Resting‐state functional connectivity of the striatum in early‐stage Parkinson's disease: Cognitive decline and motor symptomatology

Abstract

Parkinson's disease is a neurodegenerative disorder characterized by changes to dopaminergic function in the striatum and a range of cognitive and motor deficits. Neuroimaging studies have repeatedly shown differences in activation and functional connectivity patterns of the striatum between symptomatic individuals with Parkinson's disease and healthy controls. However, the presence and severity of cognitive and motor symptoms seem to differ dramatically among individuals with Parkinson's disease at the early‐stages. To investigate the neural basis of such heterogeneity, we examined the resting state functional connectivity patterns of caudate and putamen subdivisions in relation to cognitive and motor impairments among 62 early‐stage individuals with Parkinson's disease (21 females, 23 drug naive, ages 39–77 years, average UPDRS motor scores off medication = 18.56, average H&Y stage = 1.66). We also explored how changes in striatal connectivity relate to changes in symptomatology over a year. There are two main findings. First, higher motor deficit rating was associated with weaker coupling between anterior putamen and midbrain including substantia nigra. Intriguingly, steeper declines in functional connectivity between these regions were associated with greater declines in motor function over the course of 1 year. Second, decline in cognitive function, particularly in the memory and visuospatial domains, was associated with stronger coupling between the dorsal caudate and the rostral anterior cingulate cortex. These findings remained significant after controlling for age, medication, gender, and education. In sum, our findings suggest that cognitive decline and motor deficit are each associated with a differentiable pattern of functional connectivity of striatal subregions. Hum Brain Mapp 37:648–662, 2016. © 2015 Wiley Periodicals, Inc.

INTRODUCTION

Parkinson's disease is a neurological disorder with a prevalence of roughly 1% in the population over the age of 60 [Tanner and Goldman, 1996]. A hallmark of Parkinson's disease is the dramatic loss of dopamine (DA) cells in the caudal lateral substantia nigra pars compacta [SNpc; Damier et al., 1999]. With seemingly a common progression of midbrain pathology across individuals with Parkinson's disease [Braak et al., 2003; Damier et al., 1999], phenotypic expression of motor and cognitive symptoms at the early‐stages of the disease is highly variable [Dujardin et al., 2013; Muslimovic et al., 2005]. Neuroimaging studies have found metabolic and functional changes in subcortical structures, particularly in the striatum, in relation to Parkinson's disease symptom variability [Ekman et al., 2012; Lewis et al; 2003; Wu et al., 2010]. It has been suggested that striatal subregions are differentially modulated by midbrain dopaminergic neurons and form distinct circuits responsible for the flexible control of motor, cognitive, and motivational behaviors [for reviews, see Alexander and Crutcher, 1990; Gerfen, 1992; Haber, 2003]. Loss of function in this circuit thus can have profound, wide‐ranging consequences for daily behavior. Differential changes in striatal function may therefore be a critical factor underlying heterogeneous phenotypes in Parkinson's disease, making the striatum a compelling target for investigation.

Cognitive deficits in Parkinson's disease are a major detriment to quality of life [Schrag et al., 2000] and may involve attention, response control, and working memory [see reviews by Caviness et al., 2007; Monchi et al., 2012; Pagonabarraga and Kulisevsky, 2012]. Neuroimaging studies have consistently pointed to the dorsal head of caudate as a locus for cognitive symptom heterogeneity in Parkinson's disease. For instance, in studies of voxel‐based morphometry, individuals with Parkinson's disease with cognitive deficits or dementia but not cognitively unimpaired individuals with Parkinson's disease have reduced gray matter volume in the dorsal caudate in comparison to healthy controls [Apostolova et al., 2010; Melzer et al., 2012; Nagano‐Saito et al., 2005]. fMRI studies indicate that the subset of cognitively impaired individuals with Parkinson's disease under‐recruit the dorsal caudate during working memory [Ekman et al., 2012; Lewis et al., 2003] and set‐shifting [Nagano‐Saito et al., 2014] tasks, but in the cognitively unimpaired individuals with Parkinson's disease, the caudate activated at levels close to or even above that of healthy controls. In these anatomical and functional studies, no other regions of the striatum were reported to significantly relate to cognitive impairment in Parkinson's disease.

Whereas the caudate has been implicated in cognitive function, the putamen, which has dense anatomical connections with motor cortical areas [Alexander et al., 1986; Draganski et al., 2008; Joel and Weiner, 1997], is more closely associated with motor deficits of Parkinson's disease. Lee et al. [2014] found a trend towards a negative relationship between putamen volume and disease duration. Another study using shape analysis found that atrophy of the posterior putamen was associated with motor symptom severity [Nemmi et al., 2015]. In these studies, volumes of the other basal ganglia nuclei did not show a significant correlation with motor symptomatology. fMRI studies have also shown functional deficits in the putamen in association with motor symptoms of Parkinson's disease. Decreases in putamen activity during hand movements are related to more severe motor deficits [Wu et al., 2010]. In a recent fMRI meta‐analysis of studies using simple motor tasks, decreased posterior putamen activation was associated with more severe motor deficit ratings in Parkinson's disease [Herz et al., 2014].

