Amnestic mild cognitive impairment in Parkinson’s disease: White matter structural changes and mechanisms

Mild cognitive impairment (MCI) is a heterogeneous cognitive disorder that is often comorbid with Parkinson’s diseases (PD). The amnestic subtype of PD-MCI (PD-aMCI) has a higher risk to develop dementia. However, there is a lack of studies on the white matter (WM) structural changes of PD-aMCI. We characterized the WM structural changes of PD-aMCI (n = 17) with cognitively normal PD (PD-CN, n = 19) and normal controls (n = 20), using voxel-based and tract-based spatial statistics (TBSS) analyses on fractional anisotropy (FA) axial diffusivity (AD), and radial diffusivity (RD). By excluding and then including the motor performance as a covariate in the comparison analysis between PD-aMCI and PD-CN, we attempted to discern the influences of two neuropathological mechanisms on the WM structural changes of PD-aMCI. The correlation analyses between memory and voxel-based WM measures in all PD patients were also performed (n = 36). The results showed that PD-aMCI had smaller FA values than PD-CN in the diffuse WM areas, and PD-CN had higher AD and RD values than normal controls in the right caudate. Most FA difference between PD-aMCI and PD-CN could be weakened by the motor adjustment. The FA differences between PD-aMCI and PD-CN were largely spatially overlapped with the memory-correlated FA values. Our findings demonstrated that the WM structural differences between PD-aMCI and PD-CN were mainly memory-related, and the influence of motor adjustment might indicate a common mechanism underlying both motor and memory impairment in PD-aMCI, possibly reflecting a predominant influence of dopaminergic neuropathology.


Introduction
In summary, to characterize the WM structural changes of PD-aMCI and better understand its underlying neuropathological mechanisms, we performed two layers of analyses: 1) to compare the WM structural differences between PD-aMCI and PD-CN and normal controls using voxel-based and TBSS-based DTI metrics; 2) to exclude and then include the motor performance as a covariate in the WM structural comparison between PD-aMCI and PD-CN to discern the influences of two neuropathological mechanisms. Moreover, we performed the correlation analyses between memory and voxel-based WM measures in all PD patients, to localize the memory-associated WM structures.

Participants
Thirty-six PD participants were randomly recruited from the Neurological Clinics of the First Affiliated Hospital of Fujian Medical University. The diagnosis of PD followed the UK Brain Bank criteria for idiopathic Parkinson's disease [23], and the Hoehn and Yahr (H&Y) stages of PD were evaluated [24]. Motor function of the PD patients was assessed with the Movement Disorder Society (MDS) modified version of the Unified Parkinson's Disease Rating Scale motor examination (UPDRS-III) [25]. We excluded the patients who were not at H&Y stage I-III or had been diagnosed with any of the following conditions: dementia based on DSM-IV criteria [26], a history of brain surgery, stroke, epilepsy, multiple sclerosis, progressive malignancy (active cancer or receiving radiotherapy for cancer, other than prostate or mild skin cancer), schizophrenia, bipolar disorder, and developmental disability. We also excluded the PD patients who had a Mini-Mental State Examination score (MMSE) � 24 [27].
We recruited 20 healthy older adults, who did not have any known or suspected history of neurological illness or psychiatric impairments including MCI. The normal controls (n = 20) and PD patients (n = 36) were matched on age, gender, years of education and MMSE.
This study was approved by the ethics committee of Fujian Medical University and written informed consents were obtained from all participants.

MCI classification
The Chinese version of Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) [28] were administered by trained medical graduates to assess cognitive functions of the participants. The RBANS consists of twelve subtests that evaluate five cognitive domains, which are immediate memory (list learning and story memory), delayed memory (list recall, list recognition, story recall, and figure recall), language (picture naming and semantic fluency), attention (digit span and coding), and visuospatial function (figure copy and line orientation) [29]. The RBANS index score of each cognitive domain was the average T-score of the subtests that consist of that domain; raw score of each subtest of each individual was firstly standardized to z-score with the mean and standard deviation of that subtest in all participants, and then transformed to T-score with a mean of 100 and standard deviation of 15. In line with the recommendations of the Movement Disorders Society on PD-MCI [30], the diagnosis of aMCI in PD patients was operationalized by 1) no significant impairment in activities of daily living according to medical history; 2) a report of memory complaints, either from the participants or their informant; and 3) memory impairment defined by the performance on at least two tests in the immediate and/or delayed memory domains� -1.5 standard deviation (SD) of the published normative values [31].
Seventeen PD patients met the criteria for PD-aMCI. Nineteen PD patients did not fulfill the criteria for MCI and were classified as cognitively normal PD (PD-CN). No significant differences were found between PD-aMCI and PD-CN on age, gender, years of education, disease duration, H&Y stage, UPDRS-III, Levodopa equivalent daily dose and MMSE.

