Multimodal neuroimaging elucidates the posterior-to-anterior gradient of nigrostriatal degeneration in Parkinson’s disease

Participants

We enrolled 58 participants (PD, n = 48; healthy controls (HCs), n = 10). The PD cohort comprised 31 males/17 females, with an average age of 64.23 ± 9.31 years (range, 50–81 years), a disease duration of 4.53 ± 4.55 years, a UPDRS-III score of 37.52 ± 16.14, an MMSE score of 25.98 ± 3.52, and a levodopa equivalent daily dose (LEDD) of 391.09 ± 252.11 mg. Symptom lateralization was left-predominant in 16, right-predominant in 20, and non-lateralized in 12. HCs included 3 males/7 females, age 60.0 ± 7.05 years (47–78). No between-group differences were detected for age or sex (Table 1). Unless otherwise noted, all subsequent analyses used the full sample; per-analysis n and any exclusions are indicated in the figure/table legends.

Primary striatal dopaminergic deficits (DaT-PET)

Representative ¹⁸F-FPCIT DaT-PET images show preserved striatal binding in HC and markedly reduced binding in PD (Fig. 1A). Groupwise comparisons demonstrated lower DaT-specific binding ratio (SBR) in PD across all striatal subregions, including caudate (left [L]: p = 0.0015, Cohen’s d (d) = 1.2858; right [R]: p = 0.0007, d = 1.4811), anterior putamen (aPut) (L/R: p < 0.0001, d = 2.1288/2.5945), posterior putamen (pPut) (L/R: p < 0.0001, d = 2.4342/2.7987), and whole putamen (L/R: p < 0.0001, d = 2.3050/2.7716) (Fig. 1B). Within the putamen, DaT-SBR was significantly lower in the pPut compared with the aPut bilaterally in both HC (p < 0.01) and PD (p < 0.0001) (Fig. 1C). In contrast, within the PD group, pairwise statistical comparisons revealed no significant difference in DaT-SBR between the caudate and the aPut (all p > 0.05; data not shown). Normalized to HC = 100%, PD reductions were greatest in pPut (L: 57.67%; R: 63.12%), followed by aPut (L: 45.32%; R: 50.43%) (Fig. 1D). Reductions in the caudate were more modest (L: 34.09%; R: 38.24%), as derived from normalized group means and not separately visualized (data not shown). The anatomical localization of the aPut and pPut is illustrated in Supplementary Fig. 1. Whole-brain voxelwise maps corroborated bilateral striatal reductions (Supplementary Fig. 2). In addition to striatal regions, we further examined DaT binding in the SNc. SNc FPCIT-SBR was significantly reduced in PD compared with HCs (p < 0.0001). However, the magnitude of reduction in the SNc was smaller than that observed in the striatum, consistent with the known distribution of dopamine transporter density, which is highest in striatal dopaminergic terminals and relatively lower in the midbrain.

SNc iron load (QSM) and cross-species support

Exemplar QSM images illustrate higher SNc susceptibility in PD versus HC (Fig. 2A). At the group level, SNc susceptibility was significantly elevated in PD (p = 0.0365, d = −0.5701; Fig. 2B). Post-mortem human tissue showed iron-positive deposits on H&E with co-localized pS129 α-synuclein immunoreactivity in putaminal sections (Fig. 2C). Prussian blue staining of sagittal brain sections from A53T transgenic PD mice provided histochemical evidence for the presence of iron-containing deposits within the SNc, as indicated by focal blue staining typical of ferric iron (Fig. 2D). Representative ex vivo susceptibility weighted imaging (SWI) and phase images from a single A53T transgenic mouse demonstrated increased iron-related signal within the SNc (Fig. 2E). Similarly, representative SWI images from one α-synuclein PFF–injected mouse showed iron accumulation in the SNc (Fig. 2E). These mouse imaging data are presented for qualitative cross-species validation only.

