Sub-voxel QSM reveals the mechanism linking substantia nigra iron and temporal diamagnetism to PD gait

Abstract

The precise underlying mechanisms linking changes in brain iron and diamagnetic substances to abnormal gait in Parkinson’s disease (PD) patients remain elusive. This study compared paramagnetic and diamagnetic susceptibility (χ_para, χ_dia) in 29 PD patients and 35 healthy controls (HCs) using sub-voxel quantitative susceptibility mapping (QSM). The main result was in PD patients, increased χ_para values in the substantia nigra pars reticulata (SNr) and reduced χ_dia values in the temporal cortices were correlated with gait parameters (p < 0.05). This study suggests that the SNr, crucial for gait regulation, and temporal lobe diamagnetic alterations may underlie PD-related gait dysfunction.

Introduction

Parkinson’s disease (PD) is a neurodegenerative disorder of the central nervous system that is characterized by the progressive loss of dopaminergic neurons in the substantia nigra (SN) pars compacta (SNc)¹. PD predominantly affects older populations, impacting approximately 1.7% of individuals over 65, with the prevalence increasing with age^2,3. The most prominent clinical manifestations observed during disease progression are movement-related impairments, including tremors, rigidity, bradykinesia, postural instability, and gait disorders (e.g., freezing of gait (FOG))^4,5. As the disease progresses, patients experience progressive hand function and ambulation deterioration, accompanied by a significantly elevated risk of falls, making it one of the leading causes of physical disability in elderly individuals.

Iron is currently recognized as a critical factor in the neuropathological processes of neurodegenerative diseases, including PD⁶. Iron deposition has been shown to activate the microglia and astrocytes, thereby promoting the release of proinflammatory cytokines and oxidative stress factors^7,8. This cascade exacerbates the loss of dopaminergic neurons, and leads to dysfunction of the dopaminergic striatal pathway⁹. This ultimately worsens motor symptoms in PD patients, particularly gait disorders¹⁰.

Given the pivotal role of iron deposition in the pathogenesis and progression of gait disorders related to PD, noninvasive in vivo iron mapping is essential for elucidating the underlying neurodegenerative mechanisms. The currently used magnetic resonance imaging (MRI)-based brain iron detection techniques include R2*, susceptibility weighted imaging (SWI) and quantitative susceptibility mapping (QSM). R2* imaging utilizes only the amplitude information of magnetic resonance signals, and its value is closely related to the spatial distribution of iron^11,12. In contrast, QSM uses both magnitude and phase information. The tissue information contained in the phase information can indirectly reflect the difference in magnetic susceptibility between tissues, and this difference helps to identify diamagnetic substances (e.g., calcification) or paramagnetic substances (e.g., iron). Meanwhile, QSM is not affected by the microscopic spatial distribution of iron¹². Relevant research results also indicate that QSM exhibits higher sensitivity than R2* imaging for quantifying trace iron deposition¹³. Furthermore, SWI only allows qualitative analysis of magnetic sensitive substances. At the same time, in areas with significant differences in magnetic susceptibility, it affects the observation of lesions and image quality¹⁴. QSM involves sophisticated processing steps—including phase unwrapping, background field removal, and field-to-source inversion—to effectively suppress noise and artifacts, resulting in a high-quality quantitative susceptibility map^15,16. It has been widely used to evaluate brain iron deposition and distribution statuses in various pathological and physiological states^15,17–22.

