Neuron

Perspective

Hierarchy of Information Processing in the Brain:

A Novel ‘Intrinsic Ignition’ Framework
Gustavo Deco1,2,3,4 and Morten L. Kringelbach5,6,7,*
1Center for Brain and Cognition, Computational Neuroscience Group, Department of Information and Communication Technologies,

Universitat Pompeu Fabra, Roc Boronat 138, Barcelona 08018, Spain

2Institució Catalana de la Recerca i Estudis Avançats (ICREA), Passeig Lluı́s Companys 23, Barcelona 08010, Spain
3Department of Neuropsychology, Max Planck Institute for Human Cognitive and Brain Sciences, 04103 Leipzig, Germany
4School of Psychological Sciences, Monash University, Melbourne, Clayton VIC 3800, Australia
5Department of Psychiatry, University of Oxford, Oxford OX3 7JX, UK
6Center for Music in the Brain (MIB), Department of Clinical Medicine, Aarhus University, 8000 Aarhus C, Denmark
7Institut d’études avancées de Paris, 75004 Paris, France

*Correspondence: [email protected]
http://dx.doi.org/10.1016/j.neuron.2017.03.028

A general theory of brain function has to be able to explain local and non-local network computations over
space and time. We propose a new framework to capture the key principles of how local activity influences
global computation, i.e., describing the propagation of information and thus the broadness of communica-
tion driven by local activity. More specifically, we consider the diversity in space (nodes or brain regions)
over time using the concept of intrinsic ignition, which are naturally occurring intrinsic perturbations reflecting
the capability of a given brain area to propagate neuronal activity to other regions in a given brain state.
Characterizing the profile of intrinsic ignition for a given brain state provides insight into the precise nature
of hierarchical information processing. Combining this data-driven method with a causal whole-brain
computational model can provide novel insights into the imbalance of brain states found in neuropsychiatric
disorders.

Introduction the ‘‘intrinsic ignition’’ of brain activity across time and space,
Historically, within brain science, the most sophisticated and pre- i.e., the diversity of computation in space and time. This frame-
cise spatiotemporal information has come from single-neuron work allows the study not only of the propagation of brain activ-
recordings of spiking activity (Hodgkin and Huxley, 1952; Hubel ity, but also the underlying fluctuations and their functional
and Wiesel, 1959; Mountcastle, 1957). Measuring spiking activity network consequences. Informally, the concept of intrinsic igni-
could help derive the underlying mechanistic principles of tion refers to the capability of a given local node (single neuron or
brain function (notwithstanding the mounting evidence of the brain area) in space to propagate feedforward and recurrent
importance of glial processes). This local information has helped neuronal activity to other nodes in the network as measured
develop important computational models of attention, memory, by the whole-brain integration elicited. As such, intrinsic ignition
and decision making (Amit and Brunel, 1997; Brunel and Wang, is a novel concept that can be used to describe the specific pro-
2001; Deco and Rolls, 2005; Wang, 2002). Yet, non-local informa- file of the ignition capabilities of brain regions in different brain
tion is also very important given that neurons are connected to states. Furthermore, defining the variability of the ignition-driven
other neurons in sophisticated networks with specific properties propagation of activity across time, we can classify each node
(small world or not) that are crucial for efficient brain function (Mar- (neuron or brain area) according to the local degree of functional
kov et al., 2013, 2014; Sporns and Zwi, 2004; Watts and Strogatz, variability, i.e., diversity and metastability.
1998). This local versus non-local processing can be character-
ized on many different levels, from the cortical column to brain re- Ignition as Intrinsic Perturbation
gion-specific networks to whole-brain connectivity (Sporns et al., Computation is the fundamental unit of a general theory of brain
2005). Abundant evidence shows that computation happens on function, enabling a full characterization of the fundamental
all these levels and that the field of brain science could benefit neuronal principles underlying the computations involved in
from a better understanding of the computational principles at cognitive, perceptual, and motor functions in health, as well as
the network level. in disease. This is a complex problem given that there are billions
Here we propose a novel framework for measuring and under- of recurrently, synaptically coupled, non-homogeneous neurons
standing the intrinsic dynamics and communication principles in the human brain. More precisely, the non-linear character of
of brain activity across all levels of networks (from local net- these basic elements, primarily neurons and synapses, and
works of neurons in a brain region to higher-order whole-brain the coupling through both feedforward and feedback connec-
networks). The starting point is how best to characterize the tions makes the brain a complex, non-linear dynamical system
way that information propagates from a local node through the (Deco et al., 2011; Sporns, 2014). Uncovering the fundamental
network. We are proposing a conceptual framework for studying mechanisms underlying the emerging properties of a complex

