Progression of cortical thinning in early Parkinson's disease

The aim of this study was to investigate the progression of cortical thinning and gray‐matter (GM) volume loss in early Parkinson's disease (PD). MRI and neuropsychological assessment were obtained at baseline and follow‐up (mean ± standard deviation = 35.50 ± 1.88 months) in a group of 16 early‐PD patients (H & Y stage ≤II and disease duration ≤5 years) and 15 healthy controls matched for age, gender, and years of education. FreeSurfer software was used for the analysis of cortical thickness as well as for cortical and subcortical volumetric analyses. Voxel‐based morphometry analysis was performed using SPM8. Compared to controls, PD patients showed greater regional cortical thinning in bilateral frontotemporal regions as well as greater over‐time total GM loss and amygdalar volume reduction. PD patients and controls presented similar over‐time changes in cognitive functioning. In early‐PD patients, global GM loss, amygdalar atrophy, and cortical thinning in frontotemporal regions are specifically associated with the PD‐degenerative process. © 2012 Movement Disorder Society

A B S T R AC T : The aim of this study was to investigate the progression of cortical thinning and gray-matter (GM) volume loss in early Parkinson's disease (PD). MRI and neuropsychological assessment were obtained at baseline and follow-up (mean 6 standard deviation 5 35.50 6 1.88 months) in a group of 16 early-PD patients (H & Y stage II and disease duration 5 years) and 15 healthy controls matched for age, gender, and years of education. FreeSurfer software was used for the analysis of cortical thickness as well as for cortical and subcortical volumetric analyses. Voxel-based morphometry analysis was performed using SPM8. Compared to controls, PD patients showed greater regional cortical thinning in bilateral frontotemporal regions as well as greater over-time total GM loss and amygdalar volume reduction. PD patients and controls presented similar over-time changes in cognitive functioning. In early-PD patients, global GM loss, amygdalar atrophy, and cortical thinning in frontotemporal regions are specifically associated with the PDdegenerative process. V C 2012 Movement Disorder Society Key Words: Parkinson's disease; cognition; longitudinal; cortical thickness Cognitive dysfunctions appear early in the course of Parkinson's disease (PD). Cross-sectional, communitybased studies have reported on impairments in executive functions, 1,2 attention and psychomotor speed, 3,4 memory, 1-4 visuospatial functions, 3 and verbal fluency 1,2 in newly diagnosed PD patients. Furthermore, longitudinal assessment of cognition in these early-PD patients shows a faster rate of cognitive decline than in control subjects. 5,6 According to the neuropathological Braak stages, 7 alpha-synuclein pathology ascends from the brainstem and olfactory bulb, in the early stages, to the midbrain and diencephalic nuclei, and, later, to the neocortex. However, MRI findings do not fully conform to this neuropathological staging. Cross-sectional imaging studies in early PD, using different methodological approaches, have found highly variable, and usually not extensive, patterns of structural degeneration. Specifically, voxel-based morphometry (VBM) and volumetric techniques comparing early-PD patients and matched controls have shown white-matter (WM) loss in the brainstem 8 and gray-matter (GM) loss in the striatum, 9 amygdala, 10 and orbitofrontal cortex (OFC). 10 On the other hand, cortical thickness (CTh) studies have described focal cortical thinning in the OFC, ventrolateral prefrontal cortex, occipitoparietal cortex, 9 and in the supplementary motor area. 11 These discrepant findings highlight the need for longitudinal studies using comprehensive techniques to help clarify the typical patterns of evolution of structural changes that take place early in PD.
The relationship between structural MRI changes and neuropsychological deficits in PD has not been extensively investigated, and the only available results come from cross-sectional studies. 12 PD patients with mild cognitive impairment have been found to have frontotemporal neocortical reductions, compared with PD patients with normal cognition, 13 and memory deficits have been observed to correlate with the degree of hippocampal atrophy. [14][15][16] Similarly, correlations have been found between visuospatial and -perceptual deficits and parietotemporal GM reductions 17 and between verbal fluency and frontal and temporal cortical atrophy. 18 Specifically for early PD, Ibarretxe-Bilbao et al.
(2009) showed amygdalar and bilateral OFC degeneration associated with impaired recognition of emotions as well as GM volume loss in the left lateral OFC related to decision-making impairment. 10 VBM and cortical-thickness analyses are sensitive to different, though overlapping, GM parameters that are altered in neurodegenerative processes: VBM provides a mixed measure of cortical GM, including cortical surface area and/or cortical folding as well as cortical thickness, 19 whereas cortical-thickness analysis has the advantage of providing a quantitative value that represents a physical property of the cortical mantle. These MRI techniques are considered to provide complementary information in different research settings. [20][21][22] To our knowledge, no published MRI studies have focused on the progression of brain changes using cortical thickness and VBM analyses in early PD. Indeed, longitudinal MRI studies have only been carried out at more-advanced disease stages, showing progressive ventricular dilation 23,24 and progressive cerebral GM loss in nondemented PD, 25,26 PD patients at risk for dementia (patients with visual hallucinations), 27 and in patients with PD dementia (PDD). 26,28 In the present study, we have performed a prospective assessment of a group of early-PD patients and a matched control group of healthy subjects. Subjects underwent MRI and neuropsychological assessment both at study entry and after a mean follow-up interval of 35.5 months. In a previous study, we reported on MRI findings at baseline showing GM volume loss in the amygdala and the OFC in early-PD patients 10 using a region-of-interest VBM approach. In the current study, we aimed to make a global assessment of the brain changes that take place over time in the early stages of PD using cortical-thickness analysis, whole-brain VBM, and volumetric analysis of the ventricular system and subcortical GM structures. We also aimed to investigate cognitive changes over time to establish whether these brain MRI measures correspond to possible neuropsychological changes.

