Alterations of white matter integrity associated with cognitive deficits in patients with glioma

Abstract Objective This study aimed to investigate the characteristic of brain structural connections in glioma patients and further evaluate the relationship between changes in the white matter tracts and cognitive decline. Methods This retrospective study included a total of 35 subjects with glioma and 14 demographically matched healthy controls, who underwent diffusion tensor imaging scans and formal neuropsychological assessment tests. Fractional anisotropy (FA) values of white matter tracts were derived from atlas‐based analysis to compare group differences. Furthermore, subgroup‐level analysis was performed to differentiate the effects of tumor location on white matter tracts. Partial correlation analysis was used to examine the associations between neurocognitive assessments and the integrity of tracts. Region of interest‐based network analysis was performed to validate the alteration of structural brain network in subjects with glioma. Results Compared with controls, subjects with glioma exhibited reduced FA values in the right uncinate fasciculus. Besides, subjects with glioma exhibited worse performance in several cognitive assessments. Partial correlation analysis indicated that the FA value in the right superior longitudinal fasciculus temporal part was significantly positively correlated with scores of visual–spatial abilities in subjects with glioma in the right temporal lobe (r = .932, p = .002). Region of interest‐based network analysis revealed that subjects with glioma exhibited reduced FA, fiber length (FL), and fiber number (FN) between specific brain regions compared with controls. Conclusion The present study demonstrated the reduced integrity of white matter tracts and altered structural connectivity in brain networks in patients with glioma. Notably, white matter tracts in the right hemisphere might be vulnerable to the effects of a frontal or temporal lesion and might be associated with deficient cognitive function.


| INTRODUC TI ON
Gliomas represent the most frequently diagnosed primary malignant intracranial neoplasms in adults, contributing to approximately 81% of all primary central nervous system tumors (Ostrom et al., 2014).
Certainly, infiltrative growth associated with highly complex interactions between cortical and subcortical pathway damage is considered to be a hallmark of gliomas, and therefore, patients with glioma often present with a wide range of symptoms. Besides the most common clinical manifestations, including nausea, headache, epilepsy, and neurological deficits, an increasing number of studies have shown that patients with gliomas may suffer impairments in multiple cognitive domains (Maesawa et al., 2015;Mariano, Mazza, & Galzio, 2013;Miotto et al., 2011). Gliomas have been reported to cause various degrees of white matter fiber damage (Incekara, Satoer, Visch-Brink, Vincent, & Smits, 2018;Soni, Mehrotra, Behari, Kumar, & Gupta, 2017). In addition, gliomas are well recognized for their infiltrative growth along the white matter tracts (Louis, 2006).
Thus, a better understanding of the relationship between cognitive alterations and white matter integrity in patients with glioma could potentially optimize the preoperative evaluation and the rehabilitation strategies, thereby improving patients' quality of life.

