Introduction

Parkinson’s disease (PD) is a neurodegenerative disorder that involves the autonomic nervous system1,2, contributing to the frequent dysautonomic symptoms observed in patients. Notably, the cardiovascular system is affected, resulting in symptoms such as neurogenic orthostatic hypotension and cardiac rhythm disturbances3,4. Cardiac innervation is provided by the parasympathetic and sympathetic systems. Preganglionic neurons of the parasympathetic system are located in several nuclei of the brainstem and in the sacral parasympathetic nuclei of the spinal cord segments S2 to S4. The vagus nerve, which originates in the medulla oblongata, innervates the sinoatrial and atrioventricular nodes of the heart5. In PD, neurodegeneration affects the dorsal motor nucleus of the vagus nerve in the medulla oblongata1,2. Involvement of this nucleus, as seen on multimodal magnetic resonance imaging (MRI), was shown to correlate with cardiac rhythm abnormalities during sleep6. On the other hand, preganglionic neurons of the sympathetic system targeting the heart are located in the intermediolateral cell column (IML) of the upper thoracic spinal cord7. The presence of pathological α-synuclein inclusions in the spinal cord, particularly in the IML, is almost constant in patients with PD2,8,9.

Furthermore, two main models of pathology propagation have been proposed based on histological evidence by Braak et al.1 and, more recently, in vivo imaging studies10,11,12,13. In the first model, the initial pathology originates in the enteric nervous system and propagates in an ascending manner to the central nervous system via the peripheral system. In the second model, the pathology arises in the olfactory bulb or amygdala and spreads in a descending fashion10,11,12,13. Previous studies suggested that the ascending model may be associated with the presence of rapid-eye movement sleep behavior disorder (RBD) and characterized by early autonomic damage preceding involvement of the nigrostriatal dopaminergic system10,11,12,13. Conversely, the descending model has been observed in PD patients without RBD 10,11,13. Although these models remains debated11, spinal cord damage might be more severe in patients with RBD compared to those without RBD.

Recently, MRI has been used to detect spinal cord abnormalities in diseases such as spinal-muscular amyotrophy14, amyotrophic lateral sclerosis15, and spinal cord trauma16. Morphometry derived from T2-weighted images provides a measure of atrophy, diffusion imaging probes tissue microstructure by characterizing the diffusion of water molecules and can therefore inform on tissue damage, T1 relaxometry is also hypothesized to reflect microstructural tissue properties, and magnetization transfer contrast serves as a proxy for tissue myelination17. Taken together, these techniques appear promising for exploring spinal cord alterations in PD.

In the current study, we aimed to investigate spinal cord alterations in patients with PD in comparison with healthy controls (HC), and to assess its relationship with clinical markers of autonomic dysfunction, such as orthostatic hypotension, cross-sectionally and then longitudinally, using multimodal spinal cord MRI. We hypothesized that PD patients would exhibit structural alterations in the upper portion of the thoracic spinal cord, that these alterations would be more prominent in patients with RBD compared to those without RBD, and that such changes would be associated with cardiac autonomic dysfunction observed in PD.

Methods and materials

Participants

This prospective study was conducted as part of the ICEBERG study, a five-year longitudinal project aimed at identifying and validating markers to predict and monitor the progression of dopaminergic and non-dopaminergic lesions in early and prodromal PD.

Patients with PD, along with age- and sex-matched HCs, were recruited at the Paris Brain Institute (ICM) between 2020 and 2023. The diagnosis of PD was established according to the MDS clinical diagnostic criteria for PD3. Polysomnography was used to assess the presence or absence of RBD in our sample of PD patients, who were then stratified based on the presence (PDRBD(+)) or absence (PDRBD(−)) of RBD.

After visual inspection of all raw images (performed by L.C.), participants were excluded in cases of poor image quality related to motion and/or significant susceptibility artifacts, marked cervical lordosis, disk protrusion compressing the spinal cord, or any spinal cord signal abnormality.

The study was performed in accordance with the Declaration of Helsinki and was approved by the institutional ethical standard committee (CPP Paris VI/RCB: 2014-A00725-42). All participants gave written informed consent.

