Elevated caudate connectivity in cognitively normal Parkinson’s disease patients

Mild cognitive impairment (MCI) is common in Parkinson’s disease patients. However, its underlying mechanism is not well understood, which has hindered new treatment discoveries specific to MCI. The aim of this study was to investigate functional connectivity changes of the caudate nucleus in cognitively impaired Parkinson’s patients. We recruited 18 Parkinson’s disease patients—10 PDNC [normal cognition Parkinson’s disease; Montreal Cognitive Assessment (MoCA) ≥ 26], 8 PDLC (low cognition Parkinson’s disease; MoCA < 26) —and 10 age-matched healthy controls. All subjects were scanned with resting-state functional magnetic resonance imaging (MRI) and perfusion MRI. We analyzed these data for graph theory metrics and Alzheimer’s disease-like pattern score, respectively. A strong positive correlation was found between the functional connectivity of the right caudate nucleus and MoCA scores in Parkinson’s patient groups, but not in healthy control subjects. Interestingly, PDNC’s functional connectivity of the right caudate was significantly higher than both PDLC and healthy controls, while PDLC and healthy controls were not significantly different from each other. We found that Alzheimer’s disease-like metabolic/perfusion pattern score correlated with MoCA scores in healthy controls, but not in Parkinson’s disease. Increased caudate connectivity may be related to a compensatory mechanism found in cognitively normal patients with Parkinson’s disease. Our findings support and complement the dual syndrome hypothesis.

The region-to-region connectivity matrix (z-matrix) was constructed using individually masked ROIs. The z-matrix was sorted, and adjacency matrices were defined with varying cost (1-50%), e.g., at 25% cost threshold, the top 25% of the z-values were set to 1 and the rest were set to 0 excluding the diagonal elements; therefore, the graph was undirected and unweighted, and both positive and negative connectivity have been considered 31 . The minimum cost threshold that resulted in fully connected adjacency matrix for all subjects was 15%. To ensure that our analysis was not dependent on the specific cost thresholding, adjacency matrices were considered at varying cost of 15-25%. Graph theory metrics at each of these costs were averaged together for a mean value measure. These metrics included characteristic path length, clustering coefficient, smallworldness, global and mean local efficiency, and were compared between groups 32,33 . At a regional level (left and right caudate nucleus), betweenness centrality (BC) was also estimated. BC represents if a given node is within the shortest pathway connecting any other pair of nodes. BC identifies nodes that are crucial for information flow in a brain network. In other words, network hubs tend to have high BC values. For graph theory analysis, the Brain Connectivity Toolbox 34 and in-house programs running on MATLAB 8.3.0 (Mathworks, Inc.) were utilized.
Alzheimer's disease-like CBF pattern analysis. The CBF maps were derived from pCASL data as previously described 35 using ASL Perfusion MRI data processing toolbox (https ://cfn.upenn .edu/~zewan g/ asltb x.php). The resulting CBF images were co-registered to the corresponding T1-weighted image, spatial normalized by wrapping to the MNI standard space, then smoothed using 8 mm Gaussian filter. Preprocessing was done using SPM12 with the default parameters. We then used a classification method to identify Parkinson's disease patients with Alzheimer's disease-like CBF pattern expression as previously described 24 . This pattern is characterized by decreased activity in the precuneus, the medial frontal lobes, the temporal lobes, and the cingulum; and relatively increased activity in the somatosensory-motor areas, basal ganglia, thalamus, and cerebellum. If a subject has a high score, their brain metabolic pattern looks more like Alzheimer's disease patients. We have previously compared the classification accuracy of three different methods, i.e., general linear model, scaled subprofile modelling, and support vector machine (SVM). We showed that SVM-sequential minimal optimization (SMO)-based Alzheimer's disease classifier was also sensitive to Parkinson's disease dementia, and that perfusion imaging may be also useful and potentially replacing fluorodeoxyglucose positron emission tomography (FDG-PET) 24 .
Statistical analysis. Statistical analysis was performed using SPSS (IBM Corp., Armonk, NY). Shapiro-Wilk test was performed to determine the normal distribution of each variable. For normally distributed variables (age, AD-like CBF pattern score, characteristic path length, and mean local efficiency), one-way ANOVA was used to assess group differences, followed by post-hoc Bonferroni test if applicable. For variables not normally distributed (sex, caudate BC, disease duration, UPDRS-III, BDI-II, LEDD, clustering coefficient, global efficiency, and smallworldness), Kruskal-Wallis and Mann-Whitney tests were used to determine group differences. Spearman's correlation was used to examine relationships between imaging-based variables (caudate BC and AD-like CBF pattern score) and MoCA in Parkinson's disease patients. Results were considered significant at a threshold of p < 0.05.

