Significant outcomes
• The patients with BD showed disrupted whole structural topological organization and rich club connections.
• The patients with BD showed abnormal nodal parameters predominantly in regions of the default mode network and central executive network.
Introduction
Bipolar disorder (BD) is a severe chronic mood disorder characterized by episodes of mania, hypomania, and alternating or intertwining episodes of depression (Grande et al., Reference Grande2016). The increasing body of multimodal neuroimaging evidence has shown structural and functional brain abnormalities in BD, mainly in the prefrontal cortex, temporal cortex, limbic system, and subcortical structures (Vargas et al., Reference Vargas, Lopez-Jaramillo and Vieta2013; Hanford et al., Reference Hanford2016; Wise et al., Reference Wise2016). A recent major step forward in the field has been the realization that altered structural and functional connectivity between these regions is likely as crucial to the pathogenesis of BD as isolated changes in volume and functional activity (Calhoun et al., Reference Calhoun2011; O'Donoghue et al., Reference O'Donoghue2015; Wise et al., Reference Wise2016; O'Donoghue et al., Reference O'Donoghue2017a).
Diffusion-tensor imaging (DTI), the only available technique to study the microstructure of human brain white matter (WM) in vivo, has been used to detect WM abnormalities in patients with mood disorders (Sexton et al., Reference Sexton, Mackay and Ebmeier2009). Previous DTI studies have demonstrated both local and widespread fractional anisotropy (FA) reductions in the brain WM regions in BD patients compared with healthy controls, such as in the corpus callosum, anterior limb of internal capsule, temporal-parietal WM, anterior and posterior cingulum (Vederine et al., Reference Vederine2011; O'Donoghue et al., Reference O'Donoghue2017a), and these abnormalities have been attributed to disruptions in connectivity. A recent meta-analysis of DTI studies indicated that BD is most commonly associated with a WM disruption in the fronto-limbic connectivity, which may be accompanied by other WM changes spread throughout temporal and parietal regions (Bellani et al., Reference Bellani2016). However, the majority of DTI studies in BD are limited by small subject numbers and confounded by the use of psychotropic medications (i.e. lithium, mood stabilizers, antipsychotics, and antidepressant medication) (Hafeman et al., Reference Hafeman2012). In general, the literature to date supports a normalizing effect of lithium and mood stabilizers on brain structure in BD, which is consistent with the neuroprotective characteristics of these medications identified by preclinical studies (Fisher et al., Reference Fisher2015). In addition, most studies involve heterogeneous samples of BD patients combining subjects in different mood states (i.e. manic, depressive, euthymic, and mixed states), which could have different neural network activation correlates and different WM DTI patterns (Zanetti et al., Reference Zanetti2009; Bellani et al., Reference Bellani2016; Serpa et al., Reference Serpa2017). Therefore, to better understand the brain structural changes in BD patients, the aforementioned limitations need to be overcome in research.
Rather than investigating individual regions or tracts in isolation, the brain can be analyzed as a complex and integrative network known as the human connectome (Sporns, Reference Sporns2011). A graph theoretical framework enables the study of a large-scale brain structural network by characterizing the brain regions as nodes and the WM tracts as edges (Sporns et al., Reference Sporns, Tononi and Kotter2005; Puetz et al., Reference Puetz2017). Network studies of structural connectivity obtained from DTI have revealed that healthy human WM networks exhibit ‘small-world’ networks that have an optimal balance between local specialization and global integration for information processing (Sporns, Reference Sporns2011). In addition, network hubs form a high capacity central core, or rich club, and rich club connections play a key role in promoting global communication and information integration (van den Heuvel and Sporns, Reference van den Heuvel and Sporns2011; Liang et al., Reference Liang2017). Several studies have found disrupted small-world properties and abnormal regional characteristics in various neurological and psychiatric disorders, including Alzheimer's disease (Shu et al., Reference Shu2012), major depression disorder (Bai et al., Reference Bai2012; Korgaonkar et al., Reference Korgaonkar2014), and schizophrenia (van den Heuvel et al., Reference van den Heuvel2013; Griffa et al., Reference Griffa2015). Moreover, many connectivity studies hypothesized either that these diseases may target certain connections in the rich club network (Shu et al., Reference Shu2012; van den Heuvel et al., Reference van den Heuvel2013), or that network disruptions in the rich club have the most significant impact on cognition (Daianu et al., Reference Daianu2016). Until now, few studies have found changed whole-brain structural network in BD (Leow et al., Reference Leow2013; Collin et al., Reference Collin2016; O'Donoghue et al., Reference O'Donoghue2017b), and those studies that did were inconsistent in the network topological properties that they reported. A number of confounding factors might influence structural network results and, thus, could at least partly account for this heterogeneity, including medication status, mood status, age range, chronicity, and long duration of illness.
