White Matter Hyperintensities and Mild Cognitive Impairment in Parkinson’s Disease Elijah Mak, Michael G. Dwyer, Deepa P. Ramasamy, Wing Lok Au, Louis C.S. Tan, Robert Zivadinov, Nagaendran Kandiah From the Buffalo Neuroimaging Analysis Center, Department of Neurology, School of Medicine and Biomedical Sciences, University at Buffalo, State University of New York, Buffalo, NY (EM, MGD, DPR, RZ); Department of Neurology, National Neuroscience Institute, Singapore (EM, WLA, LCST, NK); Duke-NUS Graduate Medical School, Singapore (WLA, LCST, NK); and MR Imaging Clinical Translational Research Center, School of Medicine and Biomedical Sciences, University at Buffalo, State University of New York, Buffalo, NY (RZ).
ABSTRACT OBJECTIVES: The clinical implications of white matter hyperintensities (WMH) in non-demented Parkinson’s disease (PD) have not been thoroughly examined. To address this, we investigated the spatial distribution of WMH and their regional predilection in non-demented patients with mild PD. METHODS: Cognitive assessments classified the sample into patients with mild cognitive impairment (PD-MCI, n = 25) and patients with no cognitive impairment (PD-NCI, n = 65) based on the recent formal Movement Disorder Task Force diagnostic criteria. The mean age was 65.1 ± 7.7 years, disease duration was 5.3 ± 3.9 years, and Hoehn and Yahr stage was 1.9 ± .4. WMHs were outlined on T2-weighted imaging using a semi-automated technique. The spatial distribution of WMHs were compared between PD-MCI and PD-NCI using voxel-wise lesion probability maps (LPM). General linear models examined the associations between spatially specific WMHs and cognitive domains. RESULTS: LPM analyses showed significant differences in the spatial distribution of WMH in PD-MCI compared to PD-NCI in widespread regions of the brain (P < .05). PD-MCI demonstrated significantly greater total and periventricular WMHs compared to PD-NCI (P ࣘ .02). Spatial distribution of WMHs was also significantly associated with global cognition, performance on the Frontal Assessment Battery and Fruit Fluency (P < .05). CONCLUSIONS: Voxel-wise LPM analysis revealed differences in the spatial distribution of WMH between PD-MCI and PD-NCI patients, particularly in the periventricular regions. A more widespread extent of WMH might be indicative of cognitive deterioration. Our findings warrant further longitudinal investigation into the importance of WMH spatial distribution as a predictor for conversion from PD to PD with dementia.
Keywords: White matter hyperintensities, cognitive impairment, Parkinson’s disease, mild cognitive impairment, neuroimaging. Acceptance: Received December 29, 2014. Accepted for publication January 27, 2015. Correspondence: Address correspondence to Robert Zivadinov, Department of Neurology, School of Medicine and Biomedical Sciences, The Jacobs Neurological Institute, 100 High St, Buffalo, NY 14203, USA; E-mail: [email protected]
Disclosure: This research was supported by the Singapore National Research Foundation and Singapore Ministry of Health’s National Medical Research Council and by the Buffalo Neuroimaging Analysis Center neuroimaging fellowship to Elijah Mak. J Neuroimaging 2015;25:754-760. DOI: 10.1111/jon.12230
Introduction While Parkinson’s disease (PD) traditionally has been defined by its characteristic motor hallmarks, non-motor features such as cognitive impairment and dementia are increasingly recognized as part of PD. Recent studies suggest that about 30–35% of patients with early PD experience cognitive disturbances.1,2 The impact of cognitive impairment and dementia in PD is substantial, with adverse consequences for functioning3 and mortality.4 In fact, mild cognitive impairment in PD (PD-MCI) has been recognized as a transitory state between normal aging and dementia.5 At present, there is much to be elucidated with regards to the etiology of cognitive impairment in PD. Theories point to striatal dysfunction and global neurotransmitter system deficits6 and subcortical atrophy.7 At the same time, pathological explanations implicate Lewy body degeneration and Alzheimer-type changes.8
Additionally, cognitive deterioration and progression of dementia in PD has been associated with cerebrovascular diseases, including white matter hyperintensities (WMHs).9–12 A recent longitudinal study has also demonstrated that WMH burden is a significant predictor of conversion from PD-MCI to PD with dementia.13 However, the clinical picture surrounding the role of WMH in cognitive dysfunction remains a contentious topic with conflicting findings.14 We have previously reported significantly greater WMH volumes in PD-MCI compared to PD subjects with no cognitive impairment (PD-NCI).15 In addition to volumetric assessment, complementary evidence can be obtained from the analyses of lesion probability maps (LPM), which has been increasingly recognized as a powerful tool to characterize frequency patterns and probability of occurrence of WMHs. This relatively novel approach has been applied in normal aging16 and various pathologies such as multiple sclerosis,17 however only one study
◦ 2015 by the American Society of Neuroimaging C
to date has employed LPM to interrogate WMH spatial distributions in a large cohort of newly diagnosed PD,14 finding no evidence of differential spatial distributions between PD-MCI (n = 30), PD-NCI (n = 133), and healthy controls (n = 102). By comparing the spatial distributions of WMHs between PD-MCI and PD-NCI, and testing their associations with cognitive function, we sought to examine the potential role of spatial distribution of WMH as a prognostic indicator of cognitive deterioration in PD.