While a growing body of literature has documented how behavioral measures of Parkinson's disease correlate with regional changes in striatal structure and activity, few studies have examined how cortico‐striatal functional connectivity varies with motor/cognitive deficits, and most existing studies do not examine the different subdivisions of the striatal nuclei, which may have contributed to the variability in findings [Agosta et al., 2014; Hacker et al., 2012; Seibert et al., 2012]. Previous findings were also impacted by disease duration as most studies included individuals in Hoehn and Yahr stages two or three (out of five) of the disease. To address this gap of research, we took advantage of the large cohort of resting state functional connectivity (rsFC) data from newly diagnosed individuals provided by the Parkinson's Progression Markers Initiative (PPMI). rsFC measures the correlations of spontaneous, low frequency blood oxygenation level dependent (BOLD) signals between brain regions, and is considered a measure of the functional cohesiveness of neural circuits [Biswal et al., 1995; Fox and Raichle, 2007]. This method is utilized extensively because the data is relatively easy to collect and rsFC patterns are highly reliable within and across healthy individuals [Fox and Raichle, 2007]. Previous researchers have identified patterns of rsFC for striatal areas that differentiate individuals with Parkinson's disease from healthy controls [Helmich et al., 2010; Luo et al., 2014; Szewczyk‐Krolikowski et al., 2014]. Here, we sought to characterize striatal rsFC patterns that correlate with a few key clinical measures and hence identify neural activity patterns that may contribute to variable Parkinson's disease phenotypes. An additional focus was to examine in greater spatial detail how the strength of these associations vary across subregions of the caudate and putamen. We hypothesized that the rsFC of the caudate nucleus and putamen with the cortex would be respectively modulated by cognitive and motor test scores commonly used to assess individuals with Parkinson's disease.

MATERIALS AND METHODS

Behavioral and Resting State fMRI Data

Behavioral and imaging data were obtained from the Parkinson's Progression Markers Initiative (PPMI) database (http://www.ppmi-info.org/data). PPMI—a public‐private partnership—is funded by the Michael J. Fox Foundation for Parkinson's Research and funding partners, including Abbvie, Avid Radiopharmaceuticals, Biogen, Bristol‐Meyers Squibb, Covance, GE Healthcare, Genetech, GlaxoSmithKline, Eli Lilly, Lundbeck, Merck, Meso Scale Discovery, Pfizer, Piramal, Roche, and UCB. PPMI is a multi‐site study that collects standardized behavioral, neuropsychological, biospecimen and neuroimaging data for a large number of newly diagnosed individuals with Parkinson's disease. More specific details of the PPMI study can be found in a report published previously [Marek et al., 2011]. For up‐to‐date information on the study, visit the project webpage (http://www.ppmi-info.org). A total of 73 participants’ data from seven different sites (Baylor College of Medicine, Johns Hopkins, Emory, Northwestern, Tübingen, Marburg, and Cleveland Clinic) were originally obtained for analysis in September 2014. We included only individuals classified as stage one or two on the Hoehn & Yahr scale for Parkinson's disease severity [Hoehn and Yahr, 1967], who were not reported as taking medications for cognitive impairment, such as acetylcholinesterase inhibitors, and who were not reported by experimenters as having any problems during the scanning session (e.g., experience of claustrophobia). All MRI and fMRI data were acquired on Siemens Trio Tim 3 Tesla magnets (Siemens Medical Systems, Erlangen, Germany), and imaging parameters were kept constant across all sites. High resolution anatomical images (T1‐weighted MP‐RAGE) were acquired with repetition time (TR) = 2,300 ms, echo time (TE) = 2.98 ms, flip angle (FA) = 9°, matrix = 240 × 256, field of view (FOV) = 256 mm, voxel size 1 × 1 × 1 mm³, 176 sagittal slices with slice thickness = 1 mm. Resting state fMRI (rsfMRI) echo‐planar images were acquired for 8.5 min (212 volumes) with TR = 2,400 ms, TE = 25 ms, flip angle = 80°, matrix = 68 × 66, FOV = 222 mm, 40 slices (ascending direction) and voxel size 3.3 × 3.3 × 3.3 mm³. Subjects were instructed to rest quietly, to keep their eyes open and not to fall asleep.

Individual participants’ anatomical and functional images were visually inspected one by one to ensure that the whole brain was covered and screened for artifacts. For the Parkinson's disease cohort, data from four participants were excluded from analysis because dorsal portions of the brain were cut off or had severe artifacts. Data from seven other participants were excluded due to excessive head motion during rsfMRI (for details, see Head Motion section below). As a result, a total of 62 rsfMRI datasets were included in the final sample (21 females; 39 on Parkinson's disease medications). At the time of the scan, 21 individuals were taking levodopa (34%), 16 were taking DA agonists (26%), 18 were taking MAO‐B inhibitors (29%), and none were taking COMT inhibitors. Additionally, 17 individuals were classified as tremor‐dominant (27%), 26 were classified as nontremor dominant (42%), and the remaining 19 (31%) were indeterminate. Classifications were made using criteria similar to previous reports: tremor‐dominant = Unified Parkinson's Disease Rating Scale [UPDRS; Goetz et al., 2007] resting tremor score of ≥2 for at least one hand, and nontremor dominant = UPDRS resting tremor score of 0 for all four limbs [e.g., Helmich et al., 2012]. Among the 39 individuals on Parkinson's disease medications, the mean levodopa dose equivalency (LED) [Tomlinson et al., 2010 ] was 287 mg/day (standard deviation 142 mg/day). The participants on medication did not undergo a medication washout before scanning.

Seventeen of the 62 participants also had a second rsfMRI scan 1 year after the initial scan. We utilized the data to explore longitudinal changes in rsFC in association with behavioral measures. From this group, four participants were excluded due to severe artifacts and signal loss, and two participants were excluded due to excessive head motion; as a result, rsfMRI data from 11 participants were included in the longitudinal analysis (five females, nine on Parkinson's disease medications).

Table 1 summarizes the demographics of the participants with Parkinson's disease included in this report. All participants enrolled in the PPMI study within 24 months of their initial diagnosis. We assessed motor dysfunction using the total (summed) score of the UPDRS part III off medication, where higher scores indicate more severe motor dysfunction. To assess the patients' cognitive ability, we used the total (summed) score of the Montreal Cognitive Assessment [MoCA; Nasreddine and Phillips, 2005]. Higher MoCA scores indicate better cognitive ability; to make the measure comparable to the UPDRS, we reversed the MoCA scores so that in the multiple regression analysis, higher reversed MoCA scores indicate greater cognitive decline. In addition, we assessed data on individuals’ disease stage [measured by the Hoehn and Yahr scale; Hoehn and Yahr, 1967], and their total years of education. The associations between these variables are shown in Table 2.