Neuron 94, June 7, 2017 ª 2017 Elsevier Inc. 961

 Neuron

Perspective

dynamical system will rely on solving the so-called ‘‘inverse This approach can be extended in several ways by using
problem.’’ In neuroscience, the attempts of finding a solution whole-brain computational modeling of the underlying brain
has adopted three main strategies, namely: (1) correlating neural activity in a given brain state (e.g., wakefulness, sleep, or coma)
activity with task, (2) measuring spontaneous, non-task-related, measured, for example, with magnetoencephalography (MEG)
on-going brain dynamics, and (3) lesioning or perturbing the in- or fMRI (Deco and Kringelbach, 2014). Such models can then
ternal dynamics of the brain. be systematically perturbed in ways not possible experimentally
Historically, single-neuron recordings have adopted the first (Cabral et al., 2014). Crucially, rather than just measuring the
task-evoked strategy, e.g., recording activity in primary visual complexity of the elicited activity after perturbation, the whole-
cortices to simple visual cues such as orientation (Hubel and brain model allows for the possibility of generating a long-lasting,
Wiesel, 1959). Once a correlation was established, using the strong perturbation and measuring the latency of the recovery of
third perturbation strategy, these neurons could then be lesioned brain dynamics, observable over much longer time spans (on the
(or perturbed) to show a causal functional role, e.g., lesioning order of tens of seconds).
neurons to show the effects on the brain of visual orientation Similarly, rather than using extrinsic perturbation, one could
cues (Calford et al., 2000). study the effects of a naturally occurring intrinsic perturbation.
The second strategy measuring spontaneous activity is diffi- We define intrinsic ignition as the capability of a given brain
cult to apply to a local framework of such single-neuron record- area to propagate neuronal activity to other regions in a given
ings since the spontaneous activity within these recordings has brain state, describing the whole-brain integration elicited from
traditionally been seen as merely noise. The advent of optical the propagation of both feedforward and recurrent activity. As
imaging (Grinvald et al., 1986; Kenet et al., 2003) and neuroi- such, intrinsic ignition is a novel concept that can be used to
maging experiments of non-linear, baseline, resting-state activ- describe the specific profile of the ignition capabilities of regions
ity (Biswal et al., 1995; Raichle et al., 2001) has led to a shift in across the network (from local networks of neurons in a brain
focus to the second strategy of analyzing internal on-going region to higher-order whole-brain networks) in different brain
brain dynamics (Smith et al., 2009; Zhang and Raichle, 2010). states.
This has led to an emphasis on the functional significance This definition of intrinsic ignition makes it possible to create
of the non-local compared to the local computation. The first a full characterization of each ‘‘activity event’’ of the nodes
two strategies of measuring and analyzing brain dynamics (neuron or brain areas) for a given brain state. Such intrinsic ac-
with or without task are orthogonal and complementary yet tivity event can be seen as an internal intrinsic perturbation that
only allow for understanding correlative, rather than causal, could eventually elicit (or not) the propagation of activity across
relationships. the whole network. As an example, one simple way to charac-
The third perturbation strategy has been used extensively in terize an ‘‘ignition event’’ is to binarize each measurement result-
experimental animals and measuring the behavioral outcomes. ing from the thresholding of activity at different possible levels
For ethical reasons, such systematic perturbations are not in multimodal neuroimaging data (see Figure 1). This method
directly translatable to humans. Still, the field has been able lends itself naturally to spiking neuronal activity but can easily
to draw on perturbations from naturally occurring lesions linked be extended to local field potential (LFP), MEG, and fMRI sig-
to stroke and neurosurgical procedures. In addition, there is nals. However, please note that our framework could equally
a growing field of electrical brain stimulation, often as part of well use other more sophisticated mathematical methods for ex-
neurosurgical evaluation of patients with epilepsy and resec- tracting point processes (Caballero Gaudes et al., 2013; Karaha-
tion of special cases of brain tumor (Parvizi et al., 2013; nog  lu et al., 2013; Petridou et al., 2013). Such methods have
Selimbeyoglu and Parvizi, 2010; Winawer and Parvizi, 2016). been shown to be able to describe many important aspects
Furthermore, there is even more widespread use of deep brain of dynamics, such as, for example, resting-state networks and
stimulation for symptom alleviation in movement disorders, complexity (Karahanog  lu et al., 2013; Karahanog lu and Van De
such as Parkinson’s disease (Kringelbach et al., 2007). Still, Ville, 2015).
what has been missing is an understanding of how the propa- The proposed new paradigm allows for a description of how
gation of information is influenced by perturbation (Borchers the ignition capability of each node, i.e., the integration elicited,
et al., 2011). varies for different brain states by averaging the event-related
Ideally, measurements of activity across the whole brain would changes (across many occurrences of the same event) induced
be useful to characterize the underlying mechanisms, similar for each time point aligned after each event. The differences
to how perturbation of a physical system can be described in between ignition profiles of the different nodes (i.e., neurons or
physics. Massimini and colleagues pioneered studies using brain areas) provide a marker of different brain states.
electroencephalography (EEG) to characterize the degree and Note that our novel concept of intrinsic ignition is different
short-term latency of the dynamics (100–200 ms) elicited by from the ignition defined by Dehaene and colleagues (Moutard
the external perturbation by transcranial magnetic stimulation et al., 2015), as the rapid and sustained activity elicited after
(TMS) (Massimini et al., 2005). This has been used successfully stimulation in contrast to the ultra-slow (<0.1 Hz) fluctuations
for separation of brain states in healthy subjects during wakeful- of relatively low amplitude in resting state. Both modes can
ness, dreaming, and sleeping and in different levels of anesthesia emerge from the same underlying connectome as ‘‘two dynamic
(Casali et al., 2013; Ferrarelli et al., 2010), as well as characteriza- faces’’ of the strong recurrent loops built by the brain networks.
tion of brain-injured patients emerging from coma (Rosanova Indeed, the dense lateral intra- and inter-areal connections that
et al., 2012). characterize brain networks make possible the emergence of a

