Remember me
Background:
Apathy is a prevalent and debilitating neuropsychiatric syndrome in Parkinson’s disease (PD). While numerous functional and structural brain studies have investigated the neural correlates of PD with apathy (PD-A), their findings have often been inconsistent. Network neuroscience suggests that such a syndrome may be best understood as disruptions of distributed brain networks.
Methods:
We conducted a systematic review to identify whole-brain studies reporting functional or structural alterations in patients with PD-A compared to those without apathy (PD-NA), or studies correlating apathy severity. Significant peak coordinates (195 foci from 24 studies) were integrated using functional connectivity network mapping (FCNM), leveraging resting-state functional magnetic resonance imaging from 1,093 healthy Human Connectome Project (HCP) participants. We quantified spatial overlap between the PD-A-associated network and canonical brain networks.
Results:
The FCNM analysis revealed that the spatially diverse brain regions previously reported in the PD-A literature converged onto a common functional connectivity network. This network predominantly involved the bilateral inferior frontal gyrus, bilateral anterior insula, bilateral dorsolateral prefrontal cortex, bilateral caudate nucleus, and bilateral thalamus. The PD-A associated network showed the highest spatial overlap with the ventral attention network (VAN; 34.05%), subcortical network (28.47%), and frontoparietal network (FPN; 24.89%). Robustness analyses confirmed these findings.
Conclusion:
Brain functional and structural correlates of apathy in PD converge on distributed networks involving the VAN, FPN, and subcortical circuits. Our network localization approach offers a unifying neurobiological framework for apathy in PD, potentially reconciling previous inconsistencies and informing the development of network-targeted interventions.
IntroductionApathy, characterized by diminished motivation, goal-directed behavior, and emotional responsivity, is a common neuropsychiatric syndrome in Parkinson’s disease (PD) (den Brok et al., 2015; Béreau et al., 2023), with reported prevalence rates varying widely from 12 to 70% across different studies (den Brok et al., 2015; Mele et al., 2019). This syndrome profoundly impairs daily functioning (Leroi et al., 2012; Liu et al., 2024) and overall quality of life (Plant et al., 2024; Benito-León et al., 2012), and also serve as a predictor of dementia (Dujardin et al., 2009). Furthermore, it poses a significant burden on caregivers, who often perceive it as one of the most distressing symptoms of the disease (Leroi et al., 2012; Leiknes et al., 2010). While traditionally linked to dysfunction within frontostriatal circuits (Hamada et al., 2021; Morris et al., 2023), particularly the mesocorticolimbic dopaminergic pathways (Prange et al., 2022), the precise pathophysiology of PD with apathy (PD-A) remains incompletely understood.
Neuroimaging studies have sought to identify structural or functional abnormalities in isolated brain regions associated with apathy, including the prefrontal cortex (Maggi et al., 2024), anterior cingulate cortex (Prange et al., 2022), basal ganglia (Wen et al., 2022), and limbic system (Luo et al., 2022). However, these findings have been markedly inconsistent, with considerable heterogeneity in both the neuroanatomical loci and the extent of brain changes. This variability is widely attributed to a combination of factors, such as methodological differences among studies, small sample sizes, and the multifaceted nature of apathy itself (Luo et al., 2022; Levy and Dubois, 2006). Consequently, this heterogeneity has limited the elucidation of the precise brain-behavior relationships underlying apathy in PD, thereby hindering the development of reliable biomarkers and targeted interventions.