Cross-modal associations: SNc iron ↔ striatal DaT and microstructure

In PD, putaminal susceptibility was higher in pPut than aPut bilaterally (p < 0.0001; Supplementary Fig. 3A). In contrast, within the PD group, pairwise comparisons revealed no significant difference in susceptibility between the aPut and the caudate (p > 0.05; data not shown). Voxelwise partial correlations (adjusted for age and disease duration) revealed negative associations between SNc susceptibility and striatal DaT-SBR (Supplementary Fig. 3B).

After adjusting for age and disease duration, ROI-level analyses confirmed that right SNc susceptibility was inversely correlated with right caudate DaT-SBR (r = -0.370, p = 0.011) (Fig. 3A). On the left side, significant inverse correlations were also observed between SNc susceptibility and caudate DaT-SBR (r = -0.350, p = 0.017) (Fig. 3B), whereas the associations with the putamen did not reach statistical significance. Additional sensitivity analyses further adjusting for LEDD yielded similar results, indicating that the observed associations were not substantially influenced by medication dosage.

ROI-based diffusion MRI (adjusted for age and disease duration) showed hemisphere-specific coupling between SNc and striatal microstructure. On the right hemisphere, SNc free water (FW) correlated with caudate FW (r = 0.312, p = 0.049) and putamen FW (whole: r = 0.518, aPut: r = 0.539, pPut: r = 0.489; all p < 0.001); SNc FA correlated negatively with caudate FW (r = -0.362, p = 0.022) and positively with caudate FA (r = 0.360, p = 0.023); SNc RD correlated negatively with caudate FA (r = -0.321, p = 0.043) (Fig. 3C). On the left hemisphere, SNc FW correlated positively with striatal FW—putamen (whole: r = 0.729, aPut: r = 0.733, pPut: r = 0.702, all p < 0.0001) and caudate (r = 0.447, p = 0.004), while SNc FW related negatively to pPut FA (r = -0.321, p = 0.044). SNc FA correlated negatively with striatal FW (e.g., caudate: r = -0.375, p = 0.017; putamen: r = -0.365, p = 0.020; aPut: r = -0.354, p = 0.025; pPut r = -0.366, p = 0.020) (Fig. 3C). SNc diffusivity metrics (MD, RD, and AD) showed consistent positive correlations with those in the putamen (whole and pPut; r range: 0.314 to 0.380, all p < 0.05). Conversely, these SNc metrics were negatively associated with putaminal FA, particularly in the pPut (r range: −0.336 to −0.457, all p < 0.05) (Fig. 3C). Colocalization was observed between DaT, myelin basic protein (MBP), and pS129 staining of phospho-α-synuclein deposits in the ipsilateral SNc of an α-synuclein PFF mouse model (Fig. 3D). pS129 positive phospho-α-synuclein deposits colocalized with tyrosine hydroxylase in the ipsilateral SNc (Supplementary Fig. 4), and DaT signals showed spatial overlap with MBP-labeled myelinated structures in the striatum and contralateral SNc of α-synuclein PFF-injected mice (Supplementary Fig. 5). These observations are descriptive and qualitative, intended to illustrate the anatomical proximity of pathogenic α-synuclein to dopaminergic terminals and myelinated fibers, rather than to provide quantitative evidence of axonal or myelin loss.

Nigrostriatal tractography reveals pathway- and subregion-specific diffusion patterns

Probabilistic tractography was used to delineate four nigrostriatal sub-pathways in each hemisphere: SNc→caudate, SNc→putamen, SNc→aPut, and SNc→pPut. Representative probabilistic tractography from one HC and one patient with PD is shown to illustrate the spatial organization of nigrostriatal sub-pathways (Fig. 4A and Supplementary Fig. 6B). In the tractography visualizations, fiber colors represent the principal diffusion direction (red: left–right, green: anterior–posterior, blue: superior–inferior), following standard direction-encoded color conventions. At the tract level, the representative PD subject shows visually sparser and less voluminous SNc–striatal projections than the representative HC, suggesting reduced tract coherence. These qualitative features are most apparent in SNc–putaminal pathways, particularly toward the put. At the anatomical level, the underlying T1-weighted images reveal larger ventricular spaces in representative PD cases. Tractography is visualized in each subject’s native space, and these differences reflect inter-individual anatomical variability, likely related to disease-associated cerebral atrophy, rather than tractography-derived effects. Analyses were structured hierarchically to assess (i) caudate versus putamen projections, and (ii) anterior versus posterior putaminal subdivisions.