Participants
Patients were recruited from an outpatient movement disorders clinic (Parkinson's Disease and Movement Disorders Unit, Department of Neurology, Hospital Clinic, Barcelona, Spain). Healthy subjects were recruited from friends and spouses of patients and matched for age, gender, and years of education. At baseline, 24 early-PD patients and 24 healthy subjects matched for age, gender, and years of education participated in the study (the same sample used in the study by Ibarretxe-Bilbao et al., 2009 10 ). After a mean interval of 35.50 months (standard deviation [SD] ¼ 1.85; range, 31-40), 16 PD patients and 15 healthy controls from this initial sample participated in the follow-up assessment. There were no significant differences between groups in terms of age, gender, and education. Demographical data of the participants at baseline examination are summarized in Table 1.
At the time of initial evaluation (baseline evaluation), the inclusion criteria for participating in the study were (1) fulfillment of the UK PD Society Brain Bank (PDSBB) diagnostic criteria for PD, 29 (2) H & Y stage II, and (3) disease duration 5 years. Motor symptoms were assessed by means of the UPDRS-III motor section. Exclusion criteria were (1) presence of dementia diagnosed by a neurologist according to the Movement Disorder Society diagnostic criteria for PDD, 30 (2) presence of other neurological or psychiatric disorders, such as depression, defined by a Beck's Depression Inventory (BDI-II) score higher than 14, 31 and (3) the presence of visual hallucinations, as assessed by the Neuropsychiatric Inventory Questionnaire (NPI-Q). 32 This study was approved by the institutional ethics committee. Written informed consent was obtained from all study subjects after full explanation of the procedures involved.

Neuropsychological Assessment
The neuropsychological assessment included the cognitive functions usually impaired in PD. The tests used were as follows: (1) forward and backward digit span from the Wechsler Adults Intelligence Scale (WAIS-III) as measures of immediate and working memory, respectively; (2) verbal memory with Rey's Auditory Verbal Learning Test (RAVLT) as a measure of declarative memory; (3) phonemic and semantic fluencies; (4) psychomotor speed and inhibition control with the Stroop test; and (5) visuomotor tracking and cognitive flexibility with the Trail Making test (TMT) (parts A and B). The neuropsychological battery was administered by a trained neuropsychologist (N.I.-B.), and completed in a single session lasting between 1 and 2 hours. The neuropsychologist in charge of the assessment was the same at baseline and follow-up examinations.