Notably, the MRI signal components associated with brain iron deposits are inherently complex, as paramagnetic iron and diamagnetic substances (such as myelin, calcium, and proteins) coexist within the same voxel^23–28. Owing to the abundance of diamagnetic myelin, the observed increase in magnetic susceptibility may be the result of demyelination and/or iron deposition, especially in the white matter^29–31. The commonly used QSM method currently provides only the average magnetization values of voxels and cannot distinguish between positive and negative magnetic susceptibilities within individual voxels. The pursuit of optimal techniques for separating paramagnetic and diamagnetic magnetization components remains an active area of research, and various methodologies have consequently been developed in recent years^32–38. Among these, the recently introduced APART-QSM, a sub-voxel QSM method, operates by employing a more comprehensive complex signal model and an alternative fitting scheme³⁸. This technique has yielded valuable insights in both ex vivo macaque brain experiments and in vivo human brain imaging studies^38,39. Building upon this foundation, our previous application of APART-QSM also demonstrated that paramagnetic susceptibility offers enhanced effectiveness in characterizing brain iron distribution patterns in PD patients with sleep disorders⁴⁰.

Therefore, in this study, we hypothesize that APART-QSM technique can differentiate between the respective contributions of iron and diamagnetic substances in PD patients, and further elucidate their roles in the development and progression of gait disorders. This research is expected to enhance the understanding of the neural mechanisms underlying motor dysfunction in PD and provide new perspectives for related pathological investigations.

Results

Demographics and behavioral measures of the participants

This study included 29 patients and 35 HCs. The demographic and clinical data of these participants are shown in Table 1. No significant differences were observed between the two groups in terms of age, sex, or education (p > 0.05). The velocity and stride length under the SSP or FP were lower in the PD group than in the HC group (all p < 0.001). In the PD group, the MDS-UPDRS-III was higher in the OFF state than in the ON state (p = 0.005), whereas the velocity (SSP, p = 0.037; FP, p = 0.043) and stride length (SSP, p = 0.030; FP, p = 0.023) under the SSP or FP exhibited the opposite trend (Table 1).

Table 1.

Demographics, clinical characteristics and objective gait parameters of the participants

Characteristics	HC	PD
		ON state	OFF state
Age	63.71 ± 8.74	66.90 ± 6.74
Sex (male/female)	21/14	19/10
Education (year)	9.83 ± 3.60	11 ± 3.06
Duration (month)	n/a	51.83 ± 28.42
LEDD (mg)	n/a	555.17 ± 273.64
MDS-UPDRS-III	n/a	18.21 ± 7.91	29.24 ± 11.69 ##
Vel_SSP (cm/s)	117.23 ± 17.09	100.13 ± 20.06***	95.81 (79.30–110.20) ***#
Cad_SSP (steps/min)	111.59 ± 10.17	114.05 (105.57–19.69)	113.88 (106.71–20.52)
StrTime_SSP (s)	1.08 ± 0.10	1.04 (1.00–1.13)	1.05 (1.00–1.13)
StrLength_SSP (cm)	125.67 ± 16.32	106.59 ± 17.82***	103.60 (84.49–113.84) ***#
Vel_FP (cm/s)	147.21 ± 20.84	123.70 ± 25.85***	120.42 (100.23–136.49) ***#
Cad_FP (steps/min)	127.66 ± 12.69	123.18 ± 10.67	127.81 (114.67–132.81)
StrTime_FP (s)	0.94 ± 0.09	0.97 (0.91–1.04)	0.95 (0.90–1.04)
StrLength_FP (cm)	138.20 ± 19.19	120.03 ± 20.89***	117.46 (96.40–127.25) ***#

^*, p < 0.05, **, p < 0.01, ***, p < 0.001, compared with control group. ^#, p < 0.05, ^##, p < 0.01, ^###, p < 0.001, compared with ON State. Categorical variables are presented as n, normally distributed continuous variables as mean ± standard deviation, and non-normally distributed continuous variables as median (interquartile range). For PD patients, there are ON and OFF states.

HC healthy control, PD Parkinson’s disease, LEDD levodopa equivalent daily dose, MDS-UPDRS-III Movement Disorder Society-Unified Parkinson’s Disease Rating Scale Part III, SSP self-selected pace, FP fast pace, Vel velocity, Cad cadence, StrTime stride time, StrLength stride length, n/a not available.