Table 1.

Demographic data of Parkinson's disease participants obtained from the Parkinson's Progression Markers Initiative database.

	Complete Sample (n=62)		Subsample on medication (n=39)		Subsample drug naïve (n=23)		Longitudinal cohort second scan (n=11)
	Mean (SD)	Median (range)	Mean (SD)	Median (range)	Mean (SD)	Median (range)	Mean (SD)	Median (range)
Age	59.64 (9.89)	60 (39–77)	60.59 (9.54)	62 (41–77)	58.04 (10.49)	56 (39–75)	56.72 (8.20)	57 (43–72)
Sex (female:male)	21:41		10:29		11:12		5:6
UPDRS Part III	18.56 (8.57)	17 (7–41)	18.87 (8.69)	17 (8–41)	18.04 (8.55)	19 (7–35)	23.18 (8.06)	22 (12–43)
H&Y Stage	1.66 (0.51)	2 (1–2)	1.64 (0.49)	2 (1–2)	1.65 (0.49)	2 (1–2)	1.72 (0.46)	2 (1–2)
MoCA	27.03 (3.11)	28 (15–30)	26.56 (3.52)	28 (15–30)	27.83 (2.08)	28 (23–30)	28.09 (2.07)	28 (24–30)
EDU (years)	15.47 (3.00)	16 (8–22)	14.66* (3.20)	15 (8–20)	16.83* (2.06)	16 (13–22)	14.93 (3.83)	16 (8–20)
n at each site: BCM/JHU/EM/NW/EKUT/PK/CC	14/3/14/6/3/14/8		8/2/8/2/3/14/2		6/1/6/4/0/0/6		3/0/0/1/0/3/4

Years of education was the only comparison that showed a statistical difference between drug‐naive and on‐medication subsamples (two‐sample t‐test, t = 2.90; P < 0.01).

UPDRS = Unified Parkinson's Disease Rating Scale; H&Y = Hoehn and Yahr; MoCA = Montreal Cognitive Assessment; EDU = education; BCM = Baylor College of Medicine; JHU = Johns Hopkins University; EM = Emory University; NW = Northwestern University; EKUT = Eberhard Karls University Tübingen; PK = Paracelsus‐Elena Klinik, University of Marburg; CC = Cleveland Clinic.

Table 2.

Associations between variables of interest

	Age	UPDRS Part III	MoCA	EDU (yr)
Age	—
UPDRS Part III	0.241	—
MoCA	−0.371*	−0.338*	—
EDU (yr)	−0.141	0.141	0.012	–

Numbers represent the correlation strength (r).

*P < 0.01.

UPDRS = Unified Parkinson's Disease Rating Scale; H&Y = Hoehn and Yahr; MoCA = Montreal Cognitive Assessment; EDU = education.

At the time of writing, the PPMI database had only seven control subjects with rsfMRI data, an inadequate number for statistical comparison. Therefore, we supplemented data from these seven subjects with an additional 38 subjects gathered from our own data. Control subjects were matched to the Parkinson's disease cohort on age ± 5 years and gender. Nine of the 45 control datasets were excluded due to excessive head motion during scanning (for details, see Head Motion section below). As a result, 36 control subjects remained for comparison analysis (10 females, ages 37–73 years, average age 54 years). These datasets were also collected on a Siemens 3 T scanner with similar imaging parameters: anatomical images were acquired with TR = 300 ms, TE = 2.5 ms, flip angle = 60°, field of view = 220 × 220 mm, matrix = 256 × 256, 32 slices interleaved, with slice thickness = 4mm and no gap. rsfMRI echo‐planar images were acquired for 10 min (300 volumes) with TR = 2,000 ms, TE = 25 ms, flip angle = 85°, FOV= 220 × 220 mm, matrix = 64 × 64, 32 slices, and voxel size = 3.4 × 3.4 × 4.0 mm³. Standardized ratings for motor and cognitive symptom severity for these subjects were not available, and so therefore the control subjects were only used to compare to the individuals with Parkinson's disease cohort on the effects of age (see Supporting Information for Results and Discussion).

Image Data Preprocessing

Standard image preprocessing was performed using Statistical Parametric Mapping (SPM8, http://www.fil.ion.ucl.ac.uk/spm). For each individual participant, functional images were corrected for slice timing and then realigned to the first image to correct for head motion between scans. Each individual's structural image was coregistered to the mean functional image of the dataset. Each individual structural image was then segmented and normalized to an Montreal Neurological Institute (MNI) gray matter template with affine registration followed by nonlinear transformation [Ashburner and Friston, 1999, Friston et al., 1995]. The normalization parameters determined for the structural volume were then applied to the corresponding functional image volumes for each participant. Finally, the images were smoothed with a Gaussian kernel of 6 mm at full width at half maximum.

Additional preprocessing was applied to reduce spurious fMRI signal variances that were unlikely to reflect neuronal activity [Fair et al., 2007; Fox and Raichle, 2007; Fox et al., 2005; Rombouts et al., 2009]. The sources of spurious variance were regressed out by including the signal from the ventricles, white matter and whole brain, in addition to the six parameters obtained by rigid body head motion correction. First‐order derivatives of these signals were also included in the regression. Because there is debate about whether or not it is appropriate to regress out the global signal in rsfMRI analysis [e.g., Murphy et al., 2009], we analyzed the data twice, with and without global signal regression. Because the primary results were nearly identical between the two sets of analyses, only results with global signal regression are presented here.

