The protocol was approved by the University of California San Diego (UCSD) Human Research Protections Program. Written informed consent was obtained from all participants. Participants included 25 patients with idiopathic PD (14 males, 11 females) and 26 healthy adult controls (13 males, 13 females). PD patients exhibited at least two of the three cardinal features of the disorder, were levodopa responsive, and did not exhibit features of progressive supranuclear palsy, corticobasal degeneration, multiple systems atrophy, or dementia. Exclusionary criteria included metal in the head, neurological diagnoses other than PD, psychiatric diagnoses, history of alcohol or substance abuse, and major cognitive impairment as defined by a score of less than 26 on the MMSE (Table 1) (Folstein et al., 1975). All participants were right handed except for three PD patients. There were no group differences in age, years of education, and gender (Table 1). A movement disorder specialist assessed all PD patients when they were clinically on medication using the Unified Parkinson’s Disease Rating Scale (UPDRS) and the Hoehn and Yahr scale. Patients had a PD diagnosis for an average of 7.2 years. On the Hoehn and Yahr, 23 patients were stage 2 or 2.5 (mild bilateral disease) and two were stage 3.

Neuropsychological tests (Lezak, 1995) of executive functioning included measures of verbal working memory (Digits Span Forward and Backward from the WAIS III), verbal fluency [Letter Fluency test from Delis–Kaplan Executive Function System (DKEFS)], cognitive flexibility (Trail Making Test, Parts A and B), and inhibitory control (Stroop Interference). Visuospatial functioning was assessed by the Judgment of Line Orientation Test (JOLT).

Imaging was conducted at the UCSD Radiology Imaging Laboratory using a General Electric 1.5 T Signa HDx Twin Speed MRI system with an eight-channel head coil. T1-weighted high-resolution anatomic images were collected (Spoiled Gradient Recalled, TR = 10.6 ms, TE = 4.8 ms, FOV = 25.6, slice thickness = 1 mm, NEX = 1, Flip Angle = 10°). DTI images were acquired using a single-shot EPI sequence with diffusion-encoding along 51 directions, b value = 1000 s/mm², one non-diffusion weighted image (b_o), slice thickness = 2.5 mm, TR = 15.1 s, TE = minimum, matrix = 96 × 96 mm (automatically re-gridded onto a 128 × 128 matrix), FOV = 24 mms, and voxel size 1.875 × 1.875 × 2.5 mm. The enhanced number of diffusion-encoding directions relative to most other studies (6–12 gradient directions) allowed for more accurate spatial reconstruction of white-matter structure via the greater angular resolution. Subjects underwent three DTI acquisitions to increase the signal-to-noise ratio and to ensure at least one artifact-free scan. Due to discontinuation of testing, one control subject had one acquisition. All other subjects had three artifact-free scans.

Diffusion tensor imaging data were analyzed using FMRIB Software Library version 4.1.5 (FSL, http://www.fmrib.ox.ac.uk/fsl). For each subject, images from each DTI acquisition were concatenated and corrected for eddy currents by registering all brain volumes to the first non-diffusion weighted image using a rigid body transformation (no rotation). Each DTI scan was visually inspected to ensure the absence of susceptibility artifacts. Artifact-free DTI acquisitions were fit to a diffusion model for each voxel using FDT (Behrens et al., 2003). The diffusion tensor model was then diagonalized to determine the three eigenvalues of the tensor to calculate FA, MD, axial diffusivity (AD), and radial diffusivity (RD) maps. AD and RD measure tissue changes parallel (λ1) and perpendicular [(λ2 + λ3)/2] to the axonal tract, respectively.

