The most relevant human brain regions for functional connectivity: Evidence for a dynamical workspace of binding nodes from whole-brain computational modelling
Introduction
Every conscious human experience of perceptions, memories and emotions relies on the flexible integration and segregation of information (Baars, 1989, Dehaene et al., 1998, Tononi et al., 1998). Importantly, this information integration is above and beyond the information available from the sum of its parts (Balduzzi and Tononi, 2008, Griffith and Koch, 2012), and as such has been linked to consciousness (but can also proceed without awareness, Mudrik et al., 2014).
The integration of information is likely to take place in a functionally coherent, yet distributed network of brain regions, where computations are highly segregated (Power et al., 2011, Sporns, 2013). However, these computations must also be integrated globally and over time (Hansen et al., 2015). Large-scale projects such as the Human Connectome Project (Van Essen et al., 2012) and Human Brain Project (Gerstner et al., 2012) have begun to map the brain structurally and functionally, providing the experimental tools to aid a deeper understanding of how the brain integrates and segregates relevant information.
These projects use primarily non-invasive neuroimaging methods to describe the human connectome (Sporns et al., 2005) as the complete map of the brain's neural elements and their structural interactions that allow complex integration and segregation of relevant information (Sporns, 2013). Structural anatomical connectivity can be precisely extracted by diffusion weighted/tensor imaging (DWI/DTI) measuring the white-matter fiber tracts constrained by the diffusion of water molecules (Basser and Pierpaoli, 1996, Beaulieu, 2002), from which the connectivity can be reconstructed by deterministic or probabilistic tract-tracing methods (Hagmann et al., 2010, Johansen-Berg and Rushworth, 2009). Similarly, functional information can be mapped in vivo in humans on the scale of millimetres by magnetic resonance imaging (MRI) at a temporal resolution of seconds or by magnetoencephalography (MEG) allowing a much finer temporal resolution of milliseconds. Typically, functional neuroimaging studies have measured task-related activity, but in the past decade studies have started to measure spontaneous resting-state activity over several minutes (Snyder and Raichle, 2012). These resting state studies have found highly reproducible and organized patterns of brain activity (Damoiseaux et al., 2006, Greicius et al., 2003), which overlap with task-related activity patterns (Fox and Raichle, 2007, Smith et al., 2009).
Mapping the human connectome has identified some of the crucial structural topological features of human brain architecture relevant for the functional integration and segregation of information. Graph-theoretical measures have shown that the brain is structured as a small-world network (Watts and Strogatz, 1998) around a large number of spatially distributed network communities with clustered connectivity in which the local computations are likely to be highly segregated (Power et al., 2011, Sporns, 2013) (although see Markov et al., 2013). Furthermore, network hubs linking those network communities could potentially serve to ensure efficient communication and information integration (Bullmore and Sporns, 2009). Notably, some hubs show rich, dense interconnectivity and form a central core of structural connectivity (static domain) or ‘rich club’ that has been suggested to play an important role for global brain integration (Van Boven and Loewenstein, 2003, van den Heuvel and Sporns, 2011). This raises the question whether the brain regions from the rich club are sufficient for optimal information flow over time. Misic et al. (2014) have shown that the hippocampus is a critical convergence zone for information flow, despite its modest degree profile. This suggests that a region that is not densely structurally connected could still be crucial for information flow. In this paper we suggest a new degree-naïve measure to describe the functional relevance of brain regions in dynamically integrating information over time.
In order to gain more insight into functional connectivity and network measures, significant progress has been made using whole-brain computational models. Such models have shown to be able to reflect and reproduce much of the dynamics and complexity of the real brain. The models typically use various mesoscopic top-down approximations of brain complexity with dynamical networks of local brain area attractor networks (Cabral et al., 2014a), where the more advanced models use a dynamic mean-field model derived from a proper reduction of a detailed spiking neuron model (Deco et al., 2013b). Furthermore, the dynamics of whole-brain models rely on reducing the complexity of brain networks by using a given macro scale brain parcellation, which historically has been carried out based on careful studies of the properties of the underlying brain tissue (Zilles and Amunts, 2010), which has been supplemented with modern neuroimaging parcellations that typically range from tens to several hundreds of regions (Craddock et al., 2013). A current popular choice for whole brain parcellations is the automated anatomical labeling (AAL) parcellation with 116 regions in the cerebellum, cortical and subcortical regions (Tzourio-Mazoyer et al., 2002). Whole-brain computational models have been able to provide a mechanistic explanation of the origin of resting-state networks, as e.g. shown for resting-state networks derived from resting-state MRI data (Deco and Jirsa, 2012, Honey et al., 2007) and for resting-state networks derived from MEG data (Cabral et al., 2014b). Such models have been successfully used to show that both spontaneous and task-related neural activity are strongly dependent on the properties of the underlying structural connectivity and the dynamical working point, where the working point refers to the oscillatory and bifurcation properties of any given node (Deco and Corbetta, 2011).