Recent advances in network neuroscience have prompted a paradigm shift from localizing symptoms to single regions towards a framework that conceptualizes them as disorders of large-scale, distributed brain networks (Peng et al., 2024; Darby et al., 2019). Mounting evidence supports this view, suggesting that neuropsychiatric manifestations, such as apathy, arise not solely from damage to a single brain area, but rather reflect dysfunction within interconnected brain networks (Tian et al., 2025; Alfano et al., 2021). Network localization techniques, such as lesion network mapping (Ding et al., 2023), activation network mapping (Peng et al., 2022), and functional connectivity network mapping (FCNM) (Mo et al., 2024), integrate focal neuroimaging findings (e.g., sites of atrophy, hypometabolism, or dysfunction) with large normative connectomes to identify common brain networks underlying diverse clinical symptoms (Fox, 2018). In contrast to conventional coordinate-based meta-analytic methods (e.g., activation likelihood estimation, ALE) that primarily identify spatial convergence, FCNM aims to test whether spatially heterogeneous abnormalities can be localized to a shared distributed network (Mo et al., 2024). While such network-based approaches have successfully mapped the neural circuits of complex behaviors and neuropsychiatric phenomena across a range of neurological and psychiatric conditions (Fox, 2018; Xu et al., 2024; Schaper et al., 2023; Zhang et al., 2024; Tetreault et al., 2020; Boes et al., 2015; Yang et al., 2025; Song et al., 2025), including recent applications to the general pathophysiology of PD (Song et al., 2025) and specific non-motor symptoms like impulse control disorders (Yang et al., 2025), it remains largely unknown whether the neuroanatomical substrates of PD-A can be similarly elucidated from a network perspective.
Therefore, the primary objective of this study was to map the brain network associated with PD-A by integrating previous neuroimaging markers of brain abnormality with a large-scale normative functional connectome. Specifically, we hypothesized that diverse and spatially heterogeneous brain regions previously implicated in PD-A would converge onto a common network. By employing network localization techniques, we sought to test this hypothesis and determine whether a unified network can account for these seemingly disparate anatomical findings. Identifying such a network could advance the neurobiological model of PD-A and inform the development of targeted, network-based interventions.
MethodsLiterature search and study selectionFollowing the PRISMA guidelines (Page et al., 2021), we systematically searched PubMed, Embase, and Web of Science for studies reporting functional or structural alterations related to PD-A, up to April 10, 2025. (A detailed list of the search terms is provided in Supplementary material). We further identified relevant studies by manually screening the references of review articles and meta-analyses. The study selection process is detailed in Supplementary Figure S1.
Studies were included if they met the following criteria: (a) published in a peer-reviewed English-language journal as an original article; (b) involved whole-brain comparison [gray matter volume (GMV), task-related activation, and resting-state activity] between PD-A patients and PD-NA patients or analyzed the correlation between apathy severity and neuroimaging metrics; (c) reported results as coordinates in Talairach or Montreal Neurological Institute (MNI) space. For longitudinal studies, only baseline data were included.
Exclusion criteria were as follows: (a) non-original studies (e.g., review, meta-analysis, meeting abstract); (b) no coordinate system reported; (c) studies employing region-of-interest analyses; (d) studies involving animal experiments. To avoid data duplication from overlapping patient samples across different publications, only the study reporting the largest sample size and most comprehensive information was retained for analysis. The corresponding author of each included study was contacted via email when additional information was required. Two investigators (H. C. Y. and S. Y. G.) independently performed literature search and selection and data extraction. Any discrepancies were discussed with another investigator (P. L. P.) until they were resolved.
Normative resting-state functional magnetic resonance imaging data acquisition and preprocessingWe used a normative functional connectome derived from 1,093 healthy adults [594 female; mean (SD) age 28.78 (3.69) years, range 22–37] from the HCP 1200 Subjects Data Release (HCP-S1200) (Van Essen et al., 2012) following standard quality control. Individuals in the cohort were between the ages of 22 and 37. Participants in the HCP were free of MRI contraindications, current psychiatric or neurological disorders, recent use of psychiatric medication, pregnancy, and a history of head trauma. The specific demographics of the analyzed sample can be found in Supplementary Table S1. All coordinates from prior studies were converted to MNI space using established tools1 if originally reported in Talairach space.
In this study, we utilized resting-state fMRI data from the HCP (Van Essen et al., 2012), obtained using a 3 T Siemens Skyra system. This dataset was considered appropriate for comprehensive analysis due to its high quality and large sample size. The specific acquisition parameters can be found in Supplementary Table S2. Individuals were excluded from the study if their imaging data did not pass quality assessments, such as the detection of artifacts or incomplete brain coverage.
