Altered Functional Connectivity in Patients with Subcortical Vascular Cognitive Impairment—A Resting-State Functional Magnetic Resonance Imaging Study

Recent neuroimaging studies have shown that people with subcortical vascular cognitive impairment (sVCI) have structural and functional abnormalities in the frontal lobe and subcortical brain sites. In this study, we used seed-based resting-state functional connectivity (rsFC) analysis and voxel-mirrored homotopic connectivity (VMHC) techniques to investigate the alteration of rsFC in patients with sVCI. rsFC and structural magnetic resonance images were acquired for 51 patients with subcortical cerebrovascular disease. All patients were subdivided based on cognitive status into 29 with sVCI and 22 controls; patient characteristics were matched. rsFC of the posterior cingulate cortex (PCC) and VMHC were calculated separately, and rsFC of the PCC and VMHC between the two groups were compared. The regions showing abnormal rsFC of the PCC or VMHC in sVCI patients were adopted as regions of interest for correlation analyses. Our results are as follows: The patients with sVCI exhibited increases in rsFC in the left middle temporal lobe, right inferior temporal lobe and left superior frontal gyrus, and significant decreases in rsFC of the left thalamus with the PCC. sVCI patients showed a significant deficit in VMHC between the bilateral lingual gyrus, putamen, and precentral gyrus. Additionally, the z-memory score was significantly positively associated with connectivity between the left thalamus and the PCC (r = 0.41, p = 0.03, uncorrected) in the sVCI group. Our findings suggest that the frontal lobe and subcortical brain sites play an important role in the pathogenesis of sVCI. Furthermore, rsFC between the left thalamus and the PCC might indicate the severity of sVCI.


Introduction
The dramatic worldwide increase in the proportion of elderly people has brought attention to aging-related cognitive impairments. Dementia is prominent among chronic diseases affecting rsfMRI is a technique that permits assessment of inter-and intrahemispheric FC. Our previous research revealed diffuse alteration of rsFC within the posterior cingulate cortex (PCC) in patients with VCIND [27], The present study uses rsfMRI to assess the integrity of interhemispheric interaction in these participants to determine whether this technique can provide additional evidence of the neuropathological mechanism of VCI; therefore, we mainly focused on confirming whether rsFC with PCC and VMHC were altered in patients with sVCI. Given evidence for structural and functional changes in the frontal lobe and subcortical brain sites (thalamus and caudate) associated with sVCI [9,15], we expected these regions to be particularly affected. We also investigated whether rsFC and VMHC were related to clinical scores in sVCI patients.

Participants
The current study was approved by the Research Ethics Committee of Ren Ji Hospital, School of Medicine, Shanghai Jiao Tong University (China). Each patient gave his or her informed written consent. Seventy-five patients with subcortical cerebrovascular disease were recruited from the Neurology Department of Ren Ji Hospital from July 2012 to May 2014.
To be included in the study, patients with sVCI had to meet the following criteria [28]: (1) the patient or his/her caregiver made a subjective cognitive complaint; and (2) the patient had a subcortical vascular feature, including a focal neurological symptom, any suggestive sign of cerebrovascular disease, or significant white matter hyperintensities (WMH) or lacunar infarcts (as shown on MRI scans). The exclusion criteria were: cerebral hemorrhages; cortical and/or cortico-subcortical, non-lacunar territorial infarcts; watershed infarcts; specific causes of white-matter lesions (e.g., multiple sclerosis, sarcoidosis, and brain irradiation); neurodegenerative diseases (including AD and Parkinson's disease); and signs of normal pressure hydrocephalus or alcoholic encephalopathy. Patients with a low level of education (<6 years), severe depression (Hamilton Depression Rating Scale [HDRS] 18) [29], severe claustrophobia, contraindications to MRI, other psychiatric comorbidities, or severe VCI (inability to perform neuropsychological tests) were also excluded. Patients with severe brain atrophy were excluded to avoid the effect of the atrophy on processing fMRI data. Finally, 51 of the recruited righthanded patients were subdivided based on cognitive status into either the subcortical cerebrovascular disease-normal cognitive group (control group, n = 22) or the sVCI group (n = 29) based on age, sex, and education level.

Magnetic resonance imaging analysis
Two visual rating scales for WMH severity were used on T2-weighted axial scans by two trained neurologists blinded to clinical data. Using the age-related white-matter changes (ARWMC) visual grading scale, the degree of WMH was rated on a four-point scale in five different regions of the right and left hemispheres (frontal, parieto-occipital, temporal, basal ganglia, and infratentorial) [30]. Basic ARWMC scores for each brain region were summed to obtain the total (t) ARWMC score, which was used as a measure of the entire brain WMH load. [30] The Fazekas visual rating scale was applied for separate assessment of periventricular and deep subcortical WMH in all patients [31].

