Distinct neural circuits are associated with subclinical neuropsychiatric symptoms in Parkinson’s disease

Abstract

Background:

Parkinson’s disease (PD) can present with neuropsychiatric symptoms (here, anxiety, depression, and apathy) at any stage of the disease. We investigated the neural correlates of subclinical neuropsychiatric symptoms in relation to motor and cognitive symptoms in a high-functioning PD cohort.

Methods:

Brain morphometry of the cognitively intact, early-stage (Hoehn & Yahr 2) PD group (n=48) was compared to matched controls (n=37). Whole-brain, pairwise, resting-state functional connectivity measures were correlated with neuropsychiatric symptom, motor exam, and global cognitive scores of the PD group.

Results:

Factor analysis of highly collinear anxiety, depression, and apathy scores revealed a single principal component (i.e., composite neuropsychiatric symptom score) explaining 71.6% of variance. There was no collinearity between the neuropsychiatric, motor, and cognitive scores. Compared to controls, PD group showed only subcortical changes including amygdala and nucleus accumbens atrophy, and greater pallidal volume. Reduced functional connectivity in the limbic cortical-striatal circuits and increased functional connectivity between the cerebellum and occipito-temporal regions were associated with a more impaired neuropsychiatric profile. This functional connectivity pattern was distinct from those associated with motor deficits and global cognitive functioning. The individual components of the neuropsychiatric symptoms also exhibited unique connectivity patterns.

Limitations:

Patients were scanned in “on-medication” state only and a control group with similar neuropsychiatric symptoms was not included.

Conclusion:

Abnormal functional connectivity of distinct neural circuits is present even at the subclinical stage of neuropsychiatric symptoms in PD. Neuropsychiatric phenotyping is important and may facilitate early interventions to “reorganize” these circuits and delay/prevent clinical symptom onset.

1. Introduction

The neurodegenerative process in Parkinson’s disease (PD) gradually spreads throughout the whole brain affecting multiple neurotransmitter systems and networks, and causes not only motor, but also a broad spectrum of nonmotor dysfunction (Braak et al., 2003). In fact, several nonmotor features can predate the onset of motor symptoms and become increasingly prevalent as the disease progresses (Poewe, 2008). Neuropsychiatric symptoms including anxiety, depression, and apathy are common nonmotor characteristics of PD and can be present at any stage of the disease diminishing the quality of life of patients (Aarsland et al., 2009, 2012; Weintraub and Burn, 2011).

In recent years, magnetic resonance imaging (MRI) studies have demonstrated structural and functional brain changes associated with neuropsychiatric symptoms (from here on, referring to anxiety, depression, and apathy) in PD. Depression in PD has been shown to be associated with atrophy in limbic (Kostić et al., 2010; van Mierlo et al., 2015) and widespread cortical regions (Hanganu et al., 2017), however, increased anterior cingulate volume has also been found (van Mierlo et al., 2015). In resting-state functional MRI (fMRI) studies, increased functional connectivity of the limbic and motor regions (Dan et al., 2017; Hu et al., 2015; Lou et al., 2015; Wang et al., 2018; Zhang et al., 2019) and decreased functional connectivity between the cortical-limbic (Hu et al., 2015; Huang et al., 2015; Zhang et al., 2019) and cortical-cerebellar networks (Wang et al., 2018) have been observed in PD patients with depression. Furthermore, altered functional connectivity in the default mode (Lou et al., 2015; Wei et al., 2017), salience (Wang et al., 2018; Wei et al., 2017), and fronto-parietal networks (Wei et al., 2017) have been found in PD patients with compared to those without depression. Apathy in PD has been shown to be associated with atrophy in the basal ganglia (Carriere et al., 2014, Martinez-Horta et al., 2017, Ye et al., 2018), and fronto-parietal and limbic regions (Martinez-Horta et al., 2017, Reijnders et al., 2010; Ye et al., 2018), as well as with reduced functional connectivity between predominantly limbic fronto-striatal regions (Baggio et al., 2015; Dan et al., 2017, Lucas-Jiménez et al., 2018). Increased fronto-limbic functional connectivity associated with milder subclinical apathy has also been demonstrated (Lucas-Jiménez et al., 2018). Anxiety in PD has been linked to atrophy in limbic regions (Vriend et al., 2016; Wee et al., 2016) and in fronto-cingulate and parietal cortices (Carey et al., 2020, Wee et al., 2016). Anxiety in PD has also been linked to abnormal functional connectivity of the amygdala (Zhang et al., 2019) and orbitofrontal cortex (Dan et al., 2017), and increased functional connectivity between the salience network and nodes of the “fear circuit” with the amygdala as the hub (Carey et al., 2020). Higher anxiety in PD was found to correlate negatively with salience network and positively with left fronto-parietal network intrinsic connectivity (De Micco et al., 2020).

Based on these reports, one may argue that the neuropsychiatric symptoms in PD are associated with structural changes and altered functional connectivity within and between brain areas involved in higher-order cognitive regulation and emotion processing, yet, these changes are heterogeneous. An important source of heterogeneity is the selection of brain regions. Some studies focus on the functional connectivity of major resting-state networks, whereas others use a limited number of regions of interest. There is also considerable overlap between anxiety, depression, and apathy symptomatology in PD (Aarsland et al., 2012; Pagonabarraga et al., 2015; Skorvanek et al., 2015). Furthermore, many cortical (e.g., fronto-parietal) and subcortical (e.g., striatum) regions that are associated with neuropsychiatric symptoms are also involved in concomitant motor and cognitive deficits in PD. In most functional imaging studies, the overlap between neuropsychiatric symptoms and potential commonalities in neural substrates underlying the neuropsychiatric, motor and cognitive status are not always taken into account in a comprehensive way. In addition, while most studies investigate the clinically manifest neuropsychiatric problems, there is a gap in our understanding of the neural basis of subclinical neuropsychiatric symptoms in PD.

