Our results can be summarized in five main findings: (1) all participants presented typical small-world attributes in the brain functional networks, with no differences between TAI and HC groups in small-world properties and global efficiency; (2) the patients with TAI exhibited lower Eloc in brain functional networks; (3) they also exhibited aberrant nodal centralities in some regions, including the frontal lobes, parietal lobes, caudate nucleus, and cerebellum bilaterally, and right olfactory cortex; (4) the patients with TAI presented altered long-distance connections in a functional subnetwork; (5) the nodal degree of the right inferior parietal lobule and nodal clustering coefficient of the right cerebellar hemisphere (VII) were negatively correlated with the MMSE score. These results also show that graph theory analysis is a powerful tool to examine the aberrant neural topology in patients with TAI and characterize the pathophysiology at the level of brain networks.
Large-scale brain networks typically show efficient economic small-world organization with short normalized path lengths and high normalized clustering coefficients, reaching an optimal balance between global integration and local specialization (Bassett et al., 2009; Bullmore et al., 2009). In previous studies, a reduced small-world index was observed in patients with severe TBI (Nakamura et al., 2009); in contrast, Kuceyeski et al. (2016) reported a higher small-world index. However, our study showed that patients with TAI and HCs both had typical small-world attributes in brain functional networks, with no significant group differences in small-world metrics, including σ, γ, and λ, consistent with previous studies in patients with moderate-to-severe TBI (Caeyenberghs et al., 2012; Messe et al., 2013). The present findings suggest an overall preserved organization of the functional brain network in patients with TAI, which might be attributed to functional reorganization. Brain plasticity can produce new synapses after TBI to increase the number of connections between different regions. However, the small-world index should be a combination of different graph metrics capturing both integration and segregation because it may also mistakenly identify a small-world topology in poorly segregated networks that are highly integrated (Caeyenberghs et al., 2017).
The topological architecture of functional networks was previously found to be significantly altered in patients with TBI (Caeyenberghs et al., 2012). Previous studies reported a reduction in the global integration of network efficiency accompanied with an increased path length (Han et al., 2016; Nomura et al., 2010; Pandit et al., 2013). The loss of global integration is often interpreted as reduced efficiency of information processing in the networks, attributed to the disruption of long-distance connections as a result of a TAI. In contrast, the current study showed no significant difference in Eglob between patients with TAI and HCs, suggesting that the capacity for information integration between distant brain regions and the efficiency of information dissemination in the global network were not significantly altered, consistent with a previous finding in individuals with mild TBI (Spielberg et al., 2015). This finding could be explained by the global brain network restoring effective functional connections via alternate structural pathways that circumvent the impaired white matter connections (Kuceyeski et al., 2016). Furthermore, a reduction in network efficiency with a lower Eloc was observed in patients with TAI in our study, predominantly related to short-range connections between adjacent nodes. Han et al. (Han et al., 2016) reported reduced global and local efficiencies in chronic TBI, whereas Caeyenberghs et al. (2012) found higher values of local efficiency in the TBI group than in the controls. Such inconsistencies may be due to different group characteristics and methodologies. Our finding of reduced Eloc in patients with TAI could be attributed to inadequate axonal wiring or differences in metabolic running costs to provide parallel information compared to HCs. Network disruptions may lead to individual connections circumventing the impaired efficient ones (long-range, inter-hemispheric, and between-network connections) using less efficient and alternative paths.
