Brain Structural Covariance Networks in Behavioral Variant of Frontotemporal Dementia

Recent research on behavioral variant frontotemporal dementia (bvFTD) has shown that personality changes and executive dysfunctions are accompanied by a disease-specific anatomical pattern of cortical and subcortical atrophy. We investigated the structural topological network changes in patients with bvFTD in comparison to healthy controls. In particular, 25 bvFTD patients and 20 healthy controls underwent structural 3T MRI. Next, bilaterally averaged values of 34 cortical surface areas, 34 cortical thickness values, and six subcortical volumes were used to capture single-subject anatomical connectivity and investigate network organization using a graph theory approach. Relative to controls, bvFTD patients showed altered small-world properties and decreased global efficiency, suggesting a reduced ability to combine specialized information from distributed brain regions. At a local level, patients with bvFTD displayed lower values of local efficiency in the cortical thickness of the caudal and rostral middle frontal gyrus, rostral anterior cingulate, and precuneus, cuneus, and transverse temporal gyrus. A significant correlation was also found between the efficiency of caudal anterior cingulate thickness and Mini-Mental State Examination (MMSE) scores in bvFTD patients. Taken together, these findings confirm the selective disruption in structural brain networks of bvFTD patients, providing new insights on the association between cognitive decline and graph properties.


Introduction
Behavioral variant frontotemporal dementia (bvFTD) is the most common frontotemporal lobar degeneration (FTLD), accounting for more than 50% of patients with autopsyconfirmed FTLD [1]. Characterized by a progressive impairment in social function and personality [2], patients with bvFTD often show a focused atrophy in several cortical and subcortical regions, such as the anterior cingulate, insula, prefrontal cortex, anterior temporal regions, striatum, and thalamus [3][4][5]. Despite the typical clinical features and anatomical changes, bvFTD remains difficult to diagnose, and may be confused with other Eligibility criteria included no history of other neurological or psychiatric illnesses, no clinical or neuroimaging evidence of focal lesions, and no inflammatory, infectious, or vascular diseases. The control group was selected according to ADNI-3 criteria (ADNI Protocol v1.0: 24 May 2016, http://adni.loni.usc.edu/wp-content/themes/freshnews-dev-v2/documents/clinical/ADNI3_Protocol.pdf). None of the controls had a history of neurologic or psychiatric illness. The Mini-Mental State Examination (MMSE) and Frontal Assessment Battery (FAB) were administered to all participants as the screening assessment [35,36]. All participants gave written informed consent. The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Review Board (or Ethics Committee) of ASL Lecce (verbale n. 6, 25 July 2017).

MRI Acquisition and Processing
Neuroimaging data were acquired on a 3T scanner (Philips Ingenia 3T). Set acquisition was in the sagittal plane using a Fast-Field Echo (FFE) T1-weighted sequence with the following parameters: repetition time = 8.2 ms, echo time = 3.8 ms, field of view = 256 × 256 mm 2 , 200 slices, flip angle = 8 • , and isotropic 1 mm 3 voxels.
T1-weighted images were inspected visually to check for motion-related artifacts and gross neuroanatomical alterations by a consultant neuroradiologist. Next, images were analyzed using FreeSurfer (version 6.0) (http://www.nmr.mgh.harvard.edu/martinos) to extract morphological features for cortical and subcortical brain regions [37][38][39]. Briefly, the cortical surface for each participant was reconstructed from T1-weighted images by the following steps: skull stripping, segmentation of cortical gray and white matter, separation of the two hemispheres and subcortical structures, and, finally, construction of smooth representation of the gray/white interface and the pial surface [37,[39][40][41]. Next, all images were checked for reconstruction cortical surface errors, and surface inaccuracies were corrected with FreeSurfer's editing tools. Then, the surface area was calculated using triangular tessellation of the gray/white matter interface and white matter/cerebrospinal fluid boundary (pial surface) [42]. Cortical thickness was also calculated based on the distance between closest points, between gray and white matter surfaces [39]. Finally, we used the FreeSurfer parcellation scheme based on the Desikan-Killiany Atlas to extract the cortical thickness and surface area of 68 cortical regions from both hemispheres [43]. Subcortical volumetric analyses were also performed using an automated approach that estimates the probability of structure classification based on prior templates in which those structures were manually identified [44]. We considered 12 subcortical areas, including the putamen, caudate, thalamus, pallidum, hippocampus, and amygdala, for each hemisphere. A list of cortical and subcortical regions is given in Supplementary Materials (Table S1).