962 Neuron 94, June 7, 2017

Neuron

Perspective

A (e.g., wakefulness and sleep) is eliciting a propagation of activity

across the network.
The novel concept of intrinsic ignition allows us to measure
the spatial and temporal diversity of propagation of information
in the whole network using our existing integration measure
(Deco et al., 2015). Briefly, for each intrinsic ignition event, a
value can be computed that measures the integration, i.e.,
Driving events the degree of broadcasting of the information arising from
both feedforward and recurrent processing (Figure 1). This inte-
gration value is defined as the length of the largest connected
component in a binarized functional connectivity matrix of the
whole network at a given window of time following the intrinsic
ignition triggering event. Note that the binarization procedure of
NETWORK NODES

the activity in each node allows us to construct the functional

connectivity matrix in this window of time, where the largest
subcomponent is defined as the length of the connected
component of the undirected graph defined by the binarized
matrix considered as an adjacency matrix. This is the largest
sub-graph in which any two vertices are connected to each
other by paths and which connects to no additional vertices
in the super-graph. Similar to event-related potential (ERP)
analysis, the intrinsic ignition for each node in a given network
TIME is fully characterized by the mean and standard deviation
across events.
B C At the network level, the mean of the intrinsic ignition in a
network allows us to show the spatial diversity as the differences
in average intrinsic ignition profiles across the different nodes for
different brain states. Furthermore, at the network level, we can
characterize the temporal diversity by measuring the variability
of the intrinsic ignition-driven increase of integration across
events by classifying nodes according to the local degree of
functional variability. This functional variability is a measure of
Largest
diversity across time, which is related to the fundamental func-
subcomponent
tionality of a given node, linked to high or low levels of local meta-
CONN. MATRIX TOPOLOGY stability, i.e., the variability of a state that falls outside the natural
equilibrium state of the system but persists for an extended
INTEGRATION period of time. In other words, different levels of temporal diver-
sity in a given node can be thought of as a measure of local func-
Figure 1. Schematic of the Process of Measuring Intrinsic Ignition tional variability or metastability and thus describe the versatility
We measure the activity of a node in the network and the changes in subse-
of a given network node.
quent activity across the whole network.
(A) Here we show the spiking activity in a neuron (green area superimposed on
the brain and with activity in the stippled green square). For each driving event, Hierarchies of Computation
we measure the activity in the rest of the network (in stippled red area) in the There is evidence from the field of connectomics using graph
gray time window.
(B) This corresponds to a binary connectivity matrix, where spikes are co- theory to reveal a topological hierarchy in structural connectivity
occurring. of the brain (Deco et al., 2015; Sporns et al., 2000). But it is pres-
(C) In this matrix, we find the largest subcomponent as a measure of the global ently unclear whether this structural hierarchical connectivity
integration, i.e., the broadness of communication across the network. This is
repeated for each of the driving events, producing a mean and standard de-
is reflected in the dynamical processing hierarchy. Measuring
viation of the intrinsic ignition for each network node. the intrinsic ignition of nodes (mean and variability) in a network
makes it possible to investigate the important question of
whether there is a hierarchy of information processing.
reverberatory dynamic when the level of excitation exceeds Figure 2 shows the spectrum of possibilities for hierarchical
the level of inhibition, which can be propagated globally across processing. At one end (shown in Figure 2A), the intrinsic ignition
the brain. This imbalance between excitation and inhibition is equal for all nodes in the network. This would be the equivalent
could appear spontaneously (resting state) or rapidly induced of a weak, flat non-hierarchy, demonstrated by a single circle,
(ignition) by the action of an external stimulation, explaining in where each node has equal computational importance. At the
this way both modes. Nevertheless, such concept of ignition is other end (shown in Figure 2D), intrinsic ignition is uniform, linear
not covering how an intrinsic (i.e., in the absence of any kind gradation between nodes, which corresponds to a strong hierar-
of external stimulation) local activity event in a given brain state chy in which a node at the top of the linear hierarchy has the