Comparative analysis of whole-brain magnetic susceptibility values

The results regarding the whole-brain magnetic susceptibility differences are presented in Table 2 and Fig. 1. With respect to the QSM values, the PD patients exhibited significantly greater susceptibility in the right substantia nigra pars reticulata (SNr), left insula, and left thalamus ventral anterior nucleus than the HCs did (p_FWE-corr < 0.05).

Table 2.

Brain regions with magnetic susceptibility value differences between the groups

regions	peak MNI coordinate			peak t value	pFWE-corr-value	cluster size
	X	Y	Z
QSM
PD > HC
right substantia nigra pars reticulata	9	−20	−19	7.0537	<0.001	81
left insula	−37	−12	−6	4.7233	<0.001	90
left thalamus ventral anterior nucleus	−8	−12	−4	5.2786	<0.001	88
χpara
PD > HC
right substantia nigra pars reticulata	9	−21	−20	−6.8158	<0.001	107
right middle frontal gyrus	30	35	−1	−4.4058	0.049	39
left insula	−35	−21	24	−4.5011	0.007	51
left precentral gyrus	−34	−14	35	−4.5955	0.001	63
left superior parietal lobule	−14	−59	43	−4.8514	<0.001	76
left middle cingulate cortex	−13	−9	50	−4.6509	<0.001	69
PD < HC
left cerebellar lobule IX	−41	−50	−50	5.4048	0.011	48
right temporal pole	28	12	−36	4.443	<0.001	189
right rectus gyrus	9	36	−18	4.865	<0.001	470
left rectus gyrus	−9	20	−20	3.9372	<0.001	105
χdia
PD > HC
left precentral gyrus	−52	−3	25	−4.7226	0.024	47
left superior frontal gyrus, dorsolateral	−22	12	40	−5.0337	<0.001	77
PD < HC
right temporal pole	24	6	−41	4.9595	<0.001	329
left temporal pole	−21	1	−32	4.6643	< 0.001	209
left cerebellum crus 1	−15	−67	−34	4.8004	0.018	49
right rectus gyrus	7	44	−20	3.883	<0.001	77
right inferior temporal gyrus	48	−29	−22	4.2296	<0.001	99
right medial orbital frontal gyrus	4	25	−12	4.1833	<0.001	333
left inferior temporal gyrus	−48	−26	−18	3.8407	0.008	54
left rectus gyrus	−4	28	−9	5.2364	<0.001	541
left insula	−37	−12	−6	4.7241	<0.001	167
left thalamus ventral anterior nucleus	−8	−9	−2	4.588	0.024	47

All clusters survived a voxel-level of p < 0.001 and a cluster-extent threshold of k > 30 voxels (cluster-level FWE corrected, p < 0.05). Anatomical labels were assigned to peak coordinates using the Automated Anatomical Labeling 3 (AAL3) atlas. PD > HC and PD < HC indicate regions where susceptibility values were significantly higher or lower, respectively, in the Parkinson’s disease group.

MNI Montreal Neurological Institute, ppb parts per billion, QSM quantitative susceptibility mapping, χ_para paramagnetic susceptibility mapping, χ_dia diamagnetic susceptibility mapping.

Fig. 1. Brain regions with magnetic susceptibility value differences between the PD patients and HCs.

The color bar represents the absolute t value. The red areas denote the regions exhibiting differences between the two groups: a voxel-level threshold of p < 0.001 and a cluster-extent threshold of k > 30 voxels (cluster-level FWE corrected, p < 0.05). The graph explicitly labels the specific relationships regarding the magnetic susceptibility magnitudes. QSM quantitative susceptibility mapping; χ_para paramagnetic susceptibility mapping; χ_dia diamagnetic susceptibility mapping.

With respect to the χ_para values, compared with the HCs, PD patients presented increased paramagnetic susceptibility in the right SNr, right middle frontal gyrus, left insula, left precentral gyrus, left superior parietal lobule, and left middle cingulate cortex, whereas decreased paramagnetic susceptibility was observed in the left cerebellar lobule IX, right temporal pole, and bilateral rectus gyrus (p_FWE-corr < 0.05).

