Individualized cortico-basal ganglia-thalamo-cortical circuit dysfunction links striatal dopaminergic loss to motor symptom severity in Parkinson’s disease

Abstract

Motor impairment in Parkinson’s disease (PD) is classically attributed to striatal dopaminergic degeneration, yet dopamine loss alone does not fully explain symptom severity, suggesting a key role for circuit-level mechanisms. The cortico–basal ganglia–thalamo–cortical (CBGTC) system is central to dopaminergic motor modulation, but whether dysfunction within individualized CBGTC circuits mediates motor severity across disease stages remains unclear. Here, we studied 76 PD patients (40 mild, 36 moderate-to-severe) who underwent ¹⁸F-FP-CIT PET and multimodal MRI. Individualized long and short CBGTC loops were reconstructed using connectivity profile–based segmentation and probabilistic tractography, with functional connectivity (FC) quantified alongside striatal dopaminergic integrity. Compared with mild PD, moderate-to-severe PD showed a global reduction in striatal dopamine binding, most pronounced in the bilateral caudate. Critically, FC between the caudate and premotor cortex in the more-affected hemisphere was selectively reduced with increasing disease severity in both CBGTC loops (p-FDR = 0.027). The cross-sectional mediation models demonstrated that caudate–premotor FC statistically accounted for the association between caudate dopaminergic loss and motor symptom severity. These findings position individualized cortico–basal ganglia circuit connectivity as a potential mechanistically grounded biomarker linking molecular pathology to motor impairment in PD, with potential relevance for future circuit-guided therapeutic strategies.

Introduction

Parkinson’s disease (PD) is a common neurodegenerative disorder characterized by progressive loss of substantia nigra dopamine neurons and consequent striatal dopamine depletion^1–3, which underlies its hallmark motor symptoms. PD’s prevalence is rising — projected to exceed 10 million cases by 2030 — leading to substantial disability, reduced quality of life, and a growing socioeconomic burden^4,5. Striatal dopamine depletion is a key pathological driver of PD onset and progression^6–8. However, the severity and progression of motor impairment cannot be fully explained by dopamine loss alone^2,9. Patients with comparable levels of dopaminergic depletion frequently exhibit marked heterogeneity in motor symptoms and clinical trajectories, indicating that intermediate neural mechanisms critically shape how molecular pathology translates into behavioral impairment.

One major neural network implicated in PD motor dysfunction is the cortico–basal ganglia–thalamo–cortical (CBGTC) circuit, a set of parallel loops connecting cortex, striatum, pallidum, thalamus, and back to cortex^10,11. The CBGTC circuit is the substrate for motor control¹⁰, integrating cortical plans with basal ganglia processing and thalamocortical relay for movement initiation¹¹. In health, balanced activity in the direct and indirect CBGTC pathways sustains coordinated movement. In PD, dopaminergic neuron loss upsets this balance, leading to excessive inhibition of the thalamus and reduced cortical activation, manifesting as bradykinesia¹². While this framework has long been recognized, it remains unclear how dopaminergic loss disrupts specific CBGTC connections in vivo, and whether such disruption directly contributes to differences in motor symptom severity across disease stages.

Accumulating neuroimaging work links dopaminergic pathology to circuit-level dysfunction. Nigrostriatal denervation correlates with reduced functional connectivity (FC) within the cortico-subcortical sensorimotor circuit in PD¹³. Likewise, striatal regions with dopamine depletion exhibit decreased FC with the medial premotor cortex during motor tasks, while also showing broadly impaired connectivity with extensive motor areas at rest¹⁴. Microstructure damage to the nigrostriatal pathway is also closely associated with PD symptom development^15,16. Critically, putaminal dopamine levels on the more-affected side have been shown to mediate the relationship between microstructural nigrostriatal integrety and motor symptoms¹⁵. Dopamine loss is tied to connectivity deficits and even mediates structural-functional-behavior relationships. Notably, these circuit abnormalities are partly reversible, and they attenuate with dopaminergic medications or neuromodulation in PD, underscoring the functional and dynamic nature of the circuit pathology^17,18.

Despite these advances, three specific gaps limit mechanistic inference in prior research: (i) atlas dependence—most studies define subcortical regions using group atlases, which ignore individual anatomical variability. This can blur the distinct contributions of the “long” (cortico-striato-pallido-thalamo-cortical) vs. “short” (cortico-striato-thalamo-cortical) loops, misallocate signals to the wrong functional subdivisions, and dilute connection-specific effects that might be clinically relevant. (ii) seldom integrating dopamine transporter PET and fMRI in the same patients—dopamine transporter (DAT) imaging (e.g., ¹⁸F-FP-CIT PET) and functional magnetic resonance imaging (fMRI) are rarely combined in the same cohort, so molecular pathology and circuit function are usually studied separately. And (iii) limited to bivariate correlation analyses—prior studies largely report correlations between dopamine loss and connectivity or between connectivity and symptoms, without statistically examining an indirect-effect (mediation) model in which circuit connectivity statistically accounts for the association between dopaminergic loss and symptom severity^13,14,19. Therefore, investigating the interactions among dopaminergic depletion, CBGTC circuit reorganization, and motor dysfunction across different stages of PD is essential for deepening our understanding of the neural mechanisms of PD and identifying potential biomarkers of disease progression.