By analyzing fMRI signal in various frequency ranges, Cordes et al. [2001] showed that low‐frequency fluctuations (<0.1 Hz) contribute to more than 90% of the temporal correlations between cortical regions. Most previous studies used a low‐pass filter at a cut‐off of 0.08 or 0.1 Hz [Fox and Raichle, 2007]. Thus, to study low‐frequency fluctuations, we applied a temporal band‐pass filter (0.009 Hz < f < 0.08 Hz) to the time course data [Fair et al., 2007, Fox and Raichle, 2007, Fox et al., 2005].

Head Motion Correction

As reported by Van Dijk et al. [2012] and Power et al. [2012], micro head motion (>0.1 mm) is an important source of spurious correlations in resting state functional connectivity analysis. Therefore, we applied a “scrubbing” method proposed by Power et al. [2012] as successfully applied in recent studies [Power et al., 2012; Smyser et al., 2010; Tomasi and Volkow, 2014] to remove time points affected by head motions. Briefly, for every time point t, we computed the framewise displacement given by $\mathrm{F}\mathrm{D}\left(t\right)=\left|\Delta d_x\left(t\right)\right|+\left|\Delta d_y\left(t\right)\right|+\left|\Delta d_z\left(t\right)\right|+r\left|\alpha \left(t\right)\right|+r\left|\beta \left(t\right)\right|+r\left|\gamma \left(t\right)\right|$, where ${(d}_x,d_y,d_z)$ and $\left(\alpha ,\beta ,\gamma \right)$ are the translational and rotational movements, respectively, and r (=50 mm) is a constant that approximates the mean distance between center of MNI space and the cortex and transform rotations into displacements [Power et al., 2012]. The second head movement metric was the root mean square variance (DVARS) of the differences in % signal intensity I(t) between consecutive time points across all voxels, computed as follows: $\mathrm{D}\mathrm{V}\mathrm{A}\mathrm{R}\mathrm{S}\left(t\right)=\sqrt{〈{|I\left(t\right)-I(t-1\left)\right|}^2〉}$, where the brackets indicate the mean across brain voxels.

To compute each participant's functional connectivity map, we removed every time point that exceeded the head motion limit FD(t) > 0.5 mm or DVARS(t) > 0.5% via regression [Power et al., 2012; Tomasi and Volkow, 2014]. Seven of the 69 (10.14%) subjects had >10% of time points removed, and were excluded from further analysis. On average, 1.7% of data points were removed due to motion for the remaining 62 subjects. For subjects that had a second rsfMRI scan, 2 out of 13 (15.39%) were excluded; on average, 2.0% of data points were removed due to motion for the remaining 11 subjects.

Seed‐Based Functional Connectivity: Linear Correlations

We used the caudate and putamen templates from the Anatomical Automatic Labeling (AAL) atlas [Tzourio‐Mazoyer et al., 2002]. We then bisected the caudate and putamen based on boundaries from previous studies. The putamen was divided into anterior and posterior regions at the coronal slice containing the anterior commissure [i.e., y > 0 mm as the anterior putamen, and y < 0 mm as the posterior putamen; Helmich et al., 2010; Postuma and Dagher, 2006]. The caudate was separated into dorsal and ventral regions along the axial direction at the midpoint of the nucleus [i.e., z < 6 mm as the ventral caudate, z > 7 mm as the dorsal caudate; Postuma and Dagher, 2006]. To ensure that signal mixing did not impact our results, which can occur because our striatal seed regions were adjacent to one another [e.g., Choi et al., 2012], we removed an outer layer, one voxel thick, from each striatal seed region, as in our previous work [Manza et al., 2015]. This resulted in roughly a 50 to 75% reduction in volume, depending on the shape of each region. The size of each seed region before and after this procedure were: (a) ventral caudate 6,024 mm³ reduced to 2,232 mm³, (b) dorsal caudate 7,584 mm³ reduced to 3,088 mm³, (c) anterior putamen 9,144 mm³ reduced to 4,314 mm³, and (d) posterior putamen 8,592 mm³ reduced to 3,136 mm³. Hence, the seeds were significantly smaller and there was a minimum of 8 mm between seed boundaries, decreasing the likelihood of signal bleeding. Bilateral masks were merged to form a single seed for each striatal region, as in our previous work [Manza et al., 2015].

To examine the functional connectivity patterns of the striatum in a more refined manner, we also created small spherical seed regions that were equally distributed along the anterior‐posterior axis of each striatal mask. The larger seed regions from the main analysis were used for statistical power; the primary purpose of these additional smaller seeds was to describe in more detail the findings from the main analysis. This is potentially useful because of the large body of data suggesting that DA loss and Parkinson's disease‐related pathology progresses along a posterior‐anterior gradient in the striatum [Kish et al., 1988], and because multiple striatal functional circuits have been found along a gradient [Alexander et al., 1986]. A similar approach has been used in previous fMRI studies [e.g., Bell et al., 2015; di Martino et al., 2008] but we extended the investigation to span as much of each nucleus as possible. Each sphere had a 3.5 mm radius and there was a minimum distance of 7 mm between the edges of the adjacent spheres; this resulted in five total seed regions spanning the putamen, and seven total seed regions across the caudate (see Fig. 2a,c; see Table 3 for the MNI coordinates of each seed region). We did not include seed regions along the ventral bank of the putamen for simplicity, to make the anterior to posterior analysis comparable to that of the caudate. As in the main analysis, bilateral masks were used for each seed region.

Figure 2.

Regression analyses using more spatially refined striatal seed regions. Five spherical seed regions (radius = 3.5 mm) were placed in the putamen (a) and seven seed regions in the caudate (c) along the anterior‐posterior axis of each nucleus. (b) Correlation of UPDRS scores and connectivity scores between each putamen seed region and the right midbrain cluster. (d) Correlation of reversed MoCA scores and connectivity scores between each caudate seed region and the rostral anterior cingulate cortex cluster. See Table III for the coordinates of each seed region in MNI space. [Color figure can be viewed in the online issue, which is available at http://wileyonlinelibrary.com.]