Regarding the χ_dia values, compared with the HCs, the PD patients had reduced diamagnetic susceptibility in the bilateral temporal pole, left cerebellum crus 1, bilateral rectus gyrus, bilateral inferior temporal gyrus, right medial orbital frontal gyrus, left insula, and left thalamus ventral anterior nucleus (p_FWE-corr < 0.05). Compared with the HCs, the PD patients exhibited elevated diamagnetic susceptibility in the left precentral gyrus and left superior frontal gyrus dorsolateral (p_FWE-corr < 0.05).

Correlations between magnetic susceptibility values and clinical and gait indicators

Figure 2 shows the correlations between the magnetic susceptibility values and clinical as well as gait indicators. In PD patients, the QSM values in the right SNr showed negative correlations with stride length (r_SSP = −0.397, p_SSP = 0.033) during the ON state. Negative correlations were also observed with velocity (r_SSP = −0.436, p_SSP = 0.018, r_FP = −0. 414, p_FP = 0.025) and stride length (r_SSP = −0.515, p_SSP = 0.004, r_FP = −0.404, p_FP = 0.030) in the OFF state.

Fig. 2. Correlations between the magnetic susceptibility values and the clinical and gait indicators of PD patients.

As highlighted in the boxed section, the correlation analysis specifically examined the relationship between gait abnormalities and magnetic susceptibility values in the significantly relevant brain regions, namely the right substantia nigra pars reticulata, left temporal pole, and left inferior temporal gyrus.* p < 0.05, ** p < 0.01, *** p < 0.001. The color bar represents the magnitude of the correlation coefficient (r value). MDS-UPDRS-III Movement Disorder Society-Unified Parkinson’s Disease Rating Scale Part III, SSP self-selected pace, FP fast pace, Vel velocity (cm/s), Cad cadence (steps/min), StrTime stride time (s), StrLength stride length (cm), QSM quantitative susceptibility value, χ_para paramagnetic susceptibility value, χ_dia diamagnetic susceptibility value.

The χ_para values in the right SNr were positively correlated with the MDS-UPDRS-III score (r = 0.447, p = 0.015) during the OFF state. Moreover, negative correlations were found with velocity (ON: r_SSP = −0.476, p_SSP = 0.009, r_FP = −0.466, p_FP = 0.011; OFF: r_SSP = −0.559, p_SSP = 0.002, r_FP = −0.521, p_FP = 0.004) and stride length (ON: r_SSP = −0.512, p_SSP = 0.005, r_FP = −0.451, p_FP = 0.014; OFF: r_SSP = −0.609, p_SSP < 0.001, r_FP = −0.508, p_FP = 0.005) during both ON and OFF states.

In PD patients, χ_dia values in the left temporal pole were negatively correlated with stride length (r_SSP = −0.424, p_SSP = 0.022, r_FP = −0.516, p_FP = 0.004) and velocity (r_FP = −0.444, p_FP = 0.016) during the ON state, as well as with velocity (r_SSP = −0.550, p_SSP = 0.002, r_FP = −0. 517, p_FP = 0.004), stride length (r_SSP = −0.572, p_SSP = 0.001, r_FP = −0. 568, p_FP = 0.001), and cadence (r_SSP = −0.486, p_SSP = 0.008, r_FP = −0.440, p_FP = 0.017) during the OFF state. Positive correlations were observed with the stride time (r_SSP = 0.540, p_SSP = 0.002, r_FP = 0.530, p_FP = 0.003) during the OFF state.

The χ_dia values of the left inferior temporal gyrus were negatively correlated with velocity (r_SSP = −0.515, p_SSP = 0.004, r_FP = −0.576, p_FP = 0.001), stride length (r_SSP = −0.524, p_SSP = 0.004, r_FP = −0.577, p_FP = 0.001), and cadence (r_SSP = −0.674, p_SSP < 0.001, r_FP = −0.684, p_FP < 0.001) during the OFF state, and positively correlated with stride time (r_SSP = 0.735, p_SSP < 0.001, r_FP = 0.731, p_FP < 0.001) during the OFF state.