Voxelwise statistical analyses of the DTI data were carried out using Tract-Based Spatial Statistics (TBSS) (Smith et al., 2006). First, a study-specific FA template image was generated by aligning all subjects’ FA data to a common space using a non-linear registration. TBSS automatically detects the most representative subject by identifying the FA image that requires the least amount of transformation to the FA images of all other subjects. From this FA template image, a skeleton was created that represented the center of all white-matter tracts common to all subjects. Given that FA steadily declines with age (Grieve et al., 2007), the skeleton was thresholded at FA >0.15 to include an analysis of white-matter tracts in an older population, which might be masked out using a higher threshold (e.g., >0.20). Each subject’s aligned FA data was then projected onto the skeleton and was analyzed by a voxelwise statistical analysis program (Nichols and Holmes, 2002). This process was repeated for MD, AD, and RD maps. Using the FSL Glm tool, a design matrix was constructed that included group, age, and gender. Age and gender were covariates of no interest. Age was mean-centered separately for each group. Five thousand Monte-Carlo permutations were conducted to establish voxel-by-voxel p values for tests of group differences in the DTI parameters. Threshold Free Cluster Enhancement (TFCE) (Smith and Nichols, 2009) was used to identify clusters for each parameter that significantly differed between the two groups (p < 0.05 familywise threshold).

To examine the functional significance of abnormal diffusivity in PD, we correlated the cognitive measures with the mean FA within regions of interest (ROI) generated from the voxels that showed significant group differences. We focused on FA because it is a stable metric (Bisdas et al., 2008; Teipel et al., 2011) that is more sensitive to white-matter diffusivity than other DTI measures (Papadakis et al., 1999), especially when a large number of diffusion directions are acquired (Rae et al., 2012). An advantage of the ROI approach is that it produces stable averages of parameters that may be more representative of the diffusion properties than a single voxel. ROI in each subject’s diffusion space were generated by deprojecting the thresholded voxels that showed significant group differences in FA in the voxel-based approach, and then binarized to create a mask. To account for anatomical differences among subjects, the white-matter parcellations from the FreeSurfer analysis stream (Fischl et al., 2002) were transformed into the individual diffusion space of each subject. Specifically, the Desikan atlas cortical parcellations, which are based on 34 conventional neuroanatomical regions defined by cortical sulci and gyri, were used to assign a label to the underlying white-matter via an automated FreeSurfer procedure (Salat et al., 2009). White-matter parcellations from conformed space were registered to the native FA space using an affine linear registration routine, FLIRT. The binary mask of the FA group difference was then multiplied by the white-matter parcellations to identify the corresponding region label for each subject, and the mean FA value was computed for each region. To generate ROI for abnormal white-matter that was not included in the Desikan atlas parcellations, the Johns Hopkins University (JHU) White-Matter Labels atlas was used (i.e., external capsule and anterior limb of internal capsule voxels). A small-volume correction was applied to the resulting ROI (i.e., minimum volumetric threshold of 140 mm³) to reduce the likelihood of false positives in the correlation analyses.

To explore relationships between cognitive functioning and ROI that exhibited abnormal diffusivity in PD, cognitive measures were normalized to the control group by subtracting the individual raw scores for each measure from the control group mean and then dividing by the control group standard deviation. For the Trail Making Test, the raw score for Part A was subtracted from the raw score for Part B (Trails B-A), to obtain a measure of cognitive flexibility (i.e., set shifting) that adjusted for response speed. The resulting Z scores were then correlated with the FA in each ROI, adjusting for age and gender (partial correlations). As we had a priori hypotheses that measures of executive and visual spatial processing would respectively correlate with FA in tracts underlying the frontal or parietal-occipital cortices, a conventional p < 0.05 per comparison threshold was adopted. Nonetheless, the results should be considered exploratory and interpreted with caution due to the number of multiple correlations. SPSS version 18 was used for the partial correlation analyses.

Table 1 shows that MMSE scores were slightly lower in the PD than the control group. In both groups the lowest MMSE score was 27, indicating that no subject met the criteria for dementia. Significant group differences were not found on the neuropsychological tests.

Figure 1 displays the changes in tissue diffusivity in the PD group relative to the control group groups (p < 0.05 familywise threshold). The averaged FA-skeleton is shown in green. Volumes of abnormal tissue diffusivity are shown in red and were expanded to increase their visibility. The PD group showed significant reductions in FA and increases in MD and RD throughout anterior and posterior segments of the corpus callosum and white-matter tracts underlying the frontal, parietal, and occipital cortices. Table 2 shows that changes in FA, MD, and RD in PD were typically identified in both hemispheres and in the same tracts. Group differences in AD were found only in prefrontal cortex white-matter, where AD was increased in PD in the inferior fronto-occipital fasciculus, the forceps minor, and the anterior thalamic radiation.