Recent advancement of neuroimaging techniques has provided crucial support for in vivo investigations of central neural mechanisms²⁰. Multimodal magnetic resonance imaging enables non-invasive characterization of structural and FC in neural circuits, offering an effective tool for studying abnormalities in the CBGTC circuits^21,22. Most prior studies parcellate the CBGTC system using group-level atlases, which average across individuals and obscure the continuous functional topographies of the striatum, pallidum, and thalamus—structures in which subregions project to partially overlapping cortical networks and support distinct computations^11,23,24. Conventional functional connectivity analyses often ignore individual anatomy and structural connectivity constraints, risking spurious results²⁵. Advanced diffusion MRI techniques now allow subject-specific delineation of subcortical topography and its cortical projections²⁶. By using probabilistic tractography and connectivity profile-based segmentation, one can map each person’s long and short CBGTC loops. When resting-state fMRI is computed under these individualized structural constraints, FC estimates are anatomically grounded and connection-specific, enabling individual-specific circuit analyses rather than atlas-level averages.

Concurrently, hybrid PET–MR imaging now permits integrating molecular and systems data. In PD, the striatal specific-to-nonspecific binding ratio (SNBR) from ¹⁸F-FP-CIT PET provides a quantification of dopaminergic terminal loss^27,28. Acquiring PET and MRI in the same session (with identical geometry and motion tracking) improves registration fidelity, allowing us to directly relate SNBR to circuit-specific FC measures. Moreover, statistical mediation approaches can be used to explore whether circuit changes bridge molecular pathology and clinical outcomes. For instance, a recent study showed that substantia nigra neuromelanin loss (hypopigmentation on MRI) had an effect on motor symptoms that was partly mediated by connectivity between the basal ganglia and dorsal motor network²⁹. However, such analyses, especially those integrating DAT-PET with individualized circuit mapping, are still uncommon. These converging tools now make it feasible to move beyond atlas-level correlations toward mechanistic, individual-specific models of PD circuitry.

In this study, we combined ¹⁸F-FP-CIT PET and multimodal MRI in the same PD patients to construct individualized CBGTC circuits. Using probabilistic tractography and connectivity-based segmentation, we delineated each patient’s “long” and “short” cortico–striato–thalamo–cortical loops. Within these personalized circuits, we quantified resting-state functional connectivity and related it to striatal SNBR (a PET index of dopamine terminals) as well as clinical motor scores. We hypothesized that striatal dopaminergic depletion would be associated with selective alterations in functional connectivity within individualized CBGTC circuits and that these circuit measures would be related to motor symptom burden (mild vs. moderate-to-severe). The aim was to provide a mechanistically informed, circuit-level account of how dopaminergic degeneration relates to motor impairment in PD and to establish an individualized imaging framework that may support future circuit-guided therapeutic strategies.

Results

Demographics and clinical characteristics

Seventy-six participants were analyzed and stratified by disease severity into a mild PD group (n = 40) and a moderate-to-severe PD group (n = 36) according to Hoehn and Yahr (H-Y) staging. Compared with the mild PD group, patients in the moderate-to-severe PD group were significantly older (65.58 [9.60] vs. 60.70 [7.41] years; p = 0.015) and had a longer disease duration (4.61 [3.49] vs. 2.39 [2.02] years; p = 0.001). No significant difference was observed in sex distribution between groups or in the distribution of the more affected body side. The Unified Parkinson’s Disease Rating Scale (UPDRS) Part III scores were significantly higher in the moderate-to-severe PD group than in the mild PD group (44.06 [11.68] vs. 24.23 [10.49]; p < 0.001), confirming successful clinical stratification by disease severity. Additionally, no significant differences were observed between the two groups in Non-Motor Symptoms Scale (NMSS), 39-item Parkinson’s Disease Questionnaire (PDQ-39), Hamilton Depression Rating Scale (HAMD), or Mini-Mental State Examination (MMSE) scores (all p > 0.05). Detailed demographic and clinical characteristics are summarized in Table 1.

Table 1.

Demographic and clinical characteristics

Variables	Mild PD (n = 40)	Moderate-to-severe PD (n = 36)	t/χ²	p value
Age, years	60.70 (7.41)	65.58 (9.60)	-2.496	0.015
Sex (female %)	20 (50.00)	15 (41.67)	0.530	0.331
Disease duration, years	2.39 (2.02)	4.61(3.49)	-3.429	0.001
More affected body side (right %)	9 (22.50)	10 (27.78)	1.847	0.397
LEDD, mg/d	524.84 (143.87)	570.14 (196.96)	-1.153	0.253
UPDRS-Ⅲ	24.23 (10.49)	44.06 (11.68)	-7.797	<0.001
NMSS	39.48 (31.90)	40.75 (27.38)	-0.186	0.853
PDQ-39	33.05 (18.67)	40.47 (25.04)	-1.474	0.145
MMSE	26.00 (3.43)	24.58 (3.98)	1.667	0.100
HAMD	4.22 (2.12)	4.44 (4.09)	-0.298	0.766

Data are mean (SD), unless otherwise indicated.

LEDD levodopa equivalent daily dose, NMSS Non-Motor Symptoms Scale, PD Parkinson’s disease, PDQ-39 39-Item Parkinson’ s Disease Questionnaire, UPDRS Unified Parkinson’s Disease Rating Scale.