Diffusion tensor imaging (DTI) is a widely used MRI technique
that can noninvasively reveal the diffusion of water molecules within the brain tissue to detect, visualize, and quantify the integrity of brain microstructure (Carlo, Jezzard, Basser, Barnett, & Di Chiro, 1996). Parametric quantitative information on white matter (WM) tract integrity can be derived from tractography and reflected as a value of fractional anisotropy (FA). Therefore, FA can measure changes in water diffusion and provide complementary information about the white matter microstructure, especially the development of white matter fiber tracts (Alexander et al., 2011). FA is commonly referred to as a summary measure of microstructural integrity of WM tracts and is a normalized standard deviation of the eigenvalue between 0 and 1. A reduced FA in a WM tract usually indicates a change in the integrity of the fiber bundle (Alexander et al., 2011;Incekara et al., 2018). Diffusion tensor tomography (DTT) evaluates water diffusion in the WM of the brain and shows subcortical tract integrity based on indices such as FA, mean diffusivity, and radial diffusivity. DTT allows visualization of the exact location of brain tumors that may be relevant to specific WM tracts and has been widely used (Yu, KunCheng, Yun, XunMing, & Wen, 2005). Previous studies have shown that gliomas may cause alterations in WM tract integrity to vary degrees due to their different locations and infiltrative growth (Incekara et al., 2018;Ilhami et al., 2011;Soni et al., 2017).
As an intra-axial primary brain tumor, malignant glioma cells exhibit preferential invasion along WM tracts in the brain tissue (Bellail, Hunter, Brat, Tan, & Van Meir, 2004;Louis, 2006;Mandonnet, Capelle, & Duffau, 2006). Due to underlying neoplasm invasiveness and migration along WM fibers (Bellail et al., 2004;Rao et al., 2013), in addition to the specific WM tracts that are directly affected by tumors, it is also interesting to study distal WM tracts that may be potentially affected by the invasive tumors.
Previous studies have suggested that the brain structural network of patients with brain disorders (including stroke, cerebrovascular disease, and tumor-related lesions) may be disrupted, some of which are related to cognitive dysfunction (Du et al., 2019;Ille, Engel, Kelm, Meyer, & Krieg, 2018;Urbanski et al., 2011). Despite a growing body of the literature on examining tumor-affected brains through DTT technique (Abhinav, Yeh, Mansouri, Zadeh, & Fernandez-Miranda, 2015;Conti Nibali et al., 2019;Incekara et al., 2018), a few studies have endeavored to study the effects of glioma from a whole-brain perspective. Nakajima et al. demonstrated that highlevel mentalizing processing was associated with structural connectivity networks, which were influenced by the effect of right cerebral hemispheric glioma and its resection (Nakajima et al., 2018).
Zhou et al. reported altered connection density and brain anatomical networks in patients with a brain tumor (Zhou et al., 2016). Excessive attention to a specific WM tract may lead us to ignore the effect of tumors on distant tracts and whole-brain structural networks.
However, to understand the putative long-range effects of glioma may allow more accurate prediction of the effects of glioma resection and insight into mechanisms of subcortical structural network.
Moreover, such research may further contribute to the understanding of the clinical-psychological symptoms in patients with glioma.
In the present study, we acquired DTI data of 35 glioma patients to study the effect of the gliomas on multiple WM tracts and structural networks. In addition, we sought to clarify the association between changes in structural integrity and impairment to certain cognitive domains. We hypothesized that it is not only the subcortical structures subjected to anatomical damage that exhibits impairments, but also the interconnected remote regions that seem to be affected by the lesions (Duffau, 2014).
Tumor damage can affect the inherent brain structural connectome and cause network dysfunction (Sharp, Scott, & Leech, 2014). (3) no history of cerebral radiotherapy or temozolomide chemotherapy; and (4) no or slight neurological focal deficit including aphasia or paresis. Patients with recurrent glioma (n = 2), marked peritumoral edema associated with midline shift (n = 7), or without sufficient available DTI data (n = 9) were excluded from the study. A total of 35 subjects (19 males and 16 females, with an average age of 49.60 years) met the inclusion and exclusion criteria and were included for the data analysis (the demographic characteristics are summarized in the supplementary materials, Table S1).

| MRI data acquisition
The subjects included in this retrospective study were scanned Diffusion MRI data were carried out using an echo planar imaging (EPI) sequence with the following parameters for three times: diffusion encoding in 32 independent, noncollinear directions with a b-value = 1,000 s/mm 2 and one additional image with no diffusion weighting (b = 0); slice number = 62; TR = 6,500 ms; TE = 95 ms; slice gap = 3 mm; slice thickness = 3 mm; flip angle = 90°; field of view (FOV) = 120 mm × 120 mm; and acquisition matrix = 128 × 128.

| Data preprocessing
The DTI data were preprocessed and analyzed by PANDA (Pipeline for Analyzing Brain Diffusion Images) software (http://www.nitrc. org/proje cts/panda) (Cui, Zhong, Xu, He, & Gong, 2013). The preprocessing steps were performed as described previously (Bai et al., 2012;Wang et al., 2016). Briefly, (a) the DICOM files of all subjects were converted into NIfTI images using the dcm2nii tool embedded in MRIcron; (b) the extraction of brain tissue and structure, this step also yielded the brain mask for the subsequent processing steps; (c) realignment; (d) each diffusion-weighted image was coregistered to the b0 image using affine transformation to correct the head motion artifacts and distortions caused by eddy currents, and the diffusion gradient directions were also adjusted accordingly; (e) calculation of fractional anisotropy (FA); (f) reconstruction of diffusion tensor tractography; and (g) tractography and network construction to produce 3D streamlines representing fiber tract connectivity (Mori, Crain, Chacko, & van Zijl, 1999).

| Atlas-based analysis of DTI values
The atlas-based analysis was used to analyze the differences between glioma patients and HCs and subgroup levels (12 patients with frontal LGG (Ⅰ-Ⅱ) 16 HGG (Ⅲ-Ⅳ) 19 Histopathological subtype (WHO grade) Anaplastic astrocytoma (Ⅲ) 3 (8.6%) Anaplastic oligodendroglioma (Ⅲ Occipital 0 (0%) glioma and 11 right temporal lobe glioma). However, due to data limitations, the changes in DTI values in patients with glioma in the left temporal, the parietal, and the occipital lobe could not be investigated.
Briefly, we defined WM tracts into 20 tracts using the JHU WM tractography atlas (http://cmrm.med.jhmi.edu/) (the defined WM tracts and abbreviation are listed in Table S2). The WM atlas in the standard space allows for parcellation of the entire brain WM into 20 regions of interests (ROIs), each representing a labeled region in the atlas.
Subsequently, the 20 tracts were used as masks to extract the mean diffusion measures within the regions. The regional diffusion metrics (such as FA) were generated from PANDA by averaging the values within the specific region of each WM atlas. Atlas-based values can be further exported to Statistical Package for the Social Sciences (SPSS) for statistical analysis (Cui et al., 2013). The significantly altered FA of WM tracts was selected and extracted by the tract-based masks for the further correlation analysis.

