Elsevier

NeuroImage

Volume 49, Issue 4, 15 February 2010, Pages 3132-3148
NeuroImage

Identifying the brain's most globally connected regions

https://doi.org/10.1016/j.neuroimage.2009.11.001Get rights and content

Abstract

Recent advances in brain connectivity methods have made it possible to identify hubs—the brain's most globally connected regions. Such regions are essential for coordinating brain functions due to their connectivity with numerous regions with a variety of specializations. Current structural and functional connectivity methods generally agree that default mode network (DMN) regions have among the highest global brain connectivity (GBC). We developed two novel statistical approaches using resting state functional connectivity MRI—weighted and unweighted GBC (wGBC and uGBC)—to test the hypothesis that the highest global connectivity also occurs in the cognitive control network (CCN), a network anti-correlated with the DMN across a variety of tasks. High global connectivity was found in both CCN and DMN. The newly developed wGBC approach improves upon existing methods by quantifying inter-subject consistency, quantifying the highest GBC values by percentage, and avoiding arbitrarily connection strength thresholding. The uGBC approach is based on graph theory and includes many of these improvements, but still requires an arbitrary connection threshold. We found high GBC in several subcortical regions (e.g., hippocampus, basal ganglia) only with wGBC despite the regions' extensive anatomical connectivity. These results demonstrate the complementary utility of wGBC and uGBC analyses for the characterization of the most highly connected, and thus most functionally important, regions of the brain. Additionally, the high connectivity of both the CCN and the DMN demonstrates that brain regions outside primary sensory-motor networks are highly involved in coordinating information throughout the brain.

Introduction

The brain is thought to have evolved from simple reflex circuits, bestowing flexibility on behavior by integrating specialized brain regions into coordinated networks. Perhaps reflecting our especially flexible behavioral repertoire, the human brain is estimated to have hundreds of specialized brain regions (Van Essen, 2004). However, it is unknown how these specialized regions are integrated so behavior can be coordinated. Recent research has found that some regions have much higher global brain connectivity (GBC) than others, perhaps reflecting their role in integrating brain activity in order to coordinate cognition and behavior (Achard et al., 2006, Buckner et al., 2009, Hagmann et al., 2008, Heuvel et al., 2008, Salvador et al., 2005a, Sporns et al., 2007).

Existing GBC methods, using both anatomical (Hagmann et al., 2008) and functional (Buckner et al., 2009) connectivity, have identified regions in the default mode network (DMN) as having the highest GBC. This high connectivity may reflect connections necessary to implement the wide variety of cognitive functions the network is involved in. Consistent with this notion, we hypothesized that another large-scale network implementing a variety of cognitive function, the cognitive control network (CCN), also has among the highest GBC.

The CCN has been reported in many studies of cognitive control processes, and is likely involved in coordinating networks of brain regions during novel and non-routine tasks (Cole et al., 2007, Dosenbach et al., 2006). The DMN has been reported in studies of resting state activity, suggesting it is active “by default” (Raichle et al., 2001). However, the DMN is engaged by mind wandering (Mason et al., 2007), prospective and retrospective self-reflection (D'Argembeau et al., 2008), and memory retrieval (Buckner et al., 2005), suggesting that the ‘default mode’ involves ongoing processing of information for relevance to the self. The CCN is thought to consist of dorsolateral prefrontal cortex (DLPFC), rostrolateral prefrontal cortex (RLPFC), dorsal–caudal anterior cingulate cortex (ACC), pre-supplementary motor area (pre-SMA), inferior frontal junction (IFJ), posterior parietal cortex (PPC), pre-motor cortex (PMC), and anterior insula cortex (AIC). The DMN is thought to consist of posterior cingulate cortex (PCC), rostral anterior cingulate cortex (rACC), anterior temporal lobe (aTL), superior frontal cortex (SFC), and inferior parietal cortex (IPC). Importantly, the CCN and DMN are anti-correlated during task performance and uncorrelated at rest (Fox et al., 2005, Murphy et al., 2008) (Fig. 1A), suggesting they are relatively independent networks. We predicted, given their involvement in a wide variety of complex cognitive behaviors that they would both have among the highest GBC in the human brain.

In addition to these two cortical networks, a variety of subcortical brain regions have been found in animal models to have high global connectivity. We predicted that these regions would also show high global connectivity in humans. One such region is amygdala, which is thought to integrate sensory and internal-state information for limbic processing (Barbas, 2000, Jolkkonen and Pitkänen, 1998). Similarly, hippocampal cortex (HC) is thought to integrate information from a wide variety of sources in order to encode entire episodes (Eichenbaum et al., 2007). Also, several midbrain neurotransmitter (MNT) regions such as locus coeruleus and substantia nigra are thought to project to a variety of regions throughout the brain (Fig. 1B) (Herlenius and Lagercrantz, 2004) and are thought to play important roles in motivation and arousal.

