Neuroimaging evidence for a dopamine-independent association between motor cortex microstructure and Parkinson’s disease severity Authors: Praveen Honhar, Sule Tinaz, Faranak Ebrahimian Sadabad, Mark Dias, Yanghong Yang, Shannan Henry, Ansel T. Hillmer, Yasmin Zakiniaeiz, Rajiv Radhakrishnan, Robert A. Comley, Xiaomeng Zhang, Bradley A. Hooker, Yu Zhang, Sjoerd J. Finnema, Richard E. Carson, Dustin Scheinost, David Matuskey Published: 2026 ABSTRACT We determined brain microstructure alterations in early-stage Parkinson’s disease (PD) using diffusion tensor imaging and neurite orientation dispersion and density imaging (NODDI). Additionally, dopamine transporter (DAT) PET in PD was also performed to evaluate whether microstructural changes and dopaminergic losses within the brain contribute independently to PD motor severity. Mean diffusivity was significantly higher in PD in many mid-brain, nigrostriatal, sub-cortical, cortical and white matter regions. However, Viso (NODDI outcome for cerebrospinal-fluid volume fraction) in the motor cortices and parietal lobe were more strongly correlated (r ~ 0.5, p < 0.01) with PD motor severity. Most DAT PET and diffusion MRI measures were uncorrelated and stepwise multiple linear regression analysis determined a combination of DAT availability in the putamen and Viso in the precentral gyrus (motor cortex) as the best predictor of PD motor severity (55% variance explained); inclusion of Viso in precentral gyrus independently accounted for 11% variance in motor severity. Introduction Parkinson’s disease (PD) is characterized by degeneration of nigrostriatal dopaminergic neurons1. Consequently, brain imaging biomarkers of dopamine loss have been used clinically for disease diagnosis and prognosis2,3. Single photon emission computed tomography (SPECT)-based dopamine transporter (DAT) scan4 and DAT-based positron emission tomography (PET) markers5, which measure nigrostriatal dopaminergic terminal loss, are amongst the most sensitive predictors of motor impairment in PD6,7. However, there is a pressing need to develop complementary imaging markers in PD to better understand PD-related alterations in the whole-brain and specifically the nigrostriatal motor circuit. Diffusion MRI imaging can simultaneously characterize alterations in brain tissue microstructure and neurodegeneration. Diffusion tensor imaging (DTI)8 is a well-established diffusion MRI technique that models the direction-dependent diffusivity of water molecules as a diffusivity tensor. Two important outcome measures of this tensor fit include mean diffusivity (MD) and fractional anisotropy (FA)9. MD represents the directional-invariant average diffusivity of water molecules in tissue and FA is a measure of anisotropy in diffusivity along different directions. A fundamental constraint with DTI is that MD and FA are sensitive but indirect measures of tissue microstructure alterations but are not specific measures of individual microstructure features10. With advances in diffusion MRI sequences, there has been a greater push for multi-compartment models based on biophysical representation of local tissue microstructure using diffusion patterns observed with multiple gradient strengths and directions11. One such successful model developed in the last decade is the neurite orientation dispersion and density imaging (NODDI) that uses diffusion MRI to estimate three outcome measures – isotropic volume fraction (Viso), neurite density index (NDI, also known as Vicvf or Vic) and orientation dispersion index (ODI)12. Briefly, these outcomes represent the following: Viso - volume fraction of freewater/cerebrospinal fluid, (1-Viso) represents the tissue fraction; NDI - volume fraction of neurites (axons/dendrites) within the tissue-fraction compartment, (1-NDI) represents the extracellular volume fraction (soma, non-neuronal cell types) within the tissue fraction; and ODI is a metric that represents the dispersion (spread) in the orientation of neurites12. It should be noted that MD and Viso are related measures, as high values of Viso (for example within the ventricles) would lead to higher values of MD as diffusivity of water is higher in CSF than tissue. Similarly, FA and ODI are somewhat inversely related measures – while ODI measures the overall dispersion of neurites, FA (especially within white matter) is higher when the dispersion of neurites is small, since in that case water diffusion is higher along the neurite and therefore, more anisotropic. There is no equivalent measure for neurite density in the simpler DTI model. In PD, DTI has been the most used diffusion MRI protocol, yet despite numerous studies and associated meta-analyses13–15, DTI outcomes such as FA have presented not only conflicting findings16–20 but are also difficult to interpret. For example, a lower value of FA might be attributed to neurodegeneration, fanning or sprouting of neurites or demyelination21. In contrast, only a small number of studies have employed NODDI imaging in PD focusing on the alterations within the nigra and nigrostriatal pathway22–24. Recent studies have also used NODDI in data-driven algorithms (tract-based or gray matter-based spatial statistics) to characterize whole brain microstructure alterations22,25 in PD or to discriminate early-stage PD from other Parkinsonism’s based on patterns of microstructure alterations26. While at least one study in PD has reported a significant correlation between nigrostriatal microstructure alterations assessed by NODDI and motor severity27, an important gap in the literature exists in understanding the differential benefit of nigrostriatal microstructure markers when compared with dopaminergic markers. To this end, we performed multimodal diffusion MRI (DTI and NODDI) and high-resolution DAT PET imaging in a cohort of 25 PD participants. DAT PET imaging was performed using the radiotracer [18F]FE-PE2I, a high affinity28 and selective29 marker for DAT that has previously shown excellent discrimination of PD from controls30, as well as correlations with worsening motor severity in cross-sectional6 and longitudinal PD cohorts7. We recently validated the high correlation of DAT PET measure with motor severity in PD and the same DAT PET data were reanalyzed for this study31. The focus of this work lies in understanding the similarity and differences between the diffusion MRI and DAT measures in the nigrostriatal motor circuit, as well as, in assessing whether the diffusion MRI measures within the nigrostriatal motor circuit or in other brain regions can be combined with DAT PET to explain more variation in clinical motor severity, compared to DAT PET alone. Results Study participants This study included 59 participants – 39 individuals with PD (mean±SD age: 63.8±6.7 years, Hoehn-Yahr Stage: 2.0±0.0, mild bilateral disease) and 20 controls (age: 44.8±15.5 years) who underwent structural (T1) and diffusion MRI (multi-shell diffusion weighted imaging, see Materials and Methods). Figure 1 presents diffusion MRI outcome images for a representative control