In this study, we use DTI and graph theory approaches to investigate the rich club organization and structural brain connectome in unmedicated patients with BD depression. On the basis of previous findings in the functional and structural connectome (Leow et al., Reference Leow2013; Wang et al., Reference Wang2017a), we hypothesized that in BD the structural connectome would show disrupted whole topological organization and rich club connections, and abnormal regional connectivity supporting cognitive and affective functioning.
Materials and methods
Participant
This prospective study was approved by the Ethics Committee of First Affiliated Hospital of Jinan University (Guangzhou, China), the written informed consent of each participant was obtained before the study. Patients were recruited from January 2015 to January 2017. In total, 42 right-handed, nonsmokers, out- or in-patients with BD were recruited from the psychiatry department of First Affiliated Hospital of Jinan University, Guangzhou, China. The patients were aged from 18 to 55 years. All patients met DSM-IV (Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition) criteria for BD according to the diagnostic assessment by the Structured Clinical Interview for DSM-IV Patient Edition. Diagnosis of BD was determined by two experienced clinical psychiatrists. Among the 42 BD patients, seven were diagnosed with BD-I type and 35 with BD-II type. Exclusion criteria included the presence of (a) any current psychiatric disorder (with the exception of BD and anxiety disorders); (b) a history of electroconvulsive therapy; (c) any history of moderate/severe head injury, head trauma, neurological disorder, or mental retardation; (d) alcohol/substance abuse or dependence; and (e) the presence of a concurrent significant physical illness. No patients were excluded for the exclusion criteria. The clinical state was assessed using the 24-item Hamilton Depression Rating Scale (HDRS) and the Young Mania Rating Scale (YMRS) during the 3-day period prior to the imaging session. All patients with BD were suffering from depression (HDRS-24 score ⩾21, YMRS score <7). At the time of the scanning, all patients were either medication-naïve or were unmedicated for at least 6 months. Among them, 11 patients were medication-naïve; they had never been diagnosed or did not want to take medication. While for the others, the recruited patients generally visited their physicians (psychiatrist/general practitioner) because of depressive relapse after quitting the medication. These 31 patients had been treated with antidepressants (duloxetine or paroxetine), and/or mood stabilizers (lithium, sodium valproate), and/or atypical antipsychotic medications (olanzapine or risperidone), but had been off medication for at least 6 months prior to the scan.
A total of 59 right-handed, nonsmokers healthy control participants matched for sex and age were recruited via local advertisements. All healthy control participants were interviewed to rule out the presence of current or past psychiatric illness. Further exclusion criteria for healthy controls were any history of psychiatric illness in first-degree relatives, current or past significant medical or neurological illness.
MRI data acquisition and preprocessing
All MRI data were acquired on a 3.0 T MR system (Discovery MR 750 System, GE Healthcare, Milwaukee, WI) with an eight-channel phased-array head coil. The DTI data were obtained using a single-shot spin-echo diffusion-weighted EPI sequence with the following parameters: repetition time ms/echo time ms, 8000/68; flip angle, 90°; field of view, 256 × 256 mm2; 128 × 128 matrix; 30 different diffusion directions with b = 1000 s/mm2, five non-diffusion weighted b 0 volumes, slice thickness, 2 mm; no intersection gap, and 75 interleaved axial slices covering the whole brain. For each subject, the DTI scan was performed twice, and the two DTI datasets were subsequently averaged to increase the signal-to-noise ratio (SNR). A three-dimensional brain volume imaging sequence that covered the whole brain was used for structural data acquisition (8.2/3.2; flip angle, 12°; field of view, 240 × 240 mm2; 256 × 256 matrix, slice thickness = 1 mm; no intersection gap). In addition, two routine scans using axial T1-weighted fluid attenuation inversion recovery and fast spin echo T2-weighted MR sequences were also applied to obtain brain images to confirm the absence of any brain structural abnormalities. All MR images were evaluated by two neuroradiologists.
