Patients were recruited from Peking University First Hospital from November 2021 to September 2023. The diagnosis of Parkinson’s disease, multiple system atrophy, progressive supranuclear palsy were made according to the Movement Disorder Society criteria for PD (Postuma et al., 2015), MSA (Wenning et al., 2022) and PSP (Hoglinger et al., 2017), respectively. The clinical diagnosis of parkinsonism was made based on clinical presentation, and an initial diagnosis of PD, MSA or PSP were made by at least 2 experienced neurologists in agreement in outpatient settings. Clinical diagnoses were reconfirmed after inpatient assessment and expert panel review of the cases. Exclusion criteria include: (1) major psychiatric disorders including schizophrenia, major depressive disorder, schizoaffective disorders; (2) Significant CNS disorders including debilitating stroke, imaging evidence of excessive cerebral small vessel disease (Fazekas grade 3), CNS demyelination disorders; (3) safety concerns for MRI compatibility, and inability to withstand the MRI scan (e.g., specific phobia of enclosed space); (4) History of intracranial surgeries, including deep brain stimulation; (5) red flags in the diagnosis of respective diseases, (6) overt dementia as evidenced by clinical dementia rating ≥ 2.

The study is approved by institutional review board of Peking University First Hospital (IRB00001052-17043). The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent was obtained from all participants.

To validate finding from our local cohort, we also selected 13 health control (HC) and 37 PD patients from Parkinson’s progression markers initiative (PPMI) database. The selection criteria were: (1) Explicitly assigned to HC or PD category by PPMI study institutes; (2) Had complete T1 weighted MRI, rs-fMRI, MoCA test and UPDRS motor test data acquired within 6 months of whichever first assessment. Detailed clinical protocol of PPMI study can be found at https://www.ppmi-info.org/sites/default/files/docs/002_Protocol_AM4_v3.0_23July2024_Final.pdf.

After informed consent, participants were invited to have inpatient assessments. Demographic information, disease duration, and levodopa equivalent daily dose (LEDD) were recorded. Unified movement disorder assessments included Unified Parkinson’s Disease Rating Scale (UPDRS) for PD patients, Unified Multiple System Atrophy Rating Scale (UMSARS) for MSA patients, and Progressive Supranuclear Palsy Rating Scale (PSPRS) for PSP patients. Motor assessments were performed in ‘on’ medication states. Cognitive performance was assessed with Montreal Cognitive Assessment (MoCA). Other evaluations included autonomic function assessments, electronystagmography, and levodopa challenge test. Patients had out-patient follow up for at least 1 year after discharge and all diagnoses were reviewed before analysis.

For local patients, images were acquired in ‘on’ medication sate using a 3 T scanner (Ingenia 3.0 T; Philips Healthcare, Best, Netherlands). Structural MRI data were acquired using a 3D T1-weighted imaging. Turbo field echo compressed SENSE (TFE-CS) sequence on Philips unit with the following parameters: repetition time (TR) = 6.5 ms; echo time (TE) = 2.9 ms; flip angle: 8°; FOV: 240 × 240 mm; and voxel size: 1.0 × 1.0 × 1.0 mm.

Resting-state fMRI was acquired using an echo-planar imaging (EPI) sequence with the following parameters: TR = 2,970 ms, TE = 30 ms, voxel size: 2.0× 2.0× 3.0 mm³, 3 mm slice gap, and flip angle (FA) 90°. Before resting-state data acquisition, we instructed participants to close their eyes, relax, and not engage in any particular mental activity during the scan.

A multi-echo gradient-echo (GRE) sequence was used for SWI-plus data acquisition with the following parameters: TR = 41 ms, five TEs =7.5/15/22.5/30/37.5 ms, FA = 15°, voxel size = 0.67 × 0.67× 2.0mm³, and bandwidth = 217 Hz/px, number of slices: 66, slice orientation: F-H.

PPMI imaging data selected in the current analysis was acquired with SIEMENS prisma 3 T scanner. Detailed parameters can be found in https://www.ppmi-info.org/sites/default/files/docs/PPMI2.0_002_MRI_TOM_Final_v3.0_20210727_FE.pdf.