participant. All PD participants met the Movement Disorder Society’s (MDS) diagnostic criteria32 – either as clinically established (90% expected specificity of diagnosis) or clinically probable PD (80% expected sensitivity and specificity of diagnosis). Data reported in this study was collected at the Yale PET and MR research centers between 2018 and 2024 under protocols approved by the Yale University Human Investigation Committee and the Yale University Radiation Safety Committee. All participants in the study gave written informed consent after explanation of the procedures. Participants recruited to this study underwent routine laboratory tests and neurological evaluation to exclude pregnant or lactating individuals, and significant medical/neurological illness other than PD. Other exclusion criteria included a history of addiction or contraindications to MRI. Healthy participants were recruited from the community to volunteer for research studies at the Yale PET and MR Centers and met the inclusion and exclusion criteria.Fig. 1Example of diffusion MRI images in a healthy participant.Representative images (axial slices at the level of the striatum) of MD and FA (DTI outcomes), and Viso, ODI and NDI (NODDI outcomes), in a healthy participant (female, 54 years old). Example of diffusion MRI images in a healthy participant. Representative images (axial slices at the level of the striatum) of MD and FA (DTI outcomes), and Viso, ODI and NDI (NODDI outcomes), in a healthy participant (female, 54 years old). Based on the clinical subtype dominance criteria (tremor dominant vs. postural instability/gait disturbances (PIGD) subtype) reported by Stebbins et al.33, 59% (n = 23) of our PD cohort would be classified as tremor dominant (ratio of average MDS-UPDRS tremor to PIGD scores > 1.15), 31% (n = 12) would be classified PIGD subtype (tremor to PIGD ratio < 0.90), and 10% (n = 4) had indeterminate subtype (0.90 < tremor to PIGD ratio < 1.15). However, it should be noted that due to the relative early clinical staging of our cohort (Hoehn Yahr Stage 2), gait disturbance symptoms like freezing-of-gait were uncommon in our participants. Instead, almost all PD participants (36/39) scored a cumulative total of ≤ 1 on questions 3.10, 3.11 and 3.12 (questions addressing PIGD) on MDS-UPDRS-III. The most common symptomatology for our participants, which was also reflected in the major portion of their MDS-UPDRS scores, was related to bradykinesia and rigidity, followed by tremor. Twenty-five PD participants included in this study also underwent DAT PET imaging, the results of which have been reported previously31. For this study, the larger cohort of 59 participants was used for cross-sectional diffusion MRI comparisons (PD vs. controls), and the smaller cohort of 25 PD subjects (multimodal PD cohort) was used for comparisons between diffusion MRI and DAT PET measures. Clinical characteristics of both cohorts are presented in Table 1a and 1b.Table 1a Clinical characteristics for cross-sectional diffusion MRI cohort; b Clinical characteristics and radiotracer injection details for the multimodal cohortaParkinson’s diseaseHealthy controlsp-valueN3920-Age (years)63.8±6.744.8±15.5< 0.001Sex (F/M)20/1914/60.12 (χ2 test)MDS-UPDRS-III31.0±11.8Hoehn & Yahr2.0±0.0--Disease duration (years from symptom onset)5.5±4.1bParkinson’s diseaseN25Age (years)64.5±7.0Sex (F/M)14/11MDS-UPDRS-III28.2±14.4Hoehn & Yahr2.0±0.0Disease duration (years from symptom onset)6.5±4.5Injected dose (MBq)150±34Injected mass (ng/kg)7.5±5.5aReported values are Mean ± SD.bReported values are Mean ± SD. Table 1a Clinical characteristics for cross-sectional diffusion MRI cohort; b Clinical characteristics and radiotracer injection details for the multimodal cohortaParkinson’s diseaseHealthy controlsp-valueN3920-Age (years)63.8±6.744.8±15.5< 0.001Sex (F/M)20/1914/60.12 (χ2 test)MDS-UPDRS-III31.0±11.8Hoehn & Yahr2.0±0.0--Disease duration (years from symptom onset)5.5±4.1bParkinson’s diseaseN25Age (years)64.5±7.0Sex (F/M)14/11MDS-UPDRS-III28.2±14.4Hoehn & Yahr2.0±0.0Disease duration (years from symptom onset)6.5±4.5Injected dose (MBq)150±34Injected mass (ng/kg)7.5±5.5aReported values are Mean ± SD.bReported values are Mean ± SD. a Clinical characteristics for cross-sectional diffusion MRI cohort; b Clinical characteristics and radiotracer injection details for the multimodal cohort aReported values are Mean ± SD. bReported values are Mean ± SD. Diffusion MRI analyses within cross-sectional cohort: PD vs controls Across eight primary regions of interest (ROIs), MD was significantly higher in the PD group, compared to controls, in the substantia nigra (SN), putamen, ventral striatum, nigrostriatal tract (NST), corticospinal tract (CST) and corpus callosum (CC) after Bonferroni correction (p < 0.0063, Fig. 2a). The percent increase in MD in PD for these regions was about 5-8% with Cohen’s d values ranging from 0.7-1.00 (Supplementary Table 1). MD was also significantly higher in the PD group compared to controls in 12 exploratory regions including the red nucleus, pallidum, subthalamic nucleus (STN), amygdala, anterior cingulate (ACC), precentral and postcentral gyri, orbitofrontal cortex and the four cortical lobes (p < 0.0025, 6-14% higher MD, Cohen’s d: 0.7-1.1, Supplementary Table 1 and Supplementary Figure 1). In contrast, FA in the PD group was not significantly different from controls in any primary region, but significantly lower FA (p < 0.0025, 5-8% lower, Cohen’s d range: 0.8-1.2) was observed in some exploratory regions like precentral and postcentral gyri, as well as the frontal, parietal and occipital lobes (Supplementary Table 2 and Supplementary Figures 2 and 3).Fig. 2Comparison of mean diffusivity (DTI MD) and CSF volume fraction (NODDI Viso) between Parkinson's and controls.Group comparison (PD vs controls) of (a) MD and (b) Viso in primary ROIs. Error bars represent SD. * Bonferroni-adjusted p<0.0063. Comparison of mean diffusivity (DTI MD) and CSF volume fraction (NODDI Viso) between Parkinson's and controls. Group comparison (PD vs controls) of (a) MD and (b) Viso in primary ROIs. Error bars represent SD. * Bonferroni-adjusted p<0.0063. Viso was significantly higher in the PD group, compared to controls, only in the CST (p = 0.0003, 13% higher Viso, Cohen’s d = 1.1, Fig. 2b) amongst the primary regions, and in the amygdala, occipital and temporal cortices (p < 0.0025, 15-25% higher values, Cohen’s d range: 0.9-1.3, Supplementary Table 3 and Supplementary Figure 4) within the exploratory regions. Amongst the primary regions, ODI was significantly lower in PD in the caudate (p = 0.005, -11%, Cohen’s d = 0.70) and ventral striatum (p = 0.002, -15%, Cohen’s d = 0.75), while a trend for lower ODI values was also observed in the SN and putamen (7-12% lower, Cohen’s d ~ 0.5, p (uncorrected) ~ 0.05) that did not survive Bonferroni correction (Supplementary Table 4 and Supplementary Figure 5). Significantly lower ODI was also observed in the pons, ACC, precentral gyrus, supplementary motor area, ventromedial