Results
Demographic data. The relevant demographics and clinical variables of each group are presented in Table 1. There was no significant difference in age between groups. As expected, PDLC had significantly lower MoCA scores compared to both PDNC and controls (p < 0.001, Mann-Whitney U). PDNC and healthy controls were not significantly different in MoCA (p = 0.971, Mann-Whitney U). Sex was not equally distributed between groups (more female participants in the healthy control group) (χ2 = 8.08, p = 0.012). PDLC had significantly higher BDI-II than controls (p = 0.002, Mann-Whitney U), while not significantly different from PDNC (p = 0.072, Mann-Whitney U). PDNC and healthy control BDI-II scores were not significantly different from each other (p = 0.230, Mann-Whitney U). There was no significant difference between patient groups (PDNC vs. PDLC) in age, sex, disease duration, UPDRS-III motor subscale or LEDD (p > 0.3).

Caudate functional connectivity correlates with cognitive performance in Parkinson's disease.
To examine the role of the caudate nucleus as an information hub, and the effect of its dysfunction in Parkinson's disease cognitive deficits, we measured correlation between caudate BC and MoCA scores. We found a strong positive correlation between the BC of the right caudate nucleus and MoCA scores in Parkinson's disease patients (rho = 0.629, p = 0.005) but not in healthy controls (rho = − 0.19, p = 0.59; Fig. 1A). Interestingly, PDNC's BC of the right caudate was significantly different from both PDLC and normal controls (p < 0.012, Mann-Whitney), while PDLC and normal subjects were not significantly different from each other (p = 0.965, Mann-Whitney; Fig. 2). The left caudate BC was not correlated with MoCA scores in either the Parkinson's disease group (p > 0.9) or healthy controls (p > 0.5). Neither caudate BC variables nor MoCA scores were correlated with age, or medications, within each group. No significant difference between sexes was observed for MoCA and the caudate BC variables within each group (p > 0.14). There was no significant correlation between BDI scores and MoCA in Parkinson's patients (p > 0.45). However, BDI scores correlated with MoCA scores in the healthy control group (rho = 0.721, p = 0.019).
Alzheimer's disease-like CBF pattern expression and caudate betweenness centrality. As expected, Alzheimer's disease-like CBF pattern score was correlated with MoCA in healthy controls (rho = − 0.669, p = 0.034), suggesting the feasibility of using FDG-PET-based Alzheimer's disease-like pattern quantification in pCASL-based perfusion imaging. However, the correlation was not observed in patients (rho = − 0.339, p = 0.216; Fig. 1B). No group differences (PDLC vs. PDNC vs. Healthy controls) were observed either (p = 0.444, Kruskal-Wallis). No significant correlation was observed between caudate BC and AD-like CBF pattern score (p > 0.58, Fig. 1C). This score was not correlated with any other variables, and there were no group differences in AD-like CBF pattern scores (p = 0.673, one-way ANOVA).