Neuron 94, June 7, 2017 963

 Neuron

Perspective

A
intrinsic ignition

Weak nonhierarchy
nodes

B
intrinsic ignition

Staircase hierarchy
(GWS)
nodes

C
intrinsic ignition

Graded, non-uniform
hierarchy
nodes

D
intrinsic ignition

Strong, graded uniform

hierarchy
nodes

Figure 2. Spectrum of Possible Dynamical Processing Hierarchies in the Brain

(A) Weak, flat non-hierarchy is derived when the intrinsic ignition is equal for all nodes in the network, which is shown by a single level of nodes.
(B) On the other hand, a ‘‘staircase’’ hierarchy is suggestive of distinct circles (or orbits) of groups of nodes with equal computational importance. Such a scheme
would correspond to the ideas of a global workspace with a clear computational quantum jump between sensory regions and regions in the global neuronal
workspace.
(C) There are many other possibilities of graded, non-uniform hierarchies with non-uniform, clustered circular orbits for the network nodes.
(D) Strong, graded, uniform hierarchy can occur when a node at the top of the linear hierarchy has the highest intrinsic ignition, which is demonstrated by a uniform
distribution of levels (rings) for each node in the network.

highest intrinsic ignition, demonstrated by a uniform distribution brain activity from a group of 16 healthy individuals measured
of rings for each node in the network. with fMRI (van Hartevelt et al., 2015).
Between these poles of weak and strong hierarchies, there The procedure of defining events used the simple method
are many significantly different functional possibilities. One pos- of Tagliazucchi and colleagues to binarize the fMRI series
sibility is where the intrinsic ignition resembles a staircase, sug- in the following way (Tagliazucchi et al., 2012). An event is
gestive of distinct circles (or orbits) of groups of nodes with defined as a binary signal resulting from the transformed func-
equal computational importance. These groups are, however, tional time series (BOLD fMRI) into Z scores, zi(t), and imposing
clearly stratified, shown in Figure 2B by two orbits with nodes a threshold, q, such that the binary sequence si(t) = 1 if zi(t) > q
of similar intrinsic ignition. This scheme would correspond to and is crossing the threshold from below and si(t) = 0 otherwise
the idea of the global workspace, where there is a clear, compu- (shown schematically in Figure 3). If the top signal (in red) refers
tational, quantum jump between sensory regions and regions to the analysis of the ignition capability of a given brain region,
in the global neuronal workspace (Baars, 1989; Dehaene and an event is the point where the signal crosses the threshold from
Changeux, 2011; Dehaene et al., 1998). below. The x axis represents the time whereas the y ordinates
Another possibility is that the profile of network nodes shows represent the different brain areas. The brain was segmented
a non-uniform gradation in intrinsic ignition, suggestive of a hier- into 758 regions (based on running a K-means algorithm
archical organization of a non-uniform graded variety. This is run on the standard AAL atlas, Automated Anatomical Labeling,
schematically demonstrated in Figure 2C by the non-uniformity including all cortical and subcortical areas) (Tzourio-Mazoyer
circular orbits for each node. et al., 2002). Each single black vertical bar refers to an event
for the corresponding brain region.
Whole-Brain Evidence for Hierarchy of Intrinsic Ignition For each brain region, we computed the mean and standard
The key question of the dynamical processing hierarchy across deviation of the integration associated with events. Specifically,
human brain regions can be addressed using the concept of we used a window size restricted by the intra-event duration for
spatiotemporal diversity as defined by intrinsic ignition. To this observing the evolution of the elicited integration and used the
end, we here investigated this issue by using the spontaneous maximum value of integration within that given time window.