DTI data preprocessing and network construction
All DTI data were processed with PANDA package (Cui et al., Reference Cui2013) under a Linux Operating System. For each subject, the two DTI datasets were concatenated together and the effects of head motion and image distortions caused by eddy-currents were corrected using FSL/FDT (http://www.fmrib.ox.ac.uk/fsl). In detail, the ten b 0 images (two scans) were extracted and aligned to the first b 0 image. Then we averaged the ten b 0 images and affine alignment of each of 60 diffusion-weighted image volumes (two scans) to the averaged b 0 images was applied. Diffusion-sensitive gradients were adjusted to account for the rotation applied to the measurements during motion correction (Leow et al., Reference Leow2013). Afterward, the 60 diffusion-weighted image volumes were split into two DTI datasets (30 volumes for each dataset), which were averaged and merged with the averaged b 0 images. Then we estimated voxel-wise diffusion tensor and obtained the FA maps.
We took revised automated anatomical labeling v2 (AAL2) atlas (Rolls et al., Reference Rolls, Joliot and Tzourio-Mazoyer2015), which parcellates the whole brain into 120 cortical and subcortical regions of interest (ROIs), as a node and the inter-nodal WM streamlines as the edge to construct the network for each subject. The names and abbreviations of these ROIs are presented in the online Supplementary Table S1. In specific, each of the FA maps in native space was affinely coregistered to its corresponding T1-weigted images. Then, T1-weighted images were nonlinearly registered to the MNI space. The inverse transformations were obtained to the above two steps to transform the AAL2 atlas from MNI space to diffusion native space. Each ROI from AAL2 atlas represents a node of a structural brain network. A whole brain deterministic tractography approach was performed using the fiber assignment by continuous tracking (FACT) algorithm (Mori and Zijl, Reference Mori and Zijl2002). In the calculations, fiber tracking was terminated at voxels where FA < 0.2 or where the angle between two eigenvectors to be connected by the tracking was greater than 45°. We took the mean FA value along the inter-regional streamlines as the edge weight and constructed an FA-weighted 120 × 120 matrix for each subject. To reduce the potential effect of noise or other factors during diffusion tractography caused by data acquisition and pre-/post-processing, we considered two ROIs were structurally connected if at least three streamlines connecting them (Lo et al., Reference Lo2010; Shu et al., Reference Shu2011). Such a threshold selection reduced the risk of false-positive connections due to noise or the limitations in the deterministic tractography and simultaneously ensured the size of the largest connected component in the prospective network.
Graph analysis of network topology
Global and nodal parameters
All network analyses were performed using GRETNA (http://www.nitrc.org/projects/gretna/) and illustrated using BrainNet Viewer (http://www.nitrc.org/projects/bnv/). We characterized the global properties of brain structural networks by the following parameters: characteristic path length (L p), clustering coefficient (C p), global efficiency (E glob), local efficiency (E loc), normalized clustering coefficient (γ), and normalized characteristic path length (λ) (Rubinov and Sporns, Reference Rubinov and Sporns2009). Typically, a small-world network is characterized as γ ≫ 1 and λ ≈ 1, or δ = γ/λ > 1 (Taira et al., Reference Taira2014). In other words, a small-world network has not only the higher local interconnectivity but also the approximately equivalent shortest path length compared with the random networks (Cao et al., Reference Cao2013). Three nodal parameters characterizing nodal properties of brain structural networks were also calculated: nodal strength (K i), nodal efficiency (E i), and nodal betweenness (B i). The definitions and interpretations of these global and nodal parameters are listed in the online Supplementary Table S2.