Table 3.

Center coordinates for striatal regions of interest for the spatially refined analysis

Striatal nucleus	Striatal subregion	x	y	z
Putamen	P1	(±)22	16	−6
	P2	(±)24	11	2
	P3	(±)26	4	8
	P4	(±)29	−6	6
	P5	(±)30	−17	3
Caudate	C1	(±)9	10	−10
	C2	(±)10	17	−2
	C3	(±)12	15	7
	C4	(±)14	9	14
	C5	(±)15	2	19
	C6	(±)17	−7	22
	C7	(±)19	−18	22

Coordinates were defined in the MNI space. Each sphere had a 3.5 mm radius and the minimum distance between the edges of adjacent spheres was 7 mm.

The fMRI signal time courses were averaged across all voxels for each of the seed regions. We computed the correlation coefficient between the averaged time course of each seed region and the time course of each voxel in the whole brain for each individual. To assess and compare the resting state correlation maps, we converted the r values, which were not normally distributed, to z scores by Fisher's z transform [Berry and Mielke, 2000; Jenkins and Watts, 1968]: z = 0.5 log_e[(1 + r)/(1 − r)]. The z maps were used in regression analyses.

The main analysis consisted of a multiple regression of the z map of each striatal seed region. With this approach, we attempted to account for the influence of potential confounding variables on our primary measures of interest (age, motor ratings, and cognitive scores). We included seven factors into one model: (1) age, (2) the sum of UPDRS part III scores, (3) the sum of reversed MoCA scores, (4) scanner location, (5) medication status, (6) gender, and (7) years of education. In a follow‐up analysis, we replaced medication status with LED values for each subject to examine the association between striatal rsFC and medication dosage; no significant results emerged for any of the striatal seed regions and therefore these data are not presented. For the longitudinal data, we examined how differences in behavioral changes over time correlated with differences in functional connectivity, focusing on the observed rsFC patterns from the first scan that were significant in the multiple regression analyses for UPDRS scores and MoCA scores.

Previous studies have shown that the corrected voxel peak threshold of P < 0.05, based on the Gaussian random field theory, may be too restrictive, and suggested the use of a cluster threshold [Hayasaka and Nichols, 2003; Poline et al., 1997]. Thus, we present results that satisfy both a P < 0.001 uncorrected threshold at the voxel level and a P < 0.05 false discovery rate (FDR) corrected threshold at the cluster level, where the minimum cluster size = 10 voxels. All results presented satisfied this thresholding procedure.

As individuals with Parkinson's disease can present with highly variable forms of cognitive deficits [Dujardin et al., 2013; Muslimovic et al., 2005], we used additional, specific neuropsychological measures in post hoc tests to further evaluate the specificity of the rsFC pattern found associated with MoCA scores. We followed the methods of Lebedev et al. [2014] and separated the available measures from the PPMI database generally into the memory, visuospatial, and attention/executive domains. Memory function was assessed using three learning trials and the delayed recall test from the Hopkins Verbal Learning Test‐Revised [Shapiro et al., 1999], which tests the ability to recall a list of 12 words immediately and again after a delay of 20 to 25 min. Visuospatial function was assessed using the Benton judgment of line orientation test [Benton et al., 1978], in which subjects must be able to visually match pairs of angled lines. Finally, attention/executive function was assessed by a combination of the Letter‐Number Sequencing Test, Symbol Digit Modalities Test, and three semantic fluency tests. In Letter‐Number Sequencing, subjects read a jumbled list of numbers and letters and were asked to repeat the numbers in sequential order followed by the letters in alphabetical order. In the Symbol Digit Modalities test, subjects were timed and had to quickly match specific numbers with abstract symbols that were provided in a reference key. In the semantic fluency tests, subjects were asked to provide as many names as possible of animals, fruits, and vegetables, in one minute each. For each measure, raw scores were converted to z‐scores using the mean and standard deviation of the first testing of healthy control participants listed in the PPMI database (total n = 178). Final composite scores were computed by averaging the z‐scores of the different tests in each domain. As in the primary analysis of MoCA scores, these scores were reversed; therefore a higher z‐score related to poorer performance. The memory, visuospatial, and attention/executive composite scores were then correlated with the rsFC pattern identified in the multiple regression analysis that was associated with MoCA scores.

RESULTS

Figures 1 and 2 and Table 4 show the results of regression on motor deficit rating and general cognitive functioning for 62 newly diagnosed Parkinson's disease participants. Results from each set of analyses are discussed in detail below. Supporting Information Figure 1 shows the results of one‐sample t‐tests of whole‐brain functional connectivity of each seed region. Supporting Information Figure 2 shows the results of the regression with age on striatal rsFC in the healthy controls and individuals with Parkinson's disease. Supporting Information Figure 3 shows the results of age regression using the smaller, more refined seed regions.

Figure 1.

Associations between striatal functional connectivity and clinical features of Parkinson's disease. (a) UPDRS score regression on functional connectivity revealed a significant negative association between motor deficits and anterior putamen connectivity with midbrain including substantia nigra. (b) MoCA score (reversed) regression on functional connectivity revealed a significant positive association between cognitive decline and dorsal caudate connectivity with rostral anterior cingulate cortex. Scatter plots: each data point represents the functional connectivity score for one individual corresponding to the motor symptom, and cognitive decline regressions respectively in plots (a) to (b). Blue dots correspond to drug naive individuals (n = 23) at the time of the scan, red dots correspond to individuals on Parkinson's disease medications (n = 39) at the time of the scan, and the black line indicates the line of best fit for all 62 subjects. Note: two separate midbrain clusters (a left and right lateralized cluster) emerged in the multiple regression analysis in association with UPDRS scores (see Table 4 for coordinates). Because results were nearly identical between the two sides, only results from the right lateralized cluster are presented here. [Color figure can be viewed in the online issue, which is available at http://wileyonlinelibrary.com.]