964 Neuron 94, June 7, 2017

Neuron

Perspective

A
B

INTRINSIC IGNITION (MEAN)

BOLD Signal 110

100

0 200 800
BRAIN AREAS

INTRINSIC IGNITION (VARIABILITY)

30

Largest 20
subcomponent
0 200 800
CONN. MATRIX TOPOLOGY BRAIN AREAS

Figure 3. Intrinsic Ignition Hierarchical Organization Demonstrated with Empirical Data

(A) Neuroimaging signals, such as BOLD, can be treated in the same manner as spiking data by applying a threshold method to define events. This again gives rise
to a connectivity matrix for each intrinsic ignition event for which the integration, i.e., the largest subcomponent, can be measured.
(B) When applied to fMRI signals of spontaneous activity in normal healthy participants using a very fine-grained parcellation, sorted from largest the intrinsic
ignition of nodes, we found, for both mean and standard deviation, an inverted sigmoid profile. This result indicates that there is a hierarchy of function across
brain regions, compatible with the global workspace theory. Yet, this hierarchy is not stratified in a staircase manner. Still, there are clearly regions with higher
intrinsic ignition variability, which would be more computationally relevant and could play a central role in broadcasting information than the regions with low
intrinsic ignition.

Figure 3 shows the profiles of the mean and variability of the plementary question cannot be answered by simple data-driven
ignition-driven integration. This shows an inverse sigmoidal curve, methods but requires causal whole-brain models. Given such a
suggesting that the hierarchical organization is of the non-uni- model, carefully adjusted to empirical data, it is possible to lesion
form, graded variety and thus closer to the scenario described and perturb the model offline and study the consequences (van
in Figure 2C than to the other scenarios. Although, it should be Hartevelt et al., 2015). For example, it is possible to study how
noted that, with respect to the global workspace scenario, the the spatiotemporal diversity will change following the elimination
results confirm that there is a hierarchy of function across brain of any given node.
regions. However, this hierarchy is not stratified in a staircase In order to investigate these challenging questions more care-
manner as suggested by the strong version of the global work- fully, we propose future studies testing the robustness of compu-
space theory, which is a surprising and non-trivial result. Still, tational diversity using a whole-brain model fitting resting-state
there are clearly regions with higher intrinsic ignition variability, empirical data. Briefly, whole-brain models link anatomical struc-
suggesting that they are more computationally relevant and could ture with functional dynamics. Structural connectivity data can be
play a central role in broadcasting information than the regions obtained by diffusion weighted/tensor imaging (DWI/DTI) com-
with low intrinsic ignition, which are more likely to be related to bined with probabilistic tractography, which is a measure of the
sensory processing. density of fibers between brain regions. The global dynamics of
the whole-brain model results from the mutual interactions of
The Role of Whole-Brain Computational Modeling for local node dynamics coupled through the underlying empirical
Binding Information anatomical structural connectivity matrix. Typically, the temporal
The diversity of computation has to be understood at the network dynamics of local brain areas in these models is taken to be either
level, but while intrinsic ignition is an important way to uncover the asynchronous (spiking models or their respective mean-field
computational role of a given node, this does not provide informa- reduction) or oscillatory (Deco and Kringelbach, 2014). Recently,
tion about what happens when this node is eliminated. This com- evidence has emerged for a promising version using, for each

Neuron 94, June 7, 2017 965

 Neuron

Perspective

Figure 4. Using Whole-Brain Models to Determine Causal Functional Roles of Nodes

(A) Causal whole-brain methods can be used to investigate the effects on the computational diversity fitting the model to the empirical data.
(B) The removal or perturbation of the brain regions with the highest level of intrinsic ignition can provide causal information on the functional importance
of individual brain regions in terms of how the profiles of intrinsic ignition might change, similar to how cutting off the head of a hydra may sprout many more
heads. As such, this makes it possible to investigate the change of intrinsic ignition of individual nodes across different brain states and in neuropsychiatric
disorders.