Neuropsychological and Clinical Statistical Analysis
Statistical analyses of neuropsychological and clinical variables were carried out using the statistical package, PASW-18 (2009; SPSS, Inc., Chicago, IL). Changes in clinical and neuropsychological variables between baseline and follow-up were analyzed using the general linear mixed model (GLM) for repeated measures to test whether these variables differed across time between groups. A two-factor repeatedmeasures analysis of variance (2 Â 2) (RMANOVA) was performed for each of the variables. Factors were group (PD patients and controls) and time (baseline and follow-up). Because groups were matched (there were no significant intergroup differences for age, gender, and education), these variable were not included in the model. On the other hand, the interval between neuropsychological assessments in months was entered as a covariate in the RMANOVA.

Cortical Thickness Analysis
Cross-sectional evaluation (to determine whether PD patients and controls regionally differ in cortical thickness at baseline) and longitudinal analysis (to compare CTh between the two time points [baseline and follow-up]) were performed using FreeSurfer software (version 5.1; available at: http://surfer.nmr.harvard. edu). In this procedure, a cortical surface 3D model of CTh is created using intensity and continuity information, as previously described in detail. 33 The initial processing of T1 high-resolution images includes several steps, which are performed independently for each subject and each time point: removal of nonbrain tissue, 34 automated Talairach transformation, intensity normalization, 35 tessellation of the GM/WM boundary, automated topology correction, 36,37 and surface deformation to optimally place the GM/WM and GM/ cerebrospinal fluid (CSF) boundaries. 33 The resulting representation of cortical thickness is calculated as the distance between tissue boundaries (GM/WM and GM/CSF). 33 All surface models in our study were visually inspected for accuracy.
Subsequently, for the longitudinal processing, an unbiased within-subject template space and average image 38 are created using robust, inverse consistent registration. 39 Several processing steps, such as skull stripping, Talairach transformations, atlas registration, as well as spherical surface maps and parcellations are then initialized with common information from the within-subject template, significantly increasing reliability and statistical power. 40 Finally, both time points and the template are used to create the complete subjects directories, named long run. Regional differences between groups (patients and controls) in cortical measures were assessed using a vertex-by-vertex general linear model (one factor with two levels without covariates, corresponding to a Student t test statistical model). Maps were smoothed using a circularly symmetric Gaussian kernel across the surface with a full width at half maximum (FWHM) of 15 mm. Z Monte Carlo simulations with 10,000 iterations were applied to CTh maps to provide clusterwise correction for multiple comparisons, 41 and results were thresholded at a corrected P value of 0.05 (Z ¼ 1.3). To perform main longitudinal analyses, we automatically obtained a measure of change from longitudinal data called rate of thinning (mm/year).
Mean rates of thinning for significant clusters were then extracted for each subject to perform correlational analyses with cognitive measures.
We used the same methodology to perform complementary analyses with other longitudinal measures: the temporal average (the average thickness between baseline and follow-up); the percent change (the rate of thinning divided by the baseline thickness); and the symmetrized percent change (the rate of thinning divided by the average thickness).