Neuropsychological assessment
Neuropsychological assessments were performed within 1 week of the MRI. None of the patients suffered a new clinical stroke or transient ischemic attack between the MRI and the assessment. The Montreal Cognitive Assessment (MoCA) [32] and Mini Mental State Examination (MMSE) [33], and a comprehensive battery of neuropsychological tests were designed based on a review of relevant published reports. Trail making test A and B, the Stroop color word test, verbal fluency (category) test, auditory verbal learning test (short-delayed and longdelayed free recall), Rey-Osterrieth complex figure test (delayed recall), Boston naming test (30 words), Rey-Osterrieth complex figure test (copy), Lawton and Brody's activities of daily living (ADL) scale test, Barthel Index, HDRS, and neuropsychiatric inventory were included. To assess the cognitive status of patients, the scores for each measurement of normal elderly patients in Shanghai, China, were used as the normal baseline. Cognitive dysfunction was defined as −1.5 SD in at least one neuropsychological test.
Z-score calculation. To allow a direct comparison of different tests, z-scores were generated for the neuropsychological measures. A z-score defines where a score falls in the normal distribution of scores; a z-score of +1.0 corresponds to a score 1.0 SD above the mean score. Direct comparison of performance among tests was possible because z-scores for all tests were based on an identical control population. Raw scores for each of the neuropsychological tests were z-transformed. Thereafter, the z-score of each domain (memory, attention-executive, language, and visual space function) was generated by averaging z-scores of respective tests. Finally, composite Z-scores, indicating general intellect, were computed by averaging the zscores of individual cognitive domains.

Analysis of functional magnetic resonance imaging
Structural brain MRI scans (T1-and T2-weighted images) were inspected by two experienced neuroradiologists. All of the patients were found to have WMH and slight atrophy on the MRI; patients with severe brain atrophy were excluded.
As a central component of the proposed default mode network, the PCC is implicated in attentional processes [34]. A number of studies suggest that the PCC plays an essential role in spatial orientation, self-appraisal, and internal monitoring, as well as in memory processing [35]. Given that the PCC is the seed region most commonly used in studies of the default network, it was used as the region of interest (ROI) to study its altered connections with other brain regions. To calculate rsFC with the PCC, functional images were preprocessed using Data Processing Assistant for Resting-State Functional MR Imaging toolkit (DPARSF3.0 Advanced edition) [36], which synthesizes procedures in the Resting-State Functional MR imaging toolkit (REST; http://www.restfmri.net) [37], and SPM8 (www.fil.ion.ucl.ac.uk/spm). The first 10 images were excluded to ensure steady-state longitudinal magnetization; the remaining images were then corrected for temporal differences and head motion. After patient selection, neither translation nor rotation parameters in any given data set exceeded ±1.0 mm or ±1.0°. Moreover, the mean framewise displacement (FD) was computed by averaging FDi from every time point for each subject [38]. There were no differences in the mean FD between groups (p = 0.26). All normalized T1 images were averaged to generate a mean normalized T1 template. The rsFC calculation procedure was previously described [27], but our own T1 template was used in the present study. The PCC template was selected as the ROI because our previous research mentioned using WFU-Pick Atlas [27,39]. The blood oxygenation leveldependent signal time series in the voxels within the seed region were averaged to generate the reference time series. For each subject and seed region, a correlation map was produced by computing the correlation coefficients between the reference time series and the time series from all other brain voxels. Correlation coefficients were then converted to z-values using the Fisher z-transformation to improve the normality of the distribution [40]. The individual zscores were entered into SPM8 for a one-sample t-test to determine the brain regions with significant rsFC to PCC within each group.
The VMHC calculation procedure was previously described [25,26]. Briefly, all normalized T1 images were averaged to generate a mean normalized T1 image, after which a nonlinear registration was used to normalize each brain to the symmetrical template. The transformation was applied to the symmetrical brain template to the normalized functional data. The VMHC value was then computed as the Pearson correlation (Fisher z-transformation) between the time series data of every pair of symmetrical voxels. Intergroup differences were compared using voxel wise t-tests.