The resting-state fMRI data underwent preprocessing with SPM12 software2 and the Data Processing and Analysis for Brain Imaging (DPABI) toolbox.3 The initial 10 volumes of each participant’s scan were discarded to facilitate signal equilibration and adaptation to scanner noise. Slice acquisition timing disparities in the remaining volumes were corrected, and subsequently, motion correction was performed. Head motion parameters for each volume were determined by calculating translational displacements in each direction and angular rotations around each axis. All participant data met the specified motion thresholds, with maximal translational or rotational motion parameters below 2 mm or 2°. Furthermore, framewise displacement was computed as a measure of head motion between volumes. Regression analysis was conducted to account for nuisance effects, incorporating covariates such as linear drift, estimated Friston-24 motion parameters, timepoints with excessive motion (framewise displacement > 0.5 mm), and signals from global mean, white matter, and cerebrospinal fluid. In line with previous network mapping studies that seek to improve system-specific correlations, global signal regression was incorporated into the preprocessing pipeline. The datasets were bandpass filtered to preserve frequencies ranging from 0.01 to 0.1 Hz. Initially, spatial normalization began with co-registering each participant’s T1-weighted structural image with their average functional image. These structural images were then segmented and normalized to MNI space. Each filtered functional volume was then normalized to MNI space. Subsequently, all functional data were spatially smoothed using a Gaussian kernel (FWHM = 6 × 6 × 6 mm3).
FCNM analysisWe utilized the FCNM approach to determine if the reported brain regional alterations in PD-A map onto a common brain network. For each identified contrast in the literature search, spheres with a 4-mm radius were centered at the reported coordinates of significant difference. These spheres were then merged to form a contrast-specific seed mask, hereafter termed the “contrast seed.” Subsequently, functional connectivity (FC) maps were generated for each participant by deriving the FC from each contrast seed to the entire brain based on preprocessed resting-sate fMRI data from 1,093 healthy HCP participants. The procedure comprised calculating Pearson’s r between the average time course of the seed for the contrast and the time course of all remaining brain voxels, subsequently applying a Fisher Z-transform to normalize the resultant correlation coefficients. Third, the participant-level FC maps (N = 1,093) were subjected to a voxel-wise one-sample t-test to identify brain regions consistently exhibiting positive FC with each contrast seed across the normative samples. Consistent with previous FCNM studies, we focused solely on positive FC, as the biological interpretation of negative FC remains under debate (Murphy et al., 2009; Murphy and Fox, 2017). After thresholding at p < 0.05, the group-level t-maps were corrected for multiple comparisons using a voxel-level false discovery rate (FDR) method and subsequently binarized. Finally, the binarized connectivity maps corresponding to all included contrasts were overlaid to form a network probability map. This map was then thresholded at 60% overlap (i.e., regions connected to at least 60% of the contrast seeds), a threshold validated in previous FCNM studies (Peng et al., 2022; Mo et al., 2024), to delineate the PD-A associated brain networks.
Association with canonical brain networksTo enhance interpretability, we analyzed the spatial overlap between the identified PD-A-associated network and eight established canonical brain networks. The visual network, somatomotor network (SMN), dorsal attention network (DAN), ventral attention network (VAN), limbic network, frontoparietal network (FPN), and default mode network (DMN) are the seven cortical networks defined by Yeo et al. The Human Brainnetome Atlas was utilized to delineate the subcortical network. The subcortical network mainly includes the amygdala, hippocampus, basal ganglia (including the caudate nucleus, putamen, globus pallidus, nucleus accumbens), and thalamus. We quantified these spatial relationships by calculating the proportion of overlapping voxels between each PD-A-associated network and each canonical network, relative to the total number of voxels within the respective canonical network.
ResultsIncluded studies and sample characteristicsFollowing the systematic literature search and screening process detailed in Supplementary Figure S1, a total of 24 studies (providing 195 foci) were included, comprising 1,206 patients with PD, of whom 606 were diagnosed with apathy (PD-A) and 600 were without apathy (PD-NA) (Remy et al., 2005; Reijnders et al., 2010; Alzahrani et al., 2016; Le Jeune et al., 2009; Thobois et al., 2010; Terada et al., 2018; Prange et al., 2019; Robert et al., 2012; Huang et al., 2013; Skidmore et al., 2013; Robert et al., 2014; Robert et al., 2014). Regarding the analytical approaches, 10 studies performed group comparisons, 10 studies conducted correlation analyses between apathy scores and neuroimaging metrics, and 4 studies utilized both methodologies. Imaging modalities included PET (n = 9) and MRI (n = 15), with 16 studies utilizing the MNI template and 8 studies utilizing the Talairach template for spatial normalization. Among the studies, 7 utilized structural MRI, focusing on GMV or GM density alterations; 8 employed resting-state fMRI to investigate changes in spontaneous brain activity using measures such as amplitude of low-frequency fluctuations, regional homogeneity, voxel-mirrored homotopic connectivity, and FC; and one study implemented task-based fMRI. Detailed sample and imaging characteristics of the included studies are summarized in Table 1.