prefrontal and frontal cortex (p < 0.0025, 8-13% lower ODI, Cohen’s d = 0.7-1.4, Supplementary Figure 6). Finally, NDI was not significantly different between PD and controls in any primary or exploratory region (not even at the uncorrected p < 0.05 level, Cohen’s d < 0.4), except the locus coeruleus which had a 14% higher value for mean NDI in PD that survived Bonferroni correction (p = 0.002, Supplementary Table 5, Supplementary Figures 7 & 8). For the 38 PD participants with available MDS-UPDRS-III motor scores, none of the diffusion MRI measures in any of the primary regions were significantly correlated with MDS-UPDRS-III (magnitude of Pearson’s r < 0.3, Supplementary Table 6). However, amongst the exploratory regions, notable positive correlations were observed between MDS-UPDRS-III and Viso values in the precentral gyrus (r = 0.47, p = 0.003), postcentral gyrus (r = 0.50, p = 0.001), supplementary motor area (r = 0.42, p = 0.008) and parietal lobe (r = 0.47, p = 0.003, Fig. 3). MD within the same exploratory regions was also positively correlated with MDS-UPDRS-III, although the Pearson’s correlation coefficient was lesser (0.3 < r < 0.4, 0.01 < p < 0.05, Fig. 3). ODI in motor and parietal cortex was also weakly correlated with MDS-UPDRS-III (0.3 < r < 0.4, 0.01 < p < 0.05, Supplementary Figure 9).Fig. 3Associations between diffusion MRI metrics and Parkinson's disease motor severity.Correlation between MDS-UPDRS-III and Viso or MD in the exploratory motor cortical regions and parietal lobe. Correlations were stronger with Viso. Associations between diffusion MRI metrics and Parkinson's disease motor severity. Correlation between MDS-UPDRS-III and Viso or MD in the exploratory motor cortical regions and parietal lobe. Correlations were stronger with Viso. Comparative analysis of DAT PET and diffusion MRI imaging Pearson’s r for correlation between DAT BPND (non-displaceable binding potential, PET outcome measure for DAT availability) and diffusion MRI measures in PD participants in the four nigrostriatal regions are shown in Table 2. Most correlations were not significant, except for the correlation between NDI and DAT BPND in the SN (r = 0.57, p = 0.003, survived Bonferroni threshold of p < 0.0125). MD showed a trend of non-significant negative associations with DAT BPND in all four regions. On average across ROIs, diffusion MRI measures explained only 3.0% (ODI) to 10.0% (MD) of the total variability in the DAT PET measures of PD participants.Table 2Pearson’s r between regional DAT BPND and diffusion MRI measures in nigrostriatal regionsSubstantia NigraPutamenCaudateVentral StriatumMD−0.11a−0.30−0.31−0.45FA−0.37a−0.250.004−0.10Viso0.480.060.090.005ODI0.120.040.100.31NDI0.570.020.220.04aOne outlier was removed from the analysis as it was misrepresenting the general trend (see Supplementary Figure 10). Bold value indicates significance after Bonferroni correction. Table 2Pearson’s r between regional DAT BPND and diffusion MRI measures in nigrostriatal regionsSubstantia NigraPutamenCaudateVentral StriatumMD−0.11a−0.30−0.31−0.45FA−0.37a−0.250.004−0.10Viso0.480.060.090.005ODI0.120.040.100.31NDI0.570.020.220.04aOne outlier was removed from the analysis as it was misrepresenting the general trend (see Supplementary Figure 10). Bold value indicates significance after Bonferroni correction. Pearson’s r between regional DAT BPND and diffusion MRI measures in nigrostriatal regions aOne outlier was removed from the analysis as it was misrepresenting the general trend (see Supplementary Figure 10). Bold value indicates significance after Bonferroni correction. Within the nigrostriatal motor circuit, most correlations between DAT BPND in SN and putamen, and the diffusion MRI measures in the NST tract were also modest and non-significant (see Fig. 4). However, NODDI outcome measures in the NST showed strong positive correlations with each other (Viso, ODI: 0.75; Viso, NDI: 0.86; ODI, NDI: 0.83, survived Bonferroni threshold p < 0.0024). A trend towards negative associations between MD in the NST and DAT BPND in the putamen and SN was observed, with the correlation being higher with the SN BPND (r = -0.46, p = 0.03, did not survive Bonferroni correction). DAT BPND in the putamen and SN were also moderately correlated with each other (r = 0.56, p = 0.004). Finally, Viso and MD values within the NST were uncorrelated (r = 0.09).Fig. 4Correlations between DAT PET and diffusion MRI measures within the nigrostriatal motor circuit.Pearson’s correlation coefficients between all imaging outcomes relevant to the nigrostriatal motor circuit (DAT PET in SN and putamen, and diffusion MRI outcome measures in the NST). * Bonferroni-adjusted p<0.0024. Correlations between DAT PET and diffusion MRI measures within the nigrostriatal motor circuit. Pearson’s correlation coefficients between all imaging outcomes relevant to the nigrostriatal motor circuit (DAT PET in SN and putamen, and diffusion MRI outcome measures in the NST). * Bonferroni-adjusted p<0.0024. The stepwise linear regression model for MDS-UPDRS-III using PET and diffusion MRI measures of the nigrostriatal circuit as predictors identified a model with only the putamen DAT BPND (and an intercept term) as the best predictor based on F-test criteria,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{MDS}-\mathrm{UPDRS}-\mathrm{III} \sim {{\bf{a}}}_{{\bf{0}}}+{{\bf{a}}}_{{\bf{1}}}\mathrm{DAT}\,\mathrm{Putamen}-{\rm{B}}{{\rm{P}}}_{\mathrm{ND}}$$\end{document}MDS−UPDRS−III~a0+a1DATPutamen−BPND The estimated values (s.e., p-value) were 42.5 (3.47, p < 0.001) for a0 and -10.0 (2.40, p = 0.0004) for a1. This implies that a decrease in putamen DAT availability by 1 unit of BPND is related to an increase of 10 points on the MDS-UPDRS Part III motor severity scale. Figure 5a shows the cumulative R2 (y-axis), starting from the identified model on the left end of the x-axis and adding each subsequent term that was present in the previous step of the algorithm. Thus, the final identified model explained 44.0% of the total variation in MDS-UPDRS-III, while adding the next term (Viso for NST) explained an additional 5.4% variation, which was not a significant improvement relative to the loss of an additional degree of freedom. The full linear model with all other predictors explained 57.9% of the total variation in MDS-UPDRS-III scores.Fig. 5Step-wise linear regression models relating imaging predictors to Parkinson's disease motor severity.Bar plots visualizing cumulative R2 values of each term in the stepwise linear regression algorithm involving DAT PET (Putamen-DAT) with (a) diffusion MRI measures within the nigrostriatal tract (NST) or (b) data-driven diffusion MRI measures within motor and parietal cortex. * Indicates the optimal model based on F-tests. Each term (from left-to-right) on the x-axis represents the additional term considered by the model in the previous step before being dropped. The first bar on the left (Putamen-DAT) includes an intercept term. The optimal model in (b) explained about 11% more variability in MDS-UPDRS-III