Network Construction
Cortical thickness, surface area, and volumetric values were bilaterally averaged and corrected for age, sex, and individual brain size [33]. The resulting residuals were then z-score transformed using the mean and standard deviation values of each brain region calculated from healthy controls. Finally, a measure of joint variation between the 74 morphometric features (34 cortical surface area values, 34 cortical thickness values, and six subcortical volume values) represented the edge weights of the network, and was calculated using the following formula [33,45]: 1/exp{[(z-score value of ith region of interest) − (z-score value of jth region of interest)] 2 } (1)

Graph Theory Analysis
Estimation of the global and local network characteristics was performed by using the Graph Theoretical Network Analysis (GRETNA) (www.nitrc.org/projects/gretna/) packages [46]. Small world measures and global efficiency (Eglob) were used to characterize the global topological organization of the covariance structural networks in both controls and patients with bvFTD. In particular, to examine the small-world properties of a network, the normalized clustering coefficient γ = (Cp real /Cp random ) and the normalized characteristic path length λ = (Lp real /Lp random ) were first computed. Then, the small-world index was calculated as the ratio of the normalized clustering coefficient and the normalized path length (σ = γ/λ). Of note, Cp real and Lp real are the clustering and the characteristic path length of the real network, respectively, and Cp random and Lp random represent, respectively, the mean clustering coefficient and shortest path length of 1000 matched random networks that preserve the same numbers of nodes, edges, and degree distribution as the real network. A real network can be considered as a small-world network if it fulfills the following criteria: small-world index σ = λ/ γ > 1.1 [47,48]. Compared to a random network, a small-world network is thus characterized by a higher clustering coefficient. By contrast, it exhibits a short characteristic path length comparable to that of a random network.
Regional network properties were assessed using degree centrality, the clustering coefficient, local path length, and local efficiency [11,[49][50][51]. Degree centrality is a local graph measure that is able to quantify the relative importance of a node within a network [51]. The clustering coefficient represents the ability of a node to communicate with other nodes with which it shares a direct connection (segregation ability) [49]. Nodal efficiency and characteristic path length, on the other hand, quantify the ability of information propagation between a node and the remaining nodes in the network (integration ability). Local efficiency is calculated as the global efficiency of the subgraph formed by the node's neighbors. A node with high nodal efficiency or low path length indicates high capability of information transmission with other nodes. Detailed formulas and explanations of these global and local metrics can be found in previous methodological reviews [13,21,50].
As graph measures are non-trivially dependent on the density of the underlying graph [52], intra-individual structural covariance networks were thresholded in a network density range of d = 0.10-0.40, with an interval of 0.01. Connectivity thresholding is commonly used to remove noisy or spurious links, preserving the strongest structural edges. The range of density was chosen to allow small-world network properties to be properly estimated and the number of spurious edges in each network to be minimized [53,54]. Then, the network parameters were computed for each network at each density. Finally, GRETNA was used to calculate the area under the curve (AUC, i.e., the integral over the density range) for each network measure to provide a scalar that does not depend on specific threshold selection [55,56]. Of note, graph measures were calculated based on weighted structural networks. In this way, we could characterize the relative importance of each link between network nodes. The BrainNet Viewer (http://www.nitrc.org/projects/bnv/) was used to visualize the regional brain network changes between patients and healthy controls [57].

Statistical Analyses
The Shapiro-Wilk test was performed in either demographic, neuroimaging, and neuropsychological variables (i.e., age, total intracranial volume, cognitive performance), or graph measures to verify the normality of data distribution. Next, variables with a normal distribution were compared between controls and bvFTD patients using pairwise t-tests. Non-normally distributed variables were compared between groups using Wilcoxon-Mann-Whitney test. The chi-square test was used to test for differences in the sex distribution between groups. The critical statistical threshold was set to p < 0.05. A false discovery rate (FDR) correction procedure was employed to correct for multiple comparisons in the global and local network analyses [58]. The relationships between network metrics and clinical data (disease duration and cognitive performances) of patients with bvFTD were tested using the Pearson correlation (p-value < 0.05). The correlations were considered statistically significant if the relative p-values were less than 0.05 after FDR correction.

Demographic and Clinical Characteristics
No differences were found in age, sex, years of education, or intracranial volume between the bvFTD patients and healthy controls (p > 0.05). Concerning clinical data, patients with bvFTD had significantly lower MMSE and FAB scores compared with healthy control participants (p-value < 0.001) ( Table 1).

