo, B.T., D’Esposito, M., 2015. The modular and integrative functional Dosenbach, N.U., Fair, D.A., Cohen, A.L., Schlaggar, B.L., Petersen, S.E., 2008. A
architecture of the human brain. Proc. Natl. Acad. Sci. USA 112, E6798–E6807. dual-networks architecture of top-down control. Trends Cognit. Sci. 12, 99–105.
doi:10.1073/pnas.1510619112. doi:10.1016/j.tics.2008.01.001.
Bertolero, M.A., Yeo, B.T.T., D’Esposito, M., 2017. The diverse club. Nat. Commun. 8, Duncan, J., Owen, A.M., 2000. Common regions of the human frontal lobe
1277. doi:10.1038/s41467-017-01189-w. recruited by diverse cognitive demands. Trends Neurosci. 23, 475–483.
Bertolero, M.A., Yeo, B.T.T., Bassett, D.S., D’Esposito, M., 2018. A mechanistic model doi:10.1016/s0166-2236(00)01633-7.
of connector hubs, modularity and cognition. Nat. Hum. Behav. 2, 765–777. Edelman, G.M., 1993. Neural darwinism: Selection and reentrant signaling in higher brain
doi:10.1038/s41562-018-0420-6. function. Neuron 10, 115–125. doi:10.1016/0896-6273(93)90304-a.
Bolt, T., Anderson, M.L., Uddin, L.Q., 2018. Beyond the evoked/intrinsic neural process Fiser, J., Berkes, P., Orban, G., Lengyel, M., 2010. Statistically optimal perception and
dichotomy. Netw. Neurosci. 2, 1–22. doi:10.1162/NETN_a_00028. learning: From behavior to neural representations. Trends Cognit. Sci. 14, 119–130.
Bouchacourt, F., Buschman, T.J., 2019. A flexible model of working memory. Neuron 103, doi:10.1016/j.tics.2010.01.003.
147–160. doi:10.1016/j.neuron.2019.04.020, e148. Fornito, A., Zalesky, A., Breakspear, M., 2015. The connectomics of brain disorders. Nat.
Braun, U., Schafer, A., Walter, H., Erk, S., Romanczuk-Seiferth, N., Haddad, L., Rev. Neurosci. 16, 159–172. doi:10.1038/nrn3901.
Schweiger, J.I., Grimm, O., Heinz, A., Tost, H., Meyer-Lindenberg, A., Bas- Fox, M.D., Raichle, M.E., 2007. Spontaneous fluctuations in brain activity observed
sett, D.S., 2015. Dynamic reconfiguration of frontal brain networks during ex- with functional magnetic resonance imaging. Nat. Rev. Neurosci. 8, 700–711.
ecutive cognition in humans. Proc. Natl. Acad. Sci. USA 112, 11678–11683. doi:10.1038/nrn2201.
doi:10.1073/pnas.1422487112. Friston, K., 2012. The history of the future of the bayesian brain. NeuroImage 62, 1230–
Braun, U., Schafer, A., Bassett, D.S., Rausch, F., Schweiger, J.I., Bilek, E., Erk, S., 1233. doi:10.1016/j.neuroimage.2011.10.004.
Romanczuk-Seiferth, N., Grimm, O., Geiger, L.S., Haddad, L., Otto, K., Mohnke, S., Garces, P., Pereda, E., Hernandez-Tamames, J.A., Del-Pozo, F., Maestu, F., Pineda-
Heinz, A., Zink, M., Walter, H., Schwarz, E., Meyer-Lindenberg, A., Tost, H., 2016. Dy- Pardo, J.A., 2016. Multimodal description of whole brain connectivity: a com-
namic brain network reconfiguration as a potential schizophrenia genetic risk mech- parison of resting state meg, fmri, and dwi. Hum. Brain Mapp. 37, 20–34.
anism modulated by nmda receptor function. Proc. Natl. Acad. Sci. USA 113, 12568– doi:10.1002/hbm.22995.
