Different subregional metabolism patterns in patients with cerebellar ataxia by 18F-fluorodeoxyglucose positron emission tomography

We evaluated cerebellar subregional metabolic alterations in patients with cerebellar ataxia, a representative disease involving the spinocerebellum. We retrospectively analyzed 18F-fluorodeoxyglucose positron emission tomography (18F-FDG PET) images in 44 patients with multiple system atrophy of the cerebellar type (MSA-C), 9 patients with spinocerebellar ataxia (SCA) type 2, and 14 patients with SCA type 6 and compared with 15 patients with crossed cerebellar diaschisis (CCD) and 89 normal controls. Cerebellar subregional metabolism was assessed using 13 cerebellar subregions (bilateral anterior lobes [ANT], superior/mid/inferior posterior lobes [SUPP/MIDP/INFP], dentate nucleus [DN], anterior vermis [ANTV], and superior/inferior posterior vermis [SUPV/INFV]) to determine FDG uptake ratios. MSA-C and SCA type 2 showed severely decreased metabolic ratios in all cerebellar subregions compared to normal controls (ANT, 0.58 ± 0.08 and 0.50 ± 0.06 vs. 0.82 ± 0.07, respectively, p < 0.001). SCA type 6 showed lower metabolic ratios in almost all cerebellar subregions (ANT, 0.57 ± 0.06, p < 0.001) except INFV. Anterior-posterior lobe ratio measurements revealed that SCA type 2 (Right, 0.81 ± 0.05 vs. 0.88 ± 0.04, p < 0.001; Left, 0.83 ± 0.05 vs. 0.88 ± 0.04, p = 0.003) and SCA type 6 (Right, 0.72 ± 0.05 vs. 0.88 ± 0.04, p < 0.001; Left, 0.72 ± 0.05 vs. 0.88 ± 0.04, p < 0.001) showed preferential hypometabolism in the anterior lobe compared to normal controls, which was not observed in CCD and MSA-C. Asymmetric indices were higher in CCD and MSA-C than in normal controls (p < 0.001), whereas such differences were not found in SCA types 2 and 6. In summary, quantitative analysis of cerebellar subregional metabolism ratios revealed preferential involvement of the anterior lobe, corresponding to the spinocerebellum, in patients with cerebellar ataxia, whereas patients with CCD and MSA-C exhibited more asymmetric hypometabolism in the posterior lobe.

Introduction those with SCA, 9 patients with SCA type 2 and 3 patients with SCA type 6 were included in our previous work [16]. Patients with MSA were diagnosed based on the consensus statement [20], were classified as probable MSA-C by movement disorder specialists (S.J.C. and C.S.L.), underwent 18 F-FDG PET imaging between March 2006 and December 2010, and were followed for more than two years. All SCA patients were molecularly confirmed to harbor CAG trinucleotide repeat expansions of corresponding SCA genes. Age, sex, age at onset, duration from onset to 18 F-FDG PET imaging, clinical symptoms, unified Parkinson's disease (PD) rating scale (UPDRS) scores (for MSA-C patients), and number of CAG repeats (for SCA patients) were obtained from the medical records, if available (Table 1).
Fifteen patients with CCD due to brain tumors in the frontal cortex who were identified by 18 F-FDG PET imaging were also included as controls for reduced cerebellar metabolism by non-ataxic causes. Additionally, 89 nine healthy controls between the ages of 17 and 85 years were selected from the normal 18 F-FDG PET data pool maintained at our institution. Healthy controls were confirmed to have no neurological or psychological diseases. This study was approved by the institutional review board of Asan Medical Center (IRB No.2014-0023) with the need for informed consent waived and is in accordance with the declaration of Helsinki on ethical principles for medical research involving human subjects. 18