Methods Subjects The present study included 90 mild PD patients (65.1 ± 7.71 years old, with a disease duration of 5.3 ± 3.90 years, and Hoehn and Yahr = 1.9 ± .4), who were prospectively recruited from August 2011 to March 2012 from a tertiary neurology centre in Singapore.15 PD was diagnosed by neurologists trained in movement disorders according to the National Institute of Neurological Disorders and Stroke (NINDS) criteria.18 DSM-IV criteria were used to classify patients with dementia. PD patients with dementia, serious medical, and psychiatric co-morbidities were excluded.
Clinical Assessment Demographic, clinical, and vascular risk factor data were collected, and a comprehensive clinical assessment was conducted to ascertain cognitive status and functional ability. The severity and stage of the patient’s parkinsonism was evaluated using the Unified Parkinson’s Disease Rating Scale (UPDRS) motor subscore19 and the modified Hoehn and Yahr stage.20 To standardize data on medication use, we converted dosages of PD medications to total daily levodopa-equivalent doses. This calculation was based on the conversion formulae reported by Tomlinson et al.21 The study was approved by the centralized Institutional Review Board and informed consent was obtained from patients or their legal caregivers.
Neuropsychological Assessment Cognitive performance was evaluated by trained psychologists using a standardized neuropsychological battery. Global cognition was evaluated using the Mini Mental State Examination (MMSE)22 and Montreal Cognitive Assessment (MOCA).23 As per the recommendations of the Movement Disorder Society (MDS) Task Force 2012, specific cognitive domains including memory, executive function, visuospatial function, language, and attention/working memory were also assessed.5 Episodic memory was assessed using sub-tests from the Alzheimer’s Disease Assessment Scale 11,24 including Word-List Immediate, Delayed and Recognition Recall; executive function was evaluated with the Frontal Assessment Battery (FAB)25 and the 10point clock drawing test; visuospatial function was assessed with a figure copy test and number of errors made on a Maze test; language was assessed with a 20-point object naming test and semantic fluency; attention/working memory was assessed with digit span, color trails 2 and time taken on a Maze test.24,26 Performances on individual tasks were transformed into z-scores. Subsequently, a composite summary index for each cognitive domain was derived from the corresponding averages of the respective individual neuropsychological tests.
MCI Classification To qualify for MDS Level 2 criteria for PD-MCI, performance on the suggested 5 cognitive domains (attention/working memory, executive, language, episodic memory, and visuospatial) were analyzed. Cutoff scores for the various cognitive tests were based on locally validated norms when available, and for those without, international ones were used. The performance on a cognitive test was considered abnormal if the score was 1.5 SDs below the norm. Impairment on at least 2 neuropsychological tests, represented by either 2 tests showing impairment in 1 cognitive domain or 1 test showing impairment in 2 different cognitive domains, was required. Patients with PD who did not fulfill the criteria for PD-MCI or PD-dementia were classified as PD-NCI.
Image Acquisition All subjects underwent MRI imaging on a 3 T whole body MRI system (Achieva 3.0, Philips Medical Systems, Best, The Netherlands) with a SENSE 8 channels head coil. Highresolution volumetric 3D T1-weighted MPRAGE sequence (axial acquisition, TR 7.1 ms, TE 3.3 ms, TI 850 ms, matrix 256 × 256, slice thickness 1 mm, total 180 slices, voxel-size 1 mm × 1 mm × 1 mm, scan time 5:13) and whole brain 3D fluid attenuated inversion recovery (FLAIR) sequence (turbo spin echo, TR 8000 ms, TE 340 ms, T1 2400 ms, matrix 256 × 256, slice thickness 2 mm, total 170 slices, voxel-size 1 mm × 1 mm × 1 mm, scan time 10:24) were acquired for all patients. Both clinical testing and MRI scanning were done on the same day for all patients.