Rich club analysis
The rich club organization described central backbone for global communication in the brain network (van den Heuvel and Sporns, Reference van den Heuvel and Sporns2011; van den Heuvel et al., Reference van den Heuvel2012), refers to nodes with higher degrees within brain networks and a higher connectivity strength of inter-nodal connections (van den Heuvel and Sporns, Reference van den Heuvel and Sporns2011). In the present study, the group-averaged FA-weighted structural networks were constructed for the BD and the control groups separately. The weighted rich club coefficient ϕ w(k), randomly rich club coefficient 
$\phi _{{\rm rand}}^{\rm w} (k)$, and normalized rich club coefficient 
$\phi _{{\rm norm}}^{\rm w} (k)$ for each group were calculated to describe rich club organization, the details of which are given in the Supplemental Material. The rich club regions were selected at the top 12% of most consistently ranked nodes across two groups (van den Heuvel et al., Reference van den Heuvel2013). Three classes of connections were used to characterize the edge architecture of brain networks as follows: rich club connections which link rich club nodes to rich club nodes, feeder connections which link rich club nodes to non-rich club nodes, and local connections which link non-rich club nodes to non-rich club nodes.
Statistical analysis
Comparison of demographic variables
We applied χ2-test to determine the gender difference and independent two sample t test (two-tailed) to determine the age and education difference between the BD and the control groups.
Comparison of network parameters
A non-parametric permutation test (10 000 times) was used to determine the statistical significance of between-group differences in network parameters (global and nodal parameters, rich club coefficient, feeder, and local connectivity density). Non-parametric methods use the data itself to find the null distribution, allowing the consideration of non-standard test statistics (Poldrack et al., Reference Poldrack, Mumford and Nichols2011). Under the null hypothesis, the group labels are irrelevant, and thus we can reanalyze the data over and over with different permutations of the labels. To correct the multiple comparisons for nodal parameters, false positive correction for N-node statistical comparison was applied using 1/(amount of nodes) = 1/120 = 0.0083 as significance threshold (Bassett et al., Reference Bassett2009; Meng et al., Reference Meng2014), which is equivalent to saying that nodes with p < 0.0083 to be ‘significant’. Although this correction method is not as conservative as a Bonferroni or FDR correction, it has been used in many studies, particularly for comparisons of regional network metrics (Lynall et al., Reference Lynall2010; Jung et al., Reference Jung2017). When we compared group differences of those network parameters (global and nodal parameters as well as rich club organizations), we took age, gender, and education levels as nuisance covariates and regressed them out.
Relationship between network parameters and clinical variables
For the network parameters that showed significant between-group differences, we computed partial correlations between these parameters and clinical variables for the BD patients (i.e. HDRS scores, number of episodes, durations of illness, and onset age of illness) controlling for age, gender, and education levels (p ＜ 0.05, FDR-corrected).
Results
Demographic and clinical variables
Table 1 shows the demographic and clinical data of all study participants. There were no significant differences in sex and age between the BD group and the healthy control group (p > 0.05). The control group received a significantly higher level of education than the BD group (two-tailed t test, p < 0.001).
Table 1. Demographic and clinical data in patients with BD and healthy control group
HRSD-24, the 24-item Hamilton Rating Scale; YMRS, Young Mania Rating Scale.
a p value obtained using the two-tailed two sample t test.
b p value obtained using the χ2 test.
Means ± standard deviations are reported unless otherwise noted.
Network analyses
Global and nodal parameters
Figure 1 shows the global parameters of brain networks in the BD and the control groups. The small-world properties of structural networks were estimated for the BD patients and the controls. We found δ > 1 for both groups, which suggest that the structural networks of both subject groups possess small-world characteristics. In addition, we found increased L p (p = 0.037), and decreased E glob (p = 0.043) and E loc (p = 0.021) in the BD group compared with the controls.
Fig. 1. The differences in global parameters of the brain anatomical networks between the bipolar disorder (BD) patients and the controls (HC). The BD patients showed significant decreases in E glob, and E loc, but significant increases in L p compared with the controls (p < 0.05, permutation test). The symbol of ‘+’ in red color indicates outliers. Abbreviations: C p, clustering coefficient; L p, shortest path length; E glob, global efficiency; E loc, local efficiency; λ, normalized characteristic path length; γ, normalized clustering coefficient; δ = λ/γ, small-world characteristic.