F-FDG PET image acquisition
All 18 F-FDG PET scans were acquired using an ECAT HR+ scanner (CTI-Siemens, Knoxville, Tennessee, USA) in 159 subjects and using a Discovery 690 scanner (GE healthcare, Waukesha, Wisconsin, USA) in 12 subjects. All subjects fasted for a minimum of six hours before 18 F-FDG PET imaging.
In patients studied by the ECAT HR+ scanner, transmission scans of 5 minutes using a rotating pin source of germanium-68 for attenuation correction and emission scans of 15 minutes were acquired 40 minutes after intravenous injection of 370 MBq 18 F-FDG. Patients lay under resting conditions with eyes closed in a room with dim light. In-plane and axial resolutions of the scanner were 4.3 mm and 8.3 mm full width at half maximum (FWHM), Orthostatic hypotension (n) 19 1 2 Ataxia (n) 40 9 12 Tremor (n) 5 4 4 Pyramidal symptom (n) 9 1 1 Dysarthria (n) 33 9 11 Number of CAG repeats (mean ± SD) 42.1 ± 3.3 22.9 ± 1.5 respectively. Following acquisition, 18 F-FDG PET images were reconstructed using a Gaussian filter (FWHM, 2 mm) and displayed in a 128 × 128 matrix (pixel size, 1.72 × 1.72 mm with a slice thickness of 2.43 mm). Emission images were reconstructed with ordered subset expectation maximization (OSEM) using 16 subsets and 6 iterations. In patients studied by the Discovery 690 scanner, computed tomography data for attenuation correction were acquired first, followed by three-dimensional 18 F-FDG PET image acquisition. In-plane and axial resolutions of the scanner were 4.9 mm and 5.6 mm FWHM, respectively. 18 F-FDG PET images were reconstructed using a Gaussian filter (FWHM, 2 mm) and displayed in a 256 × 256 matrix (pixel size, 0.98 × 0.98 mm with a slice thickness of 3.27 mm). Emission images were reconstructed with OSEM using 24 subsets and 2 iterations. 18

F-FDG PET image analysis
We analyzed 18 F-FDG PET images using volumes of interests (VOIs) and voxelwise comparisons.
Preprocessing and spatial normalization 18 F-FDG PET images were spatially normalized using Statistical Parametric Mapping 2 (SPM2, Wellcome Department of Cognitive Neurology, Institute of Neurology, University College London, United Kingdom) in conjunction with MATLAB version 2013a (The Mathworks Inc., USA). All reconstructed 18 F-FDG PET images were spatially normalized to the Montreal Neurological Institute (MNI) reference space [21] using a standard brain 18 F-FDG PET template provided with SPM2. Slices containing only the cerebellum were separated from these 18 F-FDG PET images, normalized images of typical normal subjects were averaged to create a study-specific cerebellar template, and individual 18 F-FDG PET images containing only cerebellar slices were normalized to the study-specific cerebellar template. The detailed procedure was described in our previous work [16]. 18 F-FDG PET images showed that higher counts on the anatomically left cerebellar cortex were flipped to the contralateral side (right) during group analysis of asymmetry in cerebellar metabolism.

VOI analysis
Definition of VOI. Subregional cerebellar metabolic ratios were obtained from VOI values, which were defined using MRIcro 1 version 1.40 (Chris Rorden, Columbia, SC, USA; www.mricro.com). Circular VOIs (diameter, 8 mm) were drawn manually on coregistered, spatially normalized, single T1-weighted magnetic resonance imaging (MRI) scans and 18 F-FDG PET images of normal controls by an experienced nuclear medicine physician specializing in nuclear neurology (M.O). A 8-mm diameter circular VOI size was determined to reduce possible overestimation of metabolic activity due to spillover effects by adjacent structure [22]. VOIs for the cerebellum were defined as follows (Fig 1). The boundary between the anterior and posterior cerebellar lobes was visually determined by the primary fissure. Next, posterior cortex and posterior vermis were divided into three and two equal parts, respectively. As a result, cerebellar cortex was further divided into one anterior and three posterior lobar subregions. Bilateral anterior cortices (ANT) corresponded to anatomic lobules I-V, and superior/mid/inferior posterior cortices (SUPP/MIDP/INFP) corresponded to anatomic lobules VI-X. Cerebellar vermis was divided into anterior vermis (ANTV), and superior/inferior posterior vermis (SUPV/INFV). Dentate nuclei (DN), which were not clearly defined in T1-weighted MRI, were defined as regions with distinct metabolic activity in the middle of the cerebellar white matter (Fig 1).
Reference area for calculating metabolic ratio. To choose a reference area in voxelwise analysis, analysis of variance (ANOVA) with multiple comparison was used in SPM2, and robustness in age-related changes within cerebellar subregional and inter-subregional metabolic ratios were evaluated according to the reference areas (occipital/parietal cortex and whole gray matter). 18 F-FDG PET counts were extracted from overlapping cerebral cortical areas defined by the sum of automated anatomical labeling (AAL) template VOIs and the normalized image of each subject.
Assessment of VOI values. The VOI template in the standard MNI space was automatically applied directly to the spatially normalized individual 18 F-FDG PET images to extract VOI values using an in-house software program called ANTIQUE (AMC NM Toolkit for Image Quantification of Excellence), which was written in Interactive Data Language (version 8.2) [16]. First, we divided the mean VOI value of a specific cerebellar subregion by the mean VOI value of the reference area to calculate the metabolic ratio. We also calculated inter-subregional ratios, including anterior-posterior lobe ratio (APR) and asymmetric index (AI) using total cerebellar VOI values as follows: Anterior to posterior lobe ratio ðcortexÞ ¼ ½counts of ANT counts of ðSUPPþMIDPþINFPÞ