In HCs, diffusion properties differed across nigrostriatal sub-pathways, indicating intrinsic microstructural heterogeneity. SNc→pPut exhibited lower RD than SNc→aPut bilaterally (right: p = 0.0156; left: p = 0.0078). Additionally, the SNc→putamen pathway showed higher FA than SNc→caudate in the left hemisphere (p = 0.0234) (Supplementary Fig. 6A).

In PD, diffusion metrics showed pronounced pathway-dependent differences (Fig. 4B and Supplementary Fig. 6C). Right hemisphere: versus SNc→caudate, SNc→putamen exhibited higher FA/AD and lower RD (all p < 0.0001), with no MD difference; further subdivision revealed SNc→pPut had higher FA (p < 0.0001)/AD (p = 0.0003) and lower RD/MD (all p < 0.0001) than SNc→aPut. Left hemisphere: a similar pattern obtained—SNc→putamen > SNc→caudate in FA/AD (all p < 0.0001) with lower RD (p = 0.0273); SNc→pPut again showed higher FA (p < 0.0001)/AD (p = 0.0345) and lower RD/MD (p < 0.0001; p = 0.0001) relative to SNc→aPut. Immunohistochemistry in the α-synuclein PFF SNc-injected mouse model provided qualitative, regionally consistent support for the imaging findings. pS129-positive phospho-α-synuclein aggregates were detected in the ipsilateral SNc and extended selectively into the medial striatum (Fig. 4C), rather than diffusely throughout the striatum. This spatially restricted propagation mirrors the non-uniform vulnerability of nigrostriatal sub-pathways observed in human PD tractography. These findings are presented as qualitative observations intended to provide anatomical and biological context for the human imaging results, rather than as formal quantitative analyses.

Notably, these FA increases and RD decreases occurred in the context of PD-related degeneration and should be interpreted as pathway-specific diffusion signatures rather than indicators of preserved microstructural integrity.

Representative LFB-stained sections from one A53T mouse and one WT mouse (n = 1 per group) are shown in Supplementary Fig. 7. The A53T section displays qualitatively weaker LFB staining along the SNc–striatal pathway compared with WT. These images are presented for illustrative purposes only and do not allow quantitative or statistical inference regarding myelin or axonal integrity.

Serial mediation: SNc iron → (SNc DaT + microstructure) → striatal DaT

Serial mediation models (age/sex-adjusted) tested a directed chain from SNc iron (QSM) to striatal DaT-SBR, mediated by SNc DaT-SBR (¹⁸F-FPCIT), as a proxy of nigral dopaminergic integrity, and nigrostriatal microstructure (FA/MD/AD/RD) (Fig. 5). Analyses were restricted to PD patients to examine disease-specific pathophysiological relationships. Standardized path coefficients (β) were reported for all model paths, and completely standardized indirect effects, as defined in the PROCESS macro, were calculated to quantify the magnitude of mediation effects and enable comparison across pathways.

In the right hemisphere, significant indirect effects of SNc iron on striatal DaT-SBR were observed via SNc DaT-SBR alone (Ind1: a₁ × b₁) for the caudate (β_indirect = −0.125 to −0.127; 95% CI excluding 0), aPut (−0.213 to −0.216), pPut (−0.212 to −0.221), and whole putamen (−0.219 to −0.225) (Fig. 5A–D illustrate representative models using AD, whereas results for FA, MD, and RD are summarized in Supplementary Tables 1–5). As illustrated in Fig. 5A–D, these panels present representative serial mediation models using AD, in which the direct path (c′: SNc iron → striatal DaT-SBR) is explicitly shown and was non-significant across regions. Detailed results for all diffusion metrics (AD, FA, MD, RD) and for all indirect pathways (Ind1, Ind2 and Ind3) are summarized in Supplementary Tables 2–5.