12573. doi:10.1073/pnas.1608819113. Gavin, A.C., et al., 2002. Functional organization of the yeast proteome by systematic
Breakspear, M., 2017. Dynamic models of large-scale brain activity. Nat. Neurosci. 20, analysis of protein complexes. Nature 415, 141–147. doi:10.1038/415141a.
340–352. doi:10.1038/nn.4497. Genon, S., Reid, A., Langner, R., Amunts, K., Eickhoff, S.B., 2018. How to char-
Bressler, S.L., Kelso, J.A., 2001. Cortical coordination dynamics and cognition. Trends acterize the function of a brain region. Trends Cognit. Sci. 22, 350–364.
Cognit. Sci. 5, 26–36. doi:10.1016/s1364-6613(00)01564-3. doi:10.1016/j.tics.2018.01.010.
Buckner, R.L., Sepulcre, J., Talukdar, T., Krienen, F.M., Liu, H., Hedden, T., Andrews- Gollo, L.L., Zalesky, A., Hutchison, R.M., van den Heuvel, M., Breakspear, M., 2015.
Hanna, J.R., Sperling, R.A., Johnson, K.A., 2009. Cortical hubs revealed by intrinsic Dwelling quietly in the rich club: brain network determinants of slow cortical fluctu-
functional connectivity: mapping, assessment of stability, and relation to alzheimer’s ations. Philos. Trans. R. Soc. Lond. B Biol. Sci. 370. doi:10.1098/rstb.2014.0165.
disease. J. Neurosci. 29, 1860–1873. doi:10.1523/JNEUROSCI.5062-08.2009. Gong, G., He, Y., Concha, L., Lebel, C., Gross, D.W., Evans, A.C., Beaulieu, C., 2009.
Cabral, J., Kringelbach, M.L., Deco, G., 2017. Functional connectivity dynamically evolves Mapping anatomical connectivity patterns of human cerebral cortex using in vivo
on multiple time-scales over a static structural connectome: models and mechanisms. diffusion tensor imaging tractography. Cereb. Cortex 19, 524–536. doi:10.1093/cer-
NeuroImage 160, 84–96. doi:10.1016/j.neuroimage.2017.03.045. cor/bhn102.
Chen, T., Cai, W., Ryali, S., Supekar, K., Menon, V., 2016. Distinct global brain dynamics Gonzalez-Castillo, J., Caballero-Gaudes, C., Topolski, N., Handwerker, D.A., Pereira, F.,
and spatiotemporal organization of the salience network. PLoS Biol. 14, e1002469. Bandettini, P.A., 2019. Imaging the spontaneous flow of thought: distinct periods of
doi:10.1371/journal.pbio.1002469. cognition contribute to dynamic functional connectivity during rest. NeuroImage 202,
Cohen, J.R., D’Esposito, M., 2016. The segregation and integration of distinct brain 116129. doi:10.1016/j.neuroimage.2019.116129.
networks and their relationship to cognition. J. Neurosci. 36, 12083–12094. Gonzalez-Castillo, J., Hoy, C.W., Handwerker, D.A., Robinson, M.E., Buchanan, L.C.,
doi:10.1523/JNEUROSCI.2965-15.2016. Saad, Z.S., Bandettini, P.A., 2015. Tracking ongoing cognition in individuals using
Cole, M.W., Schneider, W., 2007. The cognitive control network: Integrated brief, whole-brain functional connectivity patterns. Proc. Natl. Acad. Sci. USA 112,
cortical regions with dissociable functions. NeuroImage 37, 343–360. 8762–8767. doi:10.1073/pnas.1501242112.
doi:10.1016/j.neuroimage.2007.03.071. Gu, S., Pasqualetti, F., Cieslak, M., Telesford, Q.K., Yu, A.B., Kahn, A.E., Medaglia, J.D.,
Cole, M.W., Pathak, S., Schneider, W., 2010. Identifying the brain’s most glob- Vettel, J.M., Miller, M.B., Grafton, S.T., Bassett, D.S., 2015. Controllability of struc-
ally connected regions. NeuroImage 49, 3132–3148. doi:10.1016/j.neuroimage. tural brain networks. Nat. Commun. 6, 8414. doi:10.1038/ncomms9414.
