A total of 84 patients diagnosed with probable or possible MSA were recruited in the Department of Neurology, Huashan Hospital, Fudan University between February 2012 and May 2019. The clinical diagnosis and categorization between MSA-P and MSA-C were made according to the second consensus statement on the diagnosis of MSA published in 2008 (Gilman et al., 2008). Demographic information including age, age at onset, sex, disease duration, and education degree were systematically collected. All patients included neither had head injury nor had other neurological or psychiatric diseases.

This study was approved by the Human Studies Institutional Review Board, Huashan Hospital, Fudan University. Written informed consents were obtained from the participants before entering this study in accordance with the Declaration of Helsinki.

Motor disability was evaluated by the Unified Parkinson’s Disease Rating Scale III (UPDRS III) in all participants in a Med-Off state, which means that the patients were off anti-parkinsonism drugs for more than 12 h. Global cognitive status and cognitive dysfunction in five domains covering attention, executive function, memory, visuospatial function, and language were assessed using Mini-Mental State Examination (MMSE) and a series of comprehensive neuropsychological tests, which were as follows: attention by the Symbol Digit Modalities Test (SDMT; Sheridan et al., 2006) and part A of Trail-Making Test (TMT-A; Zhao et al., 2013); executive function by the part C of Stroop Color-Word Test (CWT-C; Steinberg et al., 2005) and part B of Trail-Making Test (TMT-B; Zhao et al., 2013); memory function by the Auditory Verbal Learning Test (AVLT; Guo et al., 2009) and delayed recall task of the Rey-Osterrieth Complex Figure Test (CFT-delay recall) (Caffarra et al., 2002); visuospatial function by copy task of Rey-Osterrieth Complex Figure Test (CFT; Caffarra et al., 2002) and the Clock Drawing Test (CDT; Ricci et al., 2016); and language ability by Animal Verbal Fluency Test (AVFT) and Boston Naming Test (BNT; Lucas et al., 2005). To eliminate dimensional differences in tests and quantify impairments of cognition, neuropsychological results of each test of all patients were transformed into Z-score using the mean and SD of the norm data from healthy controls of similar age and education in the Shanghai area (Wu et al., 2018; Shen et al., 2021): Z-score = (test score - mean score of norm data)/SD of norm data (when data was a score) or Z-score = (mean time of norm data - test time)/SD of norm data (when data were a time metrics) (Wu et al., 2018). Z-score less than −1.5 was defined to be the impairment in each test and Z-score of each cognition domain was calculated using the average Z-score of two tests in the same cognition domain.

¹⁸F-fluorodeoxyglucose PET imaging was acquired in all patients within 1 month before or after the clinical assessment in the Huashan Hospital. Patients were instructed to be fasting for at least 6 h in an off-drug state before the scan. Forty-five minutes resting after injection of ¹⁸F-FDG (185 MBq), a short CT scan, and subsequent 10-min PET scan was performed by a Siemens Biograph 64 PET/CT (Siemens, Munich, Germany) in the three-dimensional (3D) mode in a dimly lit and quiet room. Images were subsequently reconstructed using a 3D ordered-subset expectation-maximization algorithm (4 iterations; 24 subsets; Gaussian filter, 2 mm; zoom, 3). Reconstructed images had a matrix size of 168 × 168 × 148 and a voxel size of 2 × 2 × 1.5 mm³.

Positron emission tomography images were transformed for further processing using the MRIcron tool (output format: hdr/img). Then, images were processed by statistical parametric mapping (SPM5) software operated in MATLAB platform (Mathworks Inc., Sherborn, MA, United States). First, the original images were spatially normalized to normative stereotactic Montreal Neurological Institute (MNI) space using default [¹⁵O]-H2O PET template (Della Rosa et al., 2014)¹. Second, they were smoothed through a three-dimensional Gaussian filter with 10-mm full-width at half-maximum (FWHM).

To compare the metabolic characteristics between MSA-P and MSA-C, the two-sample t-test was used with age, sex, disease duration, and UPDRS score as covariates according to the general linear model at each voxel in SPM. To assess whether metabolic characteristics in a specific brain region would correlate with Z-score in each cognitive domain, the multiple regression model established by us previously (Ge et al., 2015; Wu et al., 2018) was applied first to detect clusters with peak threshold set at P < 0.001 over whole brain regions and extent threshold set as 120 voxels (960 mm³) and further to highlight clusters survived after a false discovery rate (FDR) correction at p < 0.05. In this step, education degree and UPDRS III score were both included as covariates to eliminate the effect of these factors. The differences in global metabolic values were modeled by analysis of covariance (ANCOVA) to minimize intersubject variability and improve signal-noise ratios. Exact coordinates and corresponding brain regions of detected clusters were ascertained through an MNI-to-Talairach conversion using Talairach–Daemon software (Research Imaging Center, University of Texas Health Science Center, San Antonio, TX, United States). The SPM maps for positive and negative correlations were superpositioned on a standard T1-weighted MRI brain template. Then, the spherical volume of interest (VOI) (4 mm radius) with the center of the circle at the peak voxel of the cluster was constructed, and metabolic value in each VOI was quantified using ScAnVP software (Version 5.9.1; Center for Neuroscience, the Feinstein Institute for Medical Research, Manhasset, NY, United States). To eliminate the individual differences in global metabolic value, the normalized regional cerebral metabolic rate of glucose (rCMRglc) was calculated by the following formula: [VOI value/whole-brain metabolism] × 100%.