Table 3 lists the 28 ROI that were derived from the voxel-based analyses that uncovered group differences in FA. ROI were comprised of the corpus callosum, frontal, parietal, and occipital white-matter in both hemispheres, and subcortical fiber tracts (left external capsule and anterior limb of the internal capsule). An exception was the right hemisphere dominance for parietal and occipital ROI, owing to the criterion adopted for selecting only larger ROI (i.e., ≥140 mm³) for the correlation analyses.

Table 4 lists the ROI that showed significant partial correlations (i.e., adjusted for age and gender) with the cognitive measures. Scatter plots of selected correlations are shown in Figures 2 and 3, together with representative examples of the ROI in diffusion space for individual PD patients. Plots graph the residuals of the cognitive and FA measures, after adjusting for age and gender. Figure 2 shows that poorer forward and backward working memory span was associated with lower FA in the mid-anterior corpus callosum and the left external capsule. Poorer backward span also correlated with lower FA in left hemisphere frontoparietal white-matter (rostral middle frontal cortex, superior corona radiata, and paracentral lobule). Verbal fluency performance correlated with lower FA especially in frontal regions (mid-anterior corpus callosum, bilateral caudal anterior cingulate, left anterior corona radiata, rostral middle frontal cortex, pars opercularis) (Figure 3), but also parietal white-matter (left paracentral lobule). Poorer cognitive flexibility (higher score signifies worse performance) correlated with lower FA in the right pericalcarine cortex and the left pars triangularis and external capsule (Figure 3). Poorer inhibitory control and visuospatial ability respectively correlated with lower FA in the left external capsule and left anterior corona radiata (Figure 3).

To determine if the cognitive-FA relationships were specific to PD, post hoc partial correlation analyses were conducted in the control group for same ROI that significantly correlated with cognitive measures in the PD group. All correlations were non-significant except for the left paracentral lobule, which correlated with verbal fluency (r_partial = 0.42, p < 0.05).

We also conducted partial correlations of FA with clinical measures of disease, adjusting for age and gender. Disease duration and levodopa dosage equivalence did not correlate with FA (p > 0.05). The UPDRS total motor score correlated negatively with FA in white-matter underlying the right pericalcarine cortex (r_partial = −0.516, p < 0.01), even when disease duration (r_partial = −0.471, p < 0.02) and levodopa dosage equivalence (r_partial = −0.483, p < 0.02) were added as covariates.

The FA reductions throughout anterior and posterior segments of fiber tracts and the corpus callosum is supported by recent reports (Rae et al., 2012; Gallagher et al., 2013), but contrasts with other findings of normal FA (Matsui et al., 2007b; Boelmans et al., 2010; Wang et al., 2011; Hattori et al., 2012; Kamagata et al., 2012) or regionally circumscribed FA reductions (Karagulle Kendi et al., 2008; Zhang et al., 2011; Zhan et al., 2012) in PD without dementia. These discrepancies may partly relate to our larger sample size relative to most other studies, which is important given the considerable variability in clinical phenotypes. Our findings may also relate to the enhanced signal-to-noise of our imaging protocol and the skeleton-based analysis, which can provide a more accurate subject-by-subject alignment of tracts (Rae et al., 2012), especially in older adults where some degree of atrophy is common.

Fractional anisotropy reductions in PD were largely due to elevated RD. Changes in RD with unchanged AD have been attributed to myelin loss (Song et al., 2002; Schmierer et al., 2007; Klawiter et al., 2011), yet demyelination is not a significant pathogenic process in PD (Teismann and Schulz, 2004). Rather, alpha-synuclein pathology is characteristic of PD. However, there are also considerable early changes in glia (Halliday and Stevens, 2011). Microglia play a role in PD neuropathogenesis, including activation from alpha-synuclein aggregation (Teismann and Schulz, 2004). Activation of microglia accompanies demyelination (Song et al., 2005), suggesting that microglia may be involved in causing the changes in RD. Though histological validation is needed, RD increases in PD may reflect early stages of degeneration that are related to alpha-synuclein accumulation and the activation of microglia. This may alter the compactness of the axonal fibers, thereby increasing RD (Smith et al., 1981).