Definition of network nodes
We used an automated anatomical labeling (AAL) template (Tzourio-Mazoyer et al., 2002), an extensively used high-resolution T1weighted brain parcellation, to parcel the brain into 90 cortical and subcortical ROIs (45 in each hemisphere without cerebellar regions).
The names and abbreviations of these ROIs are summarized in supplementary materials, Table S3. These 90 ROIs were used as nodes for global network connectivity analysis.

Constructing networks using deterministic tractography
ROI-based whole-brain deterministic tractography analysis was generated using PANDA, and the streamlines would be terminated unless the voxel with an FA < 0.2 or met a fiber with turning angle >45°. For each pair of above-mentioned network nodes, fibers with two terminal points located in their respective regions were considered connected to the two nodes. Based on the linking fibers, three basic weighted matrices, FA-weighted matrix, length-weighted matrix, and numberweighted matrix, were calculated using PANDA, which resulted in three weighted matrix maps in HCs and subjects with glioma, respectively. In a resultant matrix, each column or row represents a node or brain region defined by the AAL template. Subsequently, individual-level white matter connectivity network was constructed based on the value of

| Demographic and neurocognitive characteristics
The demographic characteristics of patients with glioma and HCs are presented in Table 2. No differences were observed for age, gender, or education between HCs and patient groups or subgroups (p > .05). Eight participants with glioma failed to finish all cognitive tests. Subjects with glioma performed worse than HCs in all six cognitive tests (DST, DSST, memory, visuospatial, mapping, and similarity, all p < .001 except DST p < .0037) (Figure 1).

| Atlas-based analysis at subgroup level
According to different anatomical locations of brain tumors, subjects with gliomas were divided into two subgroups. The two subgroups include frontal lobe (n = 12) and right temporal lobe gliomas (n = 11), representing approximately 34.3% and 31.4% of the total patients, respectively. As presented in Figure

| Correlation between WM tracts and neurocognitive assessment measures
As illustrated in Figure 3, among subjects with glioma located in the right temporal lobe group, visual-spatial scores were positively correlated with the FA of SLFTP.R (r = .932, p = .002). No significant associations between the decreased FA of WM tracts and altered cognitive functions were observed in subjects with frontal glioma.

| Characteristics of the structural network in patients with glioma
Successful tractography was performed for both subjects with glioma and HCs. Figure

Note: Values are expressed as the mean (standard deviation; SD). t column is the values of two-sample t-test between all patients with glioma and
HCs. All p-values were obtained using t-test except for gender (chi-square test). p 1 representing the comparison between HCs and all patients with glioma. p 2 representing the comparison between HCs and patients with glioma located in the frontal lobe. p 3 representing the comparison between HCs and patients with glioma located in the right temporal lobe. * Significant differences were observed between HCs and patients with glioma or patients' subgroup. * p < .05, ** p < .01, *** p < .001. There were 8 patients who did not complete all cognitive tests. Furthermore, compared with HCs, a significant decline in FN between IPL.L and ANG.R was also detected in patients with glioma (p < .001). specific WM tracts (such as UF.R) as compared with HCs. Notably, a significant correlation was found between the FA in SLFTP.R and the scores of visual-spatial assessments. In a whole-brain structural network, subdued FA, FL, and FN were observed between certain brain regions in the patients with glioma. Taken together, these results suggest that patients with glioma may exhibit microstructural changes in white matter, leading to the abnormalities in the structural connectivity network, which might explain the widespread cognitive dysfunction in patients with glioma (Korgaonkar, Fornito, Williams, & Grieve, 2014;Zhou et al., 2016).

| DTI measures of WM tracts
From the atlas-based analysis, a reduced FA of CCG.R and UF.R in subjects with glioma was detected. It seems that gliomas engen- These results indicated that the integrity of particular fibers (such as UF.R) may be more sensitive to the presence of glioma located in the frontal lobe and right hemisphere. In line with this view is evidence that microstructural WM alterations of UF.R can also be detected in patients with craniopharyngioma (Fjalldal et al., 2018).

