This scientific commentary refers to ‘The hubs of the human connectome are generally implicated in the anatomy of brain disorders’, by Crossley et al. (doi:10.1093/brain/awu132).

A growing number of studies approach brain anatomy and function from the perspective of complex networks, with an explicit focus on the distributed layout of neural elements and their interconnections in the healthy, as well as diseased brain (Sporns, 2014). Network analysis has revealed a number of characteristic features of brain network organization, such as the presence of densely connected clusters or modules of brain regions, and of highly connected network hubs. In this issue of Brain, Nicolas Crossley and colleagues report the results of an extensive meta-analysis of MRI data, including a large set of observations of grey matter lesions in various brain disorders. They reveal that the disorders are associated with damage to distinct sets of hubs, and propose that disruption of network hubs may be a common factor contributing to brain dysfunction (Crossley et al., 2014).

Network approaches to brain function capitalize on the advances that have been made across a broad range of social, technological and biological systems using the concepts and computational tools of network science. Brain networks consist of nodes and connections; these are derived from brain imaging data by first applying a parcellation to derive node boundaries, and then estimating the presence and strength of connections between pairs of nodes. Structural networks capture network anatomy, based on data obtained via techniques such as diffusion imaging and tractography. In contrast, functional networks are built from estimates of statistical dependencies between time series of neuronal activity, as recorded for example with functional MRI. The complete set of structural connections is defined as the brain’s connectome (Sporns et al., 2005), a whole-brain structural network that shapes and constrains functional brain activity and connectivity.

Not all elements of brain networks are created equal. In fact, brain networks are highly heterogeneous—for example, individual nodes have unique sets of inputs and outputs (‘connectional fingerprints’) that are thought to be crucial for defining their functional specialization. Nodes also vary in terms of how many different connections they maintain (the ‘node degree’) and how important they are with respect to neuronal communication (‘node centrality’). Network theory predicts that nodes with high degree and high centrality (‘hubs’) support efficient integration by facilitating the convergence of neuronal signals from different sensory modalities or cognitive domains (van den Heuvel and Sporns, 2013). The specialized roles of hubs in integrative processing are essential for cognition and behaviour, and hence hubs contribute high value to the network as a whole. Several recent studies suggest that the high value of brain hubs comes at a cost. Hubs maintain anatomical and functional connections that span longer distances and they tend to consume metabolic energy at a higher rate than non-hub regions (Collin et al., 2013). This combination of high value and high cost suggests that hubs are not only centres of integration but also points of vulnerability, and that they may be implicated in the pathophysiology of a range of brain disorders.

This is the hypothesis that Crossley et al. set out to address, using a multi-pronged approach. First, the authors used diffusion imaging and tractography to derive whole-brain structural networks (including cortical and subcortical regions) that could be analysed in terms of connection topology (network relations of nodes and edges) and geometry (their spatial layout). In addition, the structural connection matrix was subjected to ‘attack’, simulated as the random or targeted deletion of nodes (and their edges), followed by an analysis of the efficiency of global communication. Second, the authors examined large sets of MRI data to map the location and extent of grey matter lesions associated with a total of 26 different brain disorders. Supplementing these disorder-specific meta-analytic maps were ‘disorder general’ maps designed to assess robustness of findings by excluding subsets of disorders and subsampling of studies. Third, the authors drew on a network of functionally co-active brain regions previously derived from a large corpus of brain activation studies (Crossley et al., 2013).

Network analysis of the structural data confirmed the presence of network hubs that were highly interconnected and that maintained a large proportion of high-cost long-distance projections. Moreover, simulated attack confirmed that brain networks remain resilient to progressive deletion of randomly selected nodes, but are more vulnerable (as measured by global communication efficiency) to attacks that target hub nodes. Building on this analysis of a ‘normative connectome’, Crossley et al. then explored the relationship between brain hubs and disorder-specific and disorder-general grey matter lesion maps. Voxel-wise regression analysis of a disorder-general lesion map and a map of node degree revealed a significant and positive relationship—the higher the probability that a voxel was lesioned (estimated across all disorders), the higher its associated node degree (estimated from the intact brain connectome).

Disorder-specific meta-analyses painted a more fine-grained picture of how hubs may relate to dysfunction. In 20 of 26 disorders, the median number of connections was greater for the lesioned voxels than for voxels that were unaffected. A set of nine disorders stood out after more detailed statistical analysis that took into account effect sizes and numbers of studies, including Alzheimer’s disease, frontotemporal dementia, post-traumatic stress disorder and schizophrenia. Although hubs were implicated in all of these disorders, their anatomical distribution varied—for example hubs in medial temporal and parietal regions were most impaired in Alzheimer’s disease, whereas hubs in medial frontal and cingulate regions were most affected in schizophrenia.

The robustness of these results was explored by comparing lesion maps not only to the structural connectome but also to the functional co-activation network. Once again, network hubs were found to be more strongly associated with lesioned voxels in both disorder-general and disorder-specific lesion maps. Additional analyses examined the effect of methodological variations such as including weights in the structural connectome, using different hub metrics or employing variants of how disorders were grouped or analysed. The results indicated that the study’s major findings are indeed robust against such variations.