StudySample size (PD-A/-NA)Mean age (PD-A)Apathy assessmentAnalysisComorbidity controlMedication/DBS statusThresholdScannerTemplateMeasureFociRemy et al. (2005)20/059.8 (7.2)AESCorrelationExcluded for dementia; Apathy/anxiety not controlled forMedication-OFF; No DBSP < 0.05, uncorrectedPETTalairachBinding Potential2Reijnders et al. (2010)55/062.0 (10.1)AES/LARS/NPI-ACorrelationExcluded for major depression/dementiaMedication status not specified; No DBSP < 0.05, FDR corrected3 T MRIMNIGMD28Alzahrani et al. (2016)25/4068.7 (8.4)NPI-AComparison (PD-A-PD-NA)Excluded for dementia and other neuropsychiatric comorbiditiesMedication status not specified; No DBSp < 0.001, corrected1.5 T MRITalairachGMV9Le Jeune et al. (2009)12/057.4 (8.00)AESCorrelationExcluded for dementia and major psychiatric disordersMedication-ON; Post-DBSp < 0.005, correctedPETTalairachGlucose metabolism8Thobois et al. (2010)12/1355.8 (4.9)SASComparison (PD-A-PD-NA)Excluded for baseline apathy and mod-severe depressionPost-DBS; Drug withdrawalP ≤ 0.001, uncorrectedPETMNID2/D3 receptor availability11Terada et al. (2018)40/064.7 (8.0)FrSBECorrelationExcluded for dementia & depressionMedication-OFF; No DBSP < 0.05, FWE corrected1.5 T MRIMNIGMV1Prange et al. (2019)14/1362.5 (10.2)LARS/SASComparisonExcluded for dementia; depression as covariateDrug-naïveP < 0.05, FWE corrected1.5 T MRIMNIGMV3Robert et al. (2012)45/060.5 (7.8)AESCorrelationExcluded for depression and dementiaMedication-ON; No DBSP < 0.005, uncorrectedPETMNIGlucose metabolism6Huang et al. (2013)26/066.5 (1.4)AESCorrelationExcluded for dementia; depression or anxiety not excludedMedication-OFF; No DBSP < 0.001, uncorrectedPETMNIGlucose metabolism6Skidmore et al. (2013)15/063 (9)LARSCorrelationExcluded for dementia; depression as covariatesMedication-OFF; No DBSP < 0.005, uncorrected3 T MRIMNIALFF8Robert et al. (2014)36/058.6 (7.3)AESCorrelationExcluded for dementia & major depression; subclinical depression controlledMedication-ON; No DBSP < 0.005, uncorrectedPETTalairachGlucose metabolism8Maggi et al. (2024)48/28760.98 (9.81)UPDRS-IComparison (PD-A-PD-NA)Not controlled for (depression and anxiety were measured as outcome variables)Drug- naïve at baselineP < 0.05, FWE corrected3 T MRIMNIGMD8Robert et al. (2014)44/056.3 (7.5)AESCorrelationExcluded for depression and dementiaMedication-ON; Pre-DBSP < 0.005, uncorrectedPETTalairachGlucose metabolism1Auffret et al. (2017)12/065.9 (7.2)LARSCorrelationExcluded for dementiaMedication-ON; No DBSP < 0.001, uncorrectedPETTalairachGlucose metabolism18Shin et al. (2017)10/1273.8 (4.3)ASComparison (PD-A-PD-NA) correlationExcluded for dementia & depressionMedication-OFF; No DBSP < 0.05, corrected3 T MRI PETMNIGlucose metabolismSample and imaging characteristics of the studies included.