2009.11.001. Guimera, R., Amaral, L.A.N., 2005. Functional cartography of complex metabolic net-
Cole, M.W., Ito, T., Bassett, D.S., Schultz, D.H., 2016. Activity flow over resting- works. Nature 433, 895–900. doi:10.1038/nature03288.
state networks shapes cognitive task activations. Nat. Neurosci. 19, 1718–1726. Hagmann, P., Cammoun, L., Gigandet, X., Meuli, R., Honey, C.J., Wedeen, V.J., Sporns, O.,
doi:10.1038/nn.4406. 2008. Mapping the structural core of human cerebral cortex. PLoS Biol. 6, e159.
Cole, M.W., Bassett, D.S., Power, J.D., Braver, T.S., Petersen, S.E., 2014. Intrinsic doi:10.1371/journal.pbio.0060159.
and task-evoked network architectures of the human brain. Neuron 83, 238–251. Hou, H., Zheng, Q., Zhao, Y., Pouget, A., Gu, Y., 2019. Neural correlates
doi:10.1016/j.neuron.2014.05.014. of optimal multisensory decision making under time-varying reliabilities with
Cole, M.W., Reynolds, J.R., Power, J.D., Repovs, G., Anticevic, A., Braver, T.S., 2013. an invariant linear probabilistic population code. Neuron 104, 1010–1021.
Multi-task connectivity reveals flexible hubs for adaptive task control. Nat. Neurosci. doi:10.1016/j.neuron.2019.08.038, e1010.
16, 1348–1355. doi:10.1038/nn.3470. Huk, A., Bonnen, K., He, B.J., 2018. Beyond trial-based paradigms: continuous be-
Corbetta, M., Shulman, G.L., 2002. Control of goal-directed and stimulus-driven attention havior, ongoing neural activity, and natural stimuli. J. Neurosci. 38, 7551–7558.
in the brain. Nat. Rev. Neurosci. 3, 201–215. doi:10.1038/nrn755. doi:10.1523/JNEUROSCI.1920-17.2018.

14
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

Hwang, K., Bertolero, M.A., Liu, W.B., D’Esposito, M., 2017. The human thalamus Newman, M.E., 2006. Modularity and community structure in networks. Proc. Natl. Acad.
is an integrative hub for functional brain networks. J. Neurosci. 37, 5594–5607. Sci. USA 103, 8577–8582. doi:10.1073/pnas.0601602103.
doi:10.1523/JNEUROSCI.0067-17.2017. Nijhuis, E.H., van Cappellen van Walsum, A.M., Norris, D.G., 2013. Topographic hub maps
Ito, T., Hearne, L., Mill, R., Cocuzza, C., Cole, M.W., 2020. Discovering the compu- of the human structural neocortical network. PLoS One 8, e65511. doi:10.1371/jour-
tational relevance of brain network organization. Trends Cognit. Sci. 24, 25–38. nal.pone.0065511.
doi:10.1016/j.tics.2019.10.005. Oldham, S., Fulcher, B., Parkes, L., Arnatkevic Iute, A., Suo, C., Fornito, A., 2019. Consis-
Ito, T., Kulkarni, K.R., Schultz, D.H., Mill, R.D., Chen, R.H., Solomyak, L.I., Cole, M.W., tency and differences between centrality measures across distinct classes of networks.
2017. Cognitive task information is transferred between brain regions via resting-state PLoS One 14, e0220061. doi:10.1371/journal.pone.0220061.
network topology. Nat. Commun. 8, 1027. doi:10.1038/s41467-017-01000-w. Osher, D.E., Saxe, R.R., Koldewyn, K., Gabrieli, J.D., Kanwisher, N., Saygin, Z.M.,
Kaiser, M., 2017. Mechanisms of connectome development. Trends Cognit. Sci. 21, 703– 2016. Structural connectivity fingerprints predict cortical selectivity for multiple
717. doi:10.1016/j.tics.2017.05.010. visual categories across cortex. Cereb. Cortex 26, 1668–1683. doi:10.1093/cercor/
Kaiser, M., 2020. Changing Connectomes: Evolution, Development, and Dynamics in Net- bhu303.
work Neuroscience. MIT Press. Palla, G., Derenyi, I., Farkas, I., Vicsek, T., 2005. Uncovering the overlapping commu-
Karahanoglu, F.I., Van De Ville, D., 2015. Transient brain activity disentangles fmri nity structure of complex networks in nature and society. Nature 435, 814–818.
resting-state dynamics in terms of spatially and temporally overlapping networks. doi:10.1038/nature03607.