Volumetric Analysis of Subcortical Structures
The automated procedure for volumetric measures of brain structures implemented in FreeSurfer (version 5.1; available at: http://surfer.nmr.harvard.edu) 42 was used to obtain the volumes of subcortical structures.
Volumetric measures were automatically obtained from 7 GM structures (thalamus, caudate, putamen, pallidum, nucleus accumbens, hippocampus, and amygdala), as well as from the lateral, third, and fourth ventricles and the brainstem. For simplicity, we refer to these collectively as ''subcortical structures,'' although, strictly speaking, the hippocampus and amygdala are not subcortical. Total cortical volume, total subcortical volume, and global GM volume were also obtained.
All the volumes obtained were compared using PASW-18. RMANOVA was used to longitudinally examine changes in deep GM structures, brainstem, ventricles, and total GM volumes. Scan number was specified as a within-subject factor ''time'' with two levels (baseline and follow-up MRI acquisitions) as well as a between-subject factor ''group'' with two levels (PD patients and healthy controls). All the volumetric analyses controlled for intracranial volume (ICV). No other covariables were included in the model. Partial correlation analyses were performed between change in brain volumes and cognition, also controlling for ICV.

Voxel-Based Morphometry Analysis
Image preprocessing for voxel-based morphometry analyses was done using SPM8 (http://www.fil.ion.ucl.ac.uk/spm/software/spm8/). After realignment, baseline images were rigidly coregistered to the respective subject's follow-up image. All images were then segmented into GM, WM, and CSF probability maps with the New Segment tool of SPM8, an extended version of the ''unified segmentation'' model. 43 In an attempt to obtain an accurate subject registration and, consequently, an adequate detection of within-subject longitudinal changes, subject-specific templates were created using the Diffeomorphic Anatomical Registration Through Exponentiated Lie algebra (DARTEL) 44 tool to nonlinearly register each subject's GM and WM baseline and follow-up images. Baseline and follow-up scans were subsequently normalized to their respective template, using the Jacobian determinants for modulation. A group template was then created using the baseline GM and WM images (normalized to the individual template) of the entire sample, also through DARTEL. Baseline and follow-up scans then underwent a second normalization, this time to the group template, again applying the Jacobian determinants for modulation. A third normalization was performed to bring each subject's baseline and follow-up GM images to the MNI space using an affine transformation, then smoothed with a Gaussian kernel of 12 mm FWHM. The final voxel dimension was 1.5 mm isotropic. A similar preprocessing approach was recently proposed by Asami et al. 45 The final GM maps generated as described above were analyzed in a whole-brain voxelwise fashion. For cross-sectional baseline intergroup comparisons, images were entered into a two-sample t test model. For longitudinal analysis, subjects' images were entered into a full-factorial model with two factors (the two levels in the group factor were entered as independent, and the two levels in the time factor as nonindependent variables, with assumed unequal variance) to evaluate possible group Â time interactions in GM volume.
Total intracranial volume (obtained through the sum of the native GM, WM, and CSF maps) was entered as a covariate of no interest in all these analyses. Subjects' interscan interval (in months) was also entered as a covariate of no interest in the interaction analyses.
Results were corrected for multiple comparisons using familywise-error correction, 46 with a significance threshold set at P < 0.05.

Neuropsychological Tests and Clinical Variables
In PD patients, levodopa equivalent daily dose (LEDD), H & Y, and UPDRS-III motor section scores did not differ significantly between baseline and follow-up assessments (see Table 2).
RMANOVA revealed a significant time effect on Mini-Mental State Examination (MMSE) and depression scores, but no significant interactions between  groups and time. That is, both patients and controls showed lower scores over time, but the decline did not differ between groups. NPI-Q scores remained stable over time in both groups. At baseline, PD patients showed worse cognitive performance than controls. There were significant differences in neuropsychological variables measuring working memory (backward digits: t ¼ 3.015, P ¼ 0.005), psychomotor speed (Stroop colors: t ¼ 2.755, P ¼ 0.010; words-colors: t ¼ 2.417, P ¼ 0.023), and cognitive flexibility (TMTB: t ¼ À2.146, P ¼ 0.046). Longitudinal analyses also showed a significant group effect in working memory (digit span backward), psychomotor speed (Stroop colors), and cognitive flexibility (TMTB). A significant time effect was observed for psychomotor speed assessed by the Stroop test (Stroop words). The only other cognitive variable for which a time effect was observed was phonemic fluency, as a result of both groups improving their performance at follow-up, suggesting a practice effect. There was no significant interaction between time and group for any of the variables analyzed (see Table 3).