Note that some regions from Table 2 were excluded from Fig. 2 due to a lack of significant correlation between their magnetic susceptibility values and clinical as well as gait indicators.

Fig. 3. Processing flow of QSM.

The color bar represents the range of magnetic susceptibility values. ANTs advanced normalization tools, QSM quantitative susceptibility mapping, χ_para paramagnetic susceptibility mapping, χ_dia diamagnetic susceptibility mapping, ppm parts per million.

Discussion

In this study, we employed a relatively advanced sub-voxel QSM analysis method to quantitatively investigate the distribution characteristics of the paramagnetic and diamagnetic properties across the entire brain in PD patients and conducted correlation analyses with multiple clinical gait-related parameters. The results not only confirmed the impact of iron deposition in the SNr on gait regulation, but also revealed, for the first time, significant associations between diamagnetic properties in the temporal cortices and gait parameters. Additionally, we preliminarily explored differences in these correlations and their potential mechanisms between the ON and OFF medication states.

A meta-analysis investigating iron deposition in subcortical nuclei in PD revealed increased iron content in several regions, including the SNc, SNr, putamen, globus pallidus, and red nucleus⁴¹. Although the whole-brain analysis method used in this study may have limited statistical power for smaller nuclei, we still successfully identified significant abnormal iron deposition in the SNr, which is consistent with previous literature. Both the SNc and SNr play a central role in the pathological mechanism of PD^42,43. The motor symptoms of PD arise from the degeneration of dopamine-secreting neurons in the SNc, leading to dysfunction in its target regions (including the SNr), via direct and indirect projections⁴⁴. As a critical output structure of the basal ganglia, the SNr directly connects with the thalamus and pedunculopontine nuclei through γ-aminobutyric acid (GABA) projections, participating in motor integration, planning, and execution processes⁴⁵. The current research indicates that iron deposition in the SN of PD patients spreads from the SNc to the SNr⁴⁶. From a functional perspective, this suggests that the SNr plays a more direct role in motor control imbalance during advanced stages of the disease. Furthermore, under normal physiological conditions, iron levels in the SNr are higher than those in the SNc⁴⁷. In the advanced stages of the disease, iron deposition in the SNc tends to stabilize, which has been suggested to potentially reflect the final number of surviving dopaminergic neurons in this region⁴⁸. Building on this, our findings further indicate that in late-stage disease (H-Y stage > 3), the SNr may continue to accumulate iron due to its dominant role in motor control circuitry. These mechanisms collectively help explain why abnormal iron deposition is more prominent in the SNr rather than the SNc in our patients. Moreover,the abnormal iron deposition in the SNr observed in our study showed a positive correlation with MDS-UPDRS-III scores. This finding further supports previous studies indicating a significant association between increased SNr iron deposition and moderate-to-severe motor disorders⁴⁴.

Concurrently, as SNr iron deposition increased, the degrees of gait disorder in patients gradually increased, manifested as decreased velocity and reduced stride length, that is, bradykinesia⁴⁹. A study using 7T-MRI reported that the volume of the SN was negatively correlated with the severity of bradykinesia-rigidity⁵⁰. Our findings further specified that iron deposition in the SNr is associated with bradykinesia, while another study demonstrated that SNr electrophysiological hyperactivity leads to the occurrence of bradykinesia in PD patients⁵¹. Together, these findings highlight the role of structural and functional abnormalities in the SN, particularly the SNr, in the mechanisms underlying bradykinesia in PD patients. Future research should further explore the potential of targeting the SNr for neuromodulation therapies.