Nat. Commun. 6, 7751. doi:10.1038/ncomms8751. Passingham, R.E., Stephan, K.E., Kotter, R., 2002. The anatomical basis of functional lo-
Kenet, T., Bibitchkov, D., Tsodyks, M., Grinvald, A., Arieli, A., 2003. Spontaneously calization in the cortex. Nat. Rev. Neurosci. 3, 606–616. doi:10.1038/nrn893.
emerging cortical representations of visual attributes. Nature 425, 954–956. Pauli, W.M., O’Reilly, R.C., Yarkoni, T., Wager, T.D., 2016. Regional specialization within
doi:10.1038/nature02078. the human striatum for diverse psychological functions. Proc. Natl. Acad. Sci. USA
Kitsak, M., Gallos, L.K., Havlin, S., Liljeros, F., Muchnik, L., Stanley, H.E., Makse, H.A., 113, 1907–1912. doi:10.1073/pnas.1507610113.
2010. Identification of influential spreaders in complex networks. Nat. Phys. 6, 888– Pedersen, M., Zalesky, A., Omidvarnia, A., Jackson, G.D., 2018. Multilayer network
893. doi:10.1038/Nphys1746. switching rate predicts brain performance. Proc. Natl. Acad. Sci. USA 115, 13376–
Krienen, F.M., Yeo, B.T., Buckner, R.L., 2014. Reconfigurable task-dependent functional 13381. doi:10.1073/pnas.1814785115.
coupling modes cluster around a core functional architecture. Philos. Trans. R. Soc. Pessoa, L., 2014. Understanding brain networks and brain organization. Phys. Life Rev.
Lond. B Biol. Sci. 369. doi:10.1098/rstb.2013.0526. 11, 400–435. doi:10.1016/j.plrev.2014.03.005.
Kucyi, A., Tambini, A., Sadaghiani, S., Keilholz, S., Cohen, J.R., 2018. Spontaneous cogni- Poldrack, R.A., 2006. Can cognitive processes be inferred from neuroimaging data? Trends
tive processes and the behavioral validation of time-varying brain connectivity. Netw. Cognit. Sci. 10, 59–63. doi:10.1016/j.tics.2005.12.004.
Neurosci. 2, 397–417. doi:10.1162/netn_a_00037. Ponce-Alvarez, A., He, B.J., Hagmann, P., Deco, G., 2015. Task-driven activity reduces
Lahav, N., Ksherim, B., Ben-Simon, E., Maron-Katz, A., Cohen, R., Havlin, S., 2016. K-shell the cortical activity space of the brain: experiment and whole-brain modeling. PLoS
decomposition reveals hierarchical cortical organization of the human brain. New J. Comput. Biol. 11, e1004445. doi:10.1371/journal.pcbi.1004445.
Phys. 18 https://doi.org/Artn 083013. Pouget, A., Beck, J.M., Ma, W.J., Latham, P.E., 2013. Probabilistic brains: knowns and
Li, A., et al., 2020. A neuroimaging biomarker for striatal dysfunction in schizophrenia. unknowns. Nat. Neurosci. 16, 1170–1178. doi:10.1038/nn.3495.
Nat. Med. 26, 558–565. doi:10.1038/s41591-020-0793-8. Power, J.D., Schlaggar, B.L., Lessov-Schlaggar, C.N., Petersen, S.E., 2013. Evi-
Liang, X., Zou, Q., He, Y., Yang, Y., 2013. Coupling of functional connectivity and regional dence for hubs in human functional brain networks. Neuron 79, 798–813.
cerebral blood flow reveals a physiological basis for network hubs of the human brain. doi:10.1016/j.neuron.2013.07.035.