brain area, a local dynamical model given by a normal form of bi- regions with the highest level of intrinsic ignition (see Figure 4).
furcations (e.g., a supercritical Hopf bifurcation) (Deco et al., This would provide information on the functional importance
2016; Kringelbach et al., 2015). The normal form of a Hopf bifur- of individual brain regions. As such, it would also be possible
cation can describe the transition from asynchronous noisy to investigate the change of intrinsic ignition of individual nodes
behavior to full oscillations and thus unify previous asynchronous across different brain states (Figure 4B). It would also be of
and full oscillatory scenarios. The main parameter that is manip- high interest to use direct brain stimulation of such binding
ulated for fitting the empirical data and for analyzing the model is nodes in patients undergoing neurosurgical investigations to
the so-called global coupling parameter G. The global coupling predict and investigate their causal role in cognition (Parvizi
parameter G corresponds to the conductivity of the synaptic et al., 2013).
connections, which is considered here, for simplicity, uniform
across the brain. Our preliminary results indicate that the working Conclusion
point of a whole-brain computational model of the resting state In this Perspective, we have provided evidence for a powerful
corresponds to where there is maximal hierarchical processing novel framework for understanding the underlying basis of neu-
and thus maximal entropy. Further preliminary results suggest ral communication and network organization. The concept of
that this is not the case for other dynamic brain states, such as intrinsic ignition provides a simple way to investigate the compu-
deep sleep or drug-altered states. As such, this could provide tational spatiotemporal diversity of network nodes and provides
novel insights into the dynamic processing of hierarchy in direct evidence on the hierarchical structuring of information
different brain states. processing in the network. This method can be applied to multi-
Similar to our recent work, to the effect of removing so-called modal neuronal data; from the single neuron to local field poten-
‘‘binding nodes’’ (Deco et al., 2017), future studies could use tials, MEG, and even fMRI. Combining this data-driven method
causal whole-brain methods to investigate the effects on the with a causal whole-brain model can provide strong evidence
computational diversity of the removal or perturbation of the brain on the functional role of a network node by measuring the