Clinical data

The following clinical scores were collected in PD patients at the time of MRI: disease duration, Movement Disorder Society Unified Parkinson’s Disease Rating Scale (MDS-UPDRS)18 part III, Hoehn and Yahr stage19, REM Sleep Behavior Disorder Screening Questionnaire (RBDSQ)20, and SCales for Outcomes in PArkinson’s disease—Autonomic Dysfunction (SCOPA-AUT) scores including the cardiovascular subscore21. Participants were assessed for the presence or absence of orthostatic hypotension, defined as a drop in systolic blood pressure ≥ 20 mmHg or diastolic blood pressure ≥ 10 mmHg within 3 or 5 min of standing compared to baseline blood pressure (defined as the mean of two measurements on the upper right arm with the participant in the supine position after 5 min of rest)22. In addition, systolic and diastolic blood pressure drops were also recorded at baseline and five-year follow-up visits.

MRI acquisition

Spinal MRI scans were performed at either the first- or the third-year follow-up visit. Participants were scanned in the ‘on’ state using a 3 Tesla Siemens PRISMA scanner (Siemens Healthcare, Erlangen, Germany) with a 32-channel posterior spine coil, an 18-channel flex body coil placed anteriorly and covering the neck area, and a 64-channel head-neck coil. The field of view covered the cervical and upper thoracic spinal cord from C2 to T5 vertebral levels, thus including the cardiovascular autonomic control centers located in the upper thoracic region. The MRI protocol comprised (see supplementary Table S1 for details about the acquisition parameters; supplementary Fig. S1):

  • - 3D turbo spin echo T2-weighted acquisition, voxel size: 0.8-mm isotropic.

  • - Diffusion-weighted imaging (DWI) echo planar imaging (EPI) with cardiac gating; b-value = 1000 s/mm2; 64 diffusion encoding directions; voxel size = 1.3 × 1.3 × 5 mm3; three adjacent axial slabs covering the spinal cord from C2 to T5 vertebral levels.

  • - 3D gradient echo images acquired with (MT on) and without (MT off) magnetization transfer saturation pulse; voxel size = 0.9-mm isotropic.

  • - 3D T1 Magnetization Prepared 2 Rapid Acquisition Gradient Echoes (MP2RAGE) sequence with T1 relaxometry map; voxel size = 1-mm isotropic.

MRI analysis

MRI images were processed using the Spinal Toolbox (SCT) version 6.323.

All acquired data were transformed into NIfTI file format and organized using the Brain Imaging Data Structure (BIDS) standard. For all MRI contrasts (T2w, MT on, MT off, DWI, and T1 MP2RAGE), the spinal cord was segmented using the contrast-agnostic model (sct_deepseg)24 to extract a mask for subsequent registration to a template. Vertebral discs were manually labeled on the T2w image, and these labels were used as landmarks for template registration. After virtual straightening of the spine, the T2w image was registered to the PAM50 template using affine and non-rigid registration steps25,26,27. In addition, the MT off image was registered to the MT on image, and the ratio between the two co-registered images ((MT off – MT on)/MT off) was calculated for each voxel as the magnetization transfer ratio (MTR, sct_compute_mtr). Diffusion processing involved several steps. For each slab, diffusion images were first averaged across the time dimension to obtain a 3D volume; the spinal cord was segmented on the resulting mean diffusion volume. The obtained mask was used for motion correction of the diffusion volume28. The diffusion tensor model was fitted, and quantitative maps of fractional anisotropy (FA) and mean diffusivity (MD) were computed (sct_dmri_compute_dti). Finally, diffusion slabs were merged, and metrics were extracted in the regions of interest. T1 relaxometry maps, calculated as a default output of the T1 MP2RAGE acquisition, were used to calculate regional T1 longitudinal relaxation times.

First, shape-based analysis was performed on T2-weighted images, in individual space, to compute spinal cord cross-sectional area (CSA) at each slice across the rostro-caudal axis. Next, MTR, diffusion and T1 relaxation metric extractions were conducted in the participant space using the distinct gray matter and white matter probabilistic atlases available in the SCT. Specifically, four regions of interest were selected, including two white matter tracts (ascending and descending tracts) and two gray matter regions (intermediolateral zone, which encompasses the intermediolateral columns, and ventral horns; see supplementary Table S2 for details about the regions). The analysis was restricted to these regions to mitigate the decrease in statistical power due to a relatively small sample size and a large number of variables. Metrics were extracted slice-by-slice and averaged across each vertebral level (from C2 to T5), and within each region. For each region, values from the right or left atlas were averaged.

All raw MRI images and processing outputs were visually inspected for quality control29.