Proc. Natl. Acad. Sci. USA 110, 1929–1934. doi:10.1073/pnas.1214900110. Ringach, D.L., 2009. Spontaneous and driven cortical activity: Implications for computa-
Liao, X., Yuan, L., Zhao, T., Dai, Z., Shu, N., Xia, M., Yang, Y., Evans, A., He, Y., 2015. tion. Curr. Opin. Neurobiol. 19, 439–444. doi:10.1016/j.conb.2009.07.005.
Spontaneous functional network dynamics and associated structural substrates in the Sadaghiani, S., Kleinschmidt, A., 2013. Functional interactions between
human brain. Front. Hum. Neurosci. 9, 478. doi:10.3389/fnhum.2015.00478. intrinsic brain activity and behavior. NeuroImage 80, 379–386.
Lin, Y., Ma, J., Gu, Y., Yang, S., Li, L.M.W., Dai, Z., 2018. Intrinsic overlap- doi:10.1016/j.neuroimage.2013.04.100.
ping modular organization of human brain functional networks revealed Saygin, Z.M., Osher, D.E., Koldewyn, K., Reynolds, G., Gabrieli, J.D., Saxe, R.R., 2011.
by a multiobjective evolutionary algorithm. NeuroImage 181, 430–445. Anatomical connectivity patterns predict face selectivity in the fusiform gyrus. Nat.
doi:10.1016/j.neuroimage.2018.07.019. Neurosci. 15, 321–327. doi:10.1038/nn.3001.
Liu, X., Duyn, J.H., 2013. Time-varying functional network information extracted from Schmidt, R., LaFleur, K.J., de Reus, M.A., van den Berg, L.H., van den Heuvel, M.P., 2015.
brief instances of spontaneous brain activity. Proc. Natl. Acad. Sci. USA 110, 4392– Kuramoto model simulation of neural hubs and dynamic synchrony in the human
4397. doi:10.1073/pnas.1216856110. cerebral connectome. BMC Neurosci. 16, 54. doi:10.1186/s12868-015-0193-z.
Liu, Y.Y., Slotine, J.J., Barabasi, A.L., 2011. Controllability of complex networks. Nature Schuck, N.W., Niv, Y., 2019. Sequential replay of nonspatial task states in the human
473, 167–173. doi:10.1038/nature10011. hippocampus. Science 364. doi:10.1126/science.aaw5181.
Lurie, D.J., et al., 2020. Questions and controversies in the study of time- Seidman, S.B., 1983. Network structure and minimum degree. Soc. Netw. 5, 269.
varying functional connectivity in resting fmri. Netw. Neurosci. 4, 30–69. Shine, J.M., Bissett, P.G., Bell, P.T., Koyejo, O., Balsters, J.H., Gorgolewski, K.J.,
doi:10.1162/netn_a_00116. Moodie, C.A., Poldrack, R.A., 2016. The dynamics of functional brain networks: In-
Maldjian, J.A., Davenport, E.M., Whitlow, C.T., 2014. Graph theoretical analysis of tegrated network states during cognitive task performance. Neuron 92, 544–554.
resting-state meg data: identifying interhemispheric connectivity and the default doi:10.1016/j.neuron.2016.09.018.
mode. NeuroImage 96, 88–94. doi:10.1016/j.neuroimage.2014.03.065. Shine, J.M., Breakspear, M., Bell, P.T., Ehgoetz Martens, K.A., Shine, R., Koyejo, O.,
Mante, V., Sussillo, D., Shenoy, K.V., Newsome, W.T., 2013. Context-dependent computa- Sporns, O., Poldrack, R.A., 2019. Human cognition involves the dynamic integra-
tion by recurrent dynamics in prefrontal cortex. Nature 503, 78–84. doi:10.1038/na- tion of neural activity and neuromodulatory systems. Nat. Neurosci. 22, 289–296.
ture12742. doi:10.1038/s41593-018-0312-0.