Voxelwise analysis
Spatial preprocessing and statistical analysis were performed using SPM2. After separation of slices containing the cerebellum, 18 F-FDG PET images were spatially normalized to the studyspecific cerebellar templates with a 12-parameter affine transformation. Spatially normalized images were smoothed by convolution with an isotropic Gaussian kernel of 12 mm FWHM to increase the signal-to-noise ratio [23]. Using the group comparison model in SPM2, we assessed differences in regional cerebellar glucose metabolism between patient groups and controls with a threshold of p < 0.01 (uncorrected) and an extent threshold of 100 voxels. Patient age was used as a confounding nuisance covariate during statistical analysis to minimize its impact. This statistical threshold was selected not only to capture a wide range of statistical differences but also to provide sufficient distinguishing power to identify separate clusters of metabolic differences.

Statistical analysis
Descriptive statistics included means ± standard deviation or frequencies for each clinical characteristic. Comparisons of age at 18 F-FDG PET examination and disease duration between healthy control and patient groups were performed using nonparametric statistics (Kruskal-Wallis H test) since some of the sample sizes were small and the assumption of a Gaussian distribution was not appropriate. Significance of intergroup differences was assessed by the Kruskal-Wallis test due to non-normal distribution. Post-hoc testing was used to identify statistically significant group differences, taking into account Bonferroni correction for multiple comparisons (p < 0.05/10). Correlations between age, symptom duration, and regional VOI values and ratios were estimated by parametric Pearson's product-moment correlation coefficients in normal controls and patients with MSA-C and by nonparametric Spearman's rank correlation coefficients in patients with SCA type 2 or type 6. Data storage and analysis were performed using SPSS 1 software (Statistical Package for the Social Science, version 18.0, Chicago, IL, USA).

Reference area for calculating metabolic ratio
Voxelwise analysis. Voxelwise analysis by SPM after global count normalization indicated that parts of the occipital and parietal cortices showed little variance in regional metabolism among the disease groups and normal controls (S1 Fig). Aging effect of metabolic ratio according to reference area. Metabolic ratios of cerebellar subregions using occipital cortex as a reference for count normalization did not change with age (r = 0.141, p = 0.060 for ANT; r = −0.033, p = 0.665 for SUPP). While there was a weak positive correlation in ANT (r = 0.214, p = 0.004), there was no correlation in SUPP (r = 0.044, p = 0.556) by count normalization using parietal cortex as a reference. Moderate and weak positive correlations in ANT and SUPP, respectively, were observed by global count normalization (r = 0.448, p < 0.001 and r = 0.177, p = 0.018, respectively; S2 Fig). These results revealed that occipital cortex was appropriate for count normalization. Fig 2 shows representative images that were spatially normalized to the study-specific template and count normalization using occipital cortex.

VOI analysis
Comparison of patients with healthy controls. Patients with MSA-C and SCA type 2 showed severely decreased metabolic ratios in all cerebellar subregions compared to normal controls (p < 0.001 for all subregions, Fig 3). Patients with SCA type 6 showed decreased metabolic ratios in almost all cerebellar subregions (p < 0.001) except INFV (p = 0.006).
Comparison among SCA type 2, SCA type 6, and MSA-C. A post hoc analysis revealed that metabolic ratios in patients with SCA type 2, SCA type 6, MSA-C showed different cerebellar metabolism from each other in several subregions of the cerebellum.