We report for the first time that PD patients exhibit significantly reduced diamagnetic susceptibility in the temporal cortices, which was strongly correlated with abnormalities in multiple gait domains, including velocity, cadence, stride time, and stride length. The temporal lobe plays a crucial role in visual perception, providing essential environmental information for locomotor adaptation during walking⁵². Neurophysiological studies have identified fine-grained speed tuning mechanisms in the temporal visual-motion processing areas of humans⁵³. Importantly, FOG in PD patients has been specifically associated with abnormalities in both the corticospinal pathways and the temporal visual processing regions⁵⁴. These findings suggest that impaired higher-order visual function mediated by the temporal lobe may be associated with the occurrence of gait disturbances. Nevertheless, the pathological mechanism underlying the reduction in diamagnetic susceptibility remains unclear and lacks a well-established explanation specific to PD. Potential contributing factors may include myelin degradation (reflecting white matter structural changes), decreased calcification, or reductions in certain proteins (e.g., oxyhemoglobin)⁵⁵. Therefore, the specific mechanisms by which diamagnetic alterations in the temporal lobe contribute to the regulation of gait function still require further validation through more detailed pathological and neuroimaging evidence.

Notably, an intriguing phenomenon that has not been previously reported in earlier studies is the attenuation or even disappearance of the strength of correlation between the same gait parameters and magnetic susceptibility values in the ON state compared to the OFF state. This phenomenon occurred consistently in both SNr with increased paramagnetism and temporal lobes with decreased diamagnetism. These results can be interpreted as dopaminergic medications compensating for neurotransmitter deficiencies while simultaneously weakening or masking the association between iron deposition, temporal lobe diamagnetic properties, and gait function. On one hand, this finding provides further support for the theory that iron metabolism dysfunction may serve as an underlying pathogenic factor contributing to dopaminergic neuron loss and neural circuit impairment⁵⁶. On the other hand, it also suggests that dopaminergic therapy, upon effectively restoring motor pathway function in the basal ganglia, may reduce the motor system’s reliance on alternative circuits such as those involving the temporal lobe—this is one of our hypotheses. The pathophysiological mechanisms underlying these observations warrant further in-depth investigation.

Furthermore, we observed that the differential changes in magnetic susceptibility and their correlations exhibit significant asymmetry between the cerebral hemispheres. This asymmetric pattern may originate from inherent physiological differences between the hemispheres⁵⁷. Alternatively, it could be related to the distinct asymmetric pathological spreading mechanisms specific to brain-first subtypes of PD⁵⁸. This finding suggests that future studies should incorporate larger cohorts of early-stage patients across different subtypes to enable more in-depth mechanistic exploration and independent validation of this asymmetry.

Our research has the following limitations. First, this was a cross-sectional study involving PD patients who exhibited gait disorders. Owing to the severity of gait disorders, the degree of patient cooperation was limited, which resulted in a relatively small sample size. Second, our proposed mechanism by which diamagnetic substances influence gait disorders in PD patients remains preliminary. Future studies with histological verifications are needed to validate these findings. Third, this compromise of the non-isotropic acquisition protocol was necessary to achieve whole-brain coverage, maintain sufficient signal-to-noise ratio, and keep scan times clinically feasible for PD patients, who are prone to motion artifacts during extended sequences⁵⁹. The current protocol was adopted as a practical trade-off, and the use of non-isotropic resolution has been well documented in previous studies^60–62. Finally, the long acquisition duration and elevated specific absorption rate (SAR) associated with R2 mapping rendered it impractical in patient populations, thereby limiting direct R2′ quantification. Although the current implementation does not require R2 mapping, the original method relies on R2 information. However, this adaptation may introduce reduced accuracy, as the parameters used to model the relationship between R2′ and R2* can vary across populations, age groups, and brain regions⁶³.

