Notes on the brain’s functional networks using the graph-theoretic approach

Brain’s Functional Networks

➤Fronto-Parietal Network
➤Dorsal Attention Network
➤Salience Network
➤Cingulo-Opercular Network
➤Ventral Attention Network
➤Default Mode Network
➤Auditory Network
➤Visual Network
➤➤Somatosensory-Motor Network
➤Subcortical Network

3 additional networks, including…

➤Memory retrieval Network?~
➤Post-retrieval monitoring Network?

Brain Regions and Functions

.

.

Fronto-Parietal Network

• Anterior dorso-lateral prefrontal cortex (aDLPFC)
• Intra-parietal sulcus IIPS) and inferior parietal lobule (IPL)
• Middle cingulate cortex (MCC)
• Rostral inferior temporal cortex (rITC)

Functions:  Top ­down signals for current task goals exert control by flexibly biasing information flow across multiple large-­scale functional networks overcoming conflict from previous habits. Also allows for novel task control. Part of the ‘task positive’ cognitive control network (CCN).

Dorsal Attention Network 

• pDLPFC / frontal eye fields
• Posterior parietal cortex: Superior parietal lobule (SPL) / Intra-parietal sulcus (IPS)
• rITC (above FPN region)

Functions: Selective attention.

Ventral Attention Network (VAN)

• Ventro-lateral prefrontal cortex – middle frontal gyrus (MFG) and inferior frontal gyrus (IFG)
• Temporal parietal junction (TPJ) – inferior parietal lobule (IPL) and superior temporal gyrus (STG)

Functions: Bottom-up attentional processing.

External Attention System (EAS) : DAN and VAN

.

Functions: Control of attention through flexible interaction between both systems enables the dynamic control of attention in relation to top-down goals and bottom-up sensory stimulation.  Part of the ‘task positive’ cognitive control network (CCN).

Salience Network (SN)

• Dorsal anterior cingulate cortex (dACC)
• Pre supplementary motor area (preSMA)
• Anterior Insular cortex
• Temporal pole
• Sublenticular extended amygdala (SLEA) – made up of the amygdaloid nuclei, sublenticular nuclei, and the nucleus accumbens
• SN/VTA, substantia nigra/ventral tegmental area

.

Functions: Maintains a stable ‘saliency’ or priority map of the visual environment – including surprising stimuli, and stimuli that are pleasurable and rewarding, self-relevant, or emotionally engaging (both appetitive and aversive such as threats).

Cingulo-Opercular Network (CON)

• DLPFC
• Anterior Insula
• Dorsal anterior cingulate cortex (dACC)
• Thalamus

Functions: Vigilance and sustained attention. Tonic alertness for working memory. Set maintenance in working memory related tasks. Response override  after conflict detection.

Default Mode Network

• Medial prefrontal cortex (mPFC)
• Superior prefrontal gyrus (SPG)
• Lateral / inferior parietal cortex (LPC)
• Precuneus & Posterior cingulate cortex (PCC)
• Subgenual anterior cingulate cortex (sACC)
• Middle temporal gyrus (MTG)
• Inferior temporal cortex (IT)

Functions: Recall of the past (autobiographical memory) and imagination of the future, reflection on present mental states (esp. affective) and ‘mind-reading’ (social cognition).

Comparing functional systems with graph-theoretic (sub) networks

.

Task Positive and Task Negative Systems

Task Positive System

Also called the cognitive control network (CCN). This functional system includes portions of lateral prefrontal cortex (LPFC), posterior parietal cortex (PPC), anterior insula cortex and medial prefrontal cortex.

In Power et al’s graph-theoretic analysis it decomposes into three distinct sub-networks:

➤Fronto-Parietal

➤Dorsal Attention

➤Cingulo-Opercular

Task Negative / Default Mode System

➤Default Mode  

Andrews-Hanna and colleagues’ (2010) hub account of the Default Mode network

Applying the graph-theoretic approach to spontaneous brain activity also reveals that the default mode network comprises two subsystems that interact with a common core (looks a lot like a flexible hub as in the FP network).

• A midline core (posterior cingulate and anterior medial prefrontal cortex) is active when during self-relevant activity regardless of temporal context, and shares functional properties of both subsystems. It’s activity correlates with personal significance, introspection about one’s own mental states, and evoked emotion  This ‘core’ functions in a way similar to flexible hubs in Cole et al.’s theory of the fronto-parietal network (see below)..

• A dorsal medial prefrontal cortex subsystem (dMPFC, temporo-parietal junction (TPJ), lateral temporal cortex (LTC) and temporal pole (TempP). This is active when participants reflect on their present mental states, particularly affective states – and when participants infer the mental states of other people (social cognition). (It is also active when people make moral decisions?)

• A medial temporal lobe subsystem (ventral MPFC (vMPFC), posterior inferior parietal lobule (pIPL), retrosplenial cortex (Rsp), parahippocampal cortex (PHC), and hippocampal formation (HF+).) This network becomes active during recall of the past (autobiographical memory) and imagination of the future, and when when decisions involve constructing a mental scene based on memory.