The optimal functioning of the brain in terms of information processing depends on its ability to balance the amount of spatial segregation and integration of relevant information (Deco and Kringelbach, 2014, Sporns, 2013). Although information capability (entropy) and integration are not directly related to brain function, both measures represent fundamental aspects of brain organization and are indeed related to the exploration of the dynamical repertoire (Deco and Kringelbach, 2016, Tononi et al., 1994). Moreover, emergent research has revealed that the richness of that exploration is related to task activity (Palva and Palva, 2012). Hence, measures of information capability and integration can be used to characterise the richness of the functional dynamics.
Here we used whole-brain computational modelling to clarify the functional role of the rich club and the functional impact on resting state activity. First, we compared this topological measure to a novel dynamical measure of the temporal binding of information, which we call the dynamical workspace of binding nodes. It is proposed that evaluation of temporal binding reveals which regions within the network are more integrative, or binding, across both space (over spatial segregated brain regions) and time (as defined by the grand average of functional activity over time, carefully described in Deco et al. (2015)). This measure is related to the binding problem – that is how distributed information is bound and made available for awareness and consciousness (Crick and Koch, 1990). We compared the top ranked brain regions belonging to the rich club and to the workspace of binding nodes, and used whole-brain computational modelling to measure the impact of removing the members of either clubs on the integration and information capabilities of the human brain. This direct comparison is included to illustrate the conceptual difference of the dynamical workspace of binding nodes as a measure capable of finding functionally relevant nodes that may not exhibit high structural connectivity.
Section snippets
Ethics
This study was approved by the internal research board at CFIN, Aarhus University, Denmark. Ethics approval was granted by the Research Ethics Committee of the Central Denmark Region (De Videnskabsetiske Komitéer for Region Midtjylland). Written informed consent was obtained from all participants prior to participation.
Overall analysis pipeline
In order to determine the impact of brain regions on functional dynamics, we analyzed the functional and structural connectivity of neuroimaging data together with whole-brain
Results
The main aim was to study the dynamical relevance of each brain region for the integration and segregation of information in the brain over time. In order to do this we combined functional and structural human neuroimaging data with whole-brain computational modelling, to investigate the role of brain regions in binding functional connectivity (see Experimental procedures). As shown below in detail, we implemented a new measure of temporal binding of information in the brain, which is then used
Discussion
We used whole-brain computational modelling to address a key question in neuroscience: namely what brain regions are most important for integration of information processing. While some progress has been made through defining topological measures such as the so-called rich club, the functional consequences of these brain regions have not been assessed and in particular the importance of time in brain processing has not been fully explored. Here we used whole-brain computational modelling on
Conflict of interest
The authors declare to have no conflict of interest.
Acknowledgements
GD was supported by the ERC Advanced Grant: DYSTRUCTURE (n. 295129), by the Spanish Research Project SAF2010-16085 and the FP7-ICT BrainScales. MLK was supported by the ERC Consolidator Grant: CAREGIVING (n. 615539), TrygFonden Charitable Foundation and by Center for Music in the Brain, funded by the Danish National Research Foundation (DNRF117).
References (63)
- et al.
Microstructural and physiological features of tissues elucidated by quantitative-diffusion-tensor MRI
J. Magn. Reson. B
(1996) - et al.
Probabilistic diffusion tractography with multiple fibre orientations: what can we gain?
Neuroimage
(2007) - et al.
Exploring the network dynamics underlying brain activity during rest
Prog. Neurobiol.
(2014) - et al.
Exploring mechanisms of spontaneous functional connectivity in MEG: how delayed network interactions lead to structured amplitude envelopes of band-pass filtered oscillations
Neuroimage
(2014) - et al.
Resting brains never rest: computational insights into potential cognitive architectures
Trends Neurosci.
(2013) - et al.
Great expectations: using whole-brain computational connectomics for understanding neuropsychiatric disorders
Neuron
(2014) - et al.
Metastability and coherence: extending the communication through coherence hypothesis using a whole-brain computational perspective
Trends Neurosci.
(2016) - et al.
Temporal binding and the neural correlates of sensory awareness
Trends Cogn. Sci.