In this study, we examined the whole-brain functional neuroanatomical correlates of subclinical neuropsychiatric symptoms in community-dwelling, independent, and nondemented patients with PD using resting-state fMRI. Specifically, we aimed to address the following questions: 1) What are the neural substrates of shared and distinct features of subclinical neuropsychiatric symptoms? and 2) how do these neural substrates differ from those associated with the motor and cognitive status in PD? The basal ganglia are the core structures affected by the pathological process in PD. We hypothesized that the neuropsychiatric, motor, and cognitive features will be associated with altered functional connectivity predominantly in the limbic, motor, and cognitive cortical-striatal circuits, respectively. We also hypothesized that each neuropsychiatric component will demonstrate a unique neuroanatomical signature.

Finally, to account for the role of structural changes in functional connectivity, we also examined our PD group for potential brain atrophy compared to an age- and gender-matched healthy control cohort.

2. Material and Methods

2.1. Subjects

We enrolled 51 and included 48 subjects with a diagnosis of idiopathic PD according to the Movement Disorder Society (MDS) criteria (Postuma et al., 2015). All subjects participated in the study after giving written informed consent in accordance with the procedures approved by the Human Research Protection Office of the Yale School of Medicine. We recruited the subjects primarily through the Yale Movement Disorders Clinic and the Connecticut Advocates for Parkinson’s group. We conducted the study at the Yale Magnetic Resonance Research Center. All subjects underwent an initial screening for medical history and MRI safety. We excluded subjects with PD who met any of the following criteria: They were not fully independent (n=1), had a neurological (other than PD) disorder, had a medical condition that might affect the central nervous system, had a history of alcohol or illicit drug abuse, had a history of head injury resulting in loss of consciousness, had dementia (Montreal Cognitive Assessment (MoCA) score < 21) (Nasreddine et al., 2005), had contraindications for MRI (n=1), had abnormal findings on anatomical MRI scan (n=1). To assess brain atrophy in the PD group, we used the anatomical MRI scans of 37 age- and gender-matched control subjects from the Parkinson’s Progression Markers Initiative (PPMI) database for comparison (https://www.ppmi-info.org). The PPMI is a large, international multicenter clinical study to identify a variety of biomarkers for progression of de novo PD. The database also includes healthy controls who are 30 years or older at screening. Those who have current or active clinically significant neurological disorder, a first-degree relative with idiopathic PD, a MoCA score of 26 or less, and received certain dopaminergic or antidopaminergic drugs within six months of screening are excluded.

2.2. Clinical data collection

We assessed disease severity and stage using the MDS Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) (Goetz et al., 2008) and the Hoehn and Yahr (H & Y) scale (Hoehn and Yahr, 1967). The cutoff for H & Y for inclusion was < 3 (i.e., mild bilateral disease, may have some impairment in balance) to ensure that subjects were fully independent and could tolerate being off medication. We examined the subjects in the morning (MDS-UPDRS part III motor exam) when they were off dopaminergic medications overnight as the “off” state reflects the disease severity more accurately. Only five subjects were examined in the “medication-on” state, three of whom were not on carbidopa/levodopa treatment.

Following the MDS Task Force recommendations, we administered the Spielberger Trait Anxiety Inventory (STAI-T) (Spielberger et al., 1983, suggested (Leentjens et al., 2008a)), Beck Depression Inventory-II (BDI-II) (Beck et al., 1996, recommended (Schrag et al., 2007)), and Starkstein Apathy Scale (Starkstein et al., 1992, recommended (Leentjens et al., 2008b)). In addition, we administered the MoCA, Parkinson’s Fatigue Scale (PFS-16) (Brown et al., 2005, recommended (Friedman et al., 2010)), and PD Quality of Life Questionnaire (PDQ39) (Jenkinson et al., 1997).

2.3. Statistical analysis of the clinical data

We first assessed the normality of distribution of all clinical scores using the Shapiro-Wilk test. We compared the means and standard deviations of the normally distributed, and medians and median absolute deviations of the non-normally distributed scores with the population means or cutoff scores of the respective tests, when applicable, using one-sample t-tests (p < 0.05, two-tailed) (Supplementary Material Table S1). We used the SPSS 26 software for statistical analyses.

To assess a potential overlap between clinical dimensions, we first performed pairwise Spearman rho correlation analyses between the STAI-T, BDI-II, apathy, MoCA, and MDS-UPDRS part III scores to examine the degree of collinearity between these variables (Supplementary Table S2). Given significant collinearity between the STAI-T, BDI-II, and apathy scores, but not the MoCA and MDS-UPDRS III scores, we performed dimension reduction using factor analysis in SPSS 26. Principal component analysis with oblique rotation (assumes that the variables are not orthogonal) was used as the extraction method. The cutoff for component extraction was determined based on the eigenvalue > 1. This analysis yielded a single component with the eigenvalue of 2.15 (explaining 71.6% of variance). We called this component the composite neuropsychiatric symptom (NPScomposite) variable.