Differences in the demographic information, clinical characteristics, and neuropsychological performances between MSA-P and MSA-C were analyzed through Student’s t-test in continuous and normally distributed data. Pearson’s χ² test was used in categorical data. Mann–Whitney U test was used in continuous and non-parametric data. The post-hoc correlation analysis between Z-score in the specific cognitive domain and normalized rCMRglc in each VOI was performed through Pearson correlation and linear regression analyses. In the post-hoc metabolic comparison between patients with MSA-P and MSA-C, multiple comparisons via the Benjamini–Hochberg procedure were performed to control the FDR. Statistical analyses were performed using SPSS software (SPSS for Windows, version 22.0; SPSS Inc., Chicago, IL, United States). p < 0.05 was considered statistically significant.

Demographic, clinical, and detailed neuropsychological information of the total 84 patients with MSA was depicted in Table 1. The mean age at onset was 54.89 ± 7.97 years, and the mean disease duration was 26.80 ± 20.69 months. Patients with MSA were divided into two subtypes according to their predominant clinical symptoms, including 27 patients with MSA-P and 57 patients with MSA-C. No group differences were found for age at onset and education, but differences were observed in age, sex, disease duration, and UPDRS III score (Table 1). Both subtypes had considerable and extensive cognitive impairments in each domain represented by number and percentage of impairment as follows: attention, MSA-P, 18(67%), MSA-C, 30(53%); executive function, MSA-P, 13(48%), MSA-C, 32(56%); memory, MSA-P, 18(67%), MSA-C, 25(44%); visuospatial function, MSA-P, 12(44%), MSA-C, 22(39%); language, MSA-P, 10(37%), MSA-C, 21(37%). Compared with patients with MSA-C, patients with MSA-P performed worse on the attention domain, judging by a lower Z-score of attention domain and two corresponding tests (SDMT and TMT-A) (p < 0.05). However, this difference in attention performance was not significant after adjusting age, sex, disease duration, and UPDRS score factors.

Global metabolic values did not correlate with any Z-score in five domains (regression analysis: absolute value r ≤ 0.07, p ≥ 0.40). However, the positive and negative correlations between the regional ¹⁸F-FDG metabolism and the cognitive Z-score in the five domains were detected (Figure 1).

The brain regions correlated positively to the Z-score of the separate cognitive domain were described as follows, respectively: attention with left inferior frontal gyrus, left superior frontal gyrus, left medial frontal gyrus, and right cerebellum (uvula and cerebellar tonsil); executive function with left inferior frontal gyrus, left middle frontal gyrus, left cerebellum (uvula), and right cerebellum (pyramis and inferior semilunar lobule); language with left superior frontal gyrus, left middle frontal gyrus, left cerebellum (pyramis and uvula), and right cerebellum (tuber, pyramis, cerebellar tonsil, and declive); memory with inferior frontal gyrus; and visuospatial function with left cerebellum (pyramis) (Figure 1 and Table 2).

However, the brain regions associated negatively with the Z-score of the separate cognitive domain were shown as follows, respectively: attention with right fusiform gyrus and right insula; executive function with right fusiform gyrus and right inferior temporal gyrus; language with left anterior cingulate, right middle temporal gyrus, and right fusiform gyrus; and memory with right fusiform gyrus and right inferior occipital gyrus (Figure 1 and Table 3).

Moreover, the post-hoc correlations between the Z-score in separate cognition domain and the normalized rCMRglc in each significant brain region were calculated, with the correlation coefficients listed in Tables 2, 3. The scatterplots and relationships of cognition and metabolic value in 11 representative regions are shown in Figure 2.

Compared with patients with MSA-C, increased metabolism mainly in the bilateral cerebellum while decreased metabolism in bilateral putamen was observed in MSA-P, with no difference in frontal gyrus (Supplementary Table 1).

Considering the unadjusted cognitive differences in the attention domain between MSA-P and MSA-C, the comparisons of regional metabolism in attention-related areas mentioned above between two subtypes are presented in Figure 3 and Table 4. However, no significant metabolic intergroup differences were observed in the left inferior frontal gyrus, superior frontal gyrus, medial frontal gyrus, right fusiform gyrus, right postcentral gyrus, and right insula. Only the metabolism in the right cerebellum (including uvula and cerebellar tonsil) was significantly lower in patients with MSA-C than patients with MSA-P after adjustment for confounders and post-hoc correction for multiple comparisons.

In addition, correlations between metabolism and the cognitive Z-score in the five domains separately in two subtypes are shown in Supplementary Tables 2, 3. Cognition in patients with MSA-C correlated positively with metabolism in the frontal lobe and cerebellum, negatively with that in the fusiform gyrus and insula, whereas the clusters correlated with cognition in patients with MSA-P were limited to the cerebellum (positively) and temporal lobe (negatively).