Group differences in FC within individual-specific CBGTC circuit

Comparison of head motion parameters revealed no significant difference in mean framewise displacement (FD) between the mild and moderate-to-severe PD groups (p > 0.05). After connectivity-based segmentation, between-group differences in connection-wise FC were tested within the individualized long and short CBGTC loops. Analyses were performed separately for the clinically defined more-affected and less-affected hemispheres. In the more-affected hemisphere, patients with moderate-to-severe PD exhibited significantly reduced FC within individualized CBGTC circuits compared with patients with mild PD. Specifically, in both the long and short loops, FC between the caudate subregion preferentially connected to the premotor cortex/supplementary motor area (Caudate_PreM/SMA) and the PreM/SMA cortex was significantly decreased in the moderate-to-severe PD group (long loop: p-FDR = 0.027; short loop: p-FDR = 0.027). In addition, within the short loop, the FC between the thalamic subregion connected to dorsolateral prefrontal cortex (Thalamus_DLPFC) and DLPFC was also significantly reduced in the moderate-to-severe PD group (p-FDR = 0.027) (Table 2). Sensitivity analyses using alternative thresholds (0.5% and 2%) further validated the robustness of the above findings: the between-group difference in FC between Caudate_PreM/SMA and PreM/SMA within the short loop remained consistently significant across all three thresholds, while the remaining significant between-group differences showed directionally consistent trends across thresholds, with no additional significant results emerging (Figs. S1 and S2). Besides, no significant between-group differences in FC were observed within the individualized CBGTC circuits of the less-affected hemisphere after correction for multiple comparisons.

Table 2.

Group differences in functional connectivity in CBGTC circuits based on probabilistic tracking and voxel connectivity profiles-based segmentation

Hemisphere		ROI1	Connected with		ROI2	p-FDR value
Moderate to severe PD group < Mild PD group
More affected	“Short” loop	CaudatePreM/SMA			PreM/SMA	0.027
More affected	“Short” loop	ThalamusDLPFC			DLPFC	0.027
More affected	“Long” loop	CaudatePreM/SMA			PreM/SMA	0.027

Subcortical area with subscripts denote specific subcortical subregions that are connected to all brain areas listed in the subscript.

CBGTC cortico - basal ganglia - thalamo - cortical, DLPFC dorsolateral prefrontal cortex, FDR false discovery rate, PD Parkinson’s disease, PreM/SMA premotor/supplementary motor area, ROI regions of interest.

Group comparison of striatal SNBR

After controlling for age, sex, disease duration, and levodopa equivalent daily dose (LEDD), it was found that in the less-affected hemisphere, SNBR values were significantly reduced in the moderate-to-severe PD group in the whole striatum (p = 0.002), caudate (p = 0.001), and putamen (p = 0.039). Similar reductions were observed in the more-affected hemisphere for the whole striatum (p = 0.022), caudate (p = 0.004), but not in putamen (p = 0.210). All reported results remained significant after FDR correction (Table 3, Fig. 1).

Table 3.

Comparison of striatal SNBRs between the mild PD group and the moderate to severe PD group

Hemisphere	Brain Region	SNBR, Mean(SD)		p value
		Mild PD	Moderate to severe PD
Less-affected	Whole striatum	2.36 (0.76)	1.74 (0.67)	0.002
	Caudate	2.86 (0.82)	2.02 (1.01)	0.001
	Putamen	1.90 (0.96)	1.48 (0.57)	0.039
More-affected	Whole striatum	2.23 (0.97)	1.70 (0.73)	0.022
	Caudate	2.82 (1.14)	1.97 (0.93)	0.004
	Putamen	1.68 (0.91)	1.46 (0.63)	0.210

PD Parkinson’ s disease, SNBR specific to nonspecific binding ratio.

Fig. 1. Comparison of striatal SNBR between the mild PD group and moderate to severe PD group in the more-affected and less-affected hemispheres.

**p < 0.05, FDR corrected. The central line represents the median; box boundaries indicate the interquartile range (IQR); whiskers extend to values within 1.5 × IQR. Individual data points from all participants are overlaid. Statistical analyses were adjusted for age, sex, disease duration, and levodopa equivalent daily dose (LEDD). PD Parkinson’s disease, SNBR specific to nonspecific binding ratio.

Correlation analysis results

Figure 2 summarises the results of partial correlation analyses examining the relationships among circuit-specific FC showing between-group differences in CBGTC circuits, striatal SNBR values, and clinical scale scores. Within individualized CBGTC loops of the more-affected hemisphere, higher UPDRS-Ⅲ score was significantly associated with lower FC between Caudate_PreM/SMA and the PreM/SMA cortex in both the long and short loops (long loop: r = -0.34, p = 0.004; short loop: r = -0.29, p = 0.016). In addition, FC between Caudate_PreM/SMA and the PreM/SMA cortex in both the long and short loop of the more-affected hemisphere was positively correlated with the SNBR in the bilateral striatum, ipsilateral caudate, and contralateral putamen. Moreover, no significant correlations were observed between the differential neuroimaging metrics and NMSS, HAMD, MMSE, or PDQ-39 scores (all p > 0.05). All correlation analyses were adjusted for age, sex, disease duration, and LEDD (and mean FD for FC where applicable). After additionally controlling for H-Y scale, the key association patterns remained directionally consistent, with the main effects still statistically significant (Fig. S3).