Cortical Thickness
At baseline, there were no significant differences in cortical thickness between PD patients and the healthy control group (data not shown). However, rate-ofthinning (mm/year) analyses showed that PD patients had significantly greater progressive cortical thinning in the following areas than healthy controls at the corrected level (P < 0.05) ( Fig. 1; Table 4): right superior frontal gyrus, extending to caudal middle frontal gyrus, precentral sulcus, and precentral gyrus; right frontal pars opercularis and precentral gyrus; right superior temporal gyrus extending to the adjacent temporal sulcus; left caudal middle frontal gyrus extending to rostral middle frontal gyrus and precentral region; and left middle temporal gyrus extending to parietal cortex.
At a more-conservative correction level (P < 0.01), clusters in the right hemisphere remained significant: in the caudal middle frontal gyrus (621.2 mm 2 ; x, y, and z Talairach coordinates of the maximum: 29.5, 7.5, and 47.3) and in the superior temporal gyrus (two clusters, one with 549.2 mm 2 ; x, y, and z Talairach coordinates of the maximum: 43.9, À4.2, and À16.2; and the other with 494.4 mm 2 ; x, y, and z Talairach coordinates of the maximum: 63.1, À31.1, and 5.6). The complementary longitudinal measures analyzed showed no significant group differences in the measure temporal average. However, the measures percent change and symmetrized percent change showed patterns of thinning in frontal and temporal regions similar to those obtained by means of the rate of thinning (see Supporting Fig. 1 for the results of the symmetrized percent change). There were no significant correlations between cognitive measures and mean rate of thinning (or between cognitive measures and the complementary longitudinal measures analyzed) in the clusters described above.
No areas of significantly higher rates of thinning in controls than in PD patients were observed. Table 5 shows total GM, total cortical, and subcortical GM volumes (total and by structure) as well as ventricular volumes at baseline and follow-up. At baseline, there were no significant group differences in volumetric measures (data not shown). The loss from baseline to follow-up in subcortical deep GM structures besides the amygdalae was similar in both groups. We also observed a similar volume increase in the ventricular system in both groups.

Volumetric Analyses
Total GM volume reduction over time was greater in PD patients than in healthy controls; the time-group interaction was significant. Finally, PD patients also showed specific and significant loss in amygdalar and total cortical volumes (see Table 5). Total GM volume changes correlated with Stroop words performance decline in PD patients (r ¼ 0.541; P ¼ 0.037). Neither amygdalar volume changes nor total cortical volume reduction correlated with cognitive changes.

Voxel-Based Morphometry
No significant group differences in GM volume were found between healthy controls and PD patients at baseline in whole-brain analyses. Interaction analysis did not reveal any areas of differential GM loss across groups at the corrected level.  For exploratory purposes, we report on the results without correction for multiple comparisons-at the uncorrected level, two clusters of greater GM volume loss in PD patients than in controls were observed: in the left inferior lateral occipital cortex (3.08 cm 3 ; x, y, and z MNI coordinates of the maximum: À22, À90, and À2; P < 0.001, uncorrected) and in the left angular gyrus (0.37 cm 3 ; x, y, and z MNI coordinates of the maximum: À34, À56, and 26; P < 0.001, uncorrected). We did not find any areas of significantly greater GM volume loss in healthy controls than in PD patients.