Discussion
In this study, we found that BC of the right caudate positively correlated with MoCA scores in Parkinson's disease. Interestingly, the PDNC group showed significantly higher BC scores for the right caudate than both PDLC and normal controls. We also found the Alzheimer's disease-like CBF pattern score was not correlated with MoCA scores in Parkinson's disease.
The positive correlation between the BC of the right caudate and MoCA scores supports the current literature claiming the caudate is involved with Parkinson's disease cognitive deficits 12,39,40 . To reiterate, the dual syndrome hypothesis posits the underlying dysfunction of cognitive impairment in non-demented Parkinson's disease is in the frontostriatal circuit involving the caudate nucleus and prefrontal cortex. Our patients' cognitive status is  www.nature.com/scientificreports/ mildly impaired (i.e., PDLC), and the correlation that we have found between the BC of the caudate and MoCA scores further supports this hypothesis. The association between cognitive performance and caudate BC was only seen in the right caudate nucleus, not the left. Laterality of the caudate in Parkinson's disease has been reported in many studies, using different imaging modalities. With PET imaging, an association between the right caudate and cognitive performance has been reported: the authors found decreased dopaminergic function of the right caudate was related to slow processing time, using the Stroop test in patients at an early stage of Parkinson's disease 41 . More prominent hypometabolism in the right caudate has been shown in cognitively impaired Parkinson's disease patients 42 . This may suggest that the right caudate plays a more crucial role in maintaining cognitive functions (or at least the subdomain of cognitive functions that are more accurately assessed by MoCA), and its deterioration could lead to more cognitive impairment. www.nature.com/scientificreports/ More detailed group analysis revealed PDNC had higher right caudate BC than both PDLC and controls, and similar MoCA scores to controls. We suggest this increase in "hubness" 43 to be a successful compensatory mechanism, for maintaining cognitive performance in PDNC. Similarly, Pereira and colleagues 44 demonstrated an increase in frontal hubs in PDNC, compared to both healthy controls and Parkinson's disease with MCI. While it is difficult to assess the causality for this shifted frontal connectivity, it should be noted that the caudate receives dense projections from the dorsolateral prefrontal cortex 45 , which directly controls dopamine release in the caudate 46,47 . Nevertheless, since BC is a relative measure, it is also possible that the increased BC is a result of the relative maintenance of frontal area connections compared to degenerative areas in early Parkinson's disease.
Caudate dopamine degeneration has been associated with cortical hypometabolism characterized as Parkinson's disease cognition-related metabolic pattern (PDCP) 48 , which does not emerge until after 2 years from the initial diagnosis but slowly progresses over the years 49 . This suggests that the prefrontal-caudate circuitry is relatively intact in early stage Parkinson's disease although the overall dopamine level in the caudate nucleus is already significantly lower than age-matched normal subjects 50 . The capacity for compensation in the remaining dopaminergic neurons has already been established 51 . The more spatially diffused dopaminergic innervation in the striatum 52,53 may be associated with increased functional connectivity. Increased resting-state connectivity within the striatum, particularly the caudate, has previously been suggested to aid in cognitive maintenance in Parkinson's disease 54 . Further, no age-related changes were noted in caudate BC in our healthy control individuals (n = 37, unpublished observation). This suggests the high caudate BC is a unique feature to PDNC. A minority of Parkinson's patients (approximately 20%) with a disease duration reaching 8 years do not go on to develop dementia 55 . Our PDNC cohort had an average disease duration of 9.4 ± 2.6 years. Their success in maintaining normal cognition may therefore be associated with their high levels of caudate BC.
According to the dual syndrome hypothesis, another important axis of Parkinson's cognitive deficits is the posterior cortical and temporal abnormality-traditionally thought to be relevant to Alzheimer's disease-related pathology 14 . FDG-PET is the most frequently used imaging method that complements diagnosis of Alzheimer's disease, and is often used to quantitate disease progression 56 and treatment responses 57 . We recently developed and validated automated quantification methodology to estimate how likely an individual FDG-PET scan belongs to an Alzheimer's disease patient 24 . Results suggested that this SVM-SMO-based method was sensitive to Parkinson's disease dementia as well, but not to cognitively normal Parkinson's disease patients.
MoCA and Alzheimer's disease-like CBF pattern scores were not correlated in Parkinson's disease groups. One potential explanation is that cognitive impairment in our patients (without dementia) was primarily driven by frontostriatal abnormality, and less from Alzheimer's disease-like pathology. This negative observation further aligns with the dual syndrome hypothesis 14 : demented patients' pathology is more similar to Alzheimer's, differing from MCI pathology. A carefully designed longitudinal study is warranted to confirm this hypothesis.
We did not identify any significant differences in global graph theory metrics between the groups, posing a potential limitation to this study. Metrics such as increased clustering coefficient 58 , and characteristic path length with reduced global efficiency 44 in Parkinson's MCI compared to non-demented Parkinson's disease and healthy controls have been previously reported. However, other studies have reported maintained global integration in Parkinson's disease compared to healthy controls 30,59 . Most of these studies were done on drug-naïve Parkinson's disease patients, or while off medication. Our patients in this study were examined while on their prescribed medications. This raises another limitation-the impact of antiparkinsonian medication on perfusion has been previously documented 60 . Studies suggest chronic levodopa treatment may induce angiogenesis and increase vascular sensitivity in the putamen 61,62 . It is less evident if the caudate is also involved in this hyper-vascularity phenomenon, which may have influenced its BC measurement. It is also unknown if the Alzheimer's disease-like pattern score estimation using pCASL MRI is influenced by the use of antiparkinsonian medications.  www.nature.com/scientificreports/ While MoCA provides insightful information about patients' cognitive status, it is our limitation that PDLC vs. PDNC was primarily determined using only MoCA scores, and not by thorough neuropsychological exams. Different types of assessments were used to screen dementia in our patients (e.g., DSM-4 vs. DSM-5 criteria), and specific results for assessments other than MoCA were not disclosed, as the data was not de-identified. Nevertheless, it should be noted that our primary outcome is the correlation between MoCA and caudate BC in Parkinson's disease as one group, and the division between PDLC and PDNC was only done in a post-hoc manner to understand the effect of cognitive status in relation to healthy control subjects.
Other constraints include small sample sizes for each group. This limited further subgroup analyses, involving handedness and most-affected hemisphere (motor symptoms). When included as a covariate, sex and BDI-II scores did not significantly influence results. Nevertheless, given the small group sizes, and their disparity in sex and BDI-II scores, these results should be interpreted with caution, as a preliminary investigation of neural compensatory mechanisms for PD patients.

Conclusions
Our findings support and complement the dual syndrome hypothesis, i.e., Parkinson's disease executive dysfunction is more related with dopaminergic pathways (prefronto-caudate) and visuospatial dysfunction is more related with cholinergic (parieto-temporal); the latter being more relevant for dementia, while the former is more involved with MCI. Our study suggests these two pathologies are dissociated, and increased network hubness of the right caudate nucleus may potentially involve a compensatory mechanism.

Data availability
The raw data are not publicly available as they contain patient medical data which can only be accessed under the Personal Health Information Act (PHIA). Information regarding this can be found at https ://www.gov. mb.ca/healt h/phia/.