Besides, Highley et al. have indicated that the uncinate fasciculus
in the right hemisphere is larger and comprised of more fibers than in the left hemisphere of healthy individuals. Our findings may be attributed to the asymmetry of UF.R observed in the healthy brain (Highley, Walker, Esiri, Crow, & Harrison, 2002;Shu, Liu, Duan, & Li, 2015).
In addition, for subjects with different pathological types of glio- LGG. There is still some controversy (Lu et al., 2004;Miloushev, Chow, & Filippi, 2015;Smitha, Gupta, & Jayasree, 2013). This could be explained by the systematic factors or differences in methodology and the type of datasets used. Heterogeneity of tumor distribution and the small sample size could also have minimized the difference between LGG and HGG in our study. Therefore, additional studies with larger sample size and precise subgrouping are needed for further validation.

| Association between alterations in the FA and impaired cognitive abilities
In the present study, subjects with glioma exhibited worse performance in several cognitive domains (including executive and memory functions), which is consistent with previous studies (Mariano et al., 2013;Satoer et al., 2014). Furthermore, studies have also reported the possible relationship between altered WM tracts and cognitive impairments in patients with brain tumors. Incekara et al.
indicated that language and attention deficits were significantly associated with altered FA values of several WM tracts in patients with glioma (Incekara et al., 2018). In our study, the only cognitive assessment associated with altered FA was the visual-spatial test, which has been reported to be associated with the FA values of the posterior cingulum and right frontal-parietal regions (Kantarci et al., 2011;Mabbott, Noseworthy, Bouffet, Laughlin, & Rockel, 2006). Our results revealed that among patients with glioma located in the right temporal lobe group, the reduced FA of SLFTP.R was significantly positively associated with the visual-spatial scores. Although several variations exist, we speculate that local invasive gliomas might affect cognitive function by disrupting the integrity of WM tracts and these impairments may have a wide range of effects on multiple cognitive functions.

F I G U R E 5
Each red node represents a cortical region of the AAL template. The edges between two seeds were weighted by averaged FA, FL, or FN value. Compared with HCs, the significantly reduced value of fractional anisotropy (FA) (p < .05 FDR corrected), fiber length (FL) (p < .001), and fiber number (FN) (p < .001) in patients with glioma was presented in Figure a-c, respectively.

| Alteration of the structural network in patients with glioma
Although more and more literature indicates that the brain functional network of glioma patients has changed, there are still few studies dedicated to exploring the structural network of a brain with glioma (Ghinda, Wu, Duncan, & Northoff, 2018 with primary brain tumors following fractionated radiotherapy, which might contribute to the delayed cognitive impairments observed in patients with brain tumors (Bahrami et al., 2017). Conceivably, the functional aberrations might be attributed to the impaired integrity of the interhemispheric structural connections.
Interestingly, left inferior parietal lobule (IPL.L) and right angular gyrus (ANG.R) are considered the classic brain regions in the DMN (Buckner, Andrews-Hanna, & Schacter, 2008;Raichle, 2015). It is generally speculated that the functional configurations assumed by the brain cortex are reflective of underlying anatomical connection (Honey et al., 2009;Passingham, Stephan, & Kotter, 2002). In a network-based framework, we speculated that the reduced FA and FN between IPL.L and ANG.R might be the potential structural basis for the modifications of DMN functional connectivity in patients with glioma (Ghumman, Fortin, Noel-Lamy, Cunnane, & Whittingstall, 2016;Greicius, Supekar, Menon, & Dougherty, 2009;Maesawa et al., 2015). The exact relationship between functional networks and structural networks in glioma patients remains to be elucidated.

| LI M ITATI O N S
This study has several limitations worth noting. First, similar to many imaging studies (Maesawa et al., 2015;Zhang et al., 2016), the cur-

| CON CLUS ION
In summary, the findings of the present study demonstrated altered integrity of WM tracts in patients with glioma. Fibers in the right hemisphere (particularly UF.R) seem to be more susceptible to pathological changes caused by invasive tumors in the frontal or right temporal lobe. The altered linkages (FA, FL, and FN) suggested that the reduced integrity of WM tracts may damage brain structural networks in patients with glioma, which might be associated with impaired cognitive function.

ACK N OWLED G M ENTS
This study was supported by the grant from the clinical medical scien-

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
HL, JC, and XH conceived and designed the experiments. DL, JC, and XH preprocessed and analyzed MRI data. KY, CX, JH, and ZL contributed materials/analysis tools. DL, GH, JC, XH, YL, and YZ involved in the preparation of the article, figures, and tables. DL, YL, and JC revised the manuscript. All authors read, revised, and approved the final version of the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of the current study are available from the corresponding author upon reasonable request.