Fig. 2. Results of partial correlation analysis among intra-circuit differential FC of the CBGTC circuits, striatal SNBR, and clinical scale scores.

Color indicates the partial correlation coefficient (r): deeper red signifies stronger positive correlation, deeper blue signifies stronger negative correlation. *p < 0.05, with FDR correction. CAU Caudate, CBGTC cortico - basal ganglia - thalamo - cortical, DLPFC dorsolateral prefrontal cortex, FC functional connectivity, HAMD Hamilton Depression Rating Scale, LAS less affected side, LL long loop, MAS more affected side, MMSE Mini-Mental State Examination, NMSS Non-Motor Symptoms Scale, PDQ-39 39-Item Parkinson’s Disease Questionnaire, PreM premotor cortex, PUT Putamen, SL Short loop, SMA supplementary motor area, SNBR specific to nonspecific binding ratio, STR Striatum, UPDRS Unified Parkinson’s Disease Rating Scale.

Mediation model analysis

Mediation model analyses were conducted to examine whether circuit-specific FC within individualized CBGTC loops statistically accounted for the association between striatal dopaminergic integrity and motor symptom severity in the more-affected hemisphere. Results demonstrated that FC between Caudate_PreM/SMA and the PreM/SMA cortex served as a significant indirect-effect component in the relationship between caudate SNBR and UPDRS-III scores in both the short and long CBGTC loops. For the short loop, caudate SNBR was positively associated with Caudate_PreM/SMA − PreM/SMA FC (path a = 0.091, p < 0.001), and this FC was negatively associated with UPDRS-III scores (path b = -16.980, p = 0.038). The estimated indirect effect was significant (-1.542; 95% confidence interval [CI]: -3.311 to -0.156; Fig. 3A). Similarly, caudate SNBR showed a significant positive association with Caudate_PreM/SMA − PreM/SMA FC in the long loop (path a = 0.075, p < 0.001), and this FC was negatively associated with UPDRS-III scores (path b = -21.796, p = 0.008). The indirect effect was significant (-1.637; 95% CI: -3.515 to -0.349; Fig. 3B). In both mediation models, the direct association between caudate SNBR and UPDRS-III scores was not statistically significant after inclusion of the FC term (p > 0.05), consistent with the indirect-effect pattern observed in these cross-sectional mediation models. Moreover, after controlling for H-Y stage, the above results remained significant (Figure S4).

Fig. 3. Mediation model analysis of caudate SNBR, caudate–PreM/SMA functional connectivity, and UPDRS-III motor severity within individualized CBGTC loops.

In both models, caudate SNBR was entered as the predictor and UPDRS-III score as the outcome, with functional connectivity (FC) between the caudate subregion preferentially connected to the PreM/SMA cortex (Caudate_PreM/SMA) and the PreM/SMA cortex as the statitical mediator (indirect-effect term). Path coefficients a, b, and c’ represent standardized regression coefficients from the mediation models. Indirect effects (β) and 95% confidence intervals (CIs) were estimated using nonparametric bootstrap resampling, adjusting for age, sex, disease duration, levodopa equivalent daily dose (LEDD), and mean framewise displacement. Panel (A) showed results for the short individualized CBGTC loop, and panel (B) for the long individualized CBGTC loop, both in the more affected hemisphere. *p <0.05；**p <0.01. CAU Caudate, FC functional connectivity, LL long loop, MAS more affected side, PreM premotor cortex, SL Short loop, SMA supplementary motor area, SNBR specific to nonspecific binding ratio, UPDRS Unified Parkinson’s Disease Rating Scale.

Discussion

To our knowledge, this is the first study to investigate, within the same patients and hemisphere, how striatal dopaminergic terminal loss (SNBR on ¹⁸F-FP-CIT PET) relates to connection-level abnormalities in FC in individualised long/short CBGTC loops, and how these circuit measures relate to motor symptom severity indexed by UPDRS-III in Parkinson’s disease. Methodologically, we constrained FC estimation by individual-specific white-matter anatomy and used connectivity-profile segmentation to delineate individualized subcortical topographies, enabling connection-level tests rather than atlas-averaged associations. We observed lower striatal SNBR bilaterally with disease severity and selective weakening of caudate–PreM/SMA connection, alongside a short-loop thalamus–DLPFC reduction, in the more-affected hemisphere. It is worth noting that cross-sectional mediation models revealed a significant indirect-effect pattern in which caudate–PreM/SMA connection statistically accounted for the association between caudate SNBR and UPDRS-III. While this framework does not establish temporal causality, the observed pattern is mechanistically consistent with an intermediary circuit-level account linking dopaminergic degeneration to motor impairment.

