The salience network is responsible for switching between the default mode network and the central executive network: Replication from DCM
Introduction
The development of many new analysis methods for functional neuroimaging data such as functional connectivity, independent component analysis (ICA) and effective connectivity methods including dynamic causal modelling (DCM) has lead to a lot of work focused on brain network activity and connectivity. An important ability of the human brain is to be able to rapidly switch between different tasks; it is therefore of interest to study interactions between different networks of the brain to better understand the mechanism behind switching between different tasks.
Dynamic causal modelling (DCM) as a method was introduced for effective connectivity analyses in 2003 (Friston et al., 2003). Typically several hypothesised models are specified and Bayesian model selection (BMS) is used to infer the model which represents the best fit to the data (Penny et al., 2004). Recent advances in methodology mean it is now possible to compare families of DCMs (Penny et al., 2010), to use a random effects method for model selection (Stephan et al., 2009) and to specify nonlinear and stochastic models (Li et al., 2011, Stephan et al., 2008). The introduction of nonlinear modelling means it is now possible to test for modulations of regions on connections i.e. how regions influence connection strengths between other brain regions, providing a more realistic model of brain physiology. Stochastic modelling means that the input into the model is not deterministic, more accurately accounting for noise (Daunizeau et al., 2012) and allowing the application of DCM to resting state data, which was not previously thought possible. Stochastic DCMs also include random fluctuations in the state equations that better accounts for spontaneous neuronal fluctuations. Conventional DCM, in contrast, uses deterministic differential equations, and as a result does not account for spontaneous neuronal fluctuations or state noise.
Another method that has been used extensively to examine brain networks is independent component analysis (ICA) (Bell and Sejnowski, 1995, Comon, 1994, Hyvarinen and Oja, 2000). Whilst it is possible to apply both spatial and temporal ICA, we are using spatial ICA for our analysis. ICA of fMRI data separates the data into spatially independent patterns of activity and therefore can identify brain networks engaged in a task without the use of a predefined model. ICA has also been applied to resting state data (Lee et al., 2012) and has shown differences in functional network connectivity with age (Stevens et al., 2009) and in patients with schizophrenia (Yu et al., 2011). Resting state networks found in this fashion have been shown to be reproducible, with many networks having a high interclass correlation coefficient in a repetition experiment (Chou et al., 2012, Damoiseaux et al., 2006, Shehzad et al., 2009). In order to overcome difficulties in identifying components of interest and in determining the optimum number of components in an ICA analysis, it is now possible to spatially constrain an ICA analysis to provide the components of interest (Lin et al., 2010) by supplying templates of the networks of interest. This is particularly useful if the research question relates to specific networks of the brain.
Sridharan et al. (2008) recently used Granger causality to examine the relationship between different networks. One of the networks they studied is the default mode network (DMN). This network is consistently observed in ICA analyses of resting state data and task deactivation studies (Beckmann et al., 2005, Raichle et al., 2001). The DMN comprises the posterior cingulate (BA 23 and 31), posterior parietal cortex (BA 7, 39 and 40) and the ventromedial prefrontal cortex (Buckner et al., 2008). The central executive network (CEN) comprises the dorsolateral prefrontal cortex and posterior parietal cortex and is engaged when a cognitively demanding task requiring attention is being performed (Fox et al., 2006). The third network is the salience network (SN) which includes the ventrolateral prefrontal cortex (VLPFC) and anterior insula (jointly referred to as the fronto-insular cortex; FIC) and the anterior cingulate cortex (ACC) (Seeley et al., 2007). The salience network responds to the degree of subjective salience, whether cognitive, homeostatic, or emotional. The networks are displayed in Fig. 1, where the DMN is shown in red, the SN in blue and the CEN in green. Based on this work and others, a model has been proposed of the function of the insula, including bottom-up detection of salient events, switching between other networks to facilitate access to attention and working memory upon detection of a salient event, interaction of the anterior and posterior insula for autonomic reactivity to salient stimuli and strong functional coupling with the anterior cingulate to facilitate rapid access to the motor system (Menon and Uddin, 2010).