Figure 2 and Table 2 show the differences in nodal parameters of brain networks in the BD group compared with the controls. The BD group exhibited significantly decreased K i compared with the controls in three regions, the left cuneus (CUN.L, p = 0.006), right cerebelum_7b (HVIIb.R, p = 0.002), and right cerebelum_Crus2 (Hcrus II.R, p = 0.001). And compared with the controls, the BD group showed significantly decreased E i in two regions: the left superior parietal gyrus (SPG.L, p = 0.001) and left angular gyrus (ANG.L, p = 0.004). In addition, we also found that the BD group showed significantly increased B i in the left orbital part of inferior frontal gyrus (IFGorb.L, p = 0.003), and right calcarine gyrus (CAL.R, p = 0.003) compared with the controls.
Fig. 2. Brain regions with significantly changed nodal parameters in the bipolar disorder (BD) patients. The BD patients showed significantly decreased nodal degree (K i), represented by green color, and nodal efficiency (E i), represented by purple color, in five regions and significantly increased nodal betweenness (B i), represented by blue color, in two regions compared with the healthy controls (HC).
Table 2. Brain regions showing altered nodal parameters (nodal degree K i, nodal efficiency E i, and nodal betweenness B i) of brain structural networks in the BD patients compared with the healthy controls (HC)
Rich club analysis
Figure 3a illustrates the rich club organization found in the structural networks of both the BD and control groups. For the BD and control groups, we found the normalized rich club coefficients 
$\phi _{{\rm norm}}^{\rm w} (k) \gt 1$ over the range of degrees from k = 1–16 (Fig. 3a). However, no significant differences were detected in normalized rich club coefficients 
$\phi _{{\rm norm}}^{\rm w} (k)$ between the BD and the control groups (Fig. 1a). We detected 14 rich club regions, which are located in the bilateral medial superior frontal gyrus (SFGmedial), superior frontal gyrus (SFG), calcarine (CAL), cuneus (CUN), precuneus (PCUN), putamen (PUT), and superior parietal gyrus (SPG) (Fig. 3b).
Fig. 3. Comparison of group-averaged rich club organization between the bipolar disorder (BD) patients and the healthy controls (HC). (a) Normalized rich club coefficient 
$\phi _{{\rm norm}}^{\rm w} (k)$ original rich club coefficient 
$\phi _{}^{\rm w} (k)$ and the randomly rich club coefficient 
$\phi _{{\rm rand}}^{\rm w} (k)$. Each symbol ‘o’ in red (blue) indicates the normalized rich club coefficient for the patients (controls) at specific degree k. The gray area indicates 
$\phi _{{\rm norm}}^{\rm w} (k)$ >1 for each group. The patient group did not show a significant difference in rich club organization compared with controls. (b) Rich club members included the bilateral medial superior frontal gyrus, superior frontal gyrus, calcarine, cuneus, precuneus, putamen, and superior parietal gyrus in both healthy and patient populations. (c) Edges across individual brain networks (both controls and patients) were divided into three distinct classes: rich club connections linking rich club members (white), feeder connections linking rich club members to non-rich club members (red), and local connections connecting non-rich club members (black).
When characterizing the edge architecture in the context of rich club organization, we found both the rich club (p = 0.03) and feeder connections (p = 0.04) were decreased in the BD group compared with the controls (permutation testing, 10 000 times, Fig. 3c). That is, the connections that link the group-averaged rich club regions and the connections that link the group-averaged rich club regions with non-rich club regions. No significant group effect was found in the local connections (p = 0.08), which connect the non-rich club regions.
Relationship between network parameters and clinical variables
For the correlations between network parameters and clinical variables, no significant correlation results survived after FDR correction (p ＜ 0.05, FDR-corrected). We have also displayed the correlation results without correction in the online Supplementary Fig. S1.
Discussion
Using DTI tractography, this study detected abnormal topological organization of WM networks in homogeneous unmedicated adult patients with depressed BD. We found that compared with the healthy controls, the BD patients showed (1) increased L p, and decreased E glob and E loc; (2) abnormal nodal parameters mainly in the parietal, orbitofrontal, occipital, and cerebellar regions; (3) decreased rich club connectivity density and feeder connectivity density. These findings extend our understanding of the neuropathological mechanisms underlying BD from a perspective of WM structural connectivity pattern.