The patterns of striatal dopamine depletion and the weakened FC within the CBGTC circuits observed in this study are highly consistent with previous literature. First, we found that the degree of striatal dopaminergic loss progressively increased with the advancement of PD, with the most pronounced depletion occurring in the caudate. This aligns with the characteristic spatiotemporal progression of PD pathology, which is initially characterized by dopaminergic depletion in the putamen, with subsequent involvement of the caudate nucleus as the disease advances^30,31. Furthermore, previous studies have reported significantly reduced cortico-subcortical connectivity in PD patients³². As the disease progresses, abnormalities emerge in broader brain regions, particularly within the sensorimotor cortex and basal ganglia^33,34. Consistent with these findings, our study also demonstrated significant attenuation of specific FCs within the CBGTC circuits in moderate-to-severe PD patients, including reduced connectivity between caudate and PreM/SMA, as well as between thalamus and DLPFC. Among these, the FCs between caudate and PreM/SMA showed significant correlations with UPDRS-III scores but not with HAMD, MMSE, or PDQ-39 scores, suggesting that this connectivity alteration may be primarily linked to motor symptoms, although its relationship with non-motor features warrants further confirmation with more comprehensive instruments. It is noted that our findings highlight the pivotal role of the caudate in PD progression. While motor symptoms in PD have been classically attributed to dopaminergic denervation in the posterior putamen³⁵, our results suggest that the caudate may play a more critical role in moderate to advanced disease stages. From a functional neuroanatomical perspective, the caudate nucleus is implicated not only in higher-order motor control, such as motor preparation, sequence learning, and action selection, but also in non-motor functions including cognition and motivation^36,37. As dopaminergic depletion in the caudate worsens, its FCs with cortical regions, including PreM/SMA, becomes progressively weakened. This disruption may directly impair the efficiency of motor planning and initiation, contributing to the exacerbation of bradykinesia and difficulties in executing motor sequences^14,38,39. Thus, beyond serving as a passive relay in pathology spread, the caudate nucleus acts as a key hub, integrating sensorimotor and cognitive information to maintain functional integrity of the CBGTC circuit.

Moreover, in our cross-sectional mediation models, the FC between caudate_PreM/SMA and PreM/SMA showed a significant indirect-effect pattern, statistically accounting for the association between caudate dopaminergic depletion and motor symptom severity. This finding is consistent with the hypothesis that caudate dopaminergic loss and motor impairment are linked, at least in part, through a caudate-centered circuit feature reflecting altered caudate–PreM/SMA coordination. It provides a more precise network-level framework for understanding PD pathophysiology, in which dopaminergic depletion serves as an initial variable, while circuit-level functional disconnection acts as a closely associated intermediate link, together constituting a potential pathway from molecular pathology to clinical motor burden. From a mechanistic perspective, these findings may arise from both the suppression of normal information flow by pathological β oscillations⁴⁰ and an underlying impairment in synaptic information storage and transmission efficiency⁴¹. At the network oscillation level, dopamine loss in PD may disrupt the oscillatory balance within the basal ganglia-cortical circuits, leading to pathologically enhanced β-band activity. This excessive synchronization is thought to entrain the motor network, suppressing pro-kinetic γ oscillations and their cross-frequency phase-amplitude coupling with β rhythms, thereby impeding the initiation and execution of new motor programs^40,42. At the synaptic plasticity level, severe dopaminergic depletion in the caudate causes an imbalance in D1 and D2 receptor signaling, resulting in dysregulation of the direct and indirect pathways. This state impairs synaptic plasticity, characterized by a compromised induction of long-term potentiation (LTP) and a relative facilitation of long-term depression (LTD). The net effect is a reduced efficiency of information processing and transmission within the caudate, which further weakens FC strength and contributes to the worsening of PD symptoms^43,44. Such synaptic and oscillatory alterations could plausibly contribute to weaker caudate–PreM/SMA connectivity and greater motor symptom burden. Taken together, these mechanisms provide a biologically grounded context for interpreting the indirect-effect pattern observed in our cross-sectional models, while underscoring the need for longitudinal and interventional studies to establish temporal directionality.

The reliability of our conclusions stems primarily from several key methodological innovations, which enabled a more precise delineation of circuit dysfunction throughout the progression of PD. First, anatomically constrained FC was computed on subject-specific structural scaffolds, mitigating spurious correlations that arise when BOLD synchrony is estimated without white-matter constraints and improving physiological interpretability^25,45–47. Second, we achieved the segmentation of subcortical nuclei and the delineation of CBGTC circuits at the individual level, offering superior anatomical accuracy compared to group-level template methods²⁴. While previous studies often depend on standard brain atlases to define subcortical regions, our individualized segmentation ensured the anatomical precision of circuit nodes for each participant. This allowed the FC analysis to be conducted within a truly individual-specific circuit framework. Additionally, the single-session PET–MR protocol further reduced cross-modal misregistration and session-to-session variability, improving linkages between SNBR and loop-resolved FC.

The findings of this study have potential clinical translational value, opening promising avenues for the development of novel disease-modifying therapies and imaging biomarkers. Our results suggest that the FC strength between the caudate and PreM/SMA may serve not only as a potential imaging biomarker for assessing disease stage and predicting functional deterioration but also as a precise anatomical target for neuromodulatory interventions. Importantly, the full mediating effect of this connection on the relationship between dopamine depletion and PD symptom progression indicates that, in cases where dopamine replacement therapy provides suboptimal benefit, enhancing the connectivity strength of this pathway may represent a critical therapeutic breakthrough. In recent years, multiple non-invasive brain stimulation (NIBS) techniques, including repetitive transcranial magnetic stimulation (rTMS) and transcranial direct current stimulation (tDCS), have accumulated substantial evidence for improving both motor and non-motor symptoms of PD^48–50. A systematic review and meta-analysis reported that TMS targeting the supplementary motor area could significantly improve motor function in PD patients, with potentially greater benefits when combined with pharmacotherapy⁵¹. In a randomized controlled trial, Tao-Mian Mi et al. demonstrated that high-frequency rTMS over the supplementary motor area could normalize aberrant brain FC patterns and significantly enhance motor performance⁵². Another study showed that anodal tDCS targeting the supplementary motor area enhanced cortical excitability and improved functional activities in PD patients⁵³. Besides, emerging evidence suggests that the therapeutic effects of NIBS are not confined to the local cortical site but could propagate through neural circuits to influence the activity of deep subcortical nuclei^54,55. In this context, our work provides a clear conceptual framework for future circuit-based precision neuromodulation. By integrating multimodal neuroimaging and leveraging the CBGTC circuit mapping approach employed in this study, future research should explore the efficacy and network remodeling mechanisms of individually-tailored neuromodulation targeting the critical caudate-PreM/SMA connection for improving PD motor symptoms.