In the left hemisphere, significant Ind1 indirect effects (a₁ × b₁) linked SNc iron to left caudate DaT-SBR via SNc DaT-SBR, consistently across diffusion metrics (FA, MD, AD, RD; e.g., FA β = −0.099, 95% CI [ − 0.206, −0.015]; MD β = −0.099 [ − 0.212, −0.016]; AD β = −0.096 [ − 0.206, −0.012]; RD β = −0.101 [ − 0.211, −0.014]). The corresponding estimates can be found in Supplementary Tables 2–5. Analogous Ind1 effects were observed for aPut (β = −0.137 to −0.143) and pPut (β = −0.132 to −0.136).

Correlations were consistent with model paths: higher diffusivity ↔ lower DaT-SBR (e.g., right putamen AD r = -0.321, p = 0.049; right aPut r = -0.413, p = 0.008; right caudate r = −0.476, p = 0.002; right caudate MD r = −0.377, p = 0.017; left caudate r = −0.374, p = 0.019) (Fig. 5E–I).

Region-specific diagnostic performance (ROC)

ROC analyses showed excellent discrimination by putaminal DaT-SBR: right aPut AUC 0.96 ± 0.02 (95% CI 0.92–1.00), right pPut 0.96 ± 0.03 (0.91–1.00), left pPut 0.95 ± 0.03 (0.89–1.00), left aPut 0.93 ± 0.03 (0.86–1.00) (Fig. 6A–D). Caudate accuracy was lower but robust (right 0.83 ± 0.07 [0.69–0.97]; left 0.81 ± 0.07 [0.66–0.96]) (Fig. 6E, F). The AUCs for the right aPut and pPut were significantly higher than those for the right caudate (p = 0.023 and p = 0.035, respectively), and similarly, the AUCs for the left aPut and pPut were higher than those for the left caudate (p = 0.024 and p = 0.018, respectively). No significant differences were observed between aPut and pPut within each hemisphere (right, p = 0.676; left, p = 0.211). Overall, the putamen is more effective than the caudate for PD-HC discrimination.

Participants

The workflow detailing participant recruitment, exclusion criteria, and imaging quality control is presented in Supplementary Fig. 8. A total of 48 PD patients and 10 HCs were ultimately enrolled in this retrospective study conducted at Huashan Hospital, Fudan University (January 2023–April 2024). Inclusion criteria: (1) PD diagnosis independently confirmed by two movement-disorder neurologists per 2015 MDS Clinical Diagnostic Criteria; (2) ability to complete Hoehn–Yahr, UPDRS-III, and MMSE; (3) compliance with ¹⁸F-FPCIT PET/MRI; (4) written informed consent. Exclusion criteria: (1) secondary Parkinsonism/atypical syndromes; (2) severe cardiopulmonary/metabolic/psychiatric/systemic disease; (3) cerebrovascular disease on neuroimaging; (4) history of traumatic brain injury, brain tumor, or epilepsy. Ten sex- and age-matched HCs were recruited during the same period. PD participants completed UPDRS-III and MMSE before PET/MRI. Among the 48 PD patients, disease severity was distributed as follows: 6 at Hoehn-Yahr stage 1, 17 at stage 2, 18 at stage 3, 5 at stage 4, and 2 at stage 5. This study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board (or Ethics Committee) of Huashan Hospital, Shanghai, China (approval number: KY2023712). Written informed consent was obtained from all participants.

PET/MRI imaging acquisition

All imaging was performed using an integrated 3.0T PET/MR scanner (uPMR 790, United Imaging Healthcare, Shanghai, China) with a 24-channel head coil. PD patients discontinued antiparkinsonian medication for 12 h before scanning. ¹⁸F-FPCIT (3.7 MBq/kg) was administered 90 min before scanning via the dorsal hand vein. Participants lay supine, eyes closed, with motion minimized.

PET data were acquired in list-mode over a single-bed list-mode acquisition for 45 min. Reconstruction: slice thickness 2 mm, field of view (FOV) 300 × 300 mm², 4 iterations/20 subsets, matrix 192 × 192, full width at half maximum (FWHM) 3 mm, voxel 1.56 × 1.56 × 2 mm³.