2.4. MRI data collection

We obtained the structural MRI and fMRI scans from PD subjects on a different day within two weeks of the clinical evaluation. We collected the scans in the morning after the subjects took the first dose of their dopaminergic medication to reduce discomfort in the scanner. We collected the MRI scans of 43 subjects in a 3.0 Tesla Siemens Trio TIM, and those of five subjects in a 3.0 Tesla Siemens Prisma scanner using a 32-channel head coil and identical sequences in both scanners.

We collected high-resolution T1-weighted MPRAGE images (176 slices, voxel size: 1 mm³, FoV: 250 mm, Matrix: 256 × 256, TR: 1900 ms, TE: 2.52 ms, TI: 900 ms, flip angle: 9 degrees) for an accurate localization of the fMRI data in the beginning of each scan session. Then, we obtained axial T2*-weighted, echo planar functional images at rest for 10 min and 8 s (36 slices, voxel size: 3.5×3.5×4 mm, FoV: 224 mm, Matrix: 64×64, TR: 2000 ms, TE: 25 ms, flip angle: 90 degrees, number of acquisitions: 304). We instructed the subjects to keep their eyes closed, avoid any voluntary movement, and let their mind wander. We assessed wakefulness at the end of the scan by subjects’ report.

The structural scans of all control subjects from the PPMI database were also collected in 3.0 Tesla scanners (see Supplementary Material).

2.5. Analysis of the structural MRI scans

We used the standard processing pipeline of the FreeSurfer analysis software (available at https://surfer.nmr.mgh.harvard.edu). Briefly, we used the subcortical volume-based stream (Fischl et al., 2002) and the cortical surface-based stream (Dale et al., 1999; Fischl et al, 1999) to obtain the individual subcortical volume and cortical thickness values, respectively. We performed visual quality checks and manual edits on the pre-processed data (see Supplementary Material).

2.6. Statistical analysis of the subcortical volumes

We extracted the estimated total intracranial, basal ganglia (caudate, putamen, pallidum, accumbens), thalamus, amygdala, and hippocampus volumes from the FreeSurfer automated segmentation output files. We averaged the subcortical volumes across the hemispheres. We performed all statistical analyses on volumes that were normalized to the estimated total intracranial volume in SPSS 26. We first entered the normalized volumes as the dependent variable into a general linear model (GLM) (fixed factor: group, covariate: age, nuisance variable: scanner type) to test for a potential age-by-group interaction. We then performed between-group comparisons using two-sample t-tests when the distribution of the normalized volumes was normal in both control and PD groups, and nonparametric Mann-Whitney U tests when the distribution was not normal. We set the statistical significance threshold for seven pairwise comparisons of the subcortical structures at p < 0.007 (0.05/7).

2.7. Statistical analysis of the cortical surfaces

We assessed the between-group differences in cortical thickness using the Query, Design, Estimate, Contrast (QDEC) module in FreeSurfer. The smoothed cortical surfaces of all subjects from both hemispheres using a Gaussian kernel with full-width half-maximum (FWHM) of 15 mm were the dependent variable and entered into a GLM (categorical variable: group, covariate: age, nuisance variable: scanner type). We performed correction for multiple comparisons using Monte Carlo simulations with a significance threshold of 1.3 corresponding to p < 0.05.

2.8. Analysis of the resting-state fMRI scans

We used the Connectivity toolbox for the resting-state data analysis (Whitfield-Gabrieli and Nieto-Castanon, 2012). Preprocessing steps included the removal of the first four scans to reach magnetization steady state, motion correction, outlier detection (frame-wise displacement above 0.9 mm or global signal changes above 5 standard deviations), coregistration of functional scans with the structural scan, normalization to the standard Montreal Neurological Institute (MNI) brain template, and smoothing with a Gaussian kernel with a FWHM of 8 mm to account for inter-individual anatomical variability. De-noising steps included correction for physiological and other sources of noise by regressing out the principal components of the white matter and cerebrospinal fluid signal using the CompCor method (Chai et al., 2012), regression of motion artifacts and outliers before bandpass-filtering, and quadratic detrending. Global signal was not removed. Finally, we bandpass-filtered (0.008 < f < 0.09 Hz) the data to capture the fluctuations of the blood oxygenation level-dependent (BOLD) signal that typically occur within this frequency range at rest. The voxel-wise global mean correlations after denoising were compared between subjects scanned in two different scanners (Supplementary Material).

2.9. Statistical analysis of the resting-state fMRI scans

We used the functionally defined nodes (n=268) of the whole-brain Shen atlas (Shen et al., 2013) for the functional connectivity analyses. For each subject, we extracted the average BOLD signal time courses from these nodes and correlated them with each other using Pearson correlations. The “r” values corresponded to the functional connectivity strength between node pairs. We Fisher z-transformed the “r” values and obtained group-level functional connectivity maps for statistical analyses. Finally, we used the NPScomposite (neuropsychiatric), MDS-UPDRS part III (motor), and MoCA (cognitive) scores as the covariates of interest in separate models to be correlated with these maps. In subsequent analyses, we also used the components of the NPScomposite variable (i.e., STAI-T, BDI-II, apathy scores) separately as covariates of interest for correlation with these maps. In all correlation analyses, we used age and levodopa equivalent daily dose (LEDD) of each subject as regressors. We used the false discovery rate (FDR) method for correction for multiple comparisons (p < 0.05) for each correlation map (Genovese et al., 2002).