Image Analysis Quantitative image analyses were performed at the Buffalo Neuroimaging Analysis Center, Buffalo, NY, USA.
White Matter Hyperintensities Segmentation WMHs were outlined on each axial FLAIR image slice using a reproducible, semi-automated edge detection contouring/thresholding technique with Java Image software (version 6.0, Xinapse Systems, Northants, UK, http://www.xinapse. com). All WMH masks were created by a single rater (M.F.K) and divided into individual region of interests (ROIs) ࣙ3 mm in size (equivalent to ࣙ14.1 mm3 ), with similar reproducibility as previously reported, blinded to clinical characteristics and tests results.27 The regional localization of WMHs was determined based on their presence in the periventricular, deep subcortical, juxtacortical and infratentorial WM regions. Hyperintensities in deep gray matter were also assessed. Cortical lesions were not assessed as visualization of cortical lesions was not optimal on FLAIR sequences and double inversion recovery was not performed.
White Matter Hyperintensities Voxel-Wise Analysis We employed a permutation-based, non-parametric voxel-wise statistical mapping technique to investigate spatial WMH differences between PD-MCI and PD-NCI. First, we co-registered all FLAIR images to the Montreal Neurological Institute standardspace image (MNI152 template) using a 12-parameter affine model via ANTS non-linear registration tool (http://picsl. upenn.edu/software/ants/). The resulting transformation matrices were then applied to the segmented WMH mask images to Mak et al: WMH and Cognitive Impairment in PD
align these into the same standard space. All registrations were verified visually to exclude alignment failures. Prior to statistical comparison, all standard-space WMH mask images were smoothed using a Gaussian kernel with a standard deviation of 8 mm. The resulting blurred lesion mask is referred to as the lesion probability map for that particular subject. Voxel-wise statistics were subsequently carried out to give lesion probability maps at each standard space voxel. Within these maps, the probability of finding a lesion in any given voxel is defined by the relative voxel intensity. Voxel values were then used as dependent variables in a mass-univariate application of the general linear model. Group membership (PD-MCI or PD-NCI) was entered into the model as explanatory variable, and t-contrasts were created to assess spatially specific differences between both groups while correcting for nuisance covariates such as age, sex, and education. Indices of disease severity, including UPDRS and Hoehn and Yahr stages, were not included as covariates in the model because they were similar between both groups. To test our hypotheses that WMH are associated with both global and domain-specific cognitive dysfunction, additional t-contrasts were created to assess the relationships between all cognitive domains and spatially specific presence of WMHs within the total PD sample. Age, gender, and education were included into the general linear model, and vascular risk factors as explanatory variables. To limit the number of explanatory variables in the model, vascular risk was rated as the presence of one or more of the following: hypertension, hyperlipidemia, diabetes mellitus, and smoking history. Statistical significance was assessed via non-parametric permutation testing28 rather than parametric Gaussian random field theory as the null distribution is not known. Group labels were randomly permuted 5,000 times for each contrast to build up an empirically derived null distribution against which to compare observed effects. Furthermore, we opted to use the novel threshold-free cluster enhancement (TFCE) technique29 rather than the more standard a priori specification of a cluster-forming threshold. By enhancing individual voxels proportionally while still facilitating a fundamentally voxel-wise comparison, TFCE removes the need for manual and often arbitrary selection of an initial cluster-forming threshold. Furthermore, by performing TFCE enhancement within the permutation framework, statistical rigor is still maintained.29 For all voxel-wise analyses, results with P < .05 corrected for family-wise error (FWE) rate were regarded as significant.
Statistical Analysis Group comparisons of demographics, neuropsychological variables, and WMH number and volumes were performed using STATA12 (STATA Corp). The Student’s t-test or the non-parametric Mann-Whitney rank sum tests were used to investigate differences between groups depending on the normality of the distributions. Chi-square test was used for categorical variables (gender and vascular risk factors). To compare cognitive performances and total and regional WMH volumes between PD-MCI and PD-NCI, analysis of covariance (ANCOVA) was employed, including age, gender, and education as covariates. Regional WMH analyses between PD-MCI and PD-NCI were further adjusted by 756
Benjamini-Hochberg correction for multiple comparisons. For all analyses, two-tailed P-values were used and P < .05 was considered as significant.