Mantini, D., Perrucci, M.G., Del Gratta, C., Romani, G.L., Corbetta, M., 2007. Electrophys- Smith, S.M., Fox, P.T., Miller, K.L., Glahn, D.C., Fox, P.M., Mackay, C.E., Filippini, N.,
iological signatures of resting state networks in the human brain. Proc. Natl. Acad. Watkins, K.E., Toro, R., Laird, A.R., Beckmann, C.F., 2009. Correspondence of the
Sci. USA 104, 13170–13175. doi:10.1073/pnas.0700668104. brain’s functional architecture during activation and rest. Proc. Natl. Acad. Sci. USA
Mars, R.B., Passingham, R.E., Jbabdi, S., 2018. Connectivity fingerprints: from 106, 13040–13045. doi:10.1073/pnas.0905267106.
areal descriptions to abstract spaces. Trends Cognit. Sci. 22, 1026–1037. Smith, S.M., Miller, K.L., Moeller, S., Xu, J., Auerbach, E.J., Woolrich, M.W., Beck-
doi:10.1016/j.tics.2018.08.009. mann, C.F., Jenkinson, M., Andersson, J., Glasser, M.F., Van Essen, D.C., Fein-
Medaglia, J.D., Huang, W., Karuza, E.A., Kelkar, A., Thompson-Schill, S.L., Ribeiro, A., berg, D.A., Yacoub, E.S., Ugurbil, K., 2012. Temporally-independent functional
Bassett, D.S., 2018. Functional alignment with anatomical networks is associated with modes of spontaneous brain activity. Proc. Natl. Acad. Sci. USA 109, 3131–3136.
cognitive flexibility. Nat. Hum. Behav. 2, 156–164. doi:10.1038/s41562-017-0260-9. doi:10.1073/pnas.1121329109.
Mennes, M., Kelly, C., Colcombe, S., Castellanos, F.X., Milham, M.P., 2013. The extrinsic Spielberg, J.M., Miller, G.A., Heller, W., Banich, M.T., 2015. Flexible brain network re-
and intrinsic functional architectures of the human brain are not equivalent. Cereb. configuration supporting inhibitory control. Proc. Natl. Acad. Sci. USA 112, 10020–
Cortex 23, 223–229. doi:10.1093/cercor/bhs010. 10025. doi:10.1073/pnas.1500048112.
Meyniel, F., Sigman, M., Mainen, Z.F., 2015. Confidence as bayesian probability: from Sporns, O., Chialvo, D.R., Kaiser, M., Hilgetag, C.C., 2004. Organization, develop-
neural origins to behavior. Neuron 88, 78–92. doi:10.1016/j.neuron.2015.09.039. ment and function of complex brain networks. Trends Cognit. Sci. 8, 418–425.
Miller, E.K., Cohen, J.D., 2001. An integrative theory of prefrontal cortex function. Annu. doi:10.1016/j.tics.2004.07.008.
Rev. Neurosci. 24, 167–202. doi:10.1146/annurev.neuro.24.1.167. Stam, C.J., Tewarie, P., Van Dellen, E., van Straaten, E.C., Hillebrand, A., Van
Misic, B., Sporns, O., 2016. From regions to connections and networks: new Mieghem, P., 2014. The trees and the forest: characterization of complex brain
bridges between brain and behavior. Curr. Opin. Neurobiol. 40, 1–7. networks with minimum spanning trees. Int. J. Psychophysiol. 92, 129–138.
doi:10.1016/j.conb.2016.05.003. doi:10.1016/j.ijpsycho.2014.04.001.
Misic, B., Betzel, R.F., Nematzadeh, A., Goni, J., Griffa, A., Hagmann, P., Flammini, A., Sun, J., Liu, Z., Rolls, E.T., Chen, Q., Yao, Y., Yang, W., Wei, D., Zhang, Q., Zhang, J.,
Ahn, Y.Y., Sporns, O., 2015. Cooperative and competitive spreading dynamics Feng, J., Qiu, J., 2019. Verbal creativity correlates with the temporal variabil-
on the human connectome. Neuron 86, 1518–1529. doi:10.1016/j.neuron.2015. ity of brain networks during the resting state. Cereb. Cortex 29, 1047–1058.