Table 4.

Brain regions showing a significant relationship between seed‐based functional connectivity and motor symptomatology and cognitive function, thresholded at voxel P < 0.001 uncorrected and cluster‐level P < 0.05, FDR corrected

				MNI coordinates (mm)
Covariate of interest	Seed region	Cluster size (mm3)	z‐score	x	y	z	Identified region
UPDRS Part III	Anterior putamen	1,296	4.08	−9	−19	−5	Left midbrain
		1,647	4.11	9	−13	−8	Right midbrain
MoCA	Dorsal Caudate	2,565	3.95	9	41	−8	rACC/vmPFC

rACC = rostral anterior cingulate cortex; vmPFC = ventromedial prefrontal cortex.

Changes in Functional Connectivity with Clinical Measures of Parkinson's Disease: Multiple Regression Analyses

To examine how functional connectivity of the striatum varies in association with motor and cognitive decline during early stage Parkinson's disease, we conducted a multiple regression analysis over the rsFC of the four main seed regions while controlling for confounding factors. We also conducted additional analyses to examine how the correlation strength changed along the anterior‐posterior axis of caudate and putamen, using smaller, more refined seed regions. Because we did not have an ideal control group from the same experimental protocol to compare aging effects, we have included aging results and discussion only in the Supporting Information.

With motor deficit ratings (UPDRS Part III), functional connectivity decreased significantly between the anterior putamen and the midbrain including the substantia nigra (Fig. 1a; Table 4), and between the posterior putamen and a cluster within the right putamen. Follow‐up analysis showed that this association was significantly different across the five regions (χ ²(4) = 17.24; P < 0.005), with the strongest association observed in the two most anterior/medial seeds (putamen seeds 1 and 2; Fig. 2a,b).

With declining cognitive performance, functional connectivity increased significantly between the dorsal caudate and rostral anterior cingulate cortex (rACC) including a portion of the ventromedial prefrontal cortex (vmPFC; Fig. 1b; Table 4). Three individuals MoCA scores were more than three standard deviations from the mean and appeared to be outliers in the sample. After removing those three data points, the correlation remained significant (R ² = 0.20, P < 0.001). This association was significantly different across the seven regions (χ ²(6) = 28.07; P < 0.0001), with the strongest association in the middle three caudate seed regions (seeds 3, 4, and 5; Fig. 2c,d) that correspond to the anterior part of the dorsal caudate; the association was weaker in the more anterior ventral and posterior dorsal regions of the caudate. Follow‐up correlation analysis revealed that this pattern of cognitive‐related rsFC between dorsal caudate and rACC/vmPFC rsFC is significant for subscales including the Memory (R ² = 0.18, P < 0.001) and Visuospatial (R ² = 0.21, P < 0.001) domains, but not for the Attention/Executive domain (R ² = 0.02, P > 0.30).

In voxelwise analysis, two‐sample t‐tests for differences in functional connectivity between the medicated and drug‐naive subsamples yielded no significant results at the same statistical threshold whether age was controlled for or not (P's > 0.35). Further, all correlations that were significant for the whole group (n = 62) were also significant separately for the drug naive (n = 23) and on medication (n = 39) subsamples (Ps < 0.005). The slopes of the correlations between the drug naive and on medication subsamples were not significantly different in any of the analyses (z’s < 0.85, Ps > 0.50).

Changes in Functional Connectivity With Parkinson's Disease: Exploratory Longitudinal Analysis

In a preliminary follow‐up analysis, we examined whether changes in cognitive and motor symptomatology over time are correlated with changes in rsFC over time, using results from the motor and cognitive rsFC patterns identified in multiple regression analysis. Change in MoCA scores did not significantly correlate with change in dorsal caudate/rACC connectivity across 1 year, but there was very little variation in MoCA scores from scan 1 to scan 2. However, higher motor deficit rating at scan 2 compared with scan 1 was associated with a greater drop in the connectivity between anterior putamen and midbrain over the course of 1 year (R ² = 0.32, P = 0.06, Fig. 3). One data point was 3 SD away from the mean; this individual began taking amantadine for Parkinson's disease treatment in between the two scans. Amantadine is an NMDA receptor antagonist [Bianpied et al., 2005], and this participant was the only one in the group who was taking this class of drug for Parkinson's disease treatment (all other medicated subjects were taking L‐DOPA, DA agonists, or MAO inhibitors for one or both scans). It is possible that amantadine altered symptomatology and/or anterior putamen/dorsal midbrain connectivity in a unique fashion. Removing this potential outlier made the correlation stronger and significant (R ² = 0.49, P = 0.01). Among the 11 participants, 4 began taking Parkinson's disease medication for the first time and 4 increased their medication dosage in the 1‐year gap between the two scans, 1 remained on the same level of Parkinson's disease medication, and 1 remained drug naive for both scans.

Figure 3.

Association between striatal functional connectivity and motor symptom changes over 1 year. Among a subsample of 11 subjects, larger increases in UPDRS scores (increases in motor deficits) over the course of 1 year was associated with greater decreases in connectivity between anterior putamen and midbrain, using the same regions as the main UPDRS regression analysis depicted in Figure 1a. Of the 11 individuals, nine started Parkinson's disease medication for the first time or increased their medication dosage in the 1‐year gap between the two scans (red dots), one remained on the same level of medication (gray dot), and one remained drug naive (blue dot). [Color figure can be viewed in the online issue, which is available at http://wileyonlinelibrary.com.]