In this study, we found altered global topological properties of the whole brain structural networks in the unmedicated patients with depressed BD, including increased L p, and decreased E glob and E loc, which suggests a less optimized topological organization in their WM networks. These findings were partly consistent with the BD-related brain networks studies using DTI (Leow et al., Reference Leow2013; Collin et al., Reference Collin2016; O'Donoghue et al., Reference O'Donoghue2017b), resting state fMRI (Spielberg et al., Reference Spielberg2016; Wang et al., Reference Wang2017a), and EEG (Kim et al., Reference Kim2013). For example, two structural brain network studies found that medicated patients with euthymic BD I exhibited longer L p, lower E glob, and lower C p than controls, suggesting impaired global integration in BD patients (Leow et al., Reference Leow2013; O'Donoghue et al., Reference O'Donoghue2017b). In addition, previous studies using DTI tractography and graph theory approaches proposed the disease-related increases in the L p may be attributable to the degeneration of fiber bundles used for information transmission (Bai et al., Reference Bai2012). Taken together, our findings of abnormal global topological properties in depressed BD suggest a disturbance between segregation and integration in brain structural networks, which could indicate distributed information transmission in BD.
Turning to the nodal level, we found altered nodal parameters (K i, E i, or B i) in the depressed BD patients mainly in regions of the default mode network (DMN) including the angular gyrus and cuneus, and regions of the central executive network (CEN, also referred to as the cognitive control network) including the SPG, orbital part of inferior frontal gyrus, and posterior cerebellum. This suggested disconnections of the WM in these regions in BD patients. The DMN is believed involving in affective regulation and internal thoughts (Kaiser et al., Reference Kaiser2015); and the CEN is responsible for high-level cognitive functions and external information procession (Menon, Reference Menon2011). Previous fMRI studies have reported impairments in the coordinated activity, functional connectivity within and between the DMN and CEN in BD patients (Goya-Maldonado et al., Reference Goya-Maldonado2016; Wang et al., Reference Wang2016), which might be related to cognitive impairments and clinical symptoms. For example, the functional imbalance between the DMN and CEN is correlated with depressive symptoms or trait rumination and is not present during a self-focus condition, suggesting a failure to reallocate resources away from self-referential processing toward external goal-directed behavior when this is required (Belleau et al., Reference Belleau, Taubitz and Larson2015; Kaiser et al., Reference Kaiser2015; Lois et al., Reference Lois2017). Our findings, combined with these previous reports, suggest that disrupted WM connectivity in the DMN and CEN regions may contribute to core deficits in cognitive and affective functioning in depressed BD. Unfortunately, we failed to observe the correlation between altered nodal parameters and clinical variables. In addition, several fMRI studies found impaired DMN in euthymic BD and their unaffected siblings (Alonso-Lana et al., Reference Alonso-Lana2016; Lois et al., Reference Lois2017; Wang et al., Reference Wang2017b), and suggested that it may act as a trait marker. Therefore, the present findings need confirmation in further studies combining DTI and fMRI measurement in BD across different mood phases, preferably with a longitudinal, within-subject design.