Of course, this study has several limitations. First, its cross-sectional design prevents causal inference; the mediation effects reported here should be interpreted as statistical indirect associations rather than proof of directionality⁵⁶. Longitudinal and interventional studies (e.g., NIBS or pharmacological modulation with pre-/post-PET-MR) are required to test whether within-person changes in SNBR lead to connection-specific FC changes and clinical gains. Second, although powered for the primary connection and family-wise exploratory analyses, larger, multi-centre cohorts are needed for external validity, phenotype stratification, and evaluation of additional loops (e.g., limbic). Third, we stratified disease severity only by H-Y stage into two groups (mild vs moderate–severe); this coarse, stage-based binning may obscure continuous relations between dopaminergic loss and connection-level abnormalities and does not differentiate clinical subtypes (e.g., tremor-dominant vs PIGD). Fourth, UPDRS-II (activities of daily living) scores and domain-specific cognitive assessments (e.g., executive function) were not systematically collected, limiting a comprehensive evaluation of disease severity, and the degree of specificity of the circuit alterations to motor symptoms remains to be further delineated. Future studies will incorporate these measures to more fully evaluate the impact of disease severity. Fifth, the absence of a healthy control group prevents establishment of normative baselines for SNBR and loop-resolved FC, and limits statements about absolute abnormality. Sixth, ROI-based FC estimates inherently depend on the boundaries of the ROIs used to extract BOLD time series. Although our individualized subcortical ROIs were defined using diffusion-informed connectivity profiles (independent of fMRI) and therefore do not introduce statistical circularity, this structure-guided definition constitutes an anatomical prior that may influence FC magnitudes. Seventh, while subcortical nodes were individualised, cortical regions of interest (ROIs) were atlas-defined; extending individualisation to cortex and incorporating effective-connectivity models would refine anatomical specificity. Eighth, we focused on dopaminergic terminals; integrating additional molecular markers (e.g., α-synuclein PET once available, iron/neuromelanin MRI) could link pathology classes to loop-specific dysfunction.

In conclusion, in this cross-sectional PET–MR study, we linked presynaptic dopaminergic terminal loss to connection-level abnormalities in FC within individualised CBGTC circuit and to motor impairment in PD. Using connectivity-profile segmentation and anatomically constrained functional connectivity, we identified selective weakening of caudate–PreM/SMA FC in the more-affected hemisphere; this connection statistically mediated the association between caudate dopamine depletion and UPDRS-Ⅲ. Together, these findings position PD heterogeneity within a parallel-loop, circuit-level network framework and highlight a caudate-centered corticostriatal connection as both a candidate imaging biomarker and a potential target for circuit-guided neuromodulation.Validation in longitudinal and interventional cohorts, with finer-grained severity modelling and normative comparators, is warranted to test mechanism and generalisability.

Methods

Study design and participants

This study was designed as a cross-sectional observational investigation. The study protocol was approved by the Institutional Ethics Committee (Approval No.: 2025-181), and all procedures were conducted in accordance with the Declaration of Helsinki. Prior to enrollment, all participants were fully informed of the study details and provided written informed consent. A total of 81 patients with PD were recruited. After excluding two cases due to incomplete data and three cases with poor image quality, 76 participants were ultimately included in the analysis. The study flowchart is presented in Fig. 4.

Fig. 4.

Flowchart of the study design.

The inclusion criteria for patients with PD were as follows: (1) aged between 50 and 80 years; (2) right-handed; (3) Clinically diagnosed idiopathic Parkinson’s disease, according to the UK Parkinson’s Disease Society Brain Bank diagnostic criteria, with diagnostic certainty confirmed by movement disorder specialists⁵⁷; (4) Stable antiparkinsonian medication regimen for at least 4 weeks prior to enrollment, with no initiation, withdrawal, or dose adjustment of dopaminergic agents during this period; and (5) Ability to complete PET–MRI scanning and clinical assessments, and provision of written informed consent. The exclusion criteria included: (1) Atypical Parkinsonian syndromes or secondary Parkinsonism, including but not limited to multiple system atrophy, progressive supranuclear palsy, vascular Parkinsonism, or drug-induced Parkinsonism; (2) History of neurosurgical interventions for PD, such as deep brain stimulation, lesioning procedures, or intracranial experimental therapies; (3) Severe physical illness or clinically confirmed psychiatric disorders (4) clinically significant lesions found by brain MRI; (5) History of alcohol or substance abuse; (6) Current use of medications known or suspected to interfere with DAT binding, at the time of ¹⁸F-FP-CIT PET scanning; (7) Contraindications to PET–MRI acquisition, or failure to pass imaging quality control procedures.