2.10. Verification analysis of resting-state fMRI data

We repeated the functional connectivity analyses excluding five subjects who were on antidepressant medications (Supplementary Material).

3. Results

3.1. Demographic and clinical data

The mean age of the control group (n = 37) was 65.8 ± 8.7 years (range: 45.6 – 82) (Shapiro-Wilk test p = 0.939) and that of the PD group was 65.2 ± 8.6 (range: 45.3 – 79.7) (Shapiro-Wilk test p = 0.470). There were 23 males (62%) and 14 females (38%) in the control group and 31 males (65%) and 17 females (35%) in the PD group. The groups did not differ significantly in age (two-sample t-test, p = 0.762, equal variances) or gender (Chi-Square test, p = 0.818). The H & Y score was 2 in all PD subjects. The symptom onset side was left in 21 and right in 27 PD subjects. Four PD subjects were left-handed. Four PD subjects were not on any anti-parkinsonian medication. The rest were on carbidopa/levodopa (n=31), dopamine receptor agonists (n=15), MAO-B inhibitors (n=27), amantadine (n=12), and trihexyphenidyl (n=1). Fifteen subjects were only on one medication. Two subjects were on four, nine subjects on three, and 18 subjects on two medications. Five subjects were on chronic antidepressant treatment (sertraline, paroxetine, escitalopram). Time between last dose of dopaminergic medication and motor exam was 15.0 ± 3.8 hr. Of note, five subjects were examined in “on” state (1.8 ± 0.8 hr), three of whom were not on carbidopa/levodopa. The clinical data of the PD group are summarized in Table 1. The details of the behavioral analysis results are summarized in Supplementary Material Table S1. Briefly, our PD cohort scored significantly below the anxiety, apathy, minimal depression, and fatigue cutoff scores; significantly above the MoCA cutoff score for mild cognitive impairment, and had a significantly better quality of life compared to the PD population in the same disease stage.

Table 1.

Clinical data of the PD group

	Mean ± SD (range)
Age (yr)	65.2 ± 8.6 (45.3 – 79.7)
Disease duration (yr)	5.6 ± 3.9 (0.2 – 14.6)
LEDD (mg)	463.1 ± 347.9 (0 – 1640)
MDS-UPDRS I	8.0 ± 4.3 (0 – 19)
MDS-UPDRS II	9.6 ± 5.4 (1 – 23)
MDS-UPDRS III	29.4 ± 8.4 (12 – 58)
MDS-UPDRS IV	1.8 ± 2.3 (0 – 13)
MDS-UPDRS total	48.9 ± 12.8 (25 – 83)
MoCA	27.7 ± 1.9 (23 – 30)
STAI-T	34.5 ± 10.3 (21 – 63)
BDI-II	7.4 ± 6.7 (0 – 36)
Apathy	8.8 ± 5.2 (1 – 20)
PFS-16	2.3 ± 0.9 (1 – 4)
PDQ39-SI	15.9 ± 11.2 (0–43.4)

BDI-II: Beck depression inventory-II, LEDD: Levodopa equivalent daily dose, MDS-UPDRS: Movement Disorders Society-Unified Parkinson’s Disease Rating Scale (part III: Motor exam), MoCA: Montreal cognitive assessment test, PDQ39-SI: Parkinson’s Disease Quality of Life Questionnaire - Summary Index, PFS-16: Parkinson’s fatigue scale (averaged scores), STAI-T: Spielberger State and Trait Anxiety Inventory - Trait (also see Supplementary material).

3.2. Structural MRI results

The PD group demonstrated significantly lower volume in the amygdala and nucleus accumbens, and significantly higher volume in the pallidum compared to the control group (Table 2). The raw subcortical volumes are listed in the Supplementary Material Table S3.

Table 2.

Subcortical volumetric results

			S-W p values		M-W/t-test
Volume (mm3)	Control (N = 37)	PD (N = 48)	Control	PD	p values
Total intracranial	1,519,807 ± 170,334	1,592,940 ± 153,488	0.163	0.985	0.041
Thalamus	6,851 ± 775	7,334 ± 837	0.109	0.237	0.331
Caudate	3,425 ± 396	3,438 ± 411	0.053	0.019	0.050
Putamen	4,598 ± 631	4,515 ± 617	0.258	0.559	0.016
Pallidum	1,964 ± 205	2,290 ± 328	0.010	0.000	0.000*
Hippocampus	4,004 ± 350	3,999 ± 380	0.004	0.488	0.136
Amygdala	1,669 ± 202	1,584 ± 223	0.038	0.648	0.001*
Accumbens	469 ± 87	390 ± 75	0.069	0.564	0.000*

M-W/t-test: Mann-Whitney U nonparametric test for volumes that were not normally distributed in either group and two-sample test for volumes that were normally distributed in both groups, S-W: Shapiro-Wilk test for normality of distribution. All statistical tests were performed on volumes that were averaged across both hemispheres and normalized to the total intracranial volume.

^*:Significant p values that are below the statistical threshold of p < 0.007 (0.05/7).

There were no significant group differences in cortical thickness.