Analyses of fMRI data were performed using CONN toolbox release 22.7 (Whitfield-Gabrieli and Nieto-Castanon, 2012). fMRI volumes were realigned using SPM realign & unwarp function, where all scans were coregistered to the first volume using a least squares approach and a rigid body transformation, then resampled using b-spline interpolation for motion correction. Temporal misalignment between different slices was corrected using since temporal interpolation. Potential outlier scans were identified using ART as acquisitions with framewise displacement above 0.5 mm or global BOLD signal changes above 3 standard deviations. Reference functional image was computed by averaging all fMRI volumes for each participant. Functional and anatomical data were normalized into standard MNI space, segmented and resampled to 2 mm isotropic voxels following a direct normalization procedure using SPM unified segmentation and normalization algorithm. Lastly, functional data were smoothed using spatial convolution with a 6 mm Gaussian kernel.

fMRI data were denoised using a standard denoising pipeline including the regression of potential confounding effects characterized by white matter timeseries, CSF timeseries, motion parameters and their first order derivatives, outlier scans, followed by bandpass filtering between 0.008 Hz and 0.09 Hz.

A singular value decomposition (SVD) with 64 components was used as a subject-specific dimensionality reduction. Independent component analyses were performed to estimate 40 temporally coherent networks from the fMRI data combined across all subjects. The dimensionality of the concatenated data was further reduced using SVD with 40 components, and fast-ICA was used to identify spatially independent group-level networks from the resulting components. Last, GICA3 back-projection was used to compute ICA maps associated with these same networks separately for each individual subject.

The spatial properties of ICA networks were compared against established functional networks, and dice index was used to determine the corresponding functional networks of ICA components (Supplementary Figure 1). As illustrated by previous efforts, cerebellar network, sensorimotor network, default mode network and language networks were selected a priori as networks of interest (Supplementary Figure 2).

SWI images were processed using JHU/KKI QSM toolbox (Li et al., 2019). Laplacian phase unwrapping was first performed followed by background removal using variable-kernel sophisticated harmonic artifact reduction for phase data (V-SHARP) method, with the average image of all echoes as reference (Fang et al., 2017). Images were then smoothed with an 8 mm Gaussian Kernel. Quantitative susceptibility was calculated using thresholded k-space division (TKD) method. SWI images were then coregistered with T1-weighted images normalized into standard MNI space, segmented and resampled to 2 mm isotropic voxels following direct normalization. Magnetic susceptibility value (measured in ppm) was sampled in caudate nuclei, putamen, substantia nigra and dentate nuclei.

Statistical analyses were performed with R (4.4.1) and SPM12 for clinical variables and parametric images, respectively. Kruskal Wallis test was performed for cross-group comparison, Mann Whitney U test was used for group-wise comparison of continuous variables, and Chi-square test was applied to categorical variables.

Group-level analyses of ICA maps were performed using General Linear Models (GLM). For each ICA component (network of interest), ANOVA was performed to determine if any difference existed across diagnostic groups. For those networks with positive F-test (namely cerebellar network and language network), post-hoc t-tests were performed to establish unique network changes specific to the diagnostic group. Regression analyses were further performed on brain network and susceptibility maps to interrogate the association between clinical assessments (motor performance and MoCA scores) and brain network activity or with iron deposition. Inferences were performed at the level of individual clusters (groups of contiguous voxels). Sex and disease duration were corrected in the regression models and group comparisons. Cluster-level inferences were based on parametric statistics from Gaussian Random Field theory. Results were thresholded using a combination of a cluster-forming p < 0.005 voxel-level threshold, and a family-wise corrected p-FDR < 0.05 cluster-size threshold. An illustration of analysis framework can be found in Supplementary Figure 3.

Local study data can be shared upon reasonable request to the corresponding authors. PPMI data used in the preparation of this article was obtained on [2024-11-25] from the Parkinson’s Progression Markers Initiative (PPMI) database (https://www.ppmi-info.org/access-data-specimens/download-data), RRID:SCR_006431. For up-to-date information on the study, visit https://www.ppmi-info.org.

16 PD patients, 8 MSA patients, 7 PSP patients were recruited in our local cohort, and a further 13 HC and 37 PD patients were selected form PPMI database. Sex distribution, age and disease duration did not differ significantly across the groups. In our local cohort, the median of MoCA scores were 24.5, 26 and 22 for PD, MSA and PSP patients, respectively. Median UPDRS motor score was 26.5 for PD patients, median UMSARS score was 19.5 and median PSPRS motor section score was 20. Demographic and clinical profile of study cohort is summarized in Table 1 and extended clinical information can be found in Supplementary Table 1.