(2001) - et al.
MR connectomics: principles and challenges
J. Neurosci. Methods
(2010) - et al.
Functional connectivity dynamics: modeling the switching behavior of the resting state
Neuroimage
(2015)
Improved optimization for the robust and accurate linear registration and motion correction of brain images
Neuroimage
The rediscovery of slowness: exploring the timing of cognition
TICS
Discovering oscillatory interaction networks with M/EEG: challenges and breakthroughs
Trends Cogn. Sci.
Functional network organization of the human brain
Neuron
Complex network measures of brain connectivity: uses and interpretations
Neuroimage
A brief history of the resting state: the Washington University perspective
Neuroimage
Network attributes for segregation and integration in the human brain
Curr. Opin. Neurobiol.
Complexity and coherency: integrating information in the brain
Trends Cogn. Sci.
Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain
Neuroimage
Network hubs in the human brain
Trends Cogn. Sci.
The Human Connectome Project: a data acquisition perspective
Neuroimage
A Cognitive Theory of Consciousness
Integrated information in discrete dynamical systems: motivation and theoretical framework
PLoS Comput. Biol.
The basis of anisotropic water diffusion in the nervous system - A technical review
NMR Biomed.
Probabilistic independent component analysis for functional magnetic resonance imaging
IEEE Trans. Med. Imaging
Characterization and propagation of uncertainty in diffusion-weighted MR imaging
Magn. Reson. Med.
Complex brain networks: graph theoretical analysis of structural and functional systems
Nat. Rev. Neurosci.
Automatic 3D intersubject registration of MR volumetric data in standardized Talairach space
J. Comput. Assist. Tomogr.
Imaging human connectomes at the macroscale
Nat. Methods
Towards a neurobiological theory of consciousness
Semin. Neurosci.
Cited by (38)
Complex of global functional network as the core of consciousness
2023, Neuroscience ResearchCitation Excerpt :The dynamic workspace of the binding elements is defined by a measure of entropy linked to the binding problem, i.e., the problem of how distributed information can be bound and can become available for consciousness and awareness (Crick and Koch, 1990). When removing the dynamic workspace of binding elements from the system, the integration and information capabilities of the brain critically decrease (Deco et al., 2017). The core of the networks detected in their study is consistent with our findings in that it included association areas in the frontal, parietal, and temporal lobes, but it differs in that it includes primary visual areas.
From brain-body function to conscious interactions
2022, Neuroscience and Biobehavioral ReviewsCitation Excerpt :Finally, this multilayer framework presents several advantages: (i) it may integrate different dimensions of interactions accounting for the embodiment of conscious experience (Thompson and Varela, 2001), (ii) it is a dynamic approach, accounting for the rich spatiotemporal structure of consciousness (Deco et al., 2015; Deco et al., 2017; Ipiña et al., 2020), (iii) it is pragmatic framework already applied to relevant data, giving us new insight on the complex intertwined within brain systems and their relationship with consciousness (Deco et al., 2018b; Kringelbach et al., 2020), (iv) it offers simple concepts to reason about dynamical couplings and relevant systems (Signorelli and Meling, 2021; Signorelli and Boils, 2021), (v) it inspires computational model extensions to brain and body interactions and motivates multimodal experimental paradigms, since new data sets are required to test the major hypotheses, (vi) it comes with a concrete mathematics from which further perspectives may exploit their current implementations in other fields (Kivela et al., 2014; Boccaletti et al., 2014), and eventually, vii) the semantic of multilayer networks interacting has the potential to integrate other models of consciousness which are based on, implicitly or explicitly, graph and network theory.
Time-Varying Functional Connectivity Decreases as a Function of Acute Nicotine Abstinence
2021, Biological Psychiatry: Cognitive Neuroscience and NeuroimagingStatistical complexity and connectivity relationship in cultured neural networks
2019, Chaos, Solitons and FractalsPerturbation of whole-brain dynamics in silico reveals mechanistic differences between brain states
2018, NeuroImageCitation Excerpt :Previous studies have used whole-brain computational models to simulate the effects of structural lesions, i.e. by removing links or nodes, and studying its impact in whole-brain dynamics (Cabral et al., 2012a; Vasa et al., 2015; Aerts et al., 2016; Deco et al., 2017c). In particular, we have recently shown that the removal of specific ‘binding’ regions impacts subsequent brain activity, specifically in terms of integration and information encoding capability (Deco et al., 2017c). Rather than removing regions, we perturbed regions by making them either more oscillatory or more noisy and measured the recovery of the system's global integration.