scores than achieved by unimodal DAT PET. Step-wise linear regression models relating imaging predictors to Parkinson's disease motor severity. Bar plots visualizing cumulative R2 values of each term in the stepwise linear regression algorithm involving DAT PET (Putamen-DAT) with (a) diffusion MRI measures within the nigrostriatal tract (NST) or (b) data-driven diffusion MRI measures within motor and parietal cortex. * Indicates the optimal model based on F-tests. Each term (from left-to-right) on the x-axis represents the additional term considered by the model in the previous step before being dropped. The first bar on the left (Putamen-DAT) includes an intercept term. The optimal model in (b) explained about 11% more variability in MDS-UPDRS-III scores than achieved by unimodal DAT PET. For the post hoc data-driven version of the stepwise linear regression model, we selected Viso and ODI values for the precentral, postcentral and parietal cortex as the diffusion MRI predictors along with putamen DAT BPND. Viso was selected due to its significant correlation with MDS-UPDRS-III in the cross-sectional study. MD in those regions was not selected as it would be a similar measure to Viso that could lead to known collinearity issues with multilinear models. ODI was selected as it measures a different aspect of brain microstructure and showed promising trends for correlation with MDS-UPDRS-III, unlike NDI. Like the previous model, the starting model included linear terms from all predictors and an intercept,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{MDS}-\mathrm{UPDRS}-\mathrm{III} \sim \,{{\bf{a}}}_{{\bf{0}}}+{{\bf{a}}}_{{\bf{1}}}\mathrm{Putamen}-B{P}_{\mathrm{ND}}+{{\bf{a}}}_{{\bf{2}}}\mathrm{Precentral}-\mathrm{Viso}+{{\bf{a}}}_{{\bf{3}}}\,\mathrm{Postcentral}-\mathrm{Viso}+{{\bf{a}}}_{{\bf{4}}}\,\mathrm{Parietal}-\mathrm{Viso}+{{\bf{a}}}_{{\bf{5}}}\mathrm{Precentral}-\mathrm{ODI}+{{\bf{a}}}_{{\bf{6}}}\mathrm{Postcentral}-\mathrm{ODI}+{{\bf{a}}}_{{\bf{7}}}\,\mathrm{Parietal}-\mathrm{ODI}.$$\end{document}MDS−UPDRS−III~a0+a1Putamen−BPND+a2Precentral−Viso+a3Postcentral−Viso+a4Parietal−Viso+a5Precentral−ODI+a6Postcentral−ODI+a7Parietal−ODI. The stepwise linear regression identified the following model as the best predictor of MDS-UPDRS-III,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{MDS}-\mathrm{UPDRS}-\mathrm{III} \sim {{\bf{a}}}_{{\bf{0}}}+{{\bf{a}}}_{{\bf{1}}}\mathrm{Putamen}-{\mathrm{BP}}_{\mathrm{ND}}+{{\bf{a}}}_{{\bf{2}}}\mathrm{Precentral}-\mathrm{Viso}$$\end{document}MDS−UPDRS−III~a0+a1Putamen−BPND+a2Precentral−Viso with the estimated (s.e., p-values) values of 19.5 (10.6, p=0.08) for a0, -9.8 (2.2, p=0.0002) for a1 and 95.7 (42.1, p=0.03) for a2. The slope of the putamen DAT BPND was still around -10 like the previous model, but additionally an increase of 10% in the Viso within the precentral gyrus was related to about 10 points increase in MDS-UPDRS Part III rating. Figure 5b shows the cumulative R2 plot for this model. The identified model explained 55.1% of the total variation for MDS-UPDRS-III, an increase of 11% from the model using only putamen DAT BPND as the predictor. Further, we also tested a model that included a term for interaction effect between putamen DAT BPND and precentral-Viso, but this interaction was not significant and was dropped by the stepwise linear regression algorithm, suggesting putamen DAT BPND and precentral-Viso are independent predictors of MDS-UPDRS-III. As motor symptom presentation as well as dopaminergic loss in PD generally follows an asymmetric pattern in the early phases, we repeated key multimodal imaging analyses separately in the more affected and less affected brain hemispheres. For these analyses, the more affected hemisphere was considered to be the one contralateral to the side of first symptom onset (self-reported). We have previously shown that this definition was consistent with expected dopamine loss pattern, and the asymmetry in dopamine terminal loss was correlated to the degree of asymmetry in motor symptom presentation in this cohort31. The correlation pattern between DAT and diffusion MRI measures in the four nigrostriatal regions was similar in both hemispheres, with a slightly higher value for the correlation coefficient in the less affected hemisphere. For example, Pearson’s r between DAT and NDI in the substantia nigra was 0.53 (p = 0.006) and 0.59 (p = 0.002) for the more and less affected hemispheres, respectively. Using unilateral imaging metrics for the nigrostriatal circuit (unilateral SN and putamen DAT BPND, as well as unilateral NST diffusion MRI metrics) to predict MDS-UPDRS-III motor scores resulted in a significant effect only for the putamen DAT BPND in both hemispheres, similar to bilateral analysis. Finally, we also repeated the identified model of putamen DAT BPND and precentral Viso to predict MDS-UPDRS-III scores, using values from each hemisphere. The effect of putamen DAT BPND was similar for both hemispheres (-9.4, p < 0.001 for more affected side; -9.1, p < 0.001 for less affected side) but the effect for precentral Viso was higher in the less affected side (44.1, p = 0.05 for more affected side; 70.5, p = 0.01 for less affected side), with a correspondingly lower intercept term for the less affected side. The model fits were similar (R2 = 57% for more affected side vs. 54% for less affected side). Overall, these results are consistent with the Hoehn and Yahr Stage 2 classification of our cohort that corresponds to mild bilateral disease with involvement of both hemispheres. Observed higher effects in the less affected hemisphere may point to a floor effect for the more affected side, with alterations in the less affected hemisphere being more closely associated with disease symptomatology in the current stage of the disease. However, we recommend caution with this interpretation since asymmetry in PD is heterogeneous and would be difficult to quantify with our limited multimodal sample size. Discussion We examined the nigrostriatal system in PD through multimodal imaging with two different diffusion MRI techniques and high-resolution DAT PET. The cross-sectional diffusion MRI component of the study found large-scale brain microstructure alterations in PD compared to controls with increased MD and Viso being the most notable differences. We also reported significant correlations between these measures in the motor cortex and motor severity scores. To our knowledge, this is the first multimodal study employing high resolution DAT PET and an advanced diffusion MRI sequence in PD. The multimodal analyses with DAT PET were not only useful in examining relationships between nigrostriatal dopaminergic loss and microstructure alterations, but also pivotal in establishing the degree to which microstructural alterations outside the nigrostriatal system are associated with motor impairment, independent of dopaminergic loss in the basal ganglia. Consistent with previous reviews and meta-analyses of DTI studies in PD13–15, we observed significantly higher MD in PD in the