Introduction

Late-stage PD patients have a great risk of physical and psychological problems such as cognitive impairment, anxiety, and depression¹. Non-motor symptoms (NMS) usually appear earlier than motor deficits and seriously affect the patients’ quality of life², which highlights the importance of NMS in both diagnosis and management. Subtyping research is common in discovering biological targets associated with specific symptoms. However, while how to stratify PD motor symptoms has been largely studied^{3, 4}, there is a lack of research on the complex and diverse non-motor classification⁵. The functional connectome has great potential to dig out underlying subtypes, with disruption in functional connectivity varying considerably on disease severity and phenotypes⁶. In this study, we focused on stratifying PD subtypes with varying cognitive and psychological phenotypes by using the multivariate correlation between symptoms and brain functional connectivity.

Methods

Here, we retrospectively selected 54 Parkinson's patients (PDs) and 50 healthy controls (HCs) from Xuanwu Hospital, Beijing, China. Each patient’s information consisted of two parts: clinical assessments and MRI data. The performance of PD in non-motor characteristics was quantified by six clinical assessments scores, from Unified Parkinson’s Disease Scale I (UPDRS1), Hamilton Depression Scale (HAMD), Montreal Cognitive Assessment (MoCA), Pittsburgh Sleep Quality Index (PSQI), and Autonomic Symptom (AS). All the subjects underwent Siemens Skyra 3T MR scans with MPRAGE sequences (TR/TE =2530/2.98, Fov = 224 x 256 mm² , slice thickness = 1 mm, slice number = 192, flip angle = 7°，voxel size = 1x1x1mm³ ) and bold-EPI sequences (TR/TE = 2000/30, flip angle = 90°, slice thickness = 3mm，voxel size = 3.44×3.44×3.6mm³, image size = 64×64×35, timepoints = 176).
The preprocessing of the MRI data was completed by the fMRI Data Processing Toolbox (http://rfmri.org/dpabi)⁷. Each anatomical image was divided into 90 regions by the AAL template, then registered to the functional image for fMRI segmentation. The individual functional connectivity (FC) was obtained by calculating the Pearson correlation coefficient of any two regions’ signal strength, which was represented by a 90×90 matrix. Hierarchical clustering was implemented in the average FC matrix of the HC population, to dig out brain intrinsic networks.
The characteristics in NMS of 54 Parkinson's patients were combined to form a 54×6 matrix (54 PD patients and 6 clinical assessment scores), and the brain connections were combined to form a 54×4005 matrix (number of pairwise FC). Sparse canonical correlation analysis (sCCA)⁸ was used to calculate sparse canonical variates, which linked NMS and FC. We determined the subtyping dimensions by two steps: 1. comparing the covariance explained to find the variates that contained large original clinical information, 2. performing permutation test to find the variates that significantly reflected the correlation between two data sets. Then, 54 PDs were divided into different subtypes. ANOVA and Post-hoc analysis were performed across the groups in assessments, while the two-sample t-test with False Discovery Rate (FDR) was carried between each subtype and HCs in FC matrixes.

Results

Seven intrinsic connectivity networks, including the somatosensory network (SMT), dorsal medial default mode network (DDM), frontoparietal network (FPT), temporal default mode network (TDM), auditory network (AUD), visual network (VIS), and subcortical network (SBC) were obtained. The canonical variate I and III were finally chosen as the two dimensions for subtyping, which related to depression and cognition, respectively (see Figure1). Figure2 showed the subtypes’ non-motor features, while Figure3 displayed the specific disconnections in each subtype compared with HCs. The subtypes had statistically significant differences in the first five clinical scales (p < 0.05), but not in the Autonomic Symptom (AS) scores (p=0.862). The score fluctuation on AS was not closely connected with the changes in FC. And statistically significant changes in FC patterns were found in the first three subtypes (P_FDR<0.05), with the fourth subtype showing a trend toward significance (P_uncorrected<0.05). Table 2 summarized the salient features of the subtypes in NMS and the corresponding functional networks impairments. Depression stood out in the subtypes I and II, with decreased connectivity between the DDM, FPT, TDM, VIS, and SBC. Both subtypes II and IV showed cognitive dysfunction and disconnections between the FPT, DDM, and SBC.

Discussion

By using sCCA to link the dimensions of NMS and FC, we observed the FC-related signatures for varying non-motor impairments of PDs. The weakened connection from VIS and SBC to other brain networks was common, illustrating that the affected subcortical nucleus seriously hindered the brain functional activities of PD. Though there was generally decreased FC among all subtypes, varying non-motor clinical phenotypes across groups correspond to specific functional network reorganization. The FC features related to cognition and depression both had decreased connection between the FPT, DDM, and SBC. Specifically, cognition was characterized by abnormal connectivity within DDM, in contrast to depression with decreased connectivity within TDM. The results supported the possibility that FPT and default mode network may be the neurobiological mechanism underlying vulnerability to psychological and psychiatric symptoms. In this study, by the multivariate analysis method (sCCA), we obtained the cognition and depression-related abnormal FC patterns, locating symptom-specific biological targets to promote the development of diagnosis and treatment.

Acknowledgements

This study is supported by grants from the Basic Research Foundation of Shenzhen Science and Technology Stable Support Program (GXWD20201230155427003-20200822115709001).