Inter-subregional ratios in relation with symptom duration.
In MSA-C, there was no correlation between symptom duration and asymmetry. Anterior-posterior lobe ratio showed a weak negative correlation with symptom duration (r = −0.372, p = 0.014) in left cortex.
In SCA type 2, there was no correlation between symptom duration and asymmetry. Anterior-posterior lobe ratio showed a strong negative correlation with symptom duration (r = −0.765, p = 0.016) in right cortex.
In SCA type 6, there were no correlations between duration and inter-subregional ratios.

Voxelwise analyses
Comparison of patients with normal controls. Both MSA-C and SCA type 2 groups showed significantly decreased metabolic ratios in whole cerebellar cortex and vermis, compared with normal controls. Additionally, patients with MSA-C exhibited a more asymmetric pattern compared to those with SCA type 2. Patients with SCA type 6 showed significantly decreased metabolic ratios mainly in the anterior and superior posterior lobes (Fig 4).
Comparison among disease groups. The decrease in metabolic ratios in the right anterior and left posterior lobes was more pronounced in patients with SCA type 2 than in those with MSA-C. Additionally, the decrease in metabolic ratios in the right anterior lobe of patients with SCA type 6 was more than that found in patients with MSA-C. These findings reflected the relatively more pronounced asymmetric nature of metabolic patterns observed in MSA-C compared to the SCA subtypes. Comparison between SCA subtypes revealed that the decrease in metabolic ratios of the posterior cerebellar lobes was more marked in patients with SCA type 2 than in those with SCA type 6 (S3 Fig).