Results Subject Demographic and Clinical Characteristics Within the PD group, 25 were classified as PD-MCI and 65 as PD-NCI patients. Group comparisons and clinical and demographic data between PD-MCI and the PD-NCI are shown in Table 1. The PD-MCI group was significantly older than PD-NCI, while PD-NCI group had a higher education level. Both groups were comparable in terms of duration of disease, UPDRS scores and Hoehn and Yahr staging. Group analysis of the various cardiovascular risk factors demonstrated a significantly higher prevalence of diabetes mellitus (P = .004) and hyperlipidemia (P = .031) in PD-MCI compared to PD-NCI patients.
Cognitive Performance PD-MCI patients had significantly lower scores on global cognition compared with PD-NCI (MMSE, P = .016; MOCA, P = .018; Global Index, P = .008) after correcting for age, gender, and education years. Comparisons of individual neuropsychological tests are shown in Table 1. With the exception of visuospatial ability and episodic memory domains, the PD-MCI group performed significantly poorer on executive functioning, attention, working memory and language abilities (P ࣘ .02).
White Matter Hyperintensities Comparisons Total and periventricular WMH volumes were significantly greater in PD-MCI group compared to PD-NCI group after correction for multiple comparisons (P ࣘ .02). PD-MCI also had greater volume of deep subcortical WMH volume (Table 2), but this was not statistically significant after correcting for age, gender, education, and multiple comparisons. Voxel-wise comparisons of lesion probability maps between PD-MCI and PD-NCI showed widespread areas of significant differences in the spatial distribution of WMH between PD-MCI and PD-NCI patients (Fig 1). The reverse tcontrast (PD-NCI > PD-MCI) did not yield any significant result.
Correlation between WHM Location and Cognitive Measures We found significant widespread voxel-wise relationships between spatial distribution of WMHs and global cognition (MMSE), after family-wise error correction for multiple comparisons while controlling for age, gender, education, and vascular risk factors (Fig 2). In addition, the spatial distribution of WMHs was also significantly associated with poorer performance on the FAB and Fruit Fluency tasks (Fig 3). At the uncorrected level, spatial distribution of WMHs was negatively associated with overall executive function, attention and working memory, episodic memory and language domains, with the exception of visuospatial ability after correcting for age, gender, and education and vascular risk factors (data not shown).
Journal of Neuroimaging Vol 25 No 5 September/October 2015
Table 1. Demographic, Clinical, and Neuropsychological Characteristics among Parkinson’s Disease (PD) Patients with Mild Cognitive Impairment (PD-MCI) and PD Patients with No Cognitive Impairment (PD-NCI)
Age in years, mean (SD) Sex, male n (%) Education, years, mean (SD) Hoehn and Yahr, mean (SD) Disease duration in years, mean (SD) UPDRS, mean (SD) Levodopa equivalent dose, mg, mean (SD) Cardiovascular risk factors, n (%) Diabetes Hypertension Hyperlipidemia Smoking Global cognition, mean (SD) MMSE MOCA ADAS11 Global Index Cognitive domain, mean (SD) Executive function Attention/Working Memory Visuospatial Episodic memory Language
PD-NCI (N = 65)
PD-MCI (N = 25)
63.4 (7.6) 46 (72.3) 11.0 (3.1) 1.9 (.4) 5.4 (4.3) 17.5 (7.0) 557.4 (375.7)
69.4 (6.4) 18 (76.0) 9.3 (3.5) 1.8 (.4) 5.0 (2.7) 20.0 (8.4) 510.2 (299.0)
.001a* .723c .032b* .357b .910b .167a .767b
5 (7.8) 20 (31.3) 20 (31.3) 15 (23.4)
.004c* .069c .031c* .955c
8 (32.0) 13 (52.0) 14 (56.0) 6 (24.0)
28.4 (1.6) 27.0 (2.9) 6.8 (4.2) 0.1 (0.5)
26.7 (2.6) 24.5 (2.4) 9.8 (3.5) −0.3 (0.3)
.016d* .018d* .071d* .008d*
0.4 (1.5) 0.8 (1.7) 0.0 (1.8) 0.3 (1.6) 0.4 (1.4)
−1.0 (2.0)) −2.0 (2.4) −0.1 (1.4) −0.6 (1.5) −1.0 (1.5)