Statistical analyses
All demographics and neuropsychological data analyses were performed using SPSS version 19.0 (IBM Corporation, Armonk, NY, USA). Two-sample t-tests were performed for assessing group comparisons to determine the intergroup demographic differences, and an χ 2 test was used for sex comparison. The Mann-Whitney U test was used for comparing continuous variables if data were not normally distributed. A two-tailed p-value of 0.05 was considered statistically significant for all analyses.
The statistical significance of rsFC and VMHC within groups was analyzed with a one-sample t test (p < 0.05, corrected with a single voxel height of p < 0.01 and cluster volume >594 mm 3 , p < 0.01 and cluster volume >486 mm 3 , respectively), using a software program (AFNI AlphaSim; http://afni.nimh.gov/pub/dist/doc/manual/AlphaSim.pdf).The significant differences in rsFC between groups were analyzed using a two-sample t test (p < 0.05, corrected with a single voxel height of p < 0.01 and a cluster volume >405 mm 3 ) using AFNI AlphaSim and the significant differences in VMHC between groups were analyzed using a two-sample t test (p < 0.05, corrected with a single voxel height of p < 0.01 and a cluster volume > 432 mm 3 ) using the same software program with a gray matter mask produced by the symmetric template previously outlined. Pearson correlations were performed to identify the relationship between the strength of connectivity with the clinical neuropsychological variables. To account for the potential influence of outliers, Shepherd's pi correlation analysis was used [41]. Finally, a two-tailed p value of 0.007 with Bonferroni correction was considered statistically significant.

Demographics, neuropsychological data, and magnetic resonance image analysis
The demographic characteristics, neuropsychological scores, and MRI analyses of the patients are presented in Table 1. Sex, age, and education level did not differ significantly between the two groups. The MoCA, MMSE, composite z-score, and z-scores of the each domain (memory, attention-executive, language, and visual space function) of the sVCI group were significantly lower than those of the controls. The tARWMC and Fazekas PV scores were significantly higher in the sVCI group.
Group differences in resting-state functional connectivity with posterior cingulate cortex Within-group of rsFC with PCC showed in Fig 1A and 1B. In the present study, we just focused on the positive correlations of PCC, that is, a traditional default mode network (DMN) study of the participants. As shown in Fig 1A and 1B the DMN consisted of the PCC, medial prefrontal cortex, left and right superior frontal gyrus, left and right lateral parietal cortex, left and right lateral temporal cortex, left and right parahippocampal gyrus, and thalamus, which is in agreement with the DMN identified in previous studies [42][43][44]. Then, we defined a PCC-FC mask from combing the results of within group t tests of sVCI and control groups to perform the between-group comparisons. Compared with the control group, patients with sVCI exhibited increased rsFC in the left middle temporal lobe, right inferior temporal lobe and left superior frontal gyrus. The left thalamus, exhibited decreased connectivity (Table 2 and Fig 1C).

Group differences in voxel-mirrored homotopic connectivity
Within-group results indicated that both patients (Fig 2A) and control subjects (Fig 2B) had robust homotopic functional connectivity with regional differences in strength. Although sVCI patients showed deficits in VMHC between the left and right lingual gyrus, putamen, and precentral gyrus. They did not show regional VMHC that was greater than that of the controls (Table 3 and Fig 2C).

Relationship between resting-state functional connectivity of the posterior cingulate cortex and voxel-mirrored homotopic connectivity values with clinical variables
To identify the relationship between the strength of rsFC and VMHC and the clinical scores, the mean rsFC or VMHC values were extracted and averaged within a spherical 5.0-mm ROI radius centered on the rsFC or VMHC group difference peak (reported in Figs 1 and 2 and  Tables 2 and 3). Pearson correlations were performed between rsFC or VMHC with MMSE, MoCA, composite z-score, and z-score of each domain within the sVCI group. A significantly positive correlation was observed only in rsFC between the left thalamus and PCC with zmemory score in the sVCI group (r = 0.41, p = 0.03) (Fig 3A), whereas no significant results were found using Shepherd's correlation (Pi = 0.41, p = 0.07) (Fig 3B). No significant correlations survived after Bonferroni correction.