The preponderance of RD changes in PD without dementia contrasts with Alzheimer’s disease and to a lesser extent MCI, wherein decreased FA throughout the brain often relates to changes in both RD and AD (Bosch et al., 2012). This is relevant to our finding of increased AD in some right hemisphere frontal tracts, in which RD was also increased. Pathological processes that cause changes in AD and RD often occur in close proximity (Song et al., 2003), and may signify more significant early cell loss and gliosis than RD changes by themselves. This possibility is consistent with our hypothesis that pathological changes would be present particularly in prefrontal cortex white-matter.

Individual differences in neuropsychological test performance in the PD group were associated with tissue diffusivity especially in anterior segments of the left hemisphere frontal white-matter and the external capsule, which contain cortico-cortical association fibers. These findings were not related to motor symptom severity, disease duration, and LDE, which did not correlate with tissue diffusivity. This agrees with a study of 12 non-demented PD patients, wherein a worsening of motor symptoms correlated only with FA reductions in the putamen and substantia nigra (Zhan et al., 2012), which were not evaluated in our study.

We found that verbal fluency was associated with FA in the mid-anterior corpus callosum, a main source for interhemispheric communication. Verbal fluency also correlated with FA in multiple frontal white-matter regions including the left rostral middle frontal cortex (i.e., dorsolateral prefrontal cortex) and Broca’s area (left pars opercularis). These results concur with reports of largely frontal activation during verbal fluency in young adults (Abrahams et al., 2003; Birn et al., 2010), verbal fluency deficits after frontal lobe damage, especially to the left hemisphere (Baldo et al., 2006; Robinson et al., 2012), and improved verbal fluency in PD after repetitive transcranial magnetic stimulation to the left dorsolateral prefrontal cortex (Pereira et al., 2013).

Verbal working memory also correlated with FA in the mid-anterior corpus callosum. In addition, backward span correlated with FA changes near the left rostral middle frontal cortex and the superior corona radiata and paracentral lobule, which contain fibers near the supplementary motor area, the cingulate gyrus, and the parietal cortex. Frontostriatal and parietal networks govern working memory, especially when executive functions are engaged to manipulate the contents of active memory (Lewis et al., 2004; Badre, 2008; McNab and Klingberg, 2008), as for the backward span test. The current findings are compatible with a functional imaging study that reported reduced frontostriatal activation in PD individuals with executive impairment relative to a cognitively unimpaired PD group when manipulating information in working memory (Lewis et al., 2003).

Interestingly, FA in the external capsule correlated with most executive functioning measures including forward and backward working memory span, inhibitory control (Stroop Interference), and cognitive flexibility (Trails B-A). It has been suggested that the external capsule is important for these functions, because it carries cholinergic fibers from the basal forebrain to the cerebral cortex. Indeed, a recent study reported that cholinergic denervation in PD without dementia correlated with working memory, inhibitory control, and executive functioning, but not with visuospatial processing (Bohnen et al., 2012). These findings were independent of the relationships found between dopamine denervation and the same cognitive measures (Bohnen et al., 2012), consistent with the involvement of multiple neurotransmitter systems in PD (Kehagia et al., 2010).

Trails B-A performance was uniquely associated with FA in the pathway beneath the right pericalcarine cortex and the left pars triangularis, which is near Broca’s area. These results are functionally meaningful because set shifting on this task emphasizes visual search and engages subvocalization process. Lastly, we found that visuospatial ability (Judgment of Line Orientation) correlated strongly with FA in the left anterior coronal radiata, possibly because optimal performance partly depends on prefrontal control processes that govern attention and the integration of relational information (Watson and Chatterjee, 2012). Contrary to our hypothesis, visuospatial processing was not related to the diffusivity of tissue near posterior segments of the cortex, despite significant FA abnormalities in these tracts. The reason for this is unclear, but other tests of visuospatial abilities may better probe for posterior cortical functioning.