Statistical analyses

Clinical and demographic data

Statistical analyses were performed using R version 4.3.2 (R Development Core Team, 2023). Continuous data were reported as mean ± standard deviation, and categorical variables as counts and percentages. All tests were two-sided, with significance set at p < 0.05 or false discovery rate (FDR)-adjusted p < 0.05. Clinical and demographic data were compared across the three groups (HCs, PDRBD(+), and PDRBD(−) patients) using Kruskal–Wallis tests followed by Dunn’s post hoc test for multiple comparisons for continuous variables, or Fisher’s exact tests for categorical variables.

Group comparisons

Between-group differences in MRI metrics were evaluated using linear models (LMs), with one model per region (ascending tracts, descending tracts, intermediolateral zone, ventral horns) and per MRI metric (CSA, T1, FA, MD, MTR), with ‘Group’ as the main factor and age and sex as covariates of no interest. Group effects were tested using Type II analysis of variance (ANOVA) F-tests, performed with the ‘car’ R package (v3.1–2). P-values from the F-tests were corrected for multiple comparisons across regions using FDR correction, with each MRI metric type treated separately. To assess group differences, we tested both two-group and three-group models: the first compared HCs and all PD patients; the second compared HCs, PDRBD(+), and PDRBD(–) patients. For the three-group model, when a significant group effect was detected, post hoc pairwise comparisons were performed using Tukey’s method with the ‘emmeans’ R package (v1.8.9). For each model, assumptions and model fit were checked afterwards by visually inspecting residual distribution plots using the ‘ggResidpanel’ R package (v0.3.0).

MRI metrics were first averaged across all vertebral levels from C2 to T5 to assess overall spinal cord differences, and then specifically analyzed at the C6–T1 vertebral levels, which correspond to C7-T2 spinal cord segments that encompass key centers for cardiovascular autonomic control7. Compared to vertebral levels, spinal segments provide a more accurate representation of the spinal cord’s functional organization into distinct rootlets30,31. Lower levels were excluded due to their higher susceptibility to motion artifacts from cardiac and respiratory activity.

Multivariate analysis

An exploratory multivariate analysis was conducted for illustration purposes to investigate group separation using candidate imaging features preselected from the four spinal cord regions, across all individual vertebral levels (C2 to T5) and for all MRI metrics. Variables were included in the model if the p-value from Kruskal–Wallis tests comparing HCs and PD patients was < 0.15. The discriminative ability of these features was then assessed using partial least squares discriminant analysis (PLS-DA), as implemented in the mixOmics package (v6.26.0). PLS-DA is a supervised machine learning method that enables dimensionality reduction, feature selection, and multiclass classification. The PLS-DA model consists of a small number of orthogonal components, each calculated as a weighted sum of the original imaging variables to maximize covariance with the group labels. The weight values (or loadings) of the resulting components indicate the contribution of each feature to group discrimination across different dimensions. PLS-DA is specifically designed to handle high-dimensional and potentially correlated data in small sample settings.32.

Model performance was evaluated on the training dataset using a receiver operating characteristic (ROC) analysis with two components, with the area under the ROC curve (AUC) serving as the evaluation metric. Individual-level classification accuracy was determined based on the predicted class assigned to each participant for each component using the maximum distance criterion.

Associations with clinical features

Spearman’s rank partial correlation analysis, controlling for age and sex, was performed to investigate relationships between imaging metrics and clinical features at the time of the MRI visit, including disease duration, systolic and diastolic blood pressure drops at 3 min, and the cardiovascular subscore of the SCOPA-AUT. P-values were adjusted for multiple correlation tests using the FDR method. Correlations were analyzed across all PD patients as well as within each PD subgroup (PDRBD(+) and PDRBD(−)).

Next, associations between longitudinal changes in systolic and diastolic blood pressure drops (3-min values at 5 years minus baseline) and MRI metrics were assessed using linear regression models, adjusted for age, sex, and baseline blood pressure drops. Associations were reported as standardized regression coefficients (β) with 95% confidence intervals, and p-values from the models were corrected for multiple comparisons across all MRI metric types.

Results

Participants

The final study population included 34 patients with PD and 30 HCs. Following quality check, 8 PD patients were excluded (insufficient image quality due to motion and/or susceptibility artifacts, n = 6; severe cervical lordosis, n = 1; spinal cord signal abnormalities related to cervical spondylotic myelopathy, n=1). Similarly, 8 HCs were excluded (motion and/or susceptibility artifacts, n = 5; severe cervical lordosis, n = 1; disk protrusion, n = 1; spinal cord signal abnormalities, n=1). This resulted in 26 patients with PD, subdivided into PDRBD(+) (n = 11) and PDRBD(−) (n = 15) subgroups, and 22 HCs.