AS, Apathy Scale; AES, apathy evaluation scale; LARS, Lille apathy rating scale; PD-A, Parkinson Disease with apathy; FrSBE, frontal systems behavior scale; NPI-A, neuropsychiatric inventory-apathy subscale; SAS, Starkstein apathy scale; DBS, Deep brain stimulation; UPDRS-I: Unified Parkinson’s Disease Rating Scale-I; PD-NA, Parkinson Disease without apathy; ALFF, amplitude of low frequency fluctuations; FC, functional connectivity; GMV, gray matter volume; GMD, gray matter density; VMHC, voxel-mirrored homotopic connectivity; ReHo, regional homogeneity; PET, positron emission tomography; MNI, Montreal Neurological Institute; MRI, Magnetic Resonance Imaging; fMRI, functional magnetic resonance; FDR, False discovery rate; FWE, Family wise error.
Convergent FC correlates of PD-AApplication of the FCNM approach, integrating coordinates of reported functional and structural correlates of PD-A from the 24 studies with the normative HCP connectome, revealed convergent FC associated with PD-A. These FC patterns involved widely distributed brain regions, primarily including the bilateral inferior frontal gyrus, bilateral anterior insula, bilateral dorsolateral prefrontal cortex (DLPFC), bilateral caudate nucleus, and bilateral thalamus (p < 0.05, FDR corrected) (Figure 1).
PD-A-associated FC overlap maps based on 4 mm radius sphere. Dysfunctional brain networks are shown as FC probability maps thresholded at 60%, showing brain regions functionally connected to more than 60% of the contrast seeds. PD-A, Parkinson’s disease with apathy; FC, functional connectivity; L, left; R, right.
Association with canonical brain networksAnalysis of the spatial overlap between the PD-A-related aberrant FC and established canonical brain networks indicated preferential involvement of specific systems. The network showed the highest overlap with the VAN, largely corresponding to the b bilateral inferior frontal gyrus and bilateral anterior insula (overlap proportion: 34.05%). Significant overlap was also observed with the FPN, primarily involving bilateral DLPFC (overlap proportion: 24.89%). Additionally, significant overlap was also observed with the subcortical network, primarily involving bilateral caudate nucleus and bilateral thalamus (overlap proportion: 28.47%) (Figure 2). Overlap proportions with the remaining canonical networks (visual, SMN, DAN, limbic network, and DMN) were all below 10%.
PD-A-associated FC overlap maps based on 4-mm radius sphere in association with canonical brain networks. Polar plots illustrate the proportion of overlapping voxels between each PD-A FC map and a canonical network to all voxels within the corresponding canonical network. Note: The blue dot represents brain dysfunction networks, defined as significant networks, exhibiting ≥10% overlap with canonical networks, whereas the red dot represents non-significant networks with <10% overlap. PD-A, Parkinson’s disease with apathy; FC, functional connectivity.
Robustness analysesFCNM analyses repeated using seed spheres with 1-mm and 7-mm radii yielded topographically similar network patterns to those obtained using the standard 4-mm radius sphere (Supplementary Figures S2, S3, respectively). Furthermore, the pattern of canonical network involvement remained consistent when replicating the FCNM procedure with spheres of 1-mm and 7-mm radii (Supplementary Figures S4, S5, respectively).
DiscussionTo our knowledge, this study is the first to integrate FCNM with large-scale human brain connectome data from the HCP to demonstrate that heterogeneous functional and structural correlates of apathy in PD map onto common brain networks. Our analysis revealed that aberrant brain regions previously identified in the literature indeed map onto a specific network encompassing key nodes within the VAN (notably the bilateral inferior frontal gyrus and bilateral anterior insula), FPN (bilateral DLPFC), and subcortical network (bilateral caudate nucleus and bilateral thalamus). Importantly, these findings demonstrated high robustness, with the primary network localization (derived from 4-mm radius seed spheres) being consistently replicated in validation analyses using both 1-mm and 7-mm radii.
A major finding of our study was the significant mapping of PD-A associated functional and structural alterations onto the subcortical network. The subcortical network, primarily comprising the basal ganglia, thalamus, amygdala, and select cerebellar nuclei (
Comments (0)