3.3. Resting-state fMRI results

The average translational and rotational head motion was 0.29 ± 0.36 mm and 0.28 ± 0.38 degrees, respectively. There were on average 1.9 ± 5.3 outliers in resting-state fMRI scans that were removed (see Supplementary Material Table S4 for details). These results show that our PD cohort had minimal head motion. The distribution of mean global correlations was not significantly different between the subjects scanned in different scanners suggesting that potential scanner effects were removed after denoising (see Supplementary Material).

The NPScomposite scores correlated negatively with the functional connectivity predominantly between the striatum/thalamus and frontal, limbic temporal, and posterior cingulate regions. The functional connectivity between the cerebellum and occipito-temporal regions and between the superior frontal and occipital regions correlated positively with the NPScomposite scores (Figure 1 top panel, Table 3). The MDS-UPDRS III scores correlated negatively with the functional connectivity predominantly between the caudal/middle parts of the right caudate and prefrontal regions and cerebellum. The functional connectivity between the right temporal pole and prefrontal regions correlated positively with the MDS-UPDRS III scores (Figure 1 middle panel, Table 4). MoCA scores correlated positively with the functional connectivity predominantly between the occipital and lateral and medial temporal regions; and negatively with the functional connectivity between the cerebellum and lateral prefrontal regions (Figure 1 bottom panel, Table 5).

Figure 1. Correlations between behavioral scores and pairwise functional connectivity.

. Top: NPS: Neuropsychiatric symptoms composite scores. Middle: MDS-UPDRS III: Movement Disorders Society Unified Parkinson’s Disease Rating Scale Part III motor exam scores. Bottom: MoCA: Montreal Cognitive Assessment scores. Nodes in both hemispheres (RH: Right hemisphere, LH: Left hemisphere) are displayed on the MNI template. Warm colors: positive, cool colors: negative correlations. AG: Angular gyrus, Cb: Cerebellum, Cd: Caudate, dACC: Dorsal anterior cingulate cortex, dlPFC: Dorsolateral prefrontal cortex, FG: Fusiform gyrus, FP: Frontal pole, HC: Hippocampus, ITG: Inferior temporal gyrus, Orb: Inferior frontal gyrus orbital part, PCC: Posterior cingulate cortex, PHC: Parahippocampus, Put: Putamen, SFG: Superior frontal gyrus, SMG: Supramarginal gyrus, SPL: Superior parietal lobule, TP: Temporal pole, Th: Thalamus, V1: Primary visual area, V2: Secondary visual area, VA: Visual association cortex. The box-and-line diagrams under each figure summarize the main connectivity results (red: positive and blue: negative correlations, FC: Frontal cortical nodes, Oc-Tp: Occipito-temporal nodes).

Table 3.

Correlations between NPScomposite scores and pairwise functional connectivity

Negative Correlations
Node Pairs	Pair Labels		T	p
Numbers	Node 1 (BA)	Node 2 (BA)
(123)–(11)	R caudate	R dlPFC (BA9)	−5.13	0.017
(127)–(189)	R thalamus	L TP (BA38)	−4.95	0.003
(127)–(60)	R thalamus	R ITG (BA20)	−4.75	0.003
(127)–(202)	R thalamus	L PHC (BA36)	−4.51	0.004
(264)–(189)	L thalamus	L TP (BA38)	−4.37	0.009
(260)–(189)	L caudate	L TP (BA38)	−4.26	0.009
(67)–(216)	R FG (BA37)	L V1 (BA17)	−4.50	0.013
(122)–(88)	R caudate	R PCC (BA23)	−4.32	0.023
(122)–(224)	R caudate	L PCC (BA23)	−4.10	0.023
(127)–(51)	R thalamus	R TP (BA38)	−3.86	0.025
(261)–(246)	L putamen	L cerebellum	−4.15	0.040
(123)–(9)	R caudate	R FP (BA10)	−3.89	0.045
(216)–(207)	L V1 (BA17)	L VA (BA19)	−3.88	0.046
Positive Correlations
Node Pairs	Pair Labels		T	p
Numbers	Node 1 (BA)	Node 2 (BA)
(109)–(76)	R cerebellum	R V2 (BA18)	4.62	0.009
(209)–(149)	L VA (BA19)	L SFG (BA8)	4.54	0.012
(112)–(67)	R cerebellum	R FG (BA37)	4.22	0.016
(209)–(14)	L VA (BA19)	R SFG (BA8)	4.09	0.025
(108)–(76)	R cerebellum	R V2 (BA18)	4.03	0.029
(109)–(80)	R cerebellum	R V2 (BA18)	3.98	0.034
(112)–(76)	R cerebellum	R V2 (BA18)	3.84	0.035
(109)–(215)	R cerebellum	L V1 (BA17)	3.78	0.036
(109)–(67)	R cerebellum	R FG (BA37)	3.73	0.036
(111)–(67)	R cerebellum	R FG (BA37)	3.61	0.049
(240)–(67)	L cerebellum	R FG (BA37)	3.55	0.049

BA: Brodmann area, dlPFC: Dorsolateral prefrontal cortex, FG: Fusiform gyrus, FP: Frontal pole, ITG: Inferior temporal gyrus, PCC: Posterior cingulate cortex, PHC: Parahippocampus, SFG: Superior frontal gyrus, TP: Temporal pole, V1: Primary visual cortex, V2: Secondary visual cortex, VA: Visual association cortex. Note: Node 261 extends to the left globus pallidus. See the interactive webpage https://bioimagesuiteweb.github.io/webapp/connviewer.html for the coordinates of the Shen Atlas nodes (Shen et al., 2013).