SN, putamen, CC and cortical regions. The same studies have also reported lower MD in the CST with a small effect size (Hedges’s g=-0.16) which was interpreted as potential compensatory mechanism in PD14. However, in this study MD was higher in PD in the NST and CST with large effect sizes (Cohen’s d) in the range of 0.7-1.0. Further, lower FA in the SN and cortical areas is generally reported in PD, though there is a high degree of heterogeneity between studies, especially for the SN13,14. While we did observe notably lower FA values in PD in the motor cortex, and frontal, parietal and temporal cortices (Cohen’s d: 0.8-1.2), there was no significant change in FA in any of the primary regions including the SN. Our DTI results, particularly the statistically significant elevation of MD in the SN of the PD group, are in line with many previous reports from free water (FW) imaging in PD that have consistently reported higher FW values within the SN or the posterior part of the SN26,34–39. While some studies have reported strong correlations between baseline FW values of SN and subsequent longitudinal changes in bradykinesia or Hoehn-Yahr disease staging36,40, associations between MD or Viso in the SN and clinical measures of motor burden were not significant in our cross-sectional analysis. A previous multimodal study using FW imaging and vesicular monoamine transporter type 2 (VMAT2) PET with [11C]DTBZ used a similar stepwise linear regression approach as our study to show that VMAT2 density in the putamen and FW in the SN were independent predictors of MDS-UPDRS motor scores39. In this study, due to a combination of restriction of model complexity due to smaller number of participants (available degrees of freedom in the statistical models) and avoiding statistical tests on multiple models, we did not test the exact same hypothesis with our diffusion MRI and PET dataset – instead choosing to focus on novel analyses involving diffusion MRI measures in the nigrostriatal tract (white matter of nigrostriatal motor circuit) and within the motor cortex – i.e., areas within the motor circuit with low signal on DAT PET. However, due to the ongoing nature of our investigation, we hope that in a future cohort with more participants we shall be able to evaluate more complex models of motor severity utilizing DAT PET and diffusion MRI measures in the SN, striatal regions, motor cortices and their interconnecting white matter tracts. Importantly, our cross-sectional results can also be utilized to compare the performance of DTI and NODDI measures in discrimination of PD from controls, and correlation with motor severity within PD participants. For group comparison, MD performed better than Viso with significantly higher MD values in PD in several primary ROIs. However, ODI was superior to FA in gray matter regions, with significantly lower ODI observed in striatal regions, anterior cingulate cortex, and parts of prefrontal and motor cortex. In terms of correlation with motor severity (MDS-UPDRS-III), Viso values within the precentral and postcentral gyri, and in the parietal lobe, were better correlated with MDS-UPDRS-III compared to MD, demonstrating the higher sensitivity of NODDI. One previous study performed in PD reported significant regional correlations between MDS-UPDRS-III and NODDI measures in SN-pars compacta and putamen27 which we did not observe. However, other studies have indeed reported no significant correlation between motor severity and regional NODDI measures of nigrostriatal gray and white matter22–24. Some of these discrepancies with Kamagata et al.27 may be attributed to the higher disease staging (Hoehn-Yahr: 2.5±0.7) and duration (7.3±4.5y) of their PD cohort compared to the present study (Hoehn-Yahr: 2.0±0.7, duration: 5.5±4.1y). The time-course of alterations in regional microstructure would also influence the correlations with MDS-UPDRS-III. As nigrostriatal dopaminergic degeneration precedes the clinical stage of the disease41, it is plausible that microstructural alterations occur first in those regions, with alterations in microstructural measures of cortical regions correlating better with disease severity at the clinically symptomatic stages of the disease. Lastly, DTI imaging of the nigrostriatal system also seems sensitive to medication (levodopa) effects, with studies on de novo PD participants observing correlations42,43 that seem to be absent in comparable studies with medicated participants44. A major strength of this study was multimodal DAT PET and diffusion MRI imaging that enabled us to explore relationships between them in the nigrostriatal motor circuit. Except for a significant correlation between NDI and DAT availability in the SN, none of the correlations between regional diffusion MRI measures and their respective DAT PET measures were significant, indicating that while DAT loss and microstructural alterations are both PD-related phenomena, these measures provide complementary information. An exceptionally surprising finding was a high degree of positive correlation (Pearson’s r=0.75 to 0.86) between all NODDI measures (Viso, ODI, NDI) within the NST in PD participants. As the NST is one of the primary white-matter tracts that degenerates in PD, Viso would be expected to correlate positively with ODI and negatively with NDI, with a negative correlation between ODI and NDI. We therefore suspect that the observed correlations between diffusion MRI metrics within the NST are spurious and related to specific technical challenges in fitting the NODDI model in this region. In particular, in a poorly myelinated tract like the NST that is affected by PD pathology, the diffusion contrast between the extracellular space (1-NDI) and the isotropic compartment (Viso) might be small and this can be aggravated by neuroinflammation that may further expand the extracellular space, leading to a diffusivity profile similar to the isotropic free water compartment. In this case, the NODDI model may assign diffusion from the extracellular compartment to Viso, that would lead to a decrease in extracellular fraction (lower estimate for 1-NDI), or a corresponding increase in NDI. This can also explain the lack of correlation between Viso and MD in the NST as well as the high association between Viso and NDI. Some recent studies have also reported elevated NDI in PD and multiple system atrophy, in the SN and white matter tracts25,26, and they have attributed this to either iron accumulation or gliosis25, but in the absence of histopathological confirmation these claims remain untested. In either case, the correlations between the NODDI measures in the NST, that remain underreported in literature, could be crucial for understanding why most studies have reported null findings for group differences in the NST with NODDI metrics as well as in their associations with motor severity measures (except for one study23 that indeed reported lower NDI in only the distal part of the contralateral NST). The multimodal analysis also provided evidence for the complementary