Image acquisition
MRI scans of all participants were acquired on a 3.0 T SIEMENS Magnetom Verio MRI scanner (Siemens Medical Solutions, Erlangen, Germany) at the Department of Radiology of the First Affiliated Hospital of Fujian Medical University. All PD participants and healthy controls underwent a standardized brain MRI protocol. The PD participants were assessed "ON" their usual dopaminergic medication. DTI scans were acquired using a single-shot echo-planar imaging-based sequence with the following parameters: TR = 4316 ms, TE = 95 ms, flip angle = 90˚, voxel size = 2 x 2 x 4.55 mm 3 , acquisition matrix = 128 x 128, FOV = 256 x 256 mm 2 , 62 non-linear diffusion weighting directions with b = 1,000 s/mm 2 and one image without diffusion weighting (i.e., b = 0 s/mm 2 ).

Image processing
After visually inspecting MRI scans for structural abnormalities, the DTI dataset was processed with the FSL 5.0 package (http://www.fmrib.ox.ac.uk/fsl/). The raw DTI images were firstly undergone head motion, eddy current correction, and skull stripping; then the diffusion tensor was reconstructed by fitting a diffusion tensor model for each image [16]. Diffusion metrics of FA, AD, and RD were calculated from the diffusion-weighted images. Subsequently, the individual voxel-wise maps of FA, AD, and RD were spatially transformed to the MNI standard space in a 2 x 2 x 2 mm spatial resolution; and then finally smoothed with a 6-mm FWHM Gaussian kernel for voxel-based DTI statistical analysis. The FA images were then averaged to generate a mean FA image, which was used to create an FA skeleton by selecting the voxels with the locally maximal FA values. The generated FA skeleton was then thresholded at FA > 0.2 to minimize the partial volume effect and cross-subject misregistration [15]. Subsequently, the diffusion metrics of FA, AD, and RD were individually projected onto that skeleton, and the resultant maps of FA, AD, and RD on the skeleton were fed into the TBSS-based DTI statistical analyses.

Statistical analysis
We compared the differences between PD-aMCI and PD-CN, or between PD-CN and normal controls, on voxel-based DTI metrics (FA, AD, and RD). The analysis between PD-aMCI and PD-CN could reveal the WM structural changes associated with aMCI in PD patients, while the analysis between PD-CN and normal controls could exhibit the WM structural changes associated with PD. The controlled covariates included age, gender, and disease duration (only for the comparison between PD-aMCI and PD-CN). We also performed TBSS-based group comparison analyses, controlled for the same covariates to validate the results of voxel-based analysis. UPDRS-III was then added in the comparison analyses between PD-aMCI and PD-CN on voxel-based DTI measures as a covariate.
Furthermore, correlation analyses were performed between two memory scores (immediate memory and delayed memory) and voxel-based DTI measures in all PD participants. The controlled covariates were age, gender, and disease duration. We then used the conjunction overlay method [32] to overlay the structural correlation map of memory upon the structural difference map between PD-aMCI and PD-CN, and the common areas from the two maps would indicate the correspondence. UPDRS-III was then added as a covariate in the WM structural correlation analyses of memory, to verify if the motor adjustment could also affect the memory-WM relationship.
In addition, we performed the correlation analyses for language, attention, visuospatial function, and UPDRS-III with voxel-based DTI measures in all PD participants, controlled for age, gender, and disease duration. The partial correlations between UPDRS-III and each cognitive performance were also examined in all PD patients, controlled for age, gender, and disease duration.
All statistical analyses with WM structural measures were implemented in FSL 5.0, and the correction for multiple comparisons employed the threshold-free cluster-enhancement (TFCE) method, which is a non-parametric permutation test [33,34]. We performed 5000 permutations and set the significance threshold at p<0.05 (FWE-corrected). The supra-threshold clusters of DTI metrics were superimposed on a series of brain slices in the MNI 152 T1 brain template. The significant results on TBSS-based diffusion metrics were dilated to enhance visualization.