In conclusion, we conducted a quantitative study on the mechanisms underlying abnormal gaits in PD patients via the paramagnetic and diamagnetic susceptibility derived from sub-voxel QSM. Our findings not only corroborate previous discoveries regarding iron deposition in the SN as a pathogenic factor but also reveal a reduction in the diamagnetic substances in the specific cortical regions involved in gait regulation, particularly within the temporal cortices. This reduction could represent one of the contributing mechanisms underlying gait dysfunction. These insights pave the way for future research, emphasizing the need to integrate more advanced imaging techniques with histological analyses to elucidate the precise roles of iron and diamagnetic components in the gait pathologies of PD patients.

Methods

Participant recruitment

From November 2020 to July 2021, this study recruited PD patients undergoing dopaminergic therapy at the Movement Disorders Program of Beijing Friendship Hospital (affiliated with Capital Medical University), along with healthy controls (HCs) from the local community. The study protocol was approved by the Ethics Committee of Beijing Friendship Hospital, Capital Medical University (2019-P2-283-02), and conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent.

Our criteria for recruiting subjects and performing clinical evaluations were based on our previous studies^61,64. The inclusion criteria were as follows: (1) patients aged between 40 and 80 years; (2) right-handed patients; and (3) patients whose PD diagnoses were confirmed by movement disorder specialists using the UK Parkinson’s Disease Society Brain Bank criteria, with their Hoehn and Yahr (H-Y) stages >3. The exclusion criteria included patients with (1) histories of other neurological disorders; (2) histories of psychiatric disorders; (3) contraindications to MRI; (4) significant head motion during MRI; and (5) atypical or secondary parkinsonism.

A total of 29 PD patients and 35 age-, sex-, and education-matched HCs were ultimately enrolled. Demographic data (sex, age, and education data) were recorded. For the PD patients, their motor function was evaluated with Part III of the Movement Disorder Society-Unified Parkinson’s Disease Rating Scale (MDS-UPDRS-III), along with disease duration and levodopa equivalent daily dose (LEDD) information. Objective gait parameters, including speed, stride length, stride time, and cadence, were assessed in all participants using a 20-foot computerized Zeno Walkway system (Proto Kinetics, Havertown, PA, USA) under self-selected pace (SSP) and fast-paced (FP) conditions. Clinical evaluations were performed for the PD patients in the defined OFF state (≥12 h after the last PD medication dose) and the super-ON state (150% of the usual morning levodopa equivalent dose, administered via a levodopa/carbidopa formulation).

MRI acquisition

All the participants underwent image acquisition on a 3.0-T MAGNETOM Prisma MRI system (Siemens Healthineers, Erlangen, Germany) using a 64-channel head coil. The QSM sequence was obtained using a multiecho gradient echo (ME-GRE) sequence with the following parameters: TR = 50 ms, first TE = 6.8 ms, TE interval = 4.1 ms, number of echoes = 10, flip angle = 15°, and voxel size = 0.8 × 0.8 × 2.0 mm³, generalized autocalibrating partial parallel acquisition (GRAPPA) with an acceleration factor of 2, and total scan time = 4 min 36 s. Additionally, fluid-attenuated inversion recovery (FLAIR) sequences were scanned to detect abnormalities in the brain via the following parameters: TR = 8500 ms, TE = 87 ms, TI = 2440 ms, flip angle = 150°, voxel size = 0.4 × 0.4 × 6.0 mm³.

Susceptibility source separation, reconstruction, and registration

Susceptibility maps were generated after executing the following postprocessing steps. The GRE magnitude images were first used to extract brain tissue and generate a binary brain mask for subsequent processing using the brain extraction tool (BET) in FSL⁶⁵. The GRE phase images were adjusted using Laplacian-based phase unwrapping⁶⁶. The background field was removed, and the tissue phase was extracted from the unwrapped phase using the V-SHARP method^67,68. Then, STAR-QSM⁶⁹ was used to generate the susceptibility maps³⁸. Specifically, the tissue phase across all echoes was averaged to produce a single average tissue phase, with the mean echo time used as the input for dipole inversion.