In addition, the enrolled PD patients were stratified by disease severity based on the H-Y scale. Patients with an H-Y score ≤ 2.0 were classified as the mild PD group, while those with a score > 2.0 were categorized as the moderate-to-severe PD group⁴.

Clinical symptom assessment

Each participant underwent a comprehensive review of medical history and physical examination by a board-certified neurologist, trained in movement-disorders rating and blinded to imaging data. Clinical symptom assessment was conducted after discontinuation of antiparkinsonian medications for at least 12 hours. The H–Y staging scale⁵⁸ was used to evaluate the severity of motor symptoms and disease stage, and the Movement Disorder Society Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) Part III⁵⁹ was used for motor symptoms assessment. Depressive symptoms were assessed using HAMD⁶⁰, and MMSE⁶¹ was used to evaluate cognitive function. Non-motor symptom burden was measured with NMSS⁶², and quality of life was assessed using PDQ-39⁶³. For all participants, LEDD was calculated based on their current medications.

Image acquisition

All imaging was performed on a hybrid PET/MR system (uPMR 790 T, United Imaging, Shanghai, China), which ensured identical head positioning, geometry, and motion monitoring across modalities and minimized cross-modal coregistration error. PD patients discontinued antiparkinsonian medication for 12 h before scanning. PET-MR scanning was performed approximately 90 minutes following intravenous injection of ¹⁸F-FP-CIT (3.7 MBq/kg). All PD participants underwent ¹⁸F-FP-CIT PET scan during which multimodal MRI scans—including resting-state fMRI, T1-weighted imaging, and diffusion tensor imaging (DTI) —were collected sequentially within the same session. Foam padding and earplugs were used to minimize head motion and reduce scanner noise. During scanning, participants were instructed to lie still, relax, keep their eyes closed, remain awake, and avoid focusing on any specific thoughts.

PET images were reconstructed with the following parameters: slice thickness = 2.8 mm, Field of View (FOV) = 300 × 300 mm², matrix = 256 × 256, voxel size = 2.4 × 2.4 × 2.8 mm³, 4 iterations/20 subsets, and post-reconstruction Gaussian smoothing (FWHM = 3 mm).

The multimodal MRI scanning parameters were as follows: T1: repetition time (TR) = 7.9 ms, echo time (TE) = 2.8 ms, matrix = 256 × 256, 176 slices, slice thickness = 1 mm, FOV = 256 × 256 mm². Resting-state fMRI: TR = 2000 ms, TE = 30 ms, matrix = 64 × 64, slice number = 35, slice thickness = 3.5 mm, flip angle = 90°, 240 volumes. DTI: 64 directions, b = 0/1000 s/mm², TR = 8632 ms, TE = 84.9 ms, matrix = 92 × 96, 60 slices, slice thickness = 2.5 mm, FOV = 230×240 mm².

Image preprocessing

For the preprocessing of ¹⁸F-FP-CIT PET and multimodal MRI images, all DICOM images were first converted to NIFTI format using the dcm2niix software (https://github.com/rordenlab/dcm2niix).

Resting-state fMRI data were preprocessed using SPM12 (version 12, https://www.fil.ion.ucl.ac.uk/spm/software/spm12/) running on MATLAB 2022b. The pipeline included the following steps: removal of the first 10 time points; slice-timing correction; head motion correction (with translational or rotational displacements limited to less than 3 mm or 3°)⁶⁴; spatial normalization to the MNI space; spatial smoothing with a 6-mm FWHM Gaussian kernel; linear detrending; nuisance covariate regression (including white matter, cerebrospinal fluid, and the Friston-24 motion parameters); and band-pass filtering (0.01–0.08 Hz). T1-weighted structural data were preprocessed using the Brain Extraction Tool (BET) from the FMRIB Software Library (FSL, version 6.0.7; https://fsl.fmrib.ox.ac.uk/fsl) to remove non-brain tissue. For the DTI data, processing was also carried out using tools from FSL. Data were corrected for eddy currents and motion with eddy, followed by rotation of b-vectors to compensate for motion-induced orientation changes. A brain mask was generated from the averaged B0 image using BET. Subsequently, diffusion tensors were fitted voxel-wise using the dtifit tool to produce diffusion tensor maps, which were then visually inspected for quality control. Based on these data, fractional anisotropy (FA) maps were calculated.

PET preprocessing was performed in SPM12. First, T1-weighted images were segmented into gray matter, white matter, and cerebrospinal fluid, and PET images were co-registered to the corresponding T1-weighted images. Partial volume effect correction was then performed using the PETPVE12 toolbox (http://www.fil.ion.ucl.ac.uk/spm/ext/#PETPVE12) based on the Müller–Gärtner method. Next, the PVE-corrected PET images were spatially normalized to MNI space. Finally, a Gaussian smoothing kernel (6 mm FWHM) was applied.

ROI definition within the CBGTC Circuits

Given the central role of the CBGTC circuits in motor control, cortical ROIs were defined a priori using the Automated Anatomical Labeling (AAL) atlas⁶⁵, encompassing the medial prefrontal cortex (mPFC), DLPFC, primary motor cortex (M1), PreM/SMA, and orbitofrontal cortex (OFC). To align with our individualized subcortical mapping, the striatal, pallidal, and thalamic ROIs used for circuit construction were not taken directly from a group atlas; instead, they were obtained from subject-specific connectivity-based segmentation (see below) and then paired with cortical ROIs to form the long and short loops.