The two subsystems interact when individuals are left to spontaneous thought -.typically freely wandering past recollection, future plans, and other (often emotionally laden) personal thoughts and experiences. There is also overlap with the FP network during experimentally- directed tasks emphasizing internal mentation such as autobiographical planning tasks.

Control Systems vs Processing Systems

Control Systems

Less well integrated on local scale but relatively connected to other functional systems; self-integration low, self-containment low. Flexibly adapts processing to a wide range of task sets.

➤Fronto-Parietal

The FPN is most active during the implementation of novel and non-­routine tasks (needed for fluid intelligence).

Processing Systems

Well integrated on local scale but relatively isolated from other functional systems; self-integration high, self-containment high

➤Default Mode  (although see second graph below)

➤Auditory

➤Visual

➤Somatosensory-Motor

. 

The Flexible Hub account of the FP network (FPN)

FPN is for adaptive task control.

FPN is capable of such functional adaptation because it is composed of flexible hubs: brain regions that flexibly and rapidly shift their brain­-wide functional connectivity patterns to implement cognitive control across a variety of tasks (Cole et al., 2013).

The flexible hub account builds on the guided activation framework, which itself derives from the biased competition account.

The guided activation framework (GAF) describes how top ­down signals originating in LPFC (representing current task goals) may implement cognitive control by biasing information flow across multiple large-­scale functional networks – thus current cognitive goals overcome conflict from previous habits.

The flexible hub theory builds on GAF to account for novel task control and by broadening this mechanism from just the LPFC to the entire FPN.

Two elements:

1. Global variable connectivity – brain regions of the network acting as ‘hubs’ flexibly shift their functional connectivity patterns with multiple brain networks across a wide variety of tasks.
2. Compositional coding  – a systematic relationship between connectivity patterns and task rules/operations allowing established representations to be re-used in novel contexts, allowing transfer of skills and knowledge across tasks. This is the principle behind transfer in n-back training.

“These mechanisms would likely allow the FPN to meaningfully contribute to a wide variety of task contexts by allowing rapid reconfiguration of information flow across multiple task-relevant networks via reuse of previously learned sets of connectivity patterns.”

The other cognitive control networks are proposed to contribute to a variety of tasks by implementing a number of distinct control processes, such as stable (rather than adaptive) task control and maintenance, conflict detection, arousal and salience, or spatial attention.

Anxiety

This functional network taxonomy is being used innovatively, as in Sylvester and colleague’s theory that anxiety disorders and high trait anxiety are associated with a particular pattern of functional network dysfunction: increased functioning of the cingulo-opercular and ventral attention networks as well as decreased functioning of the fronto-parietal and default mode networks.

---

Methods

The approach used for the data and models above is rs-fcMRI combined with graph-theory.

rs-fcMRI

Resting state functional connectivity (rs-fcMRI) measures spontaneous low-frequency fluctuations in blood oxygen level dependent (BOLD) signals in subjects at rest. This allows for measuring correlations in neural activity between distant brain regions. These correlations allow cognitive neuroscientists to non-invasively explore the functional network structure of the brain.

As a way of describing functional relationships in the brain this is an alternative to task-based approaches to identifying functional networks. This is based on studying spontaneous BOLD activity (blood-oxygen-level dependent contrast imaging).

Graph-theory

Brain activity on this approach is understood in terms of networks = graphs. Graphs are composed of a set of nodes and a set of pairwise ties between nodes.

A graph-theoretic framework can in principle can describe the entire brain network (e.g. small world measures of the entire graph), portions of the network (e.g. subgraphs) and individual nodes in the network (local efficiency) within a common framework.

The properties of a graph depend on how the nodes of the network are defined, so defining the nodes is critical to the enterprise.

Graphs used in the Power et al (2011) study

1. Areal graph of putative functional areas (see below) > 264 nodes.
2. A modified voxelwise network excluding short-distance correlations > 40,100 nodes.
3. A graph of ‘parcels’ from a popular brain atlas > 90 nodes.
4. A standard voxelwise graph > 40,100 nodes.

Graph type confirmation

Graph types (1) and (2) had subgraphs that were significantly more like known functional systems (e.g. dorsal attention system) than subgraphs in the atlas-based and standard voxelwise networks. And there was good between graph agreement.

Areal graph approach to graph definition

Different network definitions results in different network properties, with different consequences for the conclusions that can be drawn about the brain. The Power et al (2011) paper attempts to offer an approach that “more plausibly represents brain organization” – using an ‘areal graph’ method that uses best estimates of functional ‘units’ of the brain. This is not a voxel based way of defining functional areas, where large functional areas (made up of many voxels) will dominate in the graph over smaller voxel groupings, regardless of their roles in information processing.