The recently proposed APART-QSM method was subsequently used to reconstruct the paramagnetic and diamagnetic susceptibility maps. Briefly, this sub-voxel QSM approach employs a comprehensive complex GRE signal model that incorporates a voxel-specific magnitude decay kernel to characterize the distinct contributions of χ_para and χ_dia to the GRE signal as a function of echo time and static magnetic field strength³⁸. The R2 maps used for reconstructing χ_para and χ_dia were linearly estimated from R2* maps obtained via GRE imaging, eliminating the need for dedicated R2 mapping acquisitions⁶³. The variant is based on a linear model relating magnetic susceptibility source concentration to R2*, where R2* is expressed as a first-order approximation of reversible transverse relaxation R2′ that R2* = αR2′ + β, where α and β are the slope and intercept estimated from the linear least-squares regression between R2* and R2′ values⁶³. Despite differences in implementation, the sub-voxel QSM method demonstrates performance comparable to the original version in susceptibility separation.

The QSM, paramagnetic, and diamagnetic maps of each individual brain dataset were coregistered to the Montreal Neurological Institute (MNI) space using Advanced Normalization Tools (ANTs)⁷⁰. To improve registration accuracy, the QSM images were used to non-linearly align each subject to a QSM template in MNI space, ensuring greater similarity in both contrast and anatomical structure. Specifically, the QSM template was obtained from the MuSus-100 atlas⁷¹. The same deformation field derived from this registration was then applied to transform the corresponding χ_para and χ_dia maps into MNI space. The normalized images were then smoothed with a 6 mm full-width at half-maximum (FWHM) isotropic Gaussian kernel to reduce inter-subject variability. The resulting magnetic susceptibility maps were used for subsequent voxel-based whole-brain analyses. The processing workflow is shown in Fig. 3.

Statistical analysis

SPSS 25.0 and SPM12 software were used for statistical analysis. The Pearson chi-square test was employed to evaluate the sex distribution of the data. The single-sample Shapiro‒Wilk test was conducted to assess the normality of the demographic characteristics and clinical scale scores. For comparisons between PD patients and HCs: normally distributed data were analyzed using two-sample t-tests, while non-normally distributed data were analyzed using Mann–Whitney U tests. For comparisons between ON and OFF states in PD patients: normally distributed data were analyzed using paired t-tests, and non-normally distributed data were analyzed using Wilcoxon signed-rank tests. A voxel-wise comparison of magnetic susceptibility between groups was performed within a whole-brain gray matter mask using a General Linear Model (GLM) in SPM12, with sex, age, and years of education included as covariates. Statistical significance was assessed using a cluster-based inference approach with a voxel-level threshold of p < 0.001 (uncorrected) and a cluster-level threshold of p < 0.05 (family-wise error [FWE] corrected). Finally, a partial correlation analysis was performed between the magnetic susceptibility value and the clinical features, adjusting for sex, age, and years of education as covariates. A significance level of p < 0.05 was adopted throughout the analyses.

Author contributions

Z.W., L.D. and X.W. designed and conducted the study. L.D., R.A. and Y.Y. collected the data. Z.W., L.D. and Z.L. contributed to the data analysis and drafted the manuscript. Z.C.W., H.W. and W.W. provided advice on the data analysis, and revised the manuscript. All authors read and approved the final manuscript.

Data availability

The datasets generated and/or analyzed during the current study are not publicly available due to privacy, ethics, but are available from the corresponding author on reasonable request.

Code availability

The code for the APART-QSM method used in this study is available at https://github.com/AMRI-Lab/APART-QSM. The FSL software package https://fsl.fmrib.ox.ac.uk/fsl/fslwiki and ANTs toolkit http://stnava.github.io/ANTs/ were utilized for image processing. The T1w-QSM hybrid atlas was obtained from the publicly available repository on GitHub https://github.com/SMILE-Lab-ShanghaiTech/MuSus-100_Atlas. Custom scripts for statistical analysis are available from the corresponding authors upon request, subject to institutional data sharing agreements.

Competing interests

The authors declare no competing interests.