Recently relationships between brain networks have been increasingly studied with the advent of methods such as functional network connectivity (Jafri et al., 2008), which examines temporal relations between components. We wanted to explore whether it would be possible to use DCM to examine the relationship between different ICA components from resting state data, and therefore examine the relationship between different networks of the brain, specifically network switching.
The aim of this work is to further examine the network switching demonstrated by Sridharan et al. (2008), and to do so by applying a novel technique using nonlinear DCM with stochastic modelling of ICA components from resting state data. In order to demonstrate repeatability of our findings we also apply this analysis to an independent dataset.
Section snippets
Methods
We used two different datasets for this analysis. The first dataset was acquired at Bangor University, and the second was from an open access database.
Independent component analysis
The results of the constrained ICA analysis for the Bangor University data are displayed in Fig. 3. Component 1 (top left) is the CEN, component 2 (top right) the DMN and component 3 (bottom left) the SN. In order to test how well our extracted components represent the networks we correlated our binary masks used in the constrained ICA analysis with the extracted component to test how many voxels correspond in the mask and result. Performing a correlation with the CEN mask used in the analysis
Discussion
This work has confirmed the finding of Sridharan et al. (2008) that the SN is key for switching between the CEN and the DMN. We have obtained this result using resting state data; Sridharan and colleagues saw this effect with resting state data and also with task data. The novelty of our work is in demonstrating the same effect using a different modelling technique, DCM as opposed to Granger causality, further validating the role of the SN in switching between CEN and DMN. Our novel approach,
Acknowledgments
The authors would like to acknowledge support from the NeuroSKILL project, a joint Welsh/Irish partnership, part funded by the European Regional Development Fund through the Ireland Wales Programme 2007–13.
References (47)
- et al.
Intuition and autism: a possible role for Von Economo neurons
Trends Cogn. Sci.
(2005) A fast diffeomorphic image registration algorithm
NeuroImage
(2007)Independent component analysis, a new concept
Signal Process.
(1994)- et al.
Stochastic dynamic causal modeling of fMRI data: should we care about neural noise
NeuroImage
(2012) - et al.
A core system for the implementation of task sets
Neuron
(2006) - et al.
Dynamic causal modelling
NeuroImage
(2003) - et al.
Independent component analysis: algorithms and applications
Neural Netw.
(2000) - et al.
A method for functional network connectivity among spatially independent resting-state components in schizophrenia
NeuroImage
(2008) - et al.
Generalised filtering and stochastic DCM for fMRI
NeuroImage
(2011) Large-scale brain networks and psychopathology: a unifying triple network model
Trends Cogn. Sci.
(2011)
Comparing dynamic causal models
NeuroImage
Bayesian model selection for group studies – revisited
NeuroImage
Network modelling methods for FMRI
NeuroImage
Nonlinear dynamic causal models for fMRI
Neuroimage.
Bayesian model selection for group studies
NeuroImage
Dendritic architecture of the von Economo neurons
Neuroscience
Aberrant salience network (bilateral insula and anterior cingulate cortex) connectivity during information processing in schizophrenia
Schizophr. Res.
Functional network connectivity during rest and task: comparison of healthy controls and schizophrenic patients
Conf. Proc. IEEE Eng. Med. Biol. Soc.
Functional network connectivity during rest and task conditions: a comparative study
Hum. Brain Mapp
Investigations into resting-state connectivity using independent component analysis
Philos. Trans. R. Soc. Lond. B Biol. Sci.
An information maximization approach to blind separation and blind deconvolution
Neural Comput.
Salience network integrity predicts default mode network function after traumatic brain injury
Proc. Natl. Acad. Sci. U. S. A.
The brain's default network: anatomy, function, and relevance to disease
Ann. N. Y. Acad. Sci.
Cited by (494)
Individualized prediction of cognitive test scores from functional brain connectome in patients with first-episode late-life depression
2024, Journal of Affective DisordersAbnormal effective connectivity of reward network in first-episode schizophrenia with auditory verbal hallucinations
2024, Journal of Psychiatric ResearchMental gravity: Modelling the embodied self on the physical environment
2024, Journal of Environmental PsychologyPrenatal Exposure to Maternal Mood Entropy Is Associated With a Weakened and Inflexible Salience Network in Adolescence
2024, Biological Psychiatry: Cognitive Neuroscience and Neuroimaging