Another region, thalamus, includes several nuclei with differing connectivity profiles (Behrens et al., 2003), suggesting that only parts of it might have highly extensive connectivity. Similarly, basal ganglia (BG) and cerebellum connect with cortex via topographic loops (Kelly and Strick, 2003) (Fig. 1C), suggesting that some loops would bestow more wide-spread connectivity on parts of the structures than others. For these reasons, we predicted that amygdala and HC would have high global connectivity, as well as portions of thalamus, BG, and cerebellum.

Functional MRI (fMRI) is an increasingly important method for measuring functional connectivity non-invasively. Among the functional connectivity methods developed with fMRI, the decade-old method of resting state functional connectivity MRI (rs-fcMRI) is unique in its ability to capture functional connectivity largely independent of any particular brain state. Evidence for this comes from a study of anesthetized monkeys (Vincent et al., 2007) that showed rs-fcMRI patterns similar to humans at rest, as well as a study of rs-fcMRI during both task and rest in humans (Fair et al., 2007). Though further research is necessary, rs-fcMRI is thought to be based on very infrequent (∼0.01 to 0.1 Hz) bursts of spiking activity in cortex that drive correlated activity through brain networks (Golanov et al., 1994, Kannurpatti et al., 2008).

In previous work, we observed that, since the entire brain is a network, the term brain network was ill defined. We developed a working definition of brain network as a set of regions with greater internal connectivity than external connectivity (Cole and Schneider, 2007). We were able to show using rs-fcMRI that the CCN fits this criterion. Importantly, we also found that the CCN is significantly more globally connected than the rest of the brain on average. Here we sought to replicate this finding with more refined methods, and also to determine what other brain regions exhibit high GBC. As outlined above, we predicted that the CCN, DMN, and a variety of subcortical regions would be among the most highly globally connected in the brain, perhaps reflecting their roles in coordinating complex cognitive behaviors.

Recently, another GBC method was developed that combines graph theory and rs-fcMRI with a whole-brain (voxel-wise) analysis approach (Buckner et al., 2009). Unlike the GBC method developed by Cole and Schneider (2007), the Buckner et al. (2009) method uses binary connections in an unweighted graph. In order to implement this unweighted GBC (uGBC) method a connection strength threshold is necessary which, unlike the weighted GBC (wGBC) method, involves removing connections with lower strength. Since wGBC does not require thresholding of the connection strengths, we predicted that it might reveal globally connected regions with many low-strength connections (such as modulatory subcortical regions; e.g., locus coeruleus) that might be removed by uGBC thresholding.

In order to compare the uGBC and wGBC methods, we implemented the whole-brain uGBC method (as developed by Buckner et al., 2009) and modified the wGBC method (as developed by Cole and Schneider, 2007) to also include whole-brain maps. We also modified both methods to be more statistically quantitative and accessible to a wider variety of researchers. Specifically, we applied widely used parametric statistical methodology to quantify inter-subject consistency, as well as a novel and easily interpretable thresholding approach that identifies the top percentages of voxels in terms of global connectivity. Thresholding the maps in terms of top percentage GBC allows comparison of the methods using a common metric despite differences between them.

A major motivation behind the development of these methods was to provide alternatives to graph theory for identifying the brain's most globally connected regions. Though graph theory has been quite productive in characterizing brain networks thus far (Bullmore and Sporns, 2009), as a branch of mathematics it typically does not quantify the statistical certainty of a given finding (Deuker et al., 2009, Kramer et al., 2009). Here we used statistical methods to quantify between-subject certainty, as well as the degree to which voxels are globally connected (in terms of percentages). We see these new approaches as complementary to graph theory, with the potential to increase confidence in brain network findings by acknowledging and quantifying the variability and graded nature of the data underlying those findings.

Section snippets

Participants

We included 14 right-handed subjects (7 male, 7 female), aged 19 to 29 (mean age 22) in the study. These subjects were recruited from the University of Pittsburgh and surrounding area. Subjects were excluded if they had any medical, neurological, or psychiatric illness, any contraindications for MRI scans, or were left-handed. All subjects gave informed consent.

MRI data collection

Image acquisition was carried out on a 3T Siemens Trio MRI scanner. Thirty-eight transaxial slices were acquired every 2000 ms (FOV:

Grand mean global connectivity strength

The group mean global connectivity strength across all gray matter voxels (i.e., the group grand mean wGBC) was r = 0.035, with a standard deviation (between subjects) of 0.0198. All subjects had low, yet positive grand mean wGBC, suggesting brain regions are positively correlated on average.