DISCUSSION

Our results demonstrate that in early‐stage Parkinson's disease, functional connectivity of the basal ganglia seems to associate with motor dysfunction and cognitive decline. Individuals with more severe motor deficits generally showed more reduction in functional connectivity between the anterior putamen and the midbrain including substantia nigra. In addition, larger declines in anterior putamen‐midbrain connectivity over the course of 1 year were associated with greater motor impairment. Lastly, cognitive decline, particularly in the memory and visuospatial domains, was associated with higher functional connectivity between the dorsal caudate and the rACC. These results suggest that functional connectivity patterns of distinct subregions of the basal ganglia have implications in the variability of Parkinson's disease phenotypes that is not influenced by medication in these newly diagnosed individuals.

Changes in Motor Symptom Severity and Anterior Putamen Connectivity

The putamen has long been implicated as a locus for motor dysfunction in Parkinson's disease [e.g., Herz et al., 2014; Kish et al., 1988; Mink, 1996], but thus far only a few reports have provided insight on the association between putamen functional connectivity and motor symptom severity in Parkinson's disease. Hacker et al. [2012] found that lower connectivity between the entire striatum and a large portion of the midbrain/brainstem (compared with the mean of the healthy control group) was associated with higher UPDRS scores in a sample of 13 individuals with Parkinson's disease. Another recent study reported that higher UPDRS scores correlated with increased functional connectivity between the entire putamen and occipital/parietal cortex [Agosta et al., 2014]. However, striatal subregions have different anatomical projections [Alexander et al., 1986] and are associated with vastly different functions [for reviews, see Haber et al., 2000; Haber, 2014]. Bell et al. [2015] used a more spatially refined analysis and found that higher UPDRS scores correlated with lower interconnectivity between putamen subregions, although this was not significant after multiple comparison correction, perhaps because of the smaller sample size (39 individuals). Here, in 62 individuals, we found that reduced connectivity between the anterior putamen and the midbrain (including substantia nigra) is significantly associated with motor symptom severity. In a preliminary analysis, we also found that in a subset of 11 individuals with longitudinal data, the change in this connectivity pattern over the course of 1 year was correlated with the change in motor symptom severity. Remarkably, this pattern was observed despite the small sample size and heterogeneity in medication status across time. Overall, our analyses suggest that the relationship between motor symptomatology and anterior putamen connectivity with the midbrain in early‐stage Parkinson's disease is robust and can be observed in both between‐ and within‐subject analyses.

This result is intriguing given that the posterior putamen has more dense anatomical connections with motor circuits than the anterior putamen [Middleton and Strick, 2000]. At first glance, this also appears to challenge the traditional view that preferential DA deprivation of the posterior putamen is responsible for the motor deficits of Parkinson's disease [Brooks et al., 1990; Kish et al., 1988]. We believe these seemingly conflicting findings can be reconciled by reports estimating that 70% of DA cells from the midbrain projecting to the posterior putamen are already lost by the time clinical symptoms are visible, even before a diagnosis is made [Damier et al., 1999; Rajput et al., 2008]. If changes in dopaminergic signaling influences this functional connectivity pattern, as previous studies suggested [Carbonell et al., 2014; Cole et al., 2013; Tian et al., 2013], then perhaps dopaminergic loss is so profound at this stage that there is a floor effect. In other words, there is perhaps insufficient between‐subject variability in midbrain‐posterior putamen signaling for observing a significant correlation with motor behavior. Another possibility is that the severe posterior putamen damage causes striatal functional reorganization to support motor behaviors. Altered spatial distribution of fMRI signal has been observed in frontal/motor cortices of akinetic Parkinson's disease patients during movements [Sabatini et al., 2000]. More direct evidence for this “spatial remapping” theory comes from a study by Helmich et al. [2010], who reported that functional connectivity patterns between posterior putamen and inferior parietal lobule observed in healthy adults are instead represented in the anterior putamen in Parkinson's disease.

Some previous studies suggest anatomical remapping is possible. Haber et al. [2000] used retrograde tracing in nonhuman primates to show that striatonigral connections are organized in an “ascending spiral,” whereby the posterior putamen connections are restricted to a small area of the ventrolateral SNc devoted to motor function. However, they also showed that anterior putamen and caudate have overlapping connections with the ventrolateral SNc in addition to other midbrain regions. This separate but overlapping organization may allow for a mechanism of functional compensation of DA loss in Parkinson's disease. Indeed, multiple reports find axon collaterals sprouting into the striatum after SN lesions and suggest that the redundancy of dopaminergic striatonigral connections allows this compensation to develop functional relevance [Arkadir et al., 2014; Finkelstein et al., 2000; Song and Haber et al., 2000]. In animal models of Parkinson's disease such as the MPTP model, DA loss in the posterior putamen leads to reduced functional segregation between striatal sub‐regions [Bergman et al., 1998; Filion and Tremblay, 1991; Rothblat and Schneider, 1995]. Further, Mounayar et al. [2007] found that after MPTP administration in monkeys, remaining DA fibers in the anterior putamen/caudate compensated for severe DA depletion in posterior putamen; this compensation was strongest in monkeys who were able to recover from motor symptoms after MPTP injections. Our data largely corroborate these findings and suggest that the individuals who are relatively spared of motor deficits are probably able to compensate for posterior putamen damage with stronger connectivity between the anterior putamen and midbrain in Parkinson's disease.