Results
The demographic characteristics and cognitive performance for different PD groups and normal controls were shown in Table 1.
Compared to normal controls, PD-CN patients showed higher AD and RD values in the right caudate ( Table 2, Fig 1A and 1B). Compared to PD-CN, PD-aMCI patients showed lower FA values in the corpus callosum (splenium and body), posterior thalamic radiation, posterior corona radiata, tapetum, cingulum (cingulate gyrus) in the bilateral hemispheres, the left superior corona radiata and fornix (crux), and the right superior longitudinal fasciculus ( Table 2 and Fig 2A). TBSS-based FA values were compared between PD-aMCI and PD-CN: PD-aMCI showed lower FA values in the WM tracts of the corpus callosum (body and splenium), superior and inferior longitudinal fasciculus, cingulum (cingulate gyrus), and inferior fronto-occipital fasciculusin the bilateral hemispheres (Table 2 and Fig 2B). By visualizing Fig  2A and 2B, a similar spatial distribution was noted in the two FA-difference maps of PD- aMCI. No significant GM differences were found between PD-aMCI and PD-CN, as well as between PD-CN and controls. The adjustment of UPDRS-III in the comparison between PD-aMCI and PD-CN weakened all FA differences to non-significant at the threshold of p<0.05 (FWE-corrected); however, some FA differences could survive at the threshold of p<0.07 (FWE-corrected), located in the right posterior thalamic radiation, posterior corona radiata, and tapetum (Table 3 and Fig 3).
The structural correlation analyses of memory in all PD patients showed that delayed memory was positively correlated with voxel-based FA values in diffuse areas (Table 4, Fig 4A). The conjunction overlay demonstrated that the FA correlates of delayed memory were mostly overlapped with the FA differences between PD-aMCI and PD-CN, including the bilateral corpus callosum (splenium and body), cingulum (cingulate gyrus), posterior corona radiata, tapetum, the left fornix and superior corona radiata, as well as the right posterior thalamic radiation and superior longitudinal fasciculus (Fig 4B). The adjustment of UPDRS-III  weakened all previously significant FA correlates of delayed memory to non-significant at the threshold of p<0.05 (FWE-corrected); however, a few FA correlates could survive at the threshold of p< 0.07 (FWE-corrected), located in the bilateral corpus callosum (body and splenium), right posterior cingulum, posterior thalamic radiation, posterior corona radiata, and tapetum (S1 Table and S1 Fig).
Attention was positively correlated with the FA values in the right corpus callosum splenium and posterior corona radiata (S2 Table). The FA correlates of attention shared a small extent of common area (only 4 voxels) with the FA differences between PD-aMCI and PD-CN

FA changes in PD-aMCI
No prior study examined the WM changes of PD-aMCI. Using voxel-based and TBSS-bases analyses, we demonstrated a diffuse pattern of FA decreases of PD-aMCI compared to PD-CN, located in the corpus callosum (splenium and body), posterior thalamic radiation, posterior corona radiata, tapetum, and cingulum (cingulate gyrus). A few studies investigated the WM changes of PD-MCI, without specifying the subtype of MCI, and revealed significant FA decreases across different WM tracts in PD-MCI, compared to PD-CN [10,13]. Given that a lower FA value indicates decreased fiber integrity induced by demyelination [14,35,36], our findings possibly suggested that PD-aMCI might have an extensive fiber integrity disruption, relative to PD-CN.

FA changes of PD-aMCI mostly overlapped with the FA correlates of delayed memory
By the conjunction overlay, we demonstrated that the FA difference between PD-aMCI and PD-CN were mostly overlapped with the FA correlates of delayed memory. This finding indicated that decreased fiber integrity in PD-aMCI was mainly contributed by the WM abnormality associated with memory decline. In healthy older adults, the WM measures associated with memory were mainly localized in the fornix and cingulum [37][38][39]. However, in PD patients,    several studies showed a diffuse WM correlation pattern with memory, which have been demonstrated in the fornix, cingulum, corpus callosum and posterior corona radiate, consistent with our findings [40][41][42].