Discussion
To our knowledge, the present study includes the most comprehensive and detailed subregional analysis of cerebellar metabolism using 18 F-FDG PET to date. Specifically, we observed that cerebellar metabolism in the anterior lobe was relatively decreased compared to posterior lobe in patients with cerebellar ataxia regardless of etiology, in contrast to normal controls and patients with CCD. This observation was not affected by age, which supported distinct underlying pathophysiologies leading to cerebellar degeneration. Anterior-posterior lobe uptake ratios showed variable degrees of negative correlation with disease progression in patients with MSA-C and SCA type 2, which revealed the preferential involvement of anterior lobe, whereas asymmetry in none of the patient groups were affected by disease duration.
Anterior cerebellar lobe, functionally corresponding to spinocerebellum, plays a key role in maintaining posture, muscle tone, and gait [19,24]. The cardinal feature of spinocerebellar disease is involvement of lower extremities resulting in wide-based ataxic gait with small hesitant steps [19,25]. Therefore, reduced uptake in the anterior cerebellar lobe of patients with cerebellar ataxia is a logical consequence of the disruption in motor functions executed by spinocerebellum.
Although decreased metabolism in anterior lobe is common among patients with cerebellar ataxia, characteristic patterns of cerebellar subregional metabolism were distinct in each etiology. In patients with MSA-C, metabolic alterations were observed not only in the anterior-posterior ratio, which was consistent with the results of previous reports [13], but also in patterns of asymmetry. Tendency for asymmetry in cerebellar metabolism was a distinctive pattern of MSA-C among patients with cerebellar ataxia in this study and mirrored the asymmetry in clinical features of patients with probable MSA-C determined by diagnostic criteria [26]. These motor findings are often asymmetrical during the natural course of disease and may be pronounced [27]. Furthermore, about 50% of probable MSA patients reported asymmetric onset of symptoms [28]. Dopamine transporter (DAT) imaging of patients with a definitive postmortem diagnosis also showed a trend toward greater asymmetry of striatal DAT binding in MSA than in PD [29]. Although decreased cerebellar glucose metabolism in bilateral hemispheres and vermis is well known [13,15,30], its asymmetrical pattern was unnoticed. Excluding lateralization of individual 18 F-FDG PET images before group comparisons might result in neutralization of asymmetric trends. Considering the relatively short duration between onset of symptoms and 18 F-FDG PET imaging in this study, the asymmetric patterns of cerebellar metabolism could be a diagnostic clue in early stage of disease. Reaching milestones after disease onset takes more time, usually 20 to 30 months, in patients with MSA-C [31].
There were also some differences in subregional hypometabolism patterns between SCA type 2 and 6, despite similar patterns of preferential decreases in metabolic ratios in the anterior lobe of cerebellum. Patients with SCA type 2 showed the most extensive hypometabolism patterns in cerebellum. To our knowledge, only a few studies assessed cerebellar metabolism in SCA type 2 using 18 F-FDG PET. Inagaki et al. first reported decreased cerebellar metabolism in two symptomatic and three asymptomatic individuals who were genetically confirmed as SCA type 2 [14]. Regional patterns of cerebral glucose metabolism by voxel-based 18 F-FDG PET analysis showed that cerebellar metabolism was most severely compromised in SCA type 2, compared with SCA types 3 and 6 [32]. This was consistent with pathologic findings of marked loss of Purkinje cells in cerebellar cortex as well as the loss of myelinated fibers with gliosis in inferior and middle cerebellar peduncles, cerebellar white matter, and fasciculus cuneatus [33]. No reports comparing the differences in cerebellar metabolism of patients with MSA-C and SCA type 2 are currently available.
Patients with SCA type 6 exhibited hypometabolism mainly in the anterior lobe of cerebellar cortex and vermis, with relative preservation of the posterior lobe including inferior vermis. This finding is concordant with preferential superior vermian atrophy with relative sparing of inferior vermis in SCA type 6 reported in morphological and neuropathological studies [34,35]. While widespread reduction in glucose metabolism of cerebral cortex, basal ganglia, and cerebellum were previously reported in patients with SCA type 6 compared with healthy controls [36], no regional analyses were performed. Compared with SCA type 2, SCA type 6 demonstrated a lesser degree and extent of cerebellar hypometabolism [32], which were consistent with the results of the present study. Clinically, this form of SCA is characterized as pure cerebellar ataxia and can be accompanied by dysarthria, nystagmus, dysphagia, and even loss of proprioception and dystonia [37][38][39]. The pathology includes mainly the degeneration of Purkinje cells in cerebellar cortex, leaving dentate nucleus, deep cerebellar white matter, and cerebellar peduncles unaffected [39]. And the grossly normal cerebellum in some patients with SCA type 6 showed loss of Purkinje cells, being more marked in the superior vermis suggesting that the initial and primary change occurs in the cerebellar Purkinje cells, starting from the superior vermis [35].
In summary, distinct cerebellar subregional metabolic alterations among different types of cerebellar ataxia were attributable to the distinctive neuropathological changes in cerebellum, as reported in previous studies with acetylcholinesterase and DAT PET [40,41]. In contrast, patients with CCD, characterized by functional impairment in a region far from the site of a brain lesion that is anatomically connected by fiber tracts [42], showed relatively preserved metabolism in the anterior lobe resulting in high anterior-posterior lobe ratio. As the majority of patients with CCD in this study had brain tumors in the frontal lobe, this finding reflected alterations in the corticopontocerebellar tract passing through the anterior limb of internal capsule to ipsilateral pontine nuclei and crossing the midline via the middle cerebellar peduncle [43,44]. This finding was supported by anatomical observations of Brodal et al. who demonstrated that the majority of corticopontocerebellar fibers arose from the frontal cortex [45]. Brain tumors located in locations other than the frontal cortex would disrupt the corticopontocerebellar tract to a lesser degree, leading to different patterns in patients with CCD [46].