The rich club organization takes a central position in the brain's network topology, and connections among rich club brain hubs have been proposed to be central to the integration of information among different subsystems of the human brain (van den Heuvel and Sporns, Reference van den Heuvel and Sporns2011; van den Heuvel et al., Reference van den Heuvel2013). In this study, we detected decreased rich club connectivity density and feeder connectivity density in the BD patients. This may suggest that the rich club connections among the central hubs, acted as a central high cost and high capacity backbone for global brain communication (van den Heuvel et al., Reference van den Heuvel2012), were prone to be disrupted in BD patients. Several structural network studies found reduced rich club connectivity in schizophrenia patients (van den Heuvel et al., Reference van den Heuvel2013; Collin et al., Reference Collin2014; Klauser et al., Reference Klauser2017) and their unaffected siblings (Collin et al., Reference Collin2014), indicating a disproportionate disturbance of brain hubs and their mutual connections in schizophrenia. However, a recent DTI study found no indication that the brain hubs or ‘rich club’ connections were particularly affected in BD patients (Collin et al., Reference Collin2016). It is worthy to note that, in their study, the majority of the BD patients were euthymic, and the patients used psychotropic medications including lithium and antipsychotics. In addition, the identified rich club regions or hubs in the present study comprised the precuneus, cuneus, superior frontal cortex, superior parietal cortex, calcarine, and putamen in the BD patients and controls, consistent with previous studies (van den Heuvel and Sporns, Reference van den Heuvel and Sporns2011; Collin et al., Reference Collin2014; Tuladhar et al., Reference Tuladhar2017). Due to their central position in the topology of the network, the connections among these rich club nodes may play an important role in the efficient integration of information processing among distant brain regions (i.e. global efficiency of the network) (van den Heuvel and Sporns, Reference van den Heuvel and Sporns2011; van den Heuvel et al., Reference van den Heuvel2012). Damage specifically to the rich club connections seems to have a more severe impact on global efficiency than random damage to the network (van den Heuvel and Sporns, Reference van den Heuvel and Sporns2011). Therefore, these findings, in conjunction with the global parameters changes, suggest that the abnormal whole brain network organization in BD might, in part, be due to impaired rich club connections. Moreover, we found these altered nodal parameters specifically affecting rich club regions, such as the cuneus, calcarine, and superior parietal cortex, suggesting that specific core areas of the brain might be preferentially targeted by BD pathology.
This study has several limitations. First, the node and edge definitions and DTI acquisition protocols may affect the findings obtained from whole brain WM structural networks in BD patients. However, there is no widely agreed-on approach for calculating the brain network metrics. This suggests the measures may not be optimally robust just yet. In this study, we selected a widely applied template, the AAL-120 template, to define the network nodes. Regions on the AAL template differ in size, which may have a confounding effect on the link weight of the network nodes. Thus, we also used a high-resolution (~1000 parcels) parcellation (AAL-1024) by randomly subdividing the AAL atlas into 1024 regions with equal size both in the volume and in the average cortical surface, and the results were similar to our main findings (online Supplementary Table S4). Deterministic fiber tractography was used to define the edges of the structural connectome, which may result in the loss of existing fibers due to the ‘fiber crossing’ problem. Tensor-based deterministic tractography approaches yield sparse connectomes and contain false negatives (high specificity with comparatively low sensitivity), whereas probabilistic methods steered by crossing-fiber models yield dense connectomes and contain false positives (high sensitivity with comparatively low specificity), other groups have begun to study tradeoffs between connectome sensitivity and specificity (Zalesky et al., Reference Zalesky2016). Initial studies in this regard suggest (both empirically and theoretically) that specificity is at least twice as important as sensitivity when estimating key properties of brain networks (Zalesky et al., Reference Zalesky2016). Second, the BD patients were recruited including BD type I and type II. There were some clinical and anatomical data showing the differences between the subtypes (Abe et al., Reference Abe2016). However, we also did the same network analysis in 35 patients with only BDII and controls, and the results were similar to our main findings (online Supplementary Tables S6 and S7). Third, a significant difference in education level existed between the BD patients and the healthy controls, which may confound the statistical results. In order to minimize this effect, we took the education level as a nuisance covariate and regressed them out in all analyses. Fourth, the correlation results between network metrics and clinical variables did not survive after FDR correction, perhaps because of our relatively small sample size. Fifth, without a group of patients with BD in a euthymic episode, it is still not clear whether disrupted topological organization are specific to the depression episode of BD or shared by all episodes of the disease. Further studies using a prospective design may clarify this issue.
In conclusion, our study revealed the disrupted WM topological organization in the whole-brain, and abnormal regional connectivity supporting cognitive and affective functioning in BD, which, in part, be due to impaired rich club connectivity. Further longitudinal research is needed to examine how and whether the changes of WM connectivity, including the investigation of the rich club properties, are related to clinical deterioration in BD patients.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291718001150.
Acknowledgements
We wish to thank the participants and their families for their contribution to our research.
Financial support
This study was supported by the National Natural Science Foundation of China (81671670, 81501456, 81471650, 81428013, and 81471654); Planned Science and Technology Project of Guangdong Province, China (2014B020212022); and Planned Science and Technology Project of Guangzhou, China (20160402007 and 201604020184). The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Conflict of interest
All authors declare that they have no conflicts of interest.