There were no significant differences in age or sex distribution between PD patients and HCs, nor within the PD subgroups. As expected, PD patients had higher MDS-UPDRS III (p < 0.0001) than HCs. RBDSQ scores were significantly higher in the PDRBD(+) subgroup compared to the PDRBD(−) one (p < 0.001). There were no significant differences between PD subgroups in disease duration, MDS-UPDRS III, SCOPA-AUT scores, cardiovascular SCOPA-AUT subscore, presence of orthostatic hypotension, or systolic and diastolic drops at 3 or 5 min (all p > 0.05), although values tended to be higher in the PDRBD(+) subgroup (Table 1).

Table 1 Participants’ demographical and clinical characteristics.
Full size table

Group comparisons

All F-values are reported as Fdf1,df2, where df1 and df2 represent numerator and denominator degrees of freedom. First, our analysis, which covered the spinal cord from C2 to T5 vertebrae, revealed no significant group differences between HCs and all PD patients for any structural MRI metric taken individually. After FDR correction, a trend for higher MTR values in the ventral horn at the cervicothoracic junction (C6-T1) was observed in PD patients compared with HCs, (F1,44 = 6.54, pFDR = 0.06; Fig. 1, supplementary Table S3). When comparing HCs and the two PD subgroups, this trend remained (F2,43 = 4.11, pFDR = 0.09) with PDRBD(+) patients showing higher MTR values at C6-T1 compared to HCs (pFDR = 0.02). No other significant differences were seen after FDR correction for any other MRI metric (Fig. 1, supplementary Table S3).

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

MTR values across vertebral levels in HCs and PD patients. Mean values with shaded areas representing the 95% confidence intervals of Locally Estimated Scatterplot Smoothing (LOESS)-smoothed MTR values (y-axis) across vertebral levels (x-axis) for different regions of interest are shown for HCs, all PD patients, and PD subgroups. The gray shaded rectangle indicates the C6–T1 junction, where group differences in average MRI values were assessed using an F-test adjusted for age and sex. Abbreviations: HC, healthy controls; VH, ventral horns; IZ, intermediate zone; MTR, magnetization transfer ratio; PD; Parkinson’s disease; PDRBD(+), PD with RBD; PDRBD(−), PD without RBD; RBD, Rapid-Eye Movement Sleep Behavior Disorder.

Multivariate analysis

Overall, the PLS-DA model visualization with two components achieved group separation using 26 preselected variables that contributed most to the discrimination (supplementary Table S4).

Component 1 separated PDRBD(−) patients (12/15, 80%, correctly classified) from HCs (17/22, 77.3%) and PDRBD(+) patients (all misclassified as HCs). Component 2 separated HCs (19/22 correctly predicted, 86.4%) from PDRBD(−) (12/15, 80%) and PDRBD(+) (8/11, 72.7%) subgroups. Area under the ROC curve (AUC) values on the training dataset exceeded 0.90 for the discrimination of each group against the others (supplementary Figure S2).

Component 1 was negatively correlated with FA values in the descending tracts (C4, C6, and C7) and with MTR values in the intermediate zone (T3), while showing positive correlations in the descending tracts with T1 relaxation values at C3 and C4, and MD values at C4. Component 2 exhibited negative correlations with MTR values in the ventral horns (C2, C6, and C7) and in the descending tracts (C5 and C6), and positive correlations MD values in the descending tracts (C7) (supplementary Table S5, Fig. 2).

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

Multivariate analysis. Individual (left) and variable (right) plots illustrate the two-component PLS-DA model based on 26 preselected FA, MD, MTR, and T1 relaxation variables across vertebral levels from C2 to T5. The individual plot shows fair separation among the groups of HCs, PDRBD(+), and PDRBD(−). Percentages on the axes represent the variance explained by each component. Colors in the variable plot correspond to variable types: FA (red), MD (blue), MTR (green), and T1 relaxation (purple). Abbreviations: HC, healthy controls; VH: ventral horns; IZ, intermediate zone; MTR, magnetization transfer ratio; PD; Parkinson’s disease; PDRBD(+), PD with RBD; PDRBD(−), PD without RBD ; RBD, Rapid-Eye Movement Sleep Behavior Disorder.