Global Network Characteristics
The structural covariance network of controls and bvFTD patients demonstrated small-world network architecture over the preselected density range (1.37 < σ HC < 2.92; 1.25 < σ bvFTD < 2.76). However, the small-world index was smaller in bvFTD patients than in controls. Moreover, the normalized characteristic path length values in patients were greater than those of controls ( Table 2, p < 0.05, FDR corrected). Compared with the healthy control participants, the bvFTD group also exhibited significantly less global efficiency ( Table 2, p-value < 0.001, FDR corrected). No significant difference was found in the normalized clustering coefficient values between bvFTD patients and healthy controls.

Regional Network Characteristics
At a local level, bvFTD patients displayed a reduced local efficiency in the cortical thickness of the rostral and caudal middle frontal gyrus, pars opercularis, precuneus, cuneus, transverse temporal gyrus, and rostral anterior cingulate (p-value < 0.05, FDR corrected) (Figure 1, Table 3). Moreover, we observed a reduced clustering in the cortical thickness of the inferior temporal gyrus in bvFTD patients compared with controls (p-value < 0.001, FDR corrected) ( Table 3). No significant differences were found in the local properties of cortical surface areas and subcortical volumes between bvFTD patients and controls.

Correlation between Connectivity Metrics and Clinical Data
Significant correlations were found in bvFTD patients' MMSE scores with the local efficiency and nodal degree in the cortical thickness of the caudal anterior cingulate (Figure 2, p-value = 0.01, FDR corrected).

Correlation between Connectivity Metrics and Clinical Data
Significant correlations were found in bvFTD patients' MMSE scores with the local efficiency and nodal degree in the cortical thickness of the caudal anterior cingulate (Figure 2

Discussion
In the present study, we applied graph analysis to investigate the topological organization of structural brain networks in patients with bvFTD. We found altered graph metrics both at a global and local level. More specifically, when compared to healthy controls, bvFTD patients showed altered small-world properties (i.e., increased normalized path length) and decreased global efficiency. At the local level, patients with bvFTD displayed lower values of local efficiency in cortical thickness of the caudal and rostral middle frontal gyrus, rostral anterior cingulate, and precuneus, cuneus, and transverse temporal gyrus. Relative to controls, patients with bvFTD also displayed reduced values of clustering coefficients in the thickness of the inferior temporal gyrus. Finally, a significant correlation was found between the efficiency of caudal anterior cingulate thickness and the MMSE scores in bvFTD patients.
Our findings provide new insights into our understanding of structural changes in the organization of bvFTD brain networks. In particular, the reduced small-world index (σ) observed in bvFTD patients suggests that the covariance networks of bvFTD patients tend to have a more randomized configuration compared to the control group [53]. Moreover, the disruption of both normalized path length and global efficiency is indicative of an impaired functional integration of bvFTD networks, indicating a reduced ability to combine specialized information from distributed brain regions [10,49]. In the past years, several studies have investigated small-world property alterations in healthy individuals [58][59][60], as well as in neurological and psychiatric disorders [17,19,32,[61][62][63][64][65]. Neuroimaging studies have demonstrated that the cognitive and memory declines in Alzheimer's disease patients are often associated with the disruption of the small-world structure [17,27,66,67]. Evidence from graph theoretical studies have also observed reduced functional and structural integrity in bvFTD brain networks when compared to healthy controls [16,17,34,68,69]. In line with these findings, the bvFTD-related global property alterations observed in the present study are thus suggestive of an impaired functional integration, which might contribute to impairments in the cognitive function of patients with bvFTD. This idea is further supported by local property changes that we found in the frontotemporal regions of bvFTD networks. Compared to controls, patients with bvFTD showed reduced local efficiency and clustering coefficients in the cortical thickness of the middle frontal gyrus, pars-opercularis, anterior cingulate, and temporal cortices. All of these regions represent the most prominent sites of bvFTD-related focal atrophy [4,5,70]. Moreover, they play a crucial role in executive control, working memory, and emotion processing that are often disrupted in bvFTD [71,72]. Decreased values in the local properties (i.e., nodal centrality, nodal strength) of frontotemporal regions were previously reported in functional and structural networks of patients with bvFTD in comparison to controls [16,68,69]. In the present study, the reduced ability in integration found in key