The graph theoretical approach also allows the evaluation of regional characteristics of whole-brain functional networks using regional network measures, including nodal degree, nodal betweenness, nodal clustering coefficient, nodal efficiency, and nodal local efficiency, which are not obtainable through voxel-based analyses of regional changes in activation. Several studies on TBI have shown alterations in nodal metrics in specific brain regions. Pandit et al. (2013) showed a significant reduction in degree centrality and betweenness centrality in the posterior cingulate cortex in patients with TBI, whereas Caeyenberghs et al. (2012) reported higher betweenness centrality in the dorsal premotor cortex and dorsolateral prefrontal cortex in patients with moderate-to-severe TBI. Moreover, Messé et al. (2013) found distinct nodal changes (including degree, nodal efficiency, and centrality) in the temporal and thalamic regions of patients with mild TBI. Our study showed higher nodal efficiency values (degree and nodal efficiency) in patients with TAI in the left dorsolateral superior frontal gyrus; higher nodal betweenness, degree and nodal efficiency in the medial superior frontal gyrus bilaterally, and higher nodal betweenness in the left superior orbital gyrus. These findings on basic measures and segregation suggest a higher FC in specific brain regions, which may reflect a functional compensation; however, they may also indicate high energetic costs. This modification may be attributed to numerous biophysical mechanisms, including hyperexcitability or disinhibition of functionally related networks (Fornito et al., 2015; Hillary et al., 2014). Hyperconnectivity is substantially universal in patients with TBI, regardless of the injury phase (acute, subacute, or chronic) and severity, replacing the traditional view of a transient process (Caeyenberghs et al., 2012; Iraji et al., 2016). The finding of increased FC in segregation after TAI is similar to the results of studies on healthy aging (Heitger et al. 2013) and some neurological diseases, such as brain tumors (Bartolomei et al., 2006; Bosma et al., 2008). Nonetheless, hyperconnectivity should not be automatically interpreted as supporting the compensation hypothesis in brain injuries; this concept is likely to be metabolically costly (Fornito et al., 2015), resulting in reduced adaptability to regulate the activity levels of network nodes. Caeyenberghs et al. (2012) demonstrated that patients with moderate-to-severe TBI who showed a higher connectivity degree presented a lower switching performance. The current finding suggests that the activity levels across multiple network nodes may show a higher level of synchronization in patients with TAI compared to HCs, in response to a damaged neurobiological substrate. In fact, volume loss and diffuse axonal injury are among the most common pathophysiologic sequelae of TBI. The dorsolateral prefrontal cortex receives visual, somatosensory, and auditory information, and plays a central role in the cognitive control of motor behaviors (Miller et al., 2001). Therefore, the higher nodal efficiencies in the dorsolateral prefrontal cortex in patients with TAI may indicate a less automatic movement generation. TBI most commonly disrupt the frontal system and interrupt the executive control processes (Hillary et al., 2002). Moreover, Alstott et al., (2009) demonstrated that a targeted injury of the frontal lobes induces a severe disruption of this network.
Reduced nodal efficiency values (including nodal degree, nodal clustering coefficient, and local efficiency) were observed in the caudate nucleus bilaterally in patients with TAI compared to HCs. Previous studies have shown that the caudate is specifically associated with the executive function (Grahn et al., 2008), especially the right nucleus (Casey et al., 1997; Hart et al., 2013). Simoni et al. (2018) reported that patients with TBI showed a disruption of the FC between the caudate nucleus and several cortical regions, and in particular the reduced right caudate connectivity was associated with cognitive impairments, mainly in executive function and information processing speed. Fagerholm et al. (2015) demonstrated that betweenness centrality and eigenvector centrality were reduced in the cingulate cortex and caudate due to the impact of a TAI on network connections. Furthermore, caudate activity during executive function tasks is associated with prefrontal measures of structural connectivity (Casey et al., 2007). Disruption of the fronto-caudate interactions may underpin the common cognitive impairments observed in patients with TBI. Interestingly, our NBS results showed significantly reduced long-distance FC between the caudate and cerebellum, especially the right caudate. This finding indicates the presence of pathway and interplay disturbances between the cerebellum and cerebrum due to damaged white matter connections between network nodes caused by a TAI. These alterations can disrupt the network function and lead to cognitive impairment.