966 Neuron 94, June 7, 2017

Neuron

Perspective

consequences of its elimination. As such, this is a powerful new Deco, G., Kringelbach, M.L., Jirsa, V., and Ritter, P. (2016). The dynamics of
resting fluctuations in the brain: metastability and its dynamical core. bioRxiv.
method that can provide strong evidence on brain computation http://dx.doi.org/10.1101/065284.
on many levels from the single neuron to groups of neurons to
the whole-brain level. Deco, G., Van Hartevelt, T.J., Fernandes, H.M., Stevner, A., and Kringelbach,
M.L. (2017). The most relevant human brain regions for functional connectivity:
There are thus many important implications for using this Evidence for a dynamical workspace of binding nodes from whole-brain
framework to understand hitherto poorly understood problems computational modelling. Neuroimage 146, 197–210.
such as altered states of consciousness. But perhaps most
Dehaene, S., and Changeux, J.P. (2011). Experimental and theoretical ap-
importantly, it provides new tools for understanding the imbal- proaches to conscious processing. Neuron 70, 200–227.
ances in functional organization found in neuropsychiatric disor-
ders, and, when combined with whole-brain models, may even Dehaene, S., Kerszberg, M., and Changeux, J.P. (1998). A neuronal model of a
global workspace in effortful cognitive tasks. Proc. Natl. Acad. Sci. USA 95,
provide insights into novel ways of rebalancing the whole-brain 14529–14534.
networks in health and disease.
Ferrarelli, F., Massimini, M., Sarasso, S., Casali, A., Riedner, B.A., Angelini, G.,
Tononi, G., and Pearce, R.A. (2010). Breakdown in cortical effective connectiv-
ACKNOWLEDGMENTS ity during midazolam-induced loss of consciousness. Proc. Natl. Acad. Sci.
USA 107, 2681–2686.
G.D. is supported by the ERC Advanced Grant DYSTRUCTURE (n. 295129), by
the Spanish Research Project PSI2016-75688-P, and by the European Union’s Grinvald, A., Lieke, E., Frostig, R.D., Gilbert, C.D., and Wiesel, T.N. (1986).
Horizon 2020 research and innovation programme under grant agreement n. Functional architecture of cortex revealed by optical imaging of intrinsic sig-
nals. Nature 324, 361–364.
720270 (HBP SGA1). M.L.K. is supported by the ERC Consolidator Grant:
CAREGIVING (n. 615539) and Center for Music in the Brain, funded by the
Hodgkin, A.L., and Huxley, A.F. (1952). A quantitative description of membrane
Danish National Research Foundation (DNRF117). current and its application to conduction and excitation in nerve. J. Physiol.
117, 500–544.
REFERENCES
Hubel, D.H., and Wiesel, T.N. (1959). Receptive fields of single neurones in the
cat’s striate cortex. J. Physiol. 148, 574–591.
Amit, D.J., and Brunel, N. (1997). Model of global spontaneous activity and
local structured activity during delay periods in the cerebral cortex. Cereb.  lu, F.I., and Van De Ville, D. (2015). Transient brain activity disentan-
Karahanog
Cortex 7, 237–252.
gles fMRI resting-state dynamics in terms of spatially and temporally overlap-
Baars, B.J. (1989). A Cognitive Theory of Consciousness (Cambridge Univer- ping networks. Nat. Commun. 6, 7751.
sity Press).
Karahanog  lu, F.I., Caballero-Gaudes, C., Lazeyras, F., and Van de Ville, D.
Biswal, B., Yetkin, F.Z., Haughton, V.M., and Hyde, J.S. (1995). Functional (2013). Total activation: fMRI deconvolution through spatio-temporal regulari-
connectivity in the motor cortex of resting human brain using echo-planar zation. Neuroimage 73, 121–134.
MRI. Magn. Reson. Med. 34, 537–541.
Kenet, T., Bibitchkov, D., Tsodyks, M., Grinvald, A., and Arieli, A. (2003). Spon-
Borchers, S., Himmelbach, M., Logothetis, N., and Karnath, H.O. (2011). Direct taneously emerging cortical representations of visual attributes. Nature 425,
electrical stimulation of human cortex - the gold standard for mapping brain 954–956.
functions? Nat. Rev. Neurosci. 13, 63–70.
Kringelbach, M.L., Jenkinson, N., Owen, S.L.F., and Aziz, T.Z. (2007). Transla-
Brunel, N., and Wang, X.J. (2001). Effects of neuromodulation in a cortical tional principles of deep brain stimulation. Nat. Rev. Neurosci. 8, 623–635.
network model of object working memory dominated by recurrent inhibition.
J. Comput. Neurosci. 11, 63–85. Kringelbach, M.L., McIntosh, A.R., Ritter, P., Jirsa, V.K., and Deco, G. (2015).
The rediscovery of slowness: exploring the timing of cognition. Trends Cogn.
Caballero Gaudes, C., Petridou, N., Francis, S.T., Dryden, I.L., and Gowland, Sci. 19, 616–628.
P.A. (2013). Paradigm-free mapping with sparse regression automatically de-
tects single-trial functional magnetic resonance imaging blood oxygenation Markov, N.T., Ercsey-Ravasz, M., Van Essen, D.C., Knoblauch, K., Toroczkai,
level dependent responses. Hum. Brain Mapp. 34, 501–518. Z., and Kennedy, H. (2013). Cortical high-density counterstream architectures.
Science 342, 1238406.
Cabral, J., Kringelbach, M.L., and Deco, G. (2014). Exploring the network
dynamics underlying brain activity during rest. Prog. Neurobiol. 114, 102–131. Markov, N.T., Vezoli, J., Chameau, P., Falchier, A., Quilodran, R., Huissoud, C.,
Lamy, C., Misery, P., Giroud, P., Ullman, S., et al. (2014). Anatomy of hierarchy:
Calford, M.B., Wang, C., Taglianetti, V., Waleszczyk, W.J., Burke, W., and feedforward and feedback pathways in macaque visual cortex. J. Comp. Neu-
Dreher, B. (2000). Plasticity in adult cat visual cortex (area 17) following circum- rol. 522, 225–259.
scribed monocular lesions of all retinal layers. J. Physiol. 524, 587–602.
Massimini, M., Ferrarelli, F., Huber, R., Esser, S.K., Singh, H., and Tononi, G.
Casali, A.G., Gosseries, O., Rosanova, M., Boly, M., Sarasso, S., Casali, K.R.,
(2005). Breakdown of cortical effective connectivity during sleep. Science
Casarotto, S., Bruno, M.A., Laureys, S., Tononi, G., and Massimini, M. (2013).
309, 2228–2232.
A theoretically based index of consciousness independent of sensory pro-
cessing and behavior. Sci. Transl. Med. 5, 198ra105.
Mountcastle, V.B. (1957). Modality and topographic properties of single neu-
Deco, G., and Kringelbach, M.L. (2014). Great expectations: using whole-brain rons of cat’s somatic sensory cortex. J. Neurophysiol. 20, 408–434.
computational connectomics for understanding neuropsychiatric disorders.
Neuron 84, 892–905. Moutard, C., Dehaene, S., and Malach, R. (2015). Spontaneous fluctuations
and non-linear ignitions: two dynamic faces of cortical recurrent loops. Neuron
Deco, G., and Rolls, E.T. (2005). Attention, short-term memory, and action se- 88, 194–206.
lection: a unifying theory. Prog. Neurobiol. 76, 236–256.
Parvizi, J., Rangarajan, V., Shirer, W.R., Desai, N., and Greicius, M.D. (2013).
Deco, G., Jirsa, V.K., and McIntosh, A.R. (2011). Emerging concepts for the The will to persevere induced by electrical stimulation of the human cingulate
dynamical organization of resting-state activity in the brain. Nat. Rev. Neuro- gyrus. Neuron 80, 1359–1367.
sci. 12, 43–56.
Petridou, N., Gaudes, C.C., Dryden, I.L., Francis, S.T., and Gowland, P.A.
Deco, G., Tononi, G., Boly, M., and Kringelbach, M.L. (2015). Rethinking segre- (2013). Periods of rest in fMRI contain individual spontaneous events which
gation and integration: contributions of whole-brain modelling. Nat. Rev. Neu- are related to slowly fluctuating spontaneous activity. Hum. Brain Mapp. 34,
rosci. 16, 430–439. 1319–1329.