nature of DAT PET and diffusion MRI in the prediction of motor severity in PD. DAT PET is an extremely sensitive marker of dopaminergic degeneration in PD and has been shown to accurately track motor severity of the disease in cross-sectional and longitudinal cohorts6,7,30, including in the present cohort of PD subjects31. Therefore, it is unsurprising that DAT PET was a significant predictor of motor severity, but the goal was to investigate whether measures for microstructural alterations could be an additional independent predictor of variance not accounted for with DAT PET. This was not the case for the nigrostriatal motor circuit in our analysis which indicates that most of the contribution of the nigrostriatal circuit in PD-related motor impairment can be attributed to dopaminergic loss in the putamen. However, in a data-driven analysis where we paired DAT availability in the putamen with selected diffusion MRI measures of the primary motor cortex, the best model included Viso in the precentral gyrus along with the DAT measure. This model was able to explain an additional 11% variance in MDS-UPDRS-III scores that could not be accounted for by DAT PET alone. This demonstrates the importance of PD-induced alterations in the cortical part of the motor circuit on motor impairment that cannot be captured by dopaminergic markers in the basal ganglia. Indeed, previous studies have reported altered diffusion metrics in the pallidum-precentral pathway24 and parts of the frontal lobe22. Additionally, basal ganglia-cortex interactions are known to play a major role in motor control through two distinct pathways: the indirect pathway from the STN to the cortex through striatal connections that can be influenced by dopamine imbalance, and the hyperdirect input from the motor cortex to subthalamic nucleus45–47. The hyperdirect pathway is thought to provide inhibitory input for movement to the STN and has gained increased importance in PD as a key target for deep brain stimulation48,49. Our current finding of a significant effect of the motor cortex microstructural alterations on PD motor severity, that is independent of striatal dopaminergic terminal integrity, supports the idea of malfunction of the hyperdirect pathway due to glutamatergic losses in the motor cortex. This is also in line with recent findings of reduced vesicular glutamate transporter-2 density and lower norepinephrine supply to the motor cortex in animal models of PD50. The biological hypothesis-driven component of this study focused only on the nigrostriatal motor circuit, but future studies should investigate the entirety of the motor circuit including the cortical components and its afferent and efferent pathways. A few previous studies51–57 have reported results with DTI and SPECT-based 123I-FP-CIT imaging in PD, although only some of them performed investigations that leveraged the multimodal design of their study. Lenfeldt et al.52 reported a negative correlation between striatal MD and striatal DAT uptake, but no correlations with FA, consistent with our observations. Schuff et al.51 reported positive associations between FA within the contralateral SN and DAT specific binding ratio in the putamen. Two studies53,54 have previously attempted to predict MDS-UPDRS scores based on multimodal 123I-FP-CIT SPECT and DTI studies. Son et al.53 reported that a linear model with DAT uptake in the associative cortex, putamen, globus pallidum along with fiber density tractography measurements from four tracts involving these regions was the best predictor of MDS-UPDRS, however, their correlation analysis included both PD and HC, therefore, did not reflect the worsening motor severity within the patient group. In a similar investigation, Lorio et al.54 reported the putamen DAT, MD value in the operculum along with interaction between putamen DAT and gray matter volume in the premotor and insula as the best predictor of total MDS-UPDRS scores in the patient population. It should be noted that they selected the DTI predictors based on group differences between PD and HC, and neither motor cortex areas, nor basal ganglia structures, were used as DTI predictors in their study. To our knowledge, no previous study has utilized DAT PET or SPECT with an advanced microstructure imaging technique like NODDI. Despite a novel multimodal design, there are several limitations to this study. First, the cohort size is modest with only 39 PD participants and 20 controls. Therefore, results presented here should be considered preliminary until verified in an independent multimodal cohort in future. Further, the PD participants in this study had identical disease staging precluding the analysis of microstructure alterations across different stages. However, even with uniform disease staging we observed significant correlations between NODDI outcome measures and MDS-UPDRS Part III scores. As this study is a part of an ongoing PD study at Yale, we plan to enroll more subjects in the future, with representation from prodromal to more advanced clinical stages of the disease. The control cohort of this study had fewer participants and was not balanced for age, which was accounted for in all cross-sectional analyses. Finally, to make best use of the multimodal aspect of this study with DAT PET, we mainly relied on regional and biologically driven analysis within the nigrostriatal system with few exceptions. However, in future studies we plan to further compare the diffusion MRI measures in PD through data-driven analyses such as tract-based or gray matter-based spatial statistics. Nigrostriatal-cortical motor circuit dysfunction is the neuropathological underpinning of motor symptoms in PD. Our multimodal imaging and analysis approach allowed us to identify the deficits at various neurobiological levels ranging from loss of nigrostriatal dopaminergic terminals (DAT PET) to loss of microstructure integrity (diffusion MRI and NODDI) in key regions of this circuitry. Lack of correlations between these imaging modalities in most regions of interest highlights their complementary nature. Notably, the finding that a unique combination of these measures in the motor cortico-striatal hubs (precentral gyrus and putamen) more strongly predicts motor disease severity than each measure alone is further testament to the strength of our approach. Further, our investigation also revealed methodological limitations of applying complex models of tissue microstructure to the nigrostriatal tract in particular which need to be addressed with better modeling and data collection approaches. Future clinical studies can take advantage of this multimodal paradigm in cross-sectional and longitudinal studies with larger PD cohorts at various disease stages interrogating not only motor but also non-motor circuits underlying mood and cognitive problems that are common in PD. Methods Clinical assessment of motor severity 38/39 PD participants in the cross-sectional cohort (24/25 in the multimodal PD cohort) were assessed for motor severity of the disease using the