05.035. doi:10.1093/cercor/bhy010.
Najafi, M., McMenamin, B.W., Simon, J.Z., Pessoa, L., 2016. Overlapping communities Tambini, A., Davachi, L., 2019. Awake reactivation of prior experiences con-
reveal rich structure in large-scale brain networks during rest and task conditions. solidates memories and biases cognition. Trends Cognit. Sci. 23, 876–890.
NeuroImage 135, 92–106. doi:10.1016/j.neuroimage.2016.04.054. doi:10.1016/j.tics.2019.07.008.
Nepusz, T., Petroczi, A., Negyessy, L., Bazso, F., 2008. Fuzzy communities and the con- Tavor, I., Parker Jones, O., Mars, R.B., Smith, S.M., Behrens, T.E., Jbabdi, S., 2016. Task-
cept of bridgeness in complex networks. Phys. Rev. E 77, 016107. doi:10.1103/Phys- free MRI predicts individual differences in brain activity during task performance.
RevE.77.016107. Science 352, 216–220. doi:10.1126/science.aad8127.

15
D. Yin and M. Kaiser NeuroImage 235 (2021) 118027

Thompson, W.H., Brantefors, P., Fransson, P., 2017. From static to temporal network Wu, K., Taki, Y., Sato, K., Sassa, Y., Inoue, K., Goto, R., Okada, K., Kawashima, R., He, Y.,
theory: applications to functional brain connectivity. Netw. Neurosci. 1, 69–99. Evans, A.C., Fukuda, H., 2011. The overlapping community structure of structural
doi:10.1162/NETN_a_00011. brain network in young healthy individuals. PLoS One 6, e19608. doi:10.1371/jour-
Tognoli, E., Kelso, J.A., 2014. The metastable brain. Neuron 81, 35–48. nal.pone.0019608.
doi:10.1016/j.neuron.2013.12.022. Yan, G., Vertes, P.E., Towlson, E.K., Chew, Y.L., Walker, D.S., Schafer, W.R., Barabasi, A.L.,
Tomasi, D., Volkow, N.D., 2010. Functional connectivity density mapping. Proc. Natl. 2017. Network control principles predict neuron function in the caenorhabditis ele-
Acad. Sci. USA 107, 9885–9890. doi:10.1073/pnas.1001414107. gans connectome. Nature 550, 519–523. doi:10.1038/nature24056.
Tononi, G., Edelman, G.M., 1998. Consciousness and complexity. Science 282, 1846–1851. Yeo, B.T., Krienen, F.M., Chee, M.W., Buckner, R.L., 2014. Estimates of segregation and
doi:10.1126/science.282.5395.1846. overlap of functional connectivity networks in the human cerebral cortex. NeuroImage
Uddin, L.Q., 2020a. Brain mechanisms supporting flexible cognition and behavior 88, 212–227. doi:10.1016/j.neuroimage.2013.10.046.
in adolescents with autism spectrum disorder. Biol. Psychiatry 89, 172–183. Yeo, B.T., Krienen, F.M., Eickhoff, S.B., Yaakub, S.N., Fox, P.T., Buckner, R.L., As-
doi:10.1016/j.biopsych.2020.05.010. plund, C.L., Chee, M.W., 2015. Functional specialization and flexibility in human
Uddin, L.Q., 2020b. Bring the noise: reconceptualizing spontaneous neural activity. Trends association cortex. Cereb. Cortex 25, 3654–3672. doi:10.1093/cercor/bhu217.