Individualized mapping of parallel CBGTC loops

First, all the aforementioned ROIs derived from standard atlases were inversely transformed to individual diffusion space using both linear and nonlinear registration, and the registration results were visually inspected by experts to ensure the reliability of subsequent analyses. Second, probabilistic diffusion tractography combined with voxel connectivity profiles-based segmentation was conducted in native diffusion space. This analysis utilized the Probabilistic Index of Connectivity algorithm within the freely available Camino software package (http://www.cs.ucl.ac.uk/ research/medic/camino/)⁶⁶. In this study, the basal ganglia and thalamus were defined as seed regions, while five cortical and subcortical structures served as target regions. For each seed voxel, 5000 Monte Carlo streamline iterations were performed to model uncertainty in diffusion directions, with streamlines terminated when the curvature exceeded 80° or the fractional anisotropy (FA) value dropped below 0.1. Next, the probability of each seed voxel reaching each predefined target region within the circuit was calculated to generate binarized connectivity maps. To ensure the robustness of our findings, we used a primary connection threshold of >1%⁶⁷. Additionally, sensitivity analyses were conducted using more lenient (>0.5%) and more stringent (>2%) thresholds to validate the stability of the connectivity patterns. For every seed voxel we computed the probability of reaching each target and built voxelwise connectivity profiles (VCP). We then identified consensus maps for voxels simultaneously connected to all requisite nodes within the long (cortico–striato–pallido–thalamo–cortical) and short (cortico–striato–thalamo–cortical) loops, respectively. This yielded hemisphere-wise, subject-specific masks of parallel CBGTC circuits. This methodological framework has previously been applied in stroke cohort²². Detailed information on intra-circuit connectivity is provided in the Supplementary Materials.

For subsequent functional analyses, individualized subcortical segments were transformed to native fMRI space using the concatenated DTI-to-T1 and T1-to-fMRI transforms to avoid multiple resampling.

Functional connectivity within individualized CBGTC circuits

From preprocessed resting-state fMRI, we extracted mean time series from each individualized subcortical segment and each a priori cortical ROI (masked in the native fMRI space). Pairwise Pearson correlations were computed for all connections within the long and short loops of the CBGTC circuits and then Fisher-z transformed for statistical analysis. This procedure yielded 40 intra-circuit FC values per hemisphere (see Supplementary Table S1 for details).

DAT-PET quantification

The bilateral striatum and its subregions (caudate and putamen), as well as an occipital region (reference), predefined from the AAL atlas, were used as ROI templates. Based on these templates, quantitative analysis was performed on the preprocessed ¹⁸F-FP-CIT PET images. The activity level within each ROI was calculated, and the SNBR was defined as follows: (mean standardized uptake value of the striatal ROI − mean standardized uptake value of the occipital ROI) / mean standardized uptake value of the occipital ROI, where the occipital uptake was considered to represent nonspecific binding²⁷.

Statistical analysis

Statistical analyses were performed using SPSS 26.0 (IBM, Armonk, NY, USA), with the significance level set at 0.05 (two-tailed). Continuous variables were expressed as mean (SD) or median (IQR) according to distribution; categorical variables were presented as n (%). Between-group comparisons for demographics used independent sample t-tests or Mann-Whitney U tests (continuous) and χ²/Fisher’s exact tests (categorical).

After controlling for age, sex, disease duration, and LEDD as basic covariates, the following statistical analyses were performed. First, analysis of covariance (ANCOVA) was conducted to compare group differences in FC and SNBR. For the FC analysis, mean FD was additionally included to account for head motion. Multiple comparisons were corrected using the false discovery rate (FDR) method. To examine clinical relevance, partial correlation analysis was performed to assess the associations between the imaging metrics that showed significant group differences and the clinical scores, adjusting for the covariates above (and mean FD for FC where applicable). Correlation results were visualized using heatmaps. To further investigate whether the relationships between SNBR and motor symptom severity was statistically consistent with an indirect-effect structure through circuit-guided connectivity, we further fitted cross-sectional mediation models using using the PROCESS macro (v4.1, Model 4) developed by Hayes (2018) within SPSS 26.0 with 5000 bias-corrected bootstrap resamples. In these models, SNBR was entered as the predictor, UPDRS-III score as the outcome, and FC as the indirect-effect term. Indirect effects were considered statistically significant when the 95% CI of the indirect effect did not include zero.

As a sensitivity analysis addressing potential shared variance attributable to overall disease severity, H-Y stage was additionally included as a covariate (alongside the a priori covariates) in both the partial correlation and mediation models.

Author contributions

Methodology: SYY and XZJ; data collection: LHM and LJJ; Study design, methodology and supervision: WJJ, HXY, ZCT and XJG; manuscript draft: SYY; manuscript reviewing and editing: WJJ, MJ and XX. All authors approved the final version of the manuscript.

Data availability

The datasets generated and/or analyzed during the current study are not publicly available due to the sensitive nature of the human participant data, but are available from the corresponding author on reasonable request.

Code availability

No custom code was developed for this study. All data processing and analyses were performed using publicly available software and standard command-line tools.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41531-026-01409-5.