Discussion
In this longitudinal study, we have found that, compared to controls, our sample of early-PD patients presented a faster rate of cortical thinning with a bilateral frontotemporal pattern, total GM and total cortical volume loss, and amygdalar atrophy after an average interval of 35 months.
The first longitudinal studies in PD only evaluated global brain measures 25 or a limited number of GM areas. 28 Since then, longitudinal progression of PD has been evaluated in samples formed by older patients as well as with patients who had longer disease duration. In a previous work done by our group, after a followup of 25 months, GM loss was observed in limbic and paralimbic structures. 26 In another study, including patients with visual hallucinations, we also found limbic and paralimbic plus neocortical GM loss. 27 In this work, we also observed that the longitudinal GM loss correlated with several measures of cognitive decline.
To our knowledge, this is the first study to evaluate structural progressive changes in a cohort of early-PD patients. Our findings indicate that the degeneration of certain areas observed in other studies may not yet be detectable in nondemented, early-PD patients. The fact that our cohort was formed by younger patients could also be contributing to the absence of detectable atrophy in areas affected both by the disease and by the aging processes, such as the hippocampus, 47 because these two factors are considered to interact biologically in the process of neuronal cell loss in nondopaminergic structures. 48 In the present study, PD patients presented lower scores in some cognitive functions, compared to controls, at baseline, but both groups progressed at the same rate of cognitive decline over time. Previous studies reported that longitudinal changes in neuropsychological test results in early PD are much more significant in older patients. 5 This observed interaction, both between age and cognitive decline, in PD could be the functional manifestation of the interaction between age-related and disease-related GM atrophy observed, though this remains speculative.
In Lewy body (LB) diseases, cognitive changes are expected to precede detectable structural changes, 49 and we have found greater structural, but no greater cognitive, over-time deterioration in PD patients, compared to controls. However, a growing body of evidence suggests that cognitive deficits in PD have two general pathophysiological bases: (1) frontostriatal deficits, common in the early stages of disease, related to dopaminergic activity and not associated with a higher risk of evolution to dementia and (2) deficits with a more-posterior cortical basis, which herald  further cognitive decline, 5,50 and which are possibly related to primary cortical LB or Alzheimer's disease (AD)-type pathology. 51 Indeed, at baseline, our sample of PD patients presented deficits in neuropsychological functions related to dopamine-modulated circuits, such as working memory, 52,53 cognitive flexibility, 53,54 and psychomotor speed. 55 It is not surprising that deficits with a more functional basis might not be accompanied by detectable structural deterioration. On the other hand, the structural changes detected longitudinally could be related to the second mechanism of cognitive decline described above, and compensatory processes in this sample of young, early-PD patients might be preventing detectable functional repercussions. Cross-sectional structural MRI studies with early-PD patients, as pointed out in the introduction of the current article, show highly variable results, sometimes with limited or no significant GM changes, in comparison to controls. 8 Nonetheless, one study did find cortical changes in early-stage patients through CTh analysis. 9 In the analyses, thickness reductions were observed in frontal areas similar to the ones we have found to degenerate more rapidly in PD patients than in controls.
Overall, these results suggest that the different parameters evaluated provide complementary and related information on changes occurring in PD.
A common drawback in longitudinal studies is the loss of some of the initial subjects at follow-up. In this study, the final sample of 16 PD patients, though appropriate for longitudinal analyses, does not allow dividing them into subgroups to study relevant variables, such as the presence of dyskinesias, previously related to changes in the right inferior frontal gyrus, 56 or behavioral abnormalities. Also, we cannot exclude that subtle cognitive decline associated with the longitudinal structural changes described could be detected with a larger sample.
One caveat in both VBM and CTh analysis is that, as a result of the characteristics of the underlying neuroanatomy, GM and cortical thickness maps are nonstationary. 57 This, in turn, may lead to an overestimation of cluster sizes in areas of the images that display higher smoothing, and an underestimation in rougher areas, 58 affecting both the false-positive and false-negative rates of analyses that include cluster-size corrections. 59 Though cluster-size inference methods to correct for these effects are sometimes used in VBM analyses, 60,61 the permutation clusterlevel correction implemented in cortical thickness analyses assumes images to be stationary and may thus be susceptible to such artifacts.

Conclusion
In conclusion, as in other neurodegenerative diseases such as AD, our findings suggest that the pathophysio-logical process of PD exerts specific deleterious effects in the brain, even preceding clinical manifestations of progressive decline in cognitive function.