Quantitative susceptibility differed in putamen across groups, but did not survive FDR correction (F = 4.7, p = 0.17, p-adj = 0.17). Voxel-wise ANOVA suggested cross-group difference in left putamen [(−24, −3, −9); cluster size = 433; cluster-p-FDR < 0.001] and right globus pallidus [(17, −7, −2); cluster size = 181; cluster-p-FDR = 0.008, Figure 1A]. Post-hoc group wise comparisons showed MSA patients had increased iron deposition in left putamen compared to PD patients [(−24, −3, 9); cluster size = 374; cluster-p-FDR < 0.001, Figure 1B]; while PSP patients had increased iron deposition in right pallidum compared to PD patients [(17, −7, 2); cluster size = 186, cluster-p-FDR = 0.04, Figure 1C].

In our local cohort, analysis of variance found significant cross-group differences in cerebellar network (bilateral superior frontal gyrus; (2, 38, 50); cluster size = 112; p-FDR = 0.009) and language network (bilateral posterior cingulate gyri; (2, −34, 32); cluster size = 357; p-FDR = 0.01), but not sensorimotor network and default mode network. Post-hoc comparisons showed significant difference of cerebellar network and language networks between MSA and PD patients. MSA patients had decreased cerebellar connectivity with bilateral superior frontal gyrus [(2, 38, 50); cluster size = 266; p-FDR = 0.001, Figure 2A], and impaired language connectivity with bilateral posterior cingulate gyri [(4, −22, 34); cluster size = 440; p-FDR = 0.001] and bilateral thalamus [(−12, −30, +2) and (+8, −24, +4); cluster size = 241; p-FDR = 0.001; Figure 2B].

In PPMI cohort, PD patients showed increased cerebellar network activation in cerebellar vermis, brain stem and right thalamus; and decreased language network connectivity with bilateral frontal pole, frontal orbital cortex, precuneus and lateral occipital cortex (Supplementary Figure 4).

Iron deposition at ROI level did not correlate with cognition or motor symptoms in 3 diagnostic groups. However, in voxel-wise analysis, quantitative susceptibility in bilateral putamen was correlated with worse UMSARS motor scores in MSA patients (−26, 5, 4; cluster size = 413; p-FDR < 0.001 and 15, 5, −2; cluster size = 479; p-FDR < 0.001, Figure 3A). In PD patients, iron deposition in cerebellum was associated with worse MoCA scores (16, −64, −53; cluster size = 301, p-FDR = 0.01, Figure 3B).

In local PD cohort, cerebellar network connectivity in cerebellar vermis, brain stem and left thalamus positively correlated with UPDRSm scores, while in PPMI PD patients’ cerebellar connectivity with precentral, posterior central cortices was negatively associated with motor symptoms (Table 2 and Figure 4). Association between cerebellar network and motor symptoms was not significant in MSA and PSP patients. Interestingly, default mode network connectivity was not associated with cognitive performance in the patients. Instead, MoCA scores was related to language network connectivity in PD and PSP patients (Table 2 and Figure 4). Notably, increased involvement of primary motor cortex in language network was negatively related to cognitive performance, while language network activity in bilateral frontal gyri and left angular gyrus was associated with better cognitive scores in those patients.

In MSA patients, iron deposition in dentate nucleus and putamen was associated with decreased cerebellar network activity within the cerebellum and with thalamus. While quantitative susceptibility in globus pallidum was associated with decreased language network activity in left supramarginal gyrus and cerebellum, and increased language network connectivity with postcentral cortex, precuneus and right frontal pole.

In PD patients, substantia nigra iron deposition was associated with both decreased cerebellar connectivity with thalamus and decreased language network activity in left angular gyrus and cerebellum. In PSP patients, language network connectivity was influenced by iron deposition in substantia nigra and subthalamus nucleus, both causing decreased network activity in left angular gyrus. Interestingly, iron deposition in substantia nigra was also associated with increased precentral gyri activity in language network, which was linked to worse cognitive performance (Table 3).