MRI data were acquired simultaneously during PET acquisition and included a 3D T1-weighted spoiled gradient-recalled echo (SPGR) sequence (repetition time [TR] 7.9 ms, echo time [TE] 2.8 ms, matrix 256 × 256, 176 slices, 1 mm thickness, FOV 256 × 256 mm²), a multi-echo gradient echo sequence for QSM (TR 28.3 ms, TE 3.4 ms, matrix 256 × 256, 118 slices, 1 mm thickness, FOV 256 × 256 mm²), and DTI using echo-planar imaging (64 directions, b = 0/1000 s/mm², TR 8632 ms, TE 84.9 ms, matrix 92 × 96, 60 slices, 2.5 mm thickness, FOV 230 × 240 mm²).

A comprehensive summary of all sequence parameters for the uPMR790 (3T) scanner is provided in Supplementary Table 6.

Data processing

PET data were processed using SPM12 (https://www.fil.ion.ucl.ac.uk/spm/). PET was co-registered to native T1; T1 was segmented (GM/WM/CSF); partial-volume correction (PVC) was performed using the PETPVE12 toolbox based on tissue priors²⁶; images were smoothed (8 mm FWHM). DaT-SBR was computed using the occipital cortex as the reference region. Given the relatively low intrinsic DaT signal in the midbrain, SNc DaT-SBR should be interpreted as a relative index of nigral dopaminergic neuronal integrity rather than a primary quantitative marker of terminal density²⁷. Partial-volume correction and anatomically constrained ROIs were applied to mitigate spill-in and susceptibility effects. SNc DaT-SBR was used exclusively in mediation analyses to model upstream nigral involvement and was not used for diagnostic discrimination.

QSM images in parts per billion (ppb) units obtained during the scanning process were co-registered to T1 via magnitude images using SPM/FSL/ANTs. Susceptibility values were referenced to the posterior limb of the internal capsule (PLIC) due to its stable magnetic susceptibility²⁵; normalized maps were smoothed using SPM12 (1 mm FWHM).

DTI preprocessing followed FSL v6.0.3 protocols^28,29: TOPUP (susceptibility correction), BET (brain extraction), EDDY (motion/eddy-current with outlier replacement), then DTIFIT to derive FA/MD/AD/RD. The eddy-corrected diffusion-weighted images and corresponding B0 images were used to estimate voxel-wise free water (FW), reflecting the fractional volume of extracellular water, using the DIPY software package. FW metrics were extracted from SNc and striatal ROIs, whereas FW was not modeled along the reconstructed nigrostriatal tracts.

Probabilistic tractography was conducted using MRtrix3. Response functions for spherical deconvolution and fiber orientation distributions (FOD) were estimated using dwi2response and dwi2fod, respectively, with a maximum spherical harmonic order (lmax) of 8. Streamline generation used iFOD2 with streanmline number = 5000, seeds number = 1000 × streamline number, step = 0.5× voxel, angle = 45°, cutoff = 0.1, minlength = 20 mm, maxlength = 200 mm; seed–target masks defined SNc→caudate/putamen and SNc→aPut/pPut sub-pathways.

Bilateral caudate and putamen were segmented using FreeSurfer. Putamen masks were subdivided into aPut and pPut using a predefined anatomical boundary along the coronal plane by an equal-distance segmentation approach. The SNc ROI originated from a neuromelanin-sensitive MRI template³⁰, which was transformed into native T1 space using ANTs. For each ROI, we extracted DaT-SBR and QSM; DTI metrics were summarized per ROI/tract. ROI-based analyses were performed within anatomically defined gray matter nuclei (SNc and striatum), whereas tract-based analyses were conducted along reconstructed nigrostriatal projection pathways derived from probabilistic tractography. To address the inherent symptom laterality in PD and to augment statistical power for group-level comparisons, a hemisphere standardization procedure was applied for all primary analyses²⁵. For lateralized PD patients, imaging data were oriented so that the hemisphere contralateral to the clinically more affected body side corresponded to the left side. Data from non-lateralized PD patients and healthy controls were not flipped. We acknowledge that hemisphere flipping has methodological limitations and may obscure genuine hemispheric differences. Therefore, to confirm that the principal findings were not driven by this standardization procedure, we performed a complementary analysis without hemisphere flipping. Specifically, bilateral striatal DaT-SBR derived from unflipped ¹⁸F-FPCIT PET images in PD patients and HCs were directly compared. This analysis yielded consistent and significant group differences, supporting the robustness of our main results independent of hemisphere standardization (Supplementary Fig. 9).