Choosing the occipital cortex as a reference area was another major outcome of this study. Correct identification of a reference area is an important prerequisite for accurate semi-quantitative analysis of cerebellar subregional metabolism. For count normalization, the most widely used normalization method computes the ratio of voxel value to the mean glucose metabolism of all voxels within a reference region, commonly defined as the template of the entire gray matter [47]. Certain diseases causing altered cerebellar metabolism, however, also involve the cerebral cortex [48], which makes the detection of statistically significant group differences difficult [49,50]. Although areas affected by disease are different in their anatomical distribution [51], performing a count normalization procedure is necessary to compare relative regional metabolic ratios between healthy controls or different diagnoses. This arithmetic transformation, termed normalization, is often utilized to reduce variation and allows for the more sensitive detection of disease-dependent patterns of altered metabolism [52]. Therefore, choosing a certain supratentorial area as a reference region rather than the entire cerebral cortex for normalization is more appropriate for the assessment of cerebellar metabolism.
As emphasized by Yakushev et al. [53], in patients as well as in healthy controls, appropriate reference areas should be maximally stable, minimally susceptible to external physiological stimuli, unaffected by the disease of interest, and reliable and easy to determine. The importance of valid reference regions for differential diagnosis is well established in diseases involving cerebral cortex such as dementia [53,54]. Up to date, only a few studies investigated altered cerebellar subregional metabolism; thus, there is no consensus on optimal supratentorial areas for count normalization of cerebellar uptake ratios in patients with cerebellar ataxia. In previous studies, variable reference areas such as frontal cortex, thalamus, and whole gray matter were utilized [14,[55][56][57][58]. While these regions all succeeded in showing decreased cerebellar uptake ratios, patients were compared only with normal controls. Well-known age-related metabolic decline in frontal cortex and global mean [52,59] suggests that this approach might be inappropriate for reference. These caveats could mask subtle differences between distinct types of cerebellar ataxia in this study.
At the start of the present study, choosing a specific supratentorial region as a reference for count normalization presented a challenge as previous reports indicated that not only cerebellar but also cerebral regions were involved in SCA as determined by 18 F-FDG PET [36,40,48]. Thus, we choose an area with minimal changes among disease groups and controls using voxelwise analysis. This method guaranteed a consistent measurement of metabolic rate at least in patients and subjects included this study. As shown in our results, compared with global count normalization, count normalization with occipital or parietal cortex resulted in more robust aging effects; this was consistent with the results of a previous 18 F-FDG PET study showing that parietal cortex was relatively spared in cerebellar ataxia [60]. Thus, the aging effect observed in global count normalization was largely due to the well-known age-related decline in frontal metabolism as mentioned above [61].
The present study has several major limitations. First, due to the lack of clinical parameters such as the International Cooperative Ataxia Rating Scale (ICARS) [62] and the Scale for Assessment and Rating of Ataxia (SARA) [63] scores in the medical records of patients with cerebellar ataxia, we could not correlate clinical severity with cerebellar subregional metabolism. Further studies are needed to investigate the correlation between the severity of ataxia and cerebellar subregional metabolism, which will provide valuable information on disease progression and metabolic deterioration. Second, we could not completely separate cerebellum from other structures of the brain. On voxelwise analysis, some areas with different metabolic ratios were located outside the cerebellum, which were not included in the design of the present study. Furthermore, certain areas with minimum detectable differences in the cerebellum might be overlooked. Third, patients with MSA-C in the present study were diagnosed clinically, and a possible diagnosis of SCA was ruled out based on a set of genetic tests that included only the common SCA types at our institution. Therefore, misdiagnoses could not be ruled out in the absence of postmortem verification [8]. Although all diagnoses were based on strict diagnostic criteria, a longer period of follow-up is needed. Finally, we only focused on altered metabolic patterns in cerebellum and acknowledge the need for further studies designed to assess functional topography of cerebellar and cerebral connections.
Despite these limitations, the detailed subregional analysis of cerebellar metabolism in our study provided insights into the different pathophysiological processes not only among ataxic patients but also in other etiologies of cerebellar metabolic alterations.

Conclusions
Quantitative analysis of cerebellar subregional metabolism by 18 F-FDG PET revealed the preferential involvement of anterior lobe in patients with cerebellar ataxia, which corresponded to spinocerebellum that controls balance of gait, whereas patients with CCD showed more asymmetric and preferential involvement of posterior lobe. These results might reflect the functionally and metabolically distinct involvement of cerebellar subregions among cerebellar diseases. Metabolic ratios of cerebellar subregions with occipital cortex used as reference for count normalization did not change with age (r = 0.141, p = 0.060 for ANT; r = −0.033, p = 0.665 for SUPP) (Right, Figure A and B in S2 File). There was a weak positive correlation in ANT (r = 0.214, p = 0.004) but no correlation in SUPP (r = 0.044, p = 0.556) by count normalization with parietal cortex used as reference (Middle, Figure A and B in S2 File). Moderate and weak positive correlations in ANT and SUPP, respectively, were observed by global count normalization (r = 0.448, p < 0.001 for ANT; r = 0.177, p = 0.018 for SUPP) (Left, Figure A