Discussion
There are several causes of reduced interregional correlation, which, in turn, can be construed as dysfunctional connectivity. Some such causes are pathophysiological mechanisms operating within a specific region, such as abnormal synaptic modulation and signaling or loss in the amplitude of oscillations [45,46]. Similarly, a loss in the homogeneity of oscillations of neuronal elements composing a regional node might also manifest as a reduction in interregional correlation [46,47]. The mechanisms underlying the deficits in VMHC are still unknown, but they could be related to widespread white matter-integrity abnormalities [48], such as deficits in white-matter connectivity in the corpus callosum [49,50], and dysfunctions in local graymatter structure [22]. Additionally, alternative (e.g., subcortical) pathways are known to exist, although the callosum is the largest conduit for information transfer and coordination between the hemispheres. sVCI patients were found to have higher tARWMC and Fazekas scores than the control group. Sachdev et al. [51] reported that patients with post-stroke CI had a significantly higher load of total as well as periventricular WMH in terms of absolute volume. Sudo et al. [52] also found that mild VCI was significantly associated with the severity of the whitematter hyperintensities as graded on the Fazekas scale; the higher grades were more frequently associated with the development of cognitive impairments than the lower grades. Lin et al. [53] found that, compared with normally cognitive patients, VCIND decreased fractional anisotropy and increased mean diffusivity in all projection, association, and commissural fibers. Additionally, they found that MoCA scores correlated with diffusion tensor imaging values in all supratentorial WM tracts. It is presumed that more extensive and/or diffuse whitematter lesions have more of an effect on the rsFC and VMHC deficits, but the definite mechanisms need further study. Different rsFC of the PCC were detected in patients with sVCI. Patients with sVCI exhibit increased rsFC in the left middle temporal lobe, right inferior temporal lobe and left superior frontal gyrus. The left thalamus exhibited decreased connectivity, which is partly consistent with our previous research [27]. In addition, deficits in VMHC were detected between left and right lingual gyrus, putamen, and precentral gyrus in sVCI patients. These changes might be caused by subcortical white-matter lesions that destroyed direct and indirect fiber tract connectivity across the cerebral white matter and influenced cortical rsFC and hypoperfusion resulting from subcortical cerebrovascular disease. This increased connectivity might be evoked by compensatory recruitment and plasticity mechanisms [27].
The lingual gyrus is a brain structure that is linked to vision processing, especially that related to letters. Ghosh [54] found that a cerebellar-occipital-thalamic network is activated when the semantic relatedness of words are assessed by the brain. Semantic association tasks, in addition to being a measure of language fluency, are an indication of the extent of executive function and working memory; therefore, these areas could play a role in the transfer of semantic information to a more permanent storage format in the cortical association areas. The thalamus is a crucial brain area that processes and integrates neural activity from widespread neocortical inputs and outputs [55] and is believed to coordinate information and facilitate communication (e.g., memory, attention, and perception) in a number of areas in the cerebral cortex [56,57] The thalamus pulvinar has widespread connectivity with the posterior parietal lobe and the precuneus. Our results revealed a significantly positive correlation in rsFC of the left thalamus with a z-memory score in the sVCI group, suggesting that the alteration in rsFC might estimate the severity of cognitive impairment and that alteration in the function of the thalamus plays a role in the pathology of sVCI.
The DMN is a collection of brain regions which are typically deactivated in goal-directed tasks and activated during rest periods [58]. Providing insight into function, the default network is active when individuals are engaged in internally focused tasks including autobiographical memory retrieval, envisioning the future, and conceiving the perspectives of others  [59]. As one of the part of the DMN, lateral temporal gyrus provides information from prior experiences in the form of memories and associations that are the building blocks of mental simulation [59].Other areas associated with cognitive control were also detected in the present research. The precentral gyrus and prefrontal cortex have been implicated in personality expression, decision making, planning complex cognitive behavior, planning and executing movements, and moderating social behavior, and thus changes in these regions would not be unexpected [60]. The putamen receives inputs from the association and sensorimotor areas of the cortex while the globus pallidus projects to the dorsomedial nuclei of the thalamus [61]. Sefcsik et al. [62] suggested that the frontal/subcortical circuit between the putamen and frontal motor areas plays a role in higher cognitive processing, such as executive functions, and firstorder sequence learning.

Limitations of the Study
There were some limitations in this study. First, this study used a small sample size. Second, we used clinical and not autopsy data for patient classification. AD and vascular diseases are common in old age; therefore, we unavoidably involved AD-related disease. Third, all the patients, not only VCI patients but also controls, had lacunar infarcts and white-matter hyperintensities, and because the lesions were very small (<1.5 cm) and distributed randomly throughout the brain, it was difficult to remove all of them; therefore, we processed the data as usual and did not control or consider the total number of lacunes. Fourth, although Li et al. [11] suggested that functional abnormalities can be only partly explained by morphological changes in graymatter volume in VCI patients, the atrophy of gray-matter volume in VCI brains might have a potential impact on our results. We carefully used our own template in data preprocessing to control the effect of atrophy. Fifth, this was a cross-sectional study; therefore, it could not determine whether abnormal brain activation is a state marker or trait marker of sVCI in resting-state fMRI, and we did not follow the patients to study their functional brain changes. Finally, the correlation results were not consistent when we adopted multiple comparisons (Bonferroni correction), which means that this should only be considered an exploratory analysis. To increase the statistical power, the findings should be repeated with a larger sample of subjects.

Conclusions
We found altered connectedness of the rsFC with PCC and VMHC in the VCI patients. This change was most likely caused by subcortical white-matter lesions that destroyed direct and indirect fiber tract connectivity across the cerebral white matter and influenced the cortical rsFC and hypoperfusion that resulted from subcortical cerebrovascular disease. rsFC correlations between the left thalamus and PCC might indicate the severity of sVCI, which might help us to understand the pathogenesis of sVCI.