Several aspects of the paper by Crossley et al. stand out. The meta-analysis of MRI data collected from studies of clinical disorders is remarkably extensive, both in terms of the size of the dataset (comprising 392 case-control studies including a total of 21 376 subjects) and the range of disorders (26 altogether). This made it possible to address, for the first time, the hypothesis that network hubs are generally associated with a large number of neurological and psychiatric conditions. The study provides compelling evidence in favour of this hypothesis, while also suggesting that each disorder is associated with a specific and characteristic set of hub lesions. Finally, the study finds converging evidence to implicate hubs in both structural and functional co-activation networks, which further buttresses the generality of the hypothesis.

An appealing aspect of the study is its innovative blend of brain MRI data (meta-)analysis with concepts from network science. The major findings reported by Crossley et al. are in line with those of many other network studies that probe the vulnerability of social, technological and biological systems. For example, a seminal study on the robustness of communication networks suggested that their integrity was most imperilled when attacks were levelled against highly connected nodes (Albert et al., 2000). These findings extend into the domain of biology: numerous studies of protein networks in cells show that mutations of ‘hub proteins’ have disproportionately disruptive effects. Interestingly, these hub proteins are also attractive targets for drug design—and this observation points to a potential future opportunity for new clinical approaches that aim to improve brain network organization and function. Therapeutic strategies (e.g. brain stimulation techniques) could aim to exploit the central embedding of brain hubs to correct the negative impact of pathophysiology, for example by re-routing neuronal communication patterns and thus restoring global network functionality.

What are the implications of these findings for our understanding of the neural basis of brain disorders? The major findings of the study corroborate previous analyses of the involvement of highly connected and highly central brain regions in specific disorders, for example in neurodegenerative disease (Buckner et al., 2009), as well as in computational models of lesion effects (Alstott et al., 2009). Going beyond these earlier studies, Crossley et al. identify commonalities and differences across a range of disorders that are usually studied in isolation. While each disorder will certainly have its own set of pathogenetic mechanisms, the apparent convergence of multiple disorders onto a common network substrate is highly appealing as it opens up new avenues for studying disorders as manifestations of network disturbances. Future work is needed to establish more direct, causal links between disruptions of hub nodes and specific brain disorders. This is an area where invasive studies in model organisms will play an important role, for example by building on recent advances in mapping the networks of the mouse brain (Oh et al., 2014). Targeted manipulation of hub and non-hub nodes in the mouse connectome may offer new insights into their differential contributions to brain function and their roles in triggering human brain disorders.

References

Albert
R
Jeong
H
Barabasi
AL
Error and attack tolerance of complex networks
Nature
 , 
2000
, vol. 
406
 (pg. 
378
-
82
)
Alstott
J
Breakspear
M
Hagmann
P
Cammoun
L
Sporns
O
Modeling the impact of lesions in the human brain
PLoS Comput Biol
 , 
2009
, vol. 
5
 pg. 
e1000408
 
Buckner
RL
Sepulcre
J
Talukdar
T
Krienen
FM
Liu
H
Hedden
T
, et al.  . 
Cortical hubs revealed by intrinsic functional connectivity: mapping, assessment of stability, and relation to Alzheimer's disease
J Neurosci
 , 
2009
, vol. 
29
 (pg. 
1860
-
73
)
Collin
G
Sporns
O
Mandl
RCW
van den Heuvel
MP
Structural and functional aspects relating to cost and benefit of rich club organization in the human cerebral cortex
Cereb Cortex
 , 
2013
 
[Epub ahead of print]
Crossley
NA
Mechelli
A
Vertes
PE
Winton-Brown
TT
Patel
AX
Ginestet
CE
, et al.  . 
Cognitive relevance of the community structure of the human brain functional coactivation network
Proc Natl Acad Sci USA
 , 
2013
, vol. 
110
 (pg. 
11583
-
8
)
Crossley
N
Mechelli
A
Scott
J
Carletti
F
Fox
P
McGuire
P
, et al.  . 
The hubs of the human connectome area generally implicated in the anatomy of brain disorders
Brain
 , 
2014
, vol. 
137
 (pg. 
2382
-
95
)
Oh
SW
Harris
JA
Ng
L
Winslow
B
Cain
N
Mihalas
S
, et al.  . 
A mesoscale connectome of the mouse brain
Nature
 , 
2014
, vol. 
508
 (pg. 
207
-
14
)
Sporns
O
Tononi
G
Kotter
R
The human connectome: a structural description of the human brain
PLoS Comput Biol
 , 
2005
, vol. 
1
 pg. 
e42
 
Sporns
O
Contributions and challenges for network models in cognitive neuroscience
Nat Neurosci
 , 
2014
, vol. 
17
 (pg. 
652
-
60
)
van den Heuvel
MP
Sporns
O
Network hubs in the human brain
Trends Cogn Sci
 , 
2013
, vol. 
17
 (pg. 
683
-
96
)