Table 4.

Correlations between MDS-UPDRS III scores and pairwise functional connectivity

Negative Correlations
Node Pairs	Pair Labels
Numbers	Node 1 (BA)	Node 2 (BA)	T	p
(241)–(186)	L cerebellum	L TP (BA38)	−4.61	0.009
(122)–(142)	R caudate	L FP (BA10)	−4.38	0.019
(122)–(140)	R caudate	L FP (BA10)	−4.09	0.023
(122)–(12)	R caudate	R SFG (BA8)	−3.92	0.023
(122)–(182)	R caudate	L AG (BA39)	−3.88	0.023
(122)–(148)	R caudate	L SFG (BA8)	−3.74	0.029
(121)–(241)	R caudate	L cerebellum	−4.02	0.030
(109)–(249)	R cerebellum	L cerebellum	−4.23	0.031
(122)–(149)	R caudate	L SFG (BA8)	−3.60	0.034
(122)–(105)	R caudate	R cerebellum	−3.57	0.034
(122)–(150)	R caudate	L SFG (BA8)	−3.47	0.040
(241)–(153)	L cerebellum	L IFG (BA47)	−3.74	0.048
Positive Correlations
Node Pairs	Pair Labels
Numbers	Node 1 (BA)	Node 2 (BA)	T	p
(53)–(140)	R TP (BA38)	L FP (BA10)	4.54	0.013
(53)–(153)	R TP (BA38)	L IFG (BA47)	3.80	0.046
(53)–(83)	R TP (BA38)	R dACC (BA32)	3.75	0.046

AG: Angular gyrus, BA: Brodmann area, dACC: Dorsal anterior cingulate cortex, FP: Frontal pole, IFG: Inferior frontal gyrus, orbital part, SFG: Superior frontal gyrus, TP: Temporal pole. See the interactive webpage: https://bioimagesuiteweb.github.io/webapp/connviewer.html for the coordinates of the Shen Atlas nodes (Shen et al., 2013).

Table 5.

Correlations between MoCA scores and pairwise functional connectivity

Negative Correlations
Node Pairs	Pair Labels		T	p
Numbers	Node 1 (BA)	Node 2 (BA)
(205)–(13)	L VA (BA19)	R SFG (BA8)	−4.84	0.004
(147)–(255)	L dlPFC (BA9)	L cerebellum	−4.30	0.014
(147)–(244)	L dlPFC (BA9)	L cerebellum	−4.15	0.014
(147)–(101)	L dlPFC (BA9)	R cerebellum	−4.14	0.014
(130)–(142)	R pons	L FP (BA10)	−4.29	0.026
(147)–(105)	L dlPFC (BA9)	R cerebellum	−3.78	0.026
(147)–(106)	L dlPFC (BA9)	R cerebellum	−3.76	0.026
(147)–(254)	L dlPFC (BA9)	L cerebellum	−3.70	0.027
(196)–(48)	L ITG (BA20)	R AG (BA39)	−4.11	0.046
Positive Correlations
Node Pairs	Pair Labels		T	p
Numbers	Node 1 (BA)	Node 2 (BA)
(234)–(80)	L PHC (BA36)	R V2 (BA18)	4.88	0.004
(195)–(80)	L ITG (BA20)	R V2 (BA18)	4.22	0.023
(195)–(41)	L ITG (BA20)	R SPL (BA7)	4.11	0.023
(234)–(212)	L PHC (BA36)	L V2 (BA18)	4.08	0.025
(195)–(175)	L ITG (BA20)	L SPL (BA7)	3.86	0.027
(195)–(179)	L ITG (BA20)	L SMG (BA40)	3.82	0.027
(260)–(8)	L caudate	R FP (BA10)	4.26	0.028
(195)–(82)	L ITG (BA20)	R V1 (BA17)	3.68	0.034
(195)–(45)	L ITG (BA20)	R SMG (BA40)	3.60	0.036
(234)–(210)	L PHC (BA36)	L V2 (BA18)	3.80	0.039
(95)–(46)	R PHC (BA36)	R SMG (BA40)	4.14	0.041
(231)–(80)	L HC (BA54)	R V2 (BA18)	3.70	0.047
(235)–(80)	L PHC (BA36)	R V2 (BA18)	3.65	0.047

AG: Angular gyrus, BA: Brodmann area, dlPFC: Dorsolateral prefrontal cortex, FP: Frontal pole, HC: Hippocampus, ITG: Inferior temporal gyrus, PHC: Parahippocampus, SFG: Superior frontal gyrus, SMG: Supramarginal gyrus, SPL: Superior parietal lobule, V1: Primary visual cortex, V2: Secondary visual cortex, VA: Visual association cortex.

See the interactive webpage https://bioimagesuiteweb.github.io/webapp/connviewer.html for the coordinates of the Shen Atlas nodes (Shen et al., 2013).

The individual NPS components showed different correlation patterns with the pairwise functional connectivity across the whole brain. Apathy scores correlated negatively with the functional connectivity of the striatum with the prefrontal regions and cerebellum, as well as of the fusiform gyrus with numerous other, mainly limbic temporal regions (e.g., insula, hippocampus, parahippocampus, amygdala). Apathy scores showed positive correlations with the functional connectivity between a smaller number of occipital and fronto-parietal regions (Supplementary Material Figure S1, Table S5). Depression scores correlated positively with the functional connectivity of the anterior cingulate cortex and insula with sensorimotor regions including the cerebellum, as well as with the functional connectivity between fronto-parietal regions. Depression scores showed relatively few negative correlations with the functional connectivity between the striatum/thalamus and limbic fronto-temporal regions (see Supplementary Material Figure S2, Table S6). Anxiety scores showed a small number of correlations, negative with the functional connectivity between the thalamus and brainstem, and positive with the functional connectivity between the cerebellum and occipito-temporal regions (see Supplementary Material Figure S3, Table S7).