AD and RD changes in PD-CN
PD-CN showed significantly higher AD and RD values than normal controls in the right caudate, whilst no difference in FA values. AD and RD estimate the diffusivity of water molecule along and perpendicular to the direction of WM tracts, respectively [14]. Thus, when water diffusivity in different directions is proportional, a minimal change in FA would be expected [43]. Such patterns of changes in AD, RD, and FA have been suggested as an indication of enlarged extracellular space, which could be due to neurodegeneration involving neuronal death and fiber shrinkage [44][45][46]. The degeneration of dopaminergic neurons in the basal ganglia is a pathological feature of PD [17,18]. Caudate is an essential component of the cortico-basal ganglia-thalamo-cortical pathways to regulate movement [47]. The changes of MD and FA in the caudate have been reported in PD patients [48,49]. Moreover, several studies demonstrated the dopaminergic denervation, GM atrophy, and morphological deformation in the caudate in PD patients [50][51][52][53]. These lines of evidence were consistent with our findings on the increases of AD and RD in the caudate, suggesting a possible neurodegeneration in this area in PD patients.

Possible neuropathological mechanisms of PD-aMCI
Our study showed that the adjustment of UPDRS-III could weaken the originally significant FA differences between PD-aMCI and PD-CN and FA correlates of memory to non-significant. Motor and memory are two distinct brain functions with diverse neural substrates, which are normally not related. However, in PD-aMCI, the impairments in both functions might be subject to a common dopaminergic pathological influence [17][18][19]. Meanwhile, evidence suggested that the cholinergic mechanism was also involved in the PD-aMCI [8,20,21].
The extensive influence of motor adjustment as shown in our results might reflect a predominant influence of dopaminergic neuropathology [19]. However, we also noted a tendency towards significance for some voxel-based FA differences between PD-aMCI and PD-CN, which were located in the posterior cingulum, posterior thalamic radiation, and tapetum. These regions were adjacent to the fornix and hippocampal formation, which might suggest the specific influence of cholinergic neuropathology on memory [37][38][39].
Our findings were in accordance with the theory of dual neuropathological mechanisms in PD-aMCI. It has been suggested that the two neuropathological mechanisms could drive synergistically neurodegenerative processes to undermine brain structural integrity of PD-aMCI and accelerate its conversion to dementia [8,54].

WM structural correlation analyses of attention, language, visuospatial function in all PD patients
Only attention showed positive correlations with the FAs in the right corpus callosum splenium and posterior corona radiata, while language and visuospatial function showed null correlations. These results were expected, given that PD-aMCI and PD-CN had no significant differences on language and visuospatial function, which lead to a small inter-individual variation in all PD participants and reduced likelihood for a significant cognitive correlation. Moreover, different from memory, the FA correlates of attention had a small spatial overlap (4 voxels) with the FA differences between PD-aMCI and PD-CN, suggesting a limited contribution of attention-related WM changes to the WM abnormality of PD-aMCI.

Limitation
Despite that the number of our PD-aMCI patients is reasonable comparing to prior neuroimaging studies on PD-MCI, our sample size is small. To alleviate the concern on the robustness of the results, we performed the TBSS-based FA analysis as a complementary method to validate the results of voxel-based FA analysis. The results indicated a similar distribution pattern between the voxel-based and TBSS-based FA difference maps. Moreover, we employed the threshold-free cluster-enhancement (TFCE) method, which has been recommended as a stringent statistical method in voxel-wise analysis [33,34].

Conclusion
No previous neuroimaging studies have examined the WM structural changes of a specific PD-MCI subtype, PD-aMCI. Our study showed a diffuse FA decrease pattern for PD-aMCI compared to PD-CN. However, most FA difference between PD-aMCI and PD-CN could be weakened by the adjustment of motor performance, which might indicate a predominant influence of dopaminergic neuropathology. Yet, some FA differences of PD-aMCI adjacent to the fornix and hippocampal formation showed a tendency to survive the motor adjustment, which might reflect a memory-specific influence by the cholinergic neuropathological mechanism.
Supporting information S1 Table. Controlling for UPDRS-III in the correlation analysis between delayed memory and voxel-based FA in all PD patients.