Post-mortem human brain tissue

Tissue from one PD and one control donor (Supplementary Table 7) was obtained from the Netherlands Brain Bank with consent, and procedures complied with the Declaration of Helsinki and VU Medical Center approvals³¹.

Animal models

Mouse model imaging and histopathological analyses were performed for qualitative validation only, and no quantitative statistical comparisons were conducted.

Two male C57BL/6J mice (8 weeks old) received unilateral stereotactic injections of α-synuclein PFFs into the SNc, as previously described³¹. Briefly, mice were anesthetized and placed in a stereotaxic frame, and 550 nL of α-synuclein PFF solution (2.5 µg/µL) was injected at the following coordinates relative to bregma: anteroposterior (AP) −3.0 mm, mediolateral (ML) + 1.2 mm, dorsoventral (DV) −4.4 mm. Injections were performed at a slow rate using a Hamilton syringe, and the needle was left in place for several minutes before withdrawal to prevent reflux. Mice were sacrificed 12 weeks post-injection for ex vivo MRI and immunohistochemical analyses.

In addition, transgenic mice overexpressing the human α-synuclein gene carrying the A53T mutation under the control of the mouse prion protein promoter (M83 line; mixed C57BL/6 × C3H background) were used. A53T mice (10–11 months of age, n = 6) and age-matched non-transgenic littermates (NTLs; both sexes, n = 6) were housed under standard conditions. NTLs were included as internal reference controls to ensure comparable environmental and experimental conditions and to confirm the specificity of imaging and histopathological observations. However, NTLs were not subjected to quantitative analysis or statistical comparison and are therefore not shown in the results or figures. Perfusion and anesthesia were performed according to institutional protocols. Brains were fixed in 4% paraformaldehyde and cryoprotected in 30% sucrose prior to storage at −80 °C. All procedures complied with the Swiss Federal Act on Animal Protection and were approved by the Cantonal Veterinary Office. PFF-related procedures were approved by the Regierungspräsidium Giessen (V54–19c2015h01GI20/28).

Ex vivo MRI of mouse brains

The heads were placed in a 15 mL centrifuge tube filled with perfluoropolyether (Fomblin Y, LVAC 16/6, average molecular weight 2700, Sigma-Aldrich, USA). MRI data were acquired on a BioSpec 94/30 with a cryogenic 2 × 2 radio frequency phased-array surface coil (overall coil size 20 × 27 mm², Bruker BioSpin AG, Fällanden, Switzerland) with a coil system operating at 30 K (Bruker BioSpin AG, Fällanden, Switzerland) for reception used in combination with a circularly polarized 86 mm volume resonator for transmission. For SWI, a global and MAPSHIM protocol with a field map (default settings) was used for shimming³². A 3D gradient-recalled echo SWI sequence was recorded with the following parameters: FOV = 15 × 12 × 15 mm; image size = 248 × 199 × 36 μm, resulting in a spatial resolution = 60 × 60 × 438 μm. One echo with a echo time = 12 ms; repetition time = 250 ms; flip angle = 15°; number of averages = 4, acquisition scan time = 1 h 59 min 24 s. SWI and phase images were computed using the SWI processing module in ParaVision 6.0.1 (Bruker, Ettlingen, Germany) with Gauss broadening = 1 mm and mask weighting = 4. All SWI images were compared with their phase image counterparts. MRI images were analyzed in ITK SNAP and evaluated by two people blinded to the genotype of the mice³³.