Discussion
In the present study, we applied graph analysis to investigate the topological organization of structural brain networks in patients with bvFTD. We found altered graph metrics both at a global and local level. More specifically, when compared to healthy controls, bvFTD patients showed altered small-world properties (i.e., increased normalized path length) and decreased global efficiency. At the local level, patients with bvFTD displayed lower values of local efficiency in cortical thickness of the caudal and rostral middle frontal gyrus, rostral anterior cingulate, and precuneus, cuneus, and transverse temporal gyrus. Relative to controls, patients with bvFTD also displayed reduced values of clustering coefficients in the thickness of the inferior temporal gyrus. Finally, a significant correlation was found between the efficiency of caudal anterior cingulate thickness and the MMSE scores in bvFTD patients.
Our findings provide new insights into our understanding of structural changes in the organization of bvFTD brain networks. In particular, the reduced small-world index (σ) observed in bvFTD patients suggests that the covariance networks of bvFTD patients tend to have a more randomized configuration compared to the control group [53]. Moreover, the disruption of both normalized path length and global efficiency is indicative of an impaired functional integration of bvFTD networks, indicating a reduced ability to combine specialized information from distributed brain regions [10,49]. In the past years, several studies have investigated small-world property alterations in healthy individuals [58][59][60], as well as in neurological and psychiatric disorders [17,19,32,[61][62][63][64][65]. Neuroimaging studies have demonstrated that the cognitive and memory declines in Alzheimer's disease patients are often associated with the disruption of the small-world structure [17,27,66,67]. Evidence from graph theoretical studies have also observed reduced functional and structural integrity in bvFTD brain networks when compared to healthy controls [16,17,34,68,69]. In line with these findings, the bvFTD-related global property alterations observed in the present study are thus suggestive of an impaired functional integration, which might contribute to impairments in the cognitive function of patients with bvFTD. This idea is further supported by local property changes that we found in the frontotemporal regions of bvFTD networks. Compared to controls, patients with bvFTD showed reduced local efficiency and clustering coefficients in the cortical thickness of the middle frontal gyrus, pars-opercularis, anterior cingulate, and temporal cortices. All of these regions represent the most prominent sites of bvFTD-related focal atrophy [4,5,70]. Moreover, they play a crucial role in executive control, working memory, and emotion processing that are often disrupted in bvFTD [71,72]. Decreased values in the local properties (i.e., nodal centrality, nodal strength) of frontotemporal regions were previously reported in functional and structural networks of patients with bvFTD in comparison to controls [16,68,69]. In the present study, the reduced ability in integration found in key regions of the frontotemporal network further confirm a strong involvement of this network in bvFTD pathophysiology. Furthermore, the local efficiency and centrality degree values of the cortical thickness in Brain Sci. 2021, 11, 192 8 of 11 the caudal anterior cingulate were found to significantly and positively correlate with the MMSE score, indicating that the anterior cingulate might play a key role in driving cognitive deficits in bvFTD patients. Interestingly, we also found a reduced local efficiency in the cortical thickness of the precuneus and cuneus. Although gray matter alterations in these brain regions are not frequent in bvFTD patients, recent fMRI studies have reported functional connectivity alterations in posterior cortical areas of patients with FTD when compared to healthy controls, possibly reflecting reduced afferent input from limbic regions [73,74].
The current study has some limitations that need to be addressed. We considered a cohort of bvFTD patients without a histopathological confirmation. However, clinical examination was performed according to the most recent diagnostic criteria for FTD. Second, we examined a relatively small number of patients. Hence, a larger sample size is required to replicate our results. Third, in the calculation of intra-individual structural covariance networks, we used the bilaterally averaged values of cortical and subcortical morphological features. Therefore, we were not able to explore the homologous connectivity between the brain regions. However, bvFTD is traditionally associated with largely symmetrical atrophy of the frontal and temporal lobes [5,68]. Fourth, network measures are generally related to each other. Thus, it becomes difficult to say which of these measures is driving the others. The obtained results should therefore be interpreted with caution. Fifth, longitudinal studies are required to assess whether topological changes in the structural covariance network of bvFTD patients are predictive of clinical-pathological progression. Finally, it remains to be determined whether the local property changes that we found in the frontotemporal regions of bvFTD networks may represent a useful marker in distinguishing between FTD subtypes.

Conclusions
Our study provides new evidence for the usefulness of combining several morphometric measures to capture single-subject anatomical connectivity and then investigating bvFTD-related network organization using a graph theory approach. Compared to controls, patients with bvFTD showed altered graph metrics both at a global and local level. In particular, bvFTD patients were characterized by lower local efficiency values in the cortical thickness of several frontotemporal regions. These network alterations might contribute to cognitive impairments often observed in patients with bvFTD, as suggested by correlations between graph measures and MMSE scores.