MDS-Unified Parkinson’s Disease Rating Scale Part III (MDS-UPDRS-III)58 and the Hoehn and Yahr disease progression stage59 by a movement disorders neurologist in the medication “off ” state (subjects withheld dopaminergic medications for at least 12 hours prior to assessment). Participants with PD were also scanned in the medication “off ” state. MRI acquisition: diffusion and T1-weighted imaging MRI data were collected on a 3 T Siemens Prisma scanner with a 64-channel head and neck coil. A T1-weighted, magnetization prepared rapid acquisition gradient echo (MPRAGE, repetition time (TR) = 2530 ms, inversion time = 900 ms, echo time (TE) = 2.44 ms, flip angle = 9o, voxel size = 1 mm3 isotropic) was acquired for all participants for anatomical delineation and image registrations. Multi-shell diffusion imaging data (repetition time: 4100 ms, echo time: 88 ms, resolution: 1.7 mm3 isotropic, number of slices: 84, slice thickness: 1.7 mm, bandwidth: 1700 Hz/pixel, 30 or 64 diffusion gradients depending on the b-value, b-values: 12×0, 30 ×700, 64 ×2000 s/mm2, phase encoding direction: anterior-to-posterior) were also acquired in the interleaved multi-slice mode to enable analysis with NODDI and DTI models8,12. The b-value 700 and 2000 s/mm2 shells were acquired consecutively for a total diffusion MRI acquisition time of 8 min 5 s. High-resolution DAT PET Complete details of DAT PET were reported previously31 and are summarized here. Briefly, dynamic PET data up to 60 min post-injection of [18F]FE-PE2I were collected from 25 PD participants on the High-Resolution Research Tomograph (HRRT, Siemens CTI, Knoxville, USA) that provides a spatial resolution of about 3 mm (Full Width Half Maximum). Head motion was monitored throughout the acquisition using an optical tracking system, and PET data were reconstructed with point spread function modeling and corrections for motion, attenuation, scattered and random coincidences, and detector dead time60. Diffusion MRI data processing Diffusion MRI scans with reverse phase-encode blips were not collected, hence susceptibility distortion correction using methods like FMRIB Software Library’s (FSL) topup61 was not possible. Instead, we took a different approach to distortion correction by first linearly registering62 the participant’s MPRAGE image with the average b0 (b = 0 s/mm2) image and re-slicing that MPRAGE image to the participant’s native diffusion MRI space. This was followed by computing a constrained non-linear distortion correction (using nonlineardistortioncorrection in BioImage Suite63) between the images that only allows non-linear warping in the phase encode direction64. Distortion-corrected images were then corrected for eddy currents and head motion using FSL’s eddy tool65. Next, eddy-corrected diffusion MRI images were fit to one of two diffusion models, either NODDI (all b-value shells) or DTI (using data from only b = 2000 s/mm2 shell). The NODDI fit was achieved through the Accelerated Microstructure Imaging via Convex Optimization (AMICO-NODDI) toolbox66 implemented in Python. This implementation linearizes the traditional NODDI equations to facilitate much faster voxel-level model fits ( < 3 min per subject for our dataset using a parallel implementation on 6 CPUs), enabling feasibility for clinical translation66. Further, intrinsic diffusivity is a fixed parameter in the NODDI model that depends on age and disease status67 and can be approximated through the axial diffusivity within the corpus callosum (CC). Based on the mean axial diffusivity values in the CC in our cohort (1.0×10-3 mm2/s for both PD and controls, p [2-tailed t-test] = 0.14), and also on the reported dependence of intrinsic diffusivity with age67, the value of the parameter was changed from its default (1.7×10-3 mm2/s) to 1.0×10-3 mm2/s. The AMICO-NODDI algorithm produced three output maps – isotropic fraction or free water fraction (Viso), the neurite density index (NDI) and the orientation dispersion index (ODI). Per the recommendations of Parker et al.68, the reported NDI and ODI outcomes were weighted-averages of tissue-fraction in the respective regions/voxels. In addition, b = 2000 s/mm2 shell with 64 diffusion gradients was also fit to the DTI model8 using FSL’s dtifit to generate maps of mean diffusivity (MD) and fractional anisotropy (FA). PET data processing Dynamic PET data were corrected for partial volume effects using the Müller-Gartner algorithm69 with Computational Anatomy Toolbox (CAT) segmentation for gray matter. This correction was used for all nigrostriatal regions (caudate, putamen, ventral striatum) except the substantia nigra (SN). Dynamic PET images were then fit to the simplified reference tissue model (SRTM2)70 with a population averaged \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{2}^{{\prime} }$$\end{document}k2′ value (0.096 min-1) for the reference region (cerebellum) to generate parametric images of non-displaceable binding potential (BPND). This measure reflects the available DAT density in a brain region (or voxel) relative to the non-specific binding in the reference region. Regional PET BPND values were computed for gray matter masked nigrostriatal ROIs with Müller-Gartner partial volume correction and without any masking for the SN. Definition of regions of interest Eight subcortical and white matter ROIs were selected a priori (see, Diffusion MRI analyses with cross-sectional cohort below) for primary analyses, and 20 other brain regions (see, Table S1) were selected for exploratory analyses. The combination of primary ROIs was selected based on consistently reported alterations in previous DTI literature13–15 as well as key nigrostriatal regions where PD-related dopaminergic deficits are evident on DAT PET. Most ROIs were defined using the Anatomical Automated Labeling (AAL) for SPM2 atlas71 with the following exceptions: The ROIs for the SN, raphe nucleus and locus coeruleus were defined based on in-house templates from previous PET studies on the high-resolution HRRT scanner with radiotracers used for imaging dopamine (D2/D3) receptors72,73, serotonin transporter74 and noradrenergic transporter75, respectively. The ROIs for CC and corticospinal tract (CST) were defined using the International Consortium of Brain Mapping (ICBM) – DTI-81 atlas76, a probabilistic atlas of human white matter based on DTI imaging of 81 healthy controls. The atlas-defined mask for the CST is a predominantly white-matter ROI spanning the superior–inferior course of the tract that originates within layer 5 of the precentral gyrus but lies mostly outside it. The nigrostriatal tract (NST) is a small white matter tract that extends dorsomedially from the SN to the subthalamic nucleus (STN) and then turns dorsolateral towards the posterior putamen77. Due to the small anatomical delineations needed, coupled with poor myelination and crossing white matter tracts, brainstem tractography is difficult and most standard white matter atlases do not include the NST77. However, a recent automated method for reconstructing brainstem fiber