Cognit. Sci. 24, 734–746. doi:10.1016/j.tics.2020.06.003. Yin, D., Chen, X., Zeljic, K., Zhan, Y., Shen, X., Yan, G., Wang, Z., 2019a. A graph
Vaidya, C.J., Gordon, E.M., 2013. Phenotypic variability in resting-state functional con- representation of functional diversity of brain regions. Brain Behav. 9, e01358.
nectivity: current status. Brain Connect 3, 99–120. doi:10.1089/brain.2012.0110. doi:10.1002/brb3.1358.
van den Heuvel, M.P., Sporns, O., 2013a. An anatomical substrate for integration Yin, D., Liu, W., Zeljic, K., Wang, Z., Lv, Q., Fan, M., Cheng, W., Wang, Z., 2016.
among functional networks in human cortex. J. Neurosci. 33, 14489–14500. Dissociable changes of frontal and parietal cortices in inherent functional flexibil-
doi:10.1523/JNEUROSCI.2128-13.2013. ity across the human life span. J. Neurosci. 36, 10060–10074. doi:10.1523/JNEU-
van den Heuvel, M.P., Sporns, O., 2013b. Network hubs in the human brain. Trends Cognit. ROSCI.1476-16.2016.
Sci. 17, 683–696. doi:10.1016/j.tics.2013.09.012. Yin, D., Zhang, Z., Wang, Z., Zeljic, K., Lv, Q., Cai, D., Wang, Y., Wang, Z., 2019b. Brain
Vazquez-Rodriguez, B., Suarez, L.E., Markello, R.D., Shafiei, G., Paquola, C., Hagmann, P., map of intrinsic functional flexibility in anesthetized monkeys and awake humans.
van den Heuvel, M.P., Bernhardt, B.C., Spreng, R.N., Misic, B., 2019. Gradients of Front. Neurosci. 13, 174. doi:10.3389/fnins.2019.00174.
structure-function tethering across neocortex. Proc. Natl. Acad. Sci. USA 116, 21219– Yin, W., Li, T., Hung, S.C., Zhang, H., Wang, L., Shen, D., Zhu, H., Mucha, P.J., Cohen, J.R.,
21227. doi:10.1073/pnas.1903403116. Lin, W., 2020. The emergence of a functionally flexible brain during early infancy.
Vidaurre, D., Smith, S.M., Woolrich, M.W., 2017. Brain network dynamics are hi- Proc. Natl. Acad. Sci. USA 117, 23904–23913. doi:10.1073/pnas.2002645117.
erarchically organized in time. Proc. Natl. Acad. Sci. USA 114, 12827–12832. Zeng, G., Chen, Y., Cui, B., Yu, S., 2019. Continual learning of context-
doi:10.1073/pnas.1705120114. dependent processing in neural networks. Nat. Mach. Intell. 1, 364–372.
Vidaurre, D., Abeysuriya, R., Becker, R., Quinn, A.J., Alfaro-Almagro, F., Smith, S.M., doi:10.1038/s42256-019-0080-x.
Woolrich, M.W., 2018. Discovering dynamic brain networks from big data in rest and Zhang, J., Cheng, W., Liu, Z., Zhang, K., Lei, X., Yao, Y., Becker, B., Liu, Y., Kendrick, K.M.,
task. NeuroImage 180, 646–656. doi:10.1016/j.neuroimage.2017.06.077. Lu, G., Feng, J., 2016. Neural, electrophysiological and anatomical basis of brain-
Wang, X., Lin, Q., Xia, M., He, Y., 2018. Differentially categorized structural brain network variability and its characteristic changes in mental disorders. Brain 139,
hubs are involved in different microstructural, functional, and cognitive character- 2307–2321. doi:10.1093/brain/aww143.
istics and contribute to individual identification. Hum. Brain Mapp. 39, 1647–1663. Zuo, X.N., Ehmke, R., Mennes, M., Imperati, D., Castellanos, F.X., Sporns, O., Mil-
doi:10.1002/hbm.23941. ham, M.P., 2012. Network centrality in the human functional connectome. Cereb.
Wimmer, G.E., Liu, Y., Vehar, N., Behrens, T.E.J., Dolan, R.J., 2020. Episodic memory Cortex 22, 1862–1875. doi:10.1093/cercor/bhr269.
retrieval success is associated with rapid replay of episode content. Nat. Neurosci. 23,
1025–1033. doi:10.1038/s41593-020-0649-z.

16