3.4. Verification of resting-state fMRI results

The demographic and clinical characteristics of five subjects who were on chronic antidepressant treatment were comparable to those of the entire cohort (Supplementary Table S8). The correlation profiles between behavioral scores and pairwise functional connectivity excluding these five subjects were qualitatively identical to those of the entire cohort for the NPScomposite and MDS-UPDRS III, and differed only in the number of pairs, but not in anatomical distribution, for all other scores (Supplementary Table S9).

4. Discussion

Our PD cohort comprised independent and high-functioning individuals who, on average, had a good quality of life, no significant problems with fatigue, and only subclinical neuropsychiatric symptoms. We demonstrated significant collinearity between the apathy, depression, and anxiety scores in our PD group. The factor analysis justified the grouping of these features as a composite variable. Of note, there was no significant correlation between the neuropsychiatric, motor, and cognitive scores, therefore, significant interaction between these symptom domains was not a concern in the separate correlation analyses with the imaging data.

In summary, we found atrophy in the limbic subcortical structures including the amygdala and nucleus accumbens, and significantly higher volume in the pallidum in the PD group compared to controls. There was no significant difference in cortical thickness between the PD and control groups suggesting that the functional connectivity findings were not confounded by cortical atrophy in the PD group. We found that reduced functional connectivity between the striatum/thalamus and frontal and limbic cortical regions and increased functional connectivity between the cerebellum and occipito-temporal regions were associated with a more impaired neuropsychiatric profile. Reduced functional connectivity between the striatum and prefrontal cortical regions and cerebellum was associated with worse motor functioning. Finally, stronger functional connectivity between the posterior brain regions including the occipital, medial and lateral temporal, and parietal cortex was associated with better cognitive functioning, whereas stronger functional connectivity between the cerebellum and lateral prefrontal regions was linked to worse cognitive functioning.

4.1. Differences in neural circuits associated with neuropsychiatric, motor, and cognitive symptoms

Consistent with our hypothesis, the subclinical neuropsychiatric symptoms in our PD cohort correlated with reduced functional connectivity predominantly in the limbic cortical-striatal and in specific fronto-striatal circuits. The frontal nodes in these circuits are considered to belong to the salience network involved in detecting salient sensory and affective inputs (Menon and Uddin, 2010). The temporal pole comprises subregions with distinct functional connectivity profiles. Our temporal pole node overlaps with the anterior areas 35/36 reported by Pascual and colleagues (2015) that are thought to mediate higher-order visual and semantic processing, assessment of the value and relevance of visual information, and eye movement control (Pascual et al., 2015). The posterior cingulate cortex node in our study overlaps with the dorsal portion of BA23. In addition to being a hub in the default mode network involved in internally-oriented mental processes, the dorsal portion is also strongly connected to fronto-parietal task-positive networks important for cognitive control, and is thought to dynamically regulate the focus of attention (Leech and Sharp, 2014). Our results suggest that dysfunction in cortico-striatal circuits that mediate attention and emotional regulation is a common denominator underlying the neuropsychiatric symptoms in PD.

Motor severity mapped onto distinct cortico-striatal circuits including reduced functional connectivity not only between the motor, but interestingly also the cognitive fronto-striatal nodes implicating, perhaps compensatory, re-mapping of these circuits (Helmich et al., 2010). Of note, in contrast to the neuropsychiatric profile, increased functional connectivity between limbic fronto-temporal regions was associated with worse motor severity. Finally, contrary to our hypothesis, we did not find a strong association between cognitive functioning and the cognitive fronto-striatal circuits, but the posterior cortical brain regions. Worse cognitive performance in PD has been linked to atrophy (Uribe et al., 2016, 2018) and reduced functional connectivity in posterior cortical regions (Wolters et al., 2019).

Notably, abnormal cerebellar functional connectivity was also strongly linked to each symptom domain in different ways. Reduced cerebellar functional connectivity with the fronto-striatal nodes was associated with worse motor severity, whereas increased cerebellar functional connectivity with the occipito-temporal and lateral prefrontal cortical regions was associated with worse neuropsychiatric and cognitive functioning, respectively. Cerebellar hyperactivity and hyperconnectivity have been reported in motor task-based and resting-state fMRI studies of PD (Festini et al., 2015; Helmich et al., 2011; Wu and Hallett, 2013) and generally attributed to compensatory mechanisms. This functional pattern was most consistently exhibited in PD subjects scanned off dopaminergic medication and was either normalized or even reduced by medication (Festini et al., 2015; Solstrand Dahlberg et al., 2020). Moreover, different cerebellar lobules have been found to display altered activation and functional connectivity characteristics in PD depending on the motor, emotional, and cognitive context, as well as medication status (Solstrand Dahlberg et al., 2020). For instance, abnormal cerebellar functional connectivity has been found in cognitively impaired PD subjects (Kawabata et al., 2018; Maiti et al., 2020) and in those with depression (Wang 2018).