tracts has been successfully applied in clinical populations for studying pain78, and we utilized an updated version of the NST ROI probabilistic atlas developed in that work that is based on an expanded cohort of 62 participants. Finally, the ROI for the STN was taken from a previously reported atlas created using 7 T MRI imaging79. All ROI atlases were defined in the Montreal Neurological Institute (MNI 2 mm) space and transformed to individual participant’s diffusion MRI space through a combination of non-linear (between T1-weighted template MRI in MNI 2 mm and subject’s MPRAGE image) and linear (between subject MPRAGE and average b0 diffusion MRI image) transformations using previously described algorithms80. Statistical analyses overview Statistical analysis was performed in MATLAB (v2022a, MathWorks, Natick, USA) or R (v 4.2.2). All PET and diffusion MRI measures, as well as MDS-UPDRS-III scores, were approximately normally distributed (Shapiro-Wilk tests, p > 0.05). Inference testing for group comparisons was based on 2-tailed t-tests with different variance between groups and effect size was quantified with Cohen’s d (pooled SD). All correlations reported were assessed with Pearson’s r. To account for multiple tests in group comparisons and correlations, Bonferroni correction was used for the primary ROIs (p < 0.0063 for eight ROIs). Exploratory ROI results were also Bonferroni corrected (p < 0.0025 for twenty ROIs). Correlations between DAT PET and each diffusion MRI measure in the four nigrostriatal regions was considered significant at corrected p < 0.0125. The correlations between all imaging measures in the nigrostriatal motor circuit were corrected based on the absolute number of correlations tested. Reported p values in the text are based on univariate tests (without Bonferroni correction). The Results section clearly notes the Bonferroni threshold for each analysis and only results that meet the threshold are described as significant. As the control cohort was not well-matched for age (on average controls were 19 years younger, p < 0.001), all cross-sectional group comparisons presented in this study were adjusted for age. For this, each ROI’s diffusion MRI measures were adjusted for age (regressed towards the mean age of PD cohort) with separate estimated regional coefficients (slopes) for age in the PD and control cohorts using analysis of covariance (ANCOVA). Note that for most primary ROI and diffusion MRI outcome combinations, the coefficients for age were not significantly different (p > 0.05) between control and PD groups, the only exceptions being Viso in the NST and CC, where the slope for age in HC was close to zero ( < 1×10-16/y) but positive for PD (0.004/y for NST, 0.003/y for CC), and FA in the caudate where the slope for age in HC group was also close to zero but negative in the PD group (-0.002/y). No adjustments for age were performed for the correlational and multimodal analyses utilizing the PD-only cohort due to its relatively tight distribution in age (64.5±7.0 y). Diffusion MRI analyses with cross-sectional cohort Group differences in diffusion MRI outcomes between PD participants and controls were characterized by the Cohen’s d measure for effect size in the primary ROIs including brainstem, SN, putamen, caudate, ventral striatum, NST, CST and CC. These regions were chosen because they are involved in motor impairment and have demonstrated microstructural changes on previous diffusion MRI studies in PD13–15,23,24,27,42. Associations between diffusion MRI measures in the primary ROIs and MDS-UPDRS-III scores were investigated using Pearson’s r in the PD cohort only (using 38 participants with available MDS-UPDRS-III scores). Exploratory analyses of group differences and correlations in multiple ROIs were also performed including the pons, red nucleus, raphe nucleus, locus coeruleus, pallidum, sub-thalamic nucleus (STN), thalamus, amygdala, anterior and posterior cingulate (ACC and PCC), precentral and postcentral gyrus, supplementary motor area, olfactory cortex, orbitofrontal cortex, ventromedial prefrontal cortex, and frontal, parietal, occipital and temporal lobes. Bonferroni correction was used to correct for multiple comparisons in the primary regions for group differences and correlations, separately for each diffusion MRI measure. Comparative analyses of DAT PET and diffusion MRI imaging Within the multimodal cohort of 25 PD participants, we correlated the diffusion MRI outcome measures (MD, FA, Viso, ODI, NDI) with DAT availability in the gray matter nigrostriatal regions (putamen, caudate, ventral striatum, SN), corrected p < 0.0125 considered significant. Next, we focused on the nigrostriatal motor circuit that was defined as the combination of the putamen and SN, and the NST ROI connecting the two. We correlated (1) the different diffusion MRI measures of the NST with each other, and (2) the diffusion MRI measures for the NST with the DAT measures of the putamen and SN (total of 21 unique correlations, corrected p < 0.0024). For each correlation scatter plots were visually assessed to ensure that the correlation coefficient represents the general trend and is not biased by outliers. We also examined the relative contributions of the loss of DAT and alterations in microstructure of the NST to the motor severity in PD. An iterative stepwise linear regression model was used (stepwiseglm, MATLAB2022b) to predict MDS-UPDRS-III scores from a set of predictors that included putamen DAT BPND, SN DAT BPND, and DTI (MD, FA) and NODDI (Viso, ODI, NDI) measures within the NST. Starting with a linear model with all predictors and an intercept term,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{MDS}-\mathrm{UPDRS}-\mathrm{III} \sim \,{{\bf{a}}}_{{\bf{0}}}+{{\bf{a}}}_{{\bf{1}}}\mathrm{Putamen}-B{P}_{\mathrm{ND}}+{{\bf{a}}}_{{\bf{2}}}\mathrm{SN}-B{P}_{\mathrm{ND}}+{{\bf{a}}}_{{\bf{3}}}\mathrm{NST}-\mathrm{MD}+{{\bf{a}}}_{{\bf{4}}}\mathrm{NST}-\mathrm{FA}+{{\bf{a}}}_{{\bf{5}}}\mathrm{NST}-\mathrm{Viso}+{{\bf{a}}}_{{\bf{6}}}\mathrm{NST}-\mathrm{ODI}+{{\bf{a}}}_{{\bf{7}}}\mathrm{NST}-\mathrm{NDI}$$\end{document}MDS−UPDRS−III~a0+a1Putamen−BPND+a2SN−BPND+a3NST−MD+a4NST−FA+a5NST−Viso+a6NST−ODI+a7NST−NDI at each step the model was allowed to remove or add terms based on F-tests (sum of squared errors criteria) until the best model was identified. Apart from reporting the best model, we also quantified the cumulative R2 (percent variation explained) by each term in the model. Finally, a post hoc analysis was conducted where the predictors of the stepwise linear regression included the putamen DAT BPND (strongest DAT PET correlate of MDS-UPDRS-III) and the diffusion MRI predictors of other regions outside the nigrostriatal motor circuit that were most strongly correlated with MDS-UPDRS-III in the cross-sectional analyses (see Results). The multimodal analyses involving MDS-UPDRS-III were performed for 24 PD participants with available motor severity scores. Supplementary information Supplementary information. Supplementary information.