Moreover, we observed higher nodal efficiency values (including nodal clustering coefficient, nodal efficiency, and nodal local efficiency) in the right cerebellar hemisphere (Crus 2, VII) and cerebellar vermis (4/5, 7). However, significant reductions in nodal efficiencies emerged in the left cerebellar hemisphere (VIII) and cerebellar vermis (10). These findings suggest that the cerebellar regional network is disrupted in patients with TAI. In addition, our NBS results revealed disrupted FC between the cerebellum and cerebrum, including reduced FC between the caudate and cerebellum and higher FC between the olfactory system and cerebellum. Previous studies reported cerebellar atrophy in patients with TBI (Drijkoningen et al., 2015; Spanos et al., 2007) and lower activation within the cerebellum (Yang et al., 2012). Our previous work also showed that patients with TAI exhibited significantly reduced static fractional amplitude of low-frequency fluctuations (fALFF) in the left posterior lobe of the cerebellum (Zhou et al., 2021) and reduced interhemispheric coordination between the cerebellum posterior lobes (Li et al, 2017). Furthermore, analysis of the relationships between graph metrics and clinical measures showed that the nodal clustering coefficient of the right cerebellar hemisphere (VII) was negatively correlated with the MMSE score, suggesting a correlation with cognitive disorders in patients with TAI. These significant associations highlight the potential of network analysis to predict clinical dysfunctions in patients with TAI. Based on the above findings, we inferred that altered nodal efficiencies in the cerebellar network and disrupted FC between the cerebellum and cerebrum underlie the pathophysiology of TAI and are associated with the cognitive impairments seen in these patients.
We observed significantly reduced nodal efficiency values in the TAI group (including nodal betweenness, degree, clustering coefficient, efficiency, and local efficiency), compared to HCs, in the inferior parietal lobule, including the supramarginal gyrus bilaterally. The inferior parietal lobule is often involved in spatial perception, visuomotor integration, and sensory and memory circuitry. Additionally, this lobule is also part of the posterior default network, which has been widely investigated and exhibited alterations in nodal activity within the network and in the intrinsic connectivity between networks. Reduced homologous functional interplay was reported in the inferior parietal lobules bilaterally in patients with TAI (Li et al., 2017). Furthermore, the NBS results showed reduced FC between the left supramarginal gyrus and the right inferior parietal lobule, as well as between the right supramarginal gyrus and the right olfactory cortex. In summary, the pathophysiological alterations in these regions may contribute to cognitive impairment in patients with TAI.
Prior studies have repeatedly demonstrated that individuals with TBI often present impairments in facial and vocal affect recognition and empathy (Babbage et al., 2011; Hopkins et al., 2002; Milders et al., 2003). Emotion recognition and empathy are key components of successful interpersonal interactions and relationships. Olfactory deficits are also common after TBI, affecting more than 56% of patients (Neumann et al., 2012). However, in practice, there is a lack of awareness regarding this deficit. Hopkins et al. (2002) found that patients with a closed-head injury who presented poorer emotion recognition and arousal responses to facial expressions compared to controls also had lower olfactory test scores. Similarly, Neumann et al. (2012) showed that olfactory deficits might be indicative of impaired affect recognition and reduced empathy after TBI. Our study showed a reduction in nodal local efficiency in the right olfactory cortex in patients with TAI. Furthermore, the NBS results showed reduced FC between the right olfactory cortex and the right supramarginal gyrus, and higher FC between the right olfactory cortex and cerebellar vermis. These findings suggest that the involvement of the olfactory cortex may be closely related to emotional processing impairments in patients with TAI.
The present study has several limitations. First, considering the heterogeneity of TBI, we selected a subgroup of relatively “pure” TAI patients among the 57 available; 29 patients were not examined due to larger local hemorrhages (> 10 mL in total), and the final sample size was relatively small. Second, we only analyzed the differences in the functional brain networks between patients with TAI and HCs. Future research should examine the combination of functional and structural connectivity at the subject level. Third, this study was cross-sectional, and further longitudinal analyses are needed to investigate the predictive value of graph metrics.