Ex vivo staining of human and mouse brain tissue

H&E was performed on 3 µm sections (human/mouse). LFB was performed on 6 µm mouse sections (overnight at 60°C) to assess myelin. Prussian blue was used to assess iron in A53T and WT mice³⁴. For immunohistochemistry, human FFPE sections (3 µm) underwent deparaffinization, rehydration, and retrieval in citrate buffer (pH 6.0) at 98°C for 20 min. Mouse brains (OCT-embedded) were coronally sectioned 30–40 µm (Leica CM3050 S) and stored in cryoprotectant (ethylene glycol:glycerol:0.1 M PBS = 1:1:3) at 4°C. Primary antibodies included pS129 (Abcam), DaT, TH, and MBP (Supplementary Table 8); nuclei were counterstained with 4’, 6-diamidino-2-phenylindole (DAPI)³⁵. Whole-section imaging used Axio Observer Z1 (20×); high-resolution/co-localization used Leica SP8 (20×/63×). Quantification was performed using QuPath and ImageJ; anatomical references were based on the Allen human/mouse atlases.

Statistical analysis

Demographic categorical variables (e.g., sex) are presented as frequencies and compared using χ² tests. Continuous variables (e.g., age) are expressed as mean ± standard deviation and analyzed with independent two-sample t-tests. Effect sizes for group comparisons were quantified using Cohen’s d. Normality of continuous variables was assessed using the Shapiro–Wilk test and visual inspection of histograms.

Group differences in striatal DaT-SBR and SNc iron deposition between PD patients and HCs were assessed using Mann-Whitney U tests when the normality assumption was not met. Partial correlation analyses, controlling for age and disease duration, were conducted to evaluate the relationship between SNc iron and striatal DaT-SBR. Between-group differences in nigrostriatal pathway metrics were examined using independent two-sample t-tests, and correlations between SNc and striatal diffusion parameters were also assessed using Pearson correlation analyses when data were approximately normally distributed.

Mediation analyses were performed using a serial multiple mediation model (Model 6) implemented in the SPSS PROCESS macro to test whether nigrostriatal microstructural alterations sequentially mediate the relationship between SNc iron deposition and striatal DaT binding. SNc iron (QSM) was specified as X, SNc FPCIT-SBR as M1, SNc striatal diffusion metrics as M2, and striatal DaT-SBR as Y, with age and sex included as covariates. Standardized path coefficients (β) were estimated for each individual regression path in the serial mediation model, including a₁ (SNc iron → SNc DaT-SBR), a₂ (SNc iron → SNc–striatal diffusion metrics), d₂₁ (SNc DaT-SBR → SNc–striatal diffusion metrics), b₁ (SNc DaT-SBR → striatal DaT-SBR), b₂ (SNc–striatal diffusion metrics → striatal DaT-SBR), and the direct effect c′ (SNc iron → striatal DaT-SBR). Based on these path coefficients, three predefined indirect effects were computed using the PROCESS macro: Ind1 (SNc iron → SNc DaT-SBR → striatal DaT-SBR), quantified as a₁ × b₁; Ind2 (SNc iron → SNc–striatal diffusion metrics → striatal DaT-SBR), quantified as a₂ × b₂; and Ind3 (serial mediation via SNc DaT-SBR and diffusion metrics), quantified as a₁ × d₂₁ × b₂. The conceptual framework and path structure of the serial mediation model, including the definitions of direct and indirect effects, are illustrated in Supplementary Fig. 10. Serial mediation analyses were conducted in PD patients only, as HCs did not exhibit sufficient variability in nigrostriatal degeneration measures required for mediation modeling. The significance of indirect effects was tested using a nonparametric bootstrap approach with 5000 resamples, and 95% bias-corrected confidence intervals (CIs) were estimated to assess mediation effects given the non-normal sampling distribution of indirect effects. All mediation paths were predefined a priori based on neurobiological hypotheses rather than data-driven model selection.

ROC analysis was used to evaluate the diagnostic performance of regional DaT-SBR, and the DeLong test was applied to compare differences in the AUC between groups. Mouse model imaging and histopathological analyses, as well as post-mortem human tissue analyses, were performed for qualitative and cross-species validation purposes only due to limited sample sizes; no quantitative statistical comparisons were conducted on these data. All analyses were conducted in SPSS version 27.0 (IBM Corp.), with a two-sided p < 0.05 considered statistically significant. Given that the primary analyses were hypothesis-driven and restricted to a limited number of predefined regions of interest, no global correction for multiple comparisons was applied.