Neuron 94, June 7, 2017 967

 Neuron

Perspective
Raichle, M.E., MacLeod, A.M., Snyder, A.Z., Powers, W.J., Gusnard, D.A., and Sporns, O., Tononi, G., and Kötter, R. (2005). The human connectome: a struc-
Shulman, G.L. (2001). A default mode of brain function. Proc. Natl. Acad. Sci. tural description of the human brain. PLoS Comput. Biol. 1, e42.
USA 98, 676–682.
Tagliazucchi, E., Balenzuela, P., Fraiman, D., and Chialvo, D.R. (2012). Criti-
Rosanova, M., Gosseries, O., Casarotto, S., Boly, M., Casali, A.G., Bruno, cality in large-scale brain FMRI dynamics unveiled by a novel point process
M.A., Mariotti, M., Boveroux, P., Tononi, G., Laureys, S., and Massimini, M. analysis. Front. Physiol. 3, 15.
(2012). Recovery of cortical effective connectivity and recovery of conscious-
ness in vegetative patients. Brain 135, 1308–1320. Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O.,
Delcroix, N., Mazoyer, B., and Joliot, M. (2002). Automated anatomical labeling
of activations in SPM using a macroscopic anatomical parcellation of the MNI
Selimbeyoglu, A., and Parvizi, J. (2010). Electrical stimulation of the human MRI single-subject brain. Neuroimage 15, 273–289.
brain: perceptual and behavioral phenomena reported in the old and new liter-
ature. Front. Hum. Neurosci. 4, 46. van Hartevelt, T.J., Cabral, J., Møller, A., FitzGerald, J.J., Green, A.L., Aziz,
T.Z., Deco, G., and Kringelbach, M.L. (2015). Evidence from a rare case study
Smith, S.M., Fox, P.T., Miller, K.L., Glahn, D.C., Fox, P.M., Mackay, C.E., Fil- for Hebbian-like changes in structural connectivity induced by long-term deep
ippini, N., Watkins, K.E., Toro, R., Laird, A.R., and Beckmann, C.F. (2009). Cor- brain stimulation. Front. Behav. Neurosci. 9, 167.
respondence of the brain’s functional architecture during activation and rest.
Proc. Natl. Acad. Sci. USA 106, 13040–13045. Wang, X.J. (2002). Probabilistic decision making by slow reverberation in
cortical circuits. Neuron 36, 955–968.
Sporns, O. (2014). Contributions and challenges for network models in cogni-
tive neuroscience. Nat. Neurosci. 17, 652–660. Watts, D.J., and Strogatz, S.H. (1998). Collective dynamics of ‘small-world’
networks. Nature 393, 440–442.
Sporns, O., and Zwi, J.D. (2004). The small world of the cerebral cortex. Neuro- Winawer, J., and Parvizi, J. (2016). Linking electrical stimulation of human pri-
informatics 2, 145–162. mary visual cortex, size of affected cortical area, neuronal responses, and sub-
jective experience. Neuron 92, 1213–1219.
Sporns, O., Tononi, G., and Edelman, G.M. (2000). Theoretical neuroanatomy:
relating anatomical and functional connectivity in graphs and cortical connec- Zhang, D., and Raichle, M.E. (2010). Disease and the brain’s dark energy. Nat.
tion matrices. Cereb. Cortex 10, 127–141. Rev. Neurol. 6, 15–28.

968 Neuron 94, June 7, 2017