Top percent wGBC

As expected, all CCN and DMN regions were included in the top 5% (p < 0.00016, FDR corrected) of wGBC voxels. Additionally, several subcortical regions expected to have among the highest GBC are present. These

Discussion

The wGBC and uGBC methods developed here converge to show that brain regions in the CCN and DMN are among the most globally connected (Fig. 3, Fig. 4, Fig. 5). Several other subcortical regions, including amygdala, HC, BG, an MNT region, and cerebellum also have high wGBC (Fig. 3), as expected based on known anatomical connectivity. These findings promise to provide novel insights into the mechanisms of information integration and coordination in the brain.

The first implementation of wGBC

Acknowledgments

We thank Bruna Martins for insightful comments and suggestions. We would also like to thank Robert Cox and Kyle Simmons for developing efficient global connectivity software. This research was supported by DARPA. The findings and opinions contained here are those of the authors, not DARPA. M.C. was supported by an NSF graduate research fellowship.

References (67)

  • FairD.A. et al.

    A method for using blocked and event-related fMRI data to study “resting state” functional connectivity

    Neuroimage

    (2007)
  • FischlB. et al.

    Whole brain segmentation: automated labeling of neuroanatomical structures in the human brain

    Neuron

    (2002)
  • FischlB. et al.

    Sequence-independent segmentation of magnetic resonance images

    Neuroimage

    (2004)
  • FusterJ.M.

    Upper processing stages of the perception–action cycle

    Trends. Cogn. Sci.

    (2004)
  • FusterJ.M. et al.

    Functional interactions between inferotemporal and prefrontal cortex in a cognitive task

    Brain Res.

    (1985)
  • GenoveseC.R. et al.

    Thresholding of statistical maps in functional neuroimaging using the false discovery rate

    Neuroimage

    (2002)
  • KannurpattiS. et al.

    Spatio-temporal characteristics of low-frequency BOLD signal fluctuations in isoflurane-anesthetized rat brain

    Neuroimage

    (2008)
  • UngerleiderL.G. et al.

    ‘What’ and ‘where’ in the human brain

    Curr. Opin. Neurobiol.

    (1994)
  • Van EssenD.C.

    Surface-based approaches to spatial localization and registration in primate cerebral cortex

    Neuroimage

    (2004)
  • AchardS. et al.

    A resilient, low-frequency, small-world human brain functional network with highly connected association cortical hubs

    J. Neurosci.

    (2006)
  • AmaralL.A. et al.

    Classes of small-world networks

    Proc. Natl. Acad. Sci. U.S.A.

    (2000)
  • BehrensT.E. et al.

    Non-invasive mapping of connections between human thalamus and cortex using diffusion imaging

    Nat. Neurosci.

    (2003)
  • BucknerR.L. et al.

    Molecular, structural, and functional characterization of Alzheimer's disease: evidence for a relationship between default activity, amyloid, and memory

    J. Neurosci.

    (2005)
  • BucknerR.L. et al.

    Cortical hubs revealed by intrinsic functional connectivity: mapping, assessment of stability, and relation to Alzheimer's disease

    J. Neurosci.

    (2009)
  • BullmoreE. et al.

    Complex brain networks: graph theoretical analysis of structural and functional systems

    Nat. Rev. Neurosci.

    (2009)
  • BungeS.A. et al.

    Neural circuitry underlying rule use in humans and nonhuman primates

    J. Neurosci.

    (2005)
  • ChafeeM.V. et al.

    Inactivation of parietal and prefrontal cortex reveals interdependence of neural activity during memory-guided saccades

    J. Neurophysiol.

    (2000)
  • D'ArgembeauA. et al.

    Self-reflection across time: cortical midline structures differentiate between present and past selves

    Soc. Cogn. Affect. Neurosci.

    (2008)
  • EichenbaumH. et al.

    The medial temporal lobe and recognition memory

    Ann. Rev. Neurosci.

    (2007)
  • FoxM.D. et al.

    The human brain is intrinsically organized into dynamic, anticorrelated functional networks

    Proc. Natl. Acad. Sci. U.S.A.

    (2005)
  • FoxM. et al.

    The global signal and observed anticorrelated resting state brain networks

    J. Neurophysiol.

    (2009)
  • GigandetX. et al.

    Estimating the confidence level of white matter connections obtained with MRI tractography

    PLoS ONE

    (2008)
  • GiguereM. et al.

    Mediodorsal nucleus: areal, laminar, and tangential distribution of afferents and efferents in the frontal lobe of rhesus monkeys

    J. Comp. Neurol.

    (1988)
  • Cited by (449)

    View all citing articles on Scopus
    View full text