Areal ROI definition

1. Meta-analysis of fMRI dataset for brain regions reliably active during tasks > 151 non-overlapping meta-analytic ROIs.
2. fc-Mapping (functional connectivity mapping) assesses correlations of cortical activity (BOLD) across whole (hemispheric) cortical sheets during eyes open fixation > 193 non-overlapping ROIs.
3. Areal ROI set formation – methods 1 and 2 merged (1 given preference) to form the areal set > 264 independent ROIs.

‘Functional systems’ definition

Decades of brain imaging (PET, fMRI) studies have defined functional systems as groups of brain regions that co-activate during certain types of task.

Traditionally these systems have also been called ‘networks’ (as in the ‘dorsal attention network’).

Revised ‘network’ definition

Power et al reserve ‘network’ for the graph theoretic sense. A network is a graph on this definition.

rs-fcMRI signal is highly correlated with traditionally defined networks/systems.

Subgraphs

Subgraph detection methods (different thresholds for each graph, using Infomap for a single data set) were applied to each of the 4 graphs, to break global networks into subnetworks of highly related nodes such that nodes in the subgraphs are more highly correlated/connected to one another than the rest of the graph.The subgraphs are derived from task-free data, with no prior information about node identity.

Power et al predicted that well-formed graphs would possess well-formed subgraphs corresponding to major functional systems of the brain.

The subgraphs of graph types (1) and (2) above, that were significantly more like known functional systems such as the dorsal attention system.

Subject cohorts for rs-fcMRI graph and subgraph formation

rs-fcMRI networks were studied in continuous eyes-open fixation data from two cohorts (N=53 and 52) of healthy young adults, matched for age and sex. For a collection of N ROIs in each subject, time-course are extracted for all ROIs and an NxN correlation matrix is calculated. An average matrix is formed across all subjects in a cohort which defines a weighted graph. Different thresholds are then applied to this matrix to determine different properties (e.g. subgraphs) of the network.

Weaknesses of approach

• The methods of locating supposed functional areas (‘nodes’) may have overlooked or fabricated areas.
• Resolution is limited to 3mm voxels.
• Only BOLD is used as a signal (there are not converging results with different signals). BOLD is known to have problems in measuring functional activity in temporal and orbitofrontal cortices, and thus much remains to be discovered about the organization of the ventral surface of the brain, as well as subcortical and cerebellar organisation.

---

References

Aboitiz, F., Ossandon, T., Zamorano, F., Palma, B., & Carrasco, X. (2014). Irrelevant stimulus processing in ADHD: catecholamine dynamics and attentional networks. Developmental Psychology, 5, 183. http://doi.org/10.3389/fpsyg.2014.00183

Andrews-Hanna, J. R., Reidler, J. S., Sepulcre, J., Poulin, R., & Buckner, R. L. (2010). Functional-Anatomic Fractionation of the Brain’s Default Network. Neuron, 65(4), 550–562. http://doi.org/10.1016/j.neuron.2010.02.005

Cole, M. W., Reynolds, J. R., Power, J. D., Repovs, G., Anticevic, A., & Braver, T. S. (2013). Multi-task connectivity reveals flexible hubs for adaptive task control. Nature Neuroscience, 16(9), 1348–1355. http://doi.org/10.1038/nn.3470

Dosenbach, N. U. F., Fair, D. A., Miezin, F. M., Cohen, A. L., Wenger, K. K., Dosenbach, R. A. T., … Petersen, S. E. (2007). Distinct brain networks for adaptive and stable task control in humans. Proceedings of the National Academy of Sciences of the United States of America,104(26), 11073–11078.

Power, J. D., Cohen, A. L., Nelson, S. M., Wig, G. S., Barnes, K. A., Church, J. A., … Petersen, S. E. (2011). Functional Network Organization of the Human Brain. Neuron, 72(4), 665–678. http://doi.org/10.1016/j.neuron.2011.09.006

Seeley, W. W., Menon, V., Schatzberg, A. F., Keller, J., Glover, G. H., Kenna, H., … Greicius, M. D. (2007). Dissociable Intrinsic Connectivity Networks for Salience Processing and Executive Control. The Journal of Neuroscience : The Official Journal of the Society for Neuroscience,27(9), 2349–2356.

Sepideh Sadaghiani, M. D. (2014). Functional Characterization of the Cingulo-Opercular Network in the Maintenance of Tonic Alertness.Cerebral Cortex (New York, N.Y. : 1991), 25(9).

Spreng, R. N. (2012). The fallacy of a “task-negative” network. Cognition, 145. http://doi.org/10.3389/fpsyg.2012.00145

Sylvester, C. M., Corbetta, M., Raichle, M. E., Rodebaugh, T. L., Schlaggar, B. L., Sheline, Y. I., … Lenze, E. J. (2012). Functional network dysfunction in anxiety and anxiety disorders. Trends in Neurosciences, 35(9), 527–535. http://doi.org/10.1016/j.tins.2012.04.012

—

These notes can also be linked to in Google Docs here.