Taken together, our findings demonstrate that subclinical neuropsychiatric symptom domain is independent from motor and global cognitive functions at the behavioral level and maps onto separate neural circuits involved in attention and emotion regulation in a high-functioning cohort of PD patients.

4.2. Neural circuits associated with individual components of neuropsychiatric symptoms

Despite being strongly related at the behavioral level, the individual components of the neuropsychiatric symptoms demonstrated distinct neural correlates. Reduced functional connectivity in the fronto-striatal circuits was strongly associated with higher apathy scores consistent with previous reports (Baggio et al., 2015; Dan et al., 2017). Reduced functional connectivity of the amygdala and nucleus accumbens also correlated with higher apathy scores. Importantly, these two structures also showed significant atrophy in our PD group compared to controls consistent with previous reports of atrophy in the nucleus accumbens (Carriere et al., 2014, Martinez-Horta et al., 2017, Ye et al., 2018) and reduced nucleus accumbens-amygdala functional connectivity (Lucas-Jiménez et al., 2018) in subjects with PD and apathy.

Interestingly, apathy scores correlated significantly with reduced functional connectivity between the fusiform gyrus, which is involved in facial recognition, and limbic medial temporal regions. This finding is concordant with reports of impaired facial emotion recognition (Enrici et al., 2015; Kalampokini et al., 2018) and its relationship to apathy in nondemented subjects with PD (Martínez-Corral et al., 2010; Robert et al., 2014).

In contrast to apathy, reduced functional connectivity associated with depression symptoms was relatively rare, however, consistently involved the striatum and limbic fronto-temporal regions. Higher depression scores correlated with increased functional connectivity between the limbic anterior cingulate cortex/insula and sensorimotor networks including the cerebellum and the pallidum. Increased functional connectivity (Wang et al., 2018) and increased metabolic activity in similar regions (Wen et al., 2016) associated with depression in PD have been reported. Of note, the pallidum also showed larger volume in the PD group compared to controls. Enlarged pallidal volume in PD seems counterintuitive taken at face value, but may actually reflect an ongoing pathological process (e.g., astrogliosis (Fellner et al., 2011)). Regardless, abnormal pallidal functional connectivity associated with depression in PD is consistent with strong clinical evidence linking the pallidum to mood disorders (Lauterbach 2003).

Among all components, anxiety symptoms were associated with a paucity of functional correlations including predominantly increased functional connectivity between the cerebellum and occipito-temporal regions. This connectivity pattern was similar to the pattern underlying depression symptoms.

Taken together, these results demonstrate that each neuropsychiatric subdomain exhibits a different neural signature and contributes to that of the composite neuropsychiatric symptomatology in common and unique ways. Evidence from metabolic and neurotransmitter imaging studies supports these neuroanatomical distinctions (Thobois et al., 2017, Valli et al., 2019, Yousaf et al., 2017). For example, the anterior cingulate and orbitofrontal cortices demonstrate reduced metabolism in depression and anxiety, but increased metabolism in apathy in PD (Thobois et al., 2017). Similarly, the anatomical distribution and gradient of the dopaminergic, serotoninergic, and noradrenergic denervation vary for each neuropsychiatric symptom and according to disease stage in PD. For instance, in PD with apathy, serotoninergic deficits play a prominent role in early stages, whereas dopaminergic mesolimbic deficits become more important in advanced stages (Thobois et al., 2017). Therefore, we think that a detailed characterization of the neuropsychiatric symptomatology starting early in the disease course followed by longitudinal assessments has important implications regarding the clinical management of these symptoms and for developing therapeutic interventions.

Limitations

We scanned our subjects in “on” medication state. Our results should be interpreted with caution and not be generalized as the “off” medication functional connectivity would be expected to show considerable differences. Due to the lack of separate control groups who do not have PD, but exhibit similar neuropsychiatric symptoms, we cannot disentangle the contributions of PD-related pathological processes to the connectivity findings from those underlying the specific neuropsychiatric symptoms. Therefore, we can only claim correlations with regard to the degree of these symptoms.

5. Conclusions

Our study shows that subclinical neuropsychiatric symptoms in a high-functioning cohort of PD patients are linked to abnormal functional connectivity in neural circuits that are distinct from those underlying motor deficits and global cognitive functioning. While the functional connectivity profiles associated with the individual components of the neuropsychiatric symptoms are unique, they also converge on limbic cortical and subcortical, fronto-striatal, and cerebellar brain regions. Our findings have direct clinical implications: Abnormal functional connectivity of specific neural circuits even before the neuropsychiatric symptoms become clinically manifest suggests that there may be a window of opportunity for early interventions to “reorganize” these circuits and delay/prevent clinical symptom onset. This further emphasizes the importance of not only motor but also careful neuropsychiatric phenotyping even at the relatively early disease stages in PD. Future research examining the relationship between the longitudinal changes in functional connectivity and neuropsychiatric symptomatology could inform potential prediction models of progression and treatment response.

Highlights.

• Apathy, depression, anxiety symptoms yielded a subclinical neuropsychiatric (NPS) score

• NPS mapped to reduced functional connectivity (FC) in limbic cortico-striatal circuits

• NPS mapped to increased FC between the cerebellum and occipito-temporal regions

• The NPS FC pattern was distinct from that of motor deficits and cognitive function

• Individual components of the NPS also showed distinct neural signatures.