Changes in Cognitive Ability and Dorsal Caudate Connectivity

The dorsal caudate is viewed as critical for cognitive processing [Cools, 2008], as it receives strong connections with cortical regions implicated in cognition such as the dlPFC and dmPFC including the ACC [Selemon and Goldman‐Rakic, 1985; Vogt and Pandya 1987; Yeterian and Van Hoesen, 1978; Zhang et al., 2012]. Many neuroimaging studies have identified local changes to caudate structure and function that are associated with cognitive impairment in Parkinson's disease [e.g., Apostolova et al., 2010; Ekman et al., 2012; Kwak et al., 2010; Lewis et al., 2003; Melzer et al., 2012; Nagano‐Saito et al., 2005, 2014]. Here, we demonstrate that functional connectivity of the dorsal caudate (but not other striatal subregions) is particularly associated with Parkinson's disease‐related cognitive deficits, especially in the memory and visuospatial domains. Specifically, greater connectivity between the dorsal caudate and the rACC, including part of the vmPFC was associated with poorer cognitive performance, regardless of age, medication status, or education. It is possible that increases in striatal rsFC reflect compensation [Agosta et al., 2014; Helmich et al., 2010], but to date there is little evidence relating compensation to cognitive impairment in Parkinson's disease. Alternatively, it is possible that increased caudate‐rACC connectivity may stem from reduced inhibitory dopaminergic signaling in Parkinson's disease. Animal studies demonstrate that medial frontal cortex and dorsal striatum are anatomically connected [Chiba et al., 2001; Ferry et al., 2000; Öngür and Price, 2000] and that inhibitory D₂ signaling predominates in this circuit [Goto and Grace, 2005; Okubo et al., 1999; West and Grace, 2002]. Recent neuroimaging work also suggests DA plays largely an inhibitory role between these regions. For instance, compared with placebo, l‐DOPA administration decreases connectivity between the dorsal caudate and the rACC in healthy adults [Kelly et al., 2009]. When DA agonists are administered to individuals with Parkinson's disease during cognitive task performance, task‐related metabolic activity decreased specifically in the rACC/vmPFC [Nagano‐Saito et al., 2009]. Furthermore, Lebedev et al. [2014] used graph theory analysis and reported higher executive function scores in Parkinson's disease in association with both increased caudate DAT binding and decreased rsFC of the striatum with ventral PFC. These studies, along with the present findings, suggest that reduced DA functioning and increased dorsal caudate‐rACC connectivity may be crucial for cognitive outcomes in early‐stage Parkinson's disease. Further studies would be needed to directly test the effects of DA medications on cognitive performance and dorsal caudate‐rACC connectivity in a within‐subject analysis.

Diminished inhibitory dopaminergic signaling between dorsal caudate and rACC may also relate to broader, network‐level changes in Parkinson's disease. Researchers typically consider the rACC as part of the “Default Mode Network” (DMN) of brain regions [Fox et al., 2005, 2007], and the dorsal caudate as part of the “Dorsal Attention Network” [DAN; Choi et al., 2012; Robinson et al., 2009]. During cognitive tasks DAN activity increases while DMN activity decreases, and this pattern reverses in the resting state. Maintaining strong positive correlations within networks and strong negative correlations between networks seems to be critical for cognitive success [Kelly et al., 2008], especially for aging and age‐related disorders [Grady et al., 2010]. Reduced inhibitory DA signaling in Parkinson's disease may disrupt the balance of these potentially opposing networks. Accordingly, individuals with Parkinson's disease show aberrant DMN/DAN activity during cognitive task performance [Argyelan et al., 2008; Delaveau et al., 2010; Eimeren and van Monchi, 2009] and during resting state [Amboni et al., 2015; Baggio et al., 2015] compared with controls. Further, increased DMN‐DAN resting‐state connectivity differentiated individuals with and without cognitive impairment in Parkinson's disease [Baggio et al., 2015]. In sum, the current results may lend additional weight to the hypothesis that altered functional relationships between networks during resting state may be a neural correlate of cognitive impairment in neurological illness.

Limitations and Conclusion

This study had some limitations. Firstly, while the sample size was larger than many previous resting‐state fMRI investigations in Parkinson's disease, there was variability in factors such as medication status that could potentially influence the results. Of note, we used available data from UPDRS ratings in the “off” medication state, although medicated participants underwent fMRI only in the “on” medication state. Future work should directly compare clinical ratings and imaging measures collected prior to and after Parkinson's disease medication in a large cohort. Such data would provide valuable information on how therapeutic treatment alters the relationship between striatal functional connectivity and clinical features of Parkinson's disease. Nevertheless, in the present study, the main findings related to cognitive and motor deficits remained significant after controlling for medication status, and the correlations remained significant when the drug naive and on medication groups were analyzed separately. In addition, age, cognitive function, and motor symptoms of Parkinson's disease are often correlated [e.g., Chaudhuri et al., 2005; Mortimer et al., 1982]. It is noteworthy that statistically significant, separable rsFC patterns emerged for each of these factors in our multiple regression analysis. Second, the longitudinal results reported here were limited to a small sample and should be interpreted cautiously. This result must be replicated in a larger cohort. Our findings nonetheless provide a preliminary understanding on how rsFC changes over time with changes in motor deficit. Third, fMRI has a limited ability to resolve the spatial boundaries of small subcortical structures that are clustered closely together. Thus, we cautiously label the cluster associated with UPDRS ratings “midbrain including substantia nigra,” because it is possible that signals from adjacent subcortical regions such as subthalamic nucleus may be contributing to this result. Lastly, the lack of a formal control group limited our analysis to correlations within the patient population, thus future work is needed to assess between‐group differences in striatal rsFC with a well‐matched control population. Still, correlation analyses of symptom severity measures provide their own advantages that may highlight unique neurophysiological changes useful for future treatment and monitoring of outcomes in Parkinson's disease.

Individuals with Parkinson's disease display considerable heterogeneity of cognitive and motor symptoms. In this study, we demonstrated how resting state functional connectivity profiles of distinct striatal subregions may contribute to these variable phenotypes. We suggest that differences in cognitive decline and motor symptom severity are each associated with a particular connectivity pattern from a separable, distinct region of the striatum. Both highlight the potential effects of aberrant DA signaling on separate nigro‐striatal‐cortical circuits and how this relates to important clinical features of the disease.

Supporting information

Supporting Information