Associations with clinical features

Analyses were performed at the cervicothoracic junction (C6–T1), where group differences showed trends after FDR correction. For the PDRBD(+) subgroup, we found significant positive correlations between systolic drop at 3 min and T1 relaxation values in the ascending (ρ = 0.78, pFDR = 0.03) and descending tracts (ρ = 0.81, pFDR = 0.03). There were also positive correlations between systolic drop at 3 min and MD values in the ascending tracts (ρ = 0.77, pFDR = 0.04) and ventral horns (ρ = 0.75, pFDR = 0.04), with a trend in the descending tracts (ρ = 0.62, pFDR < 0.10) and intermediate zone (ρ = 0.64, p < 0.10). MD values in the descending tracts were significantly correlated with the cardiovascular subscore of the SCOPA-AUT (ρ = 0.75, pFDR = 0.04). No significant correlation was found with disease duration, though MD values in the intermediate zone showed a positive trend (ρ = 0.62, pFDR < 0.10) (Fig. 3). No significant correlations were observed in the entire PD group, nor in the PDRBD(−) subgroup.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

Associations between clinical variables and MRI measurements at the cervicothoracic junction of the spinal cord in the PDRBD(+) subgroup. (A, C) Spearman’s rank partial correlation matrices (rho coefficients) between clinical variables and averaged T1 relaxation (A) and MD (C) values across the C6-T1 vertebral levels in the PDRBD(+) subgroup. Asterisks indicate correlations that were significant before correction for multiple comparisons, while symbols in parentheses indicate those that remain significant after FDR correction. Circles specifically denote non-significant trends (p < 0.10). Significance levels are indicated as follows: p < 0.10 (°), p < 0.05 (*), p < 0.01 (**), and p < 0.001 (***). (B, D) Scatterplots with linear regression lines showing significant correlations (after FDR correction) between systolic blood pressure drop values at 3 min (adjusted for age and sex) and T1 relaxation values in the descending tracts (B), as well as MD values in the ascending tracts (D), also adjusted for age and sex, in the PDRBD(+) subgroup. These correlations were not significant in the PDRBD(−) subgroup. The shaded area around each regression line represents the 95% confidence interval of the fitted model. Marginal density plots along the top and right margins illustrate the distributions of systolic blood pressure drops at 3 min, T1 relaxation values in the descending tracts (B), and MD values in the ascending tracts (D) for each subgroup, respectively. Raw Spearman’s rho coefficients and p-values without correction are indicated on each plot. Abbreviations: FDR, false discovery rate; MD, mean diffusivity; PD; Parkinson’s disease; PDRBD(+), PD with RBD; RBD, Rapid-Eye Movement Sleep Behavior Disorder; SCOPA-AUT, SCales for Outcomes in PArkinson’s disease—Autonomic Dysfunction; T1, longitudinal relaxation time.

Linear models using MRI metrics and baseline blood pressure drop, and age and sex as covariates in the PDRBD(+) subgroup showed that MD values in the descending (standardized β = 0.94, 95% CI [0.42. 1.45], pFDR = 0.02) and ascending (standardized β = 1.21, 95% CI [0.40, 2.02], pFDR = 0.02) tracts were significantly associated with 5-year changes in systolic blood pressure drop. Diastolic changes were also significantly associated with MD values in the ascending (standardized β = 1.08, 95% CI [0.37, 1.79], pFDR = 0.04), with trends observed for MD (standardized β = 0.78, 95% CI [0.11, 1.45], pFDR = 0.06) and T1 relaxation (standardized β = 0.73, 95% CI [0.17, 1.29], pFDR = 0.07) values in the descending tracts (supplementary Table S6). Together, these associations suggest that microstructural alterations in the descending and ascending tracts may be associated with changes in systolic and diastolic blood pressure over time.

Discussion

In this study, we used multimodal MRI to investigate spinal cord involvement in patients with PD in comparison with HCs, and to explore its association cross-sectionally and longitudinally with clinical features of autonomic dysfunction. While group comparisons between PD and HCs, and between PD subgroups, showed no significant differences in MRI metrics, multivariate analysis combining white and gray matter measurements discriminated between PD subgroups and HCs. Within the PDRBD(+) subgroup, we found that blood pressure drops were positively correlated with T1 relaxation and MD values in the ascending and descending tracts, as well as with MD values in the ventral horns and intermediate zone at the cervicothoracic junction, indicating that longer T1 relaxation and MD values were associated with more severe orthostatic hypotension. SCOPA-AUT cardiovascular subscores were positively correlated with MD values in the descending tracts. In addition, longitudinal changes in systolic and diastolic blood pressure drops from baseline to the five-year follow-up were associated with MRI metrics. These findings suggest that subtle microstructural changes in the examined regions may be associated with cardiovascular dysautonomia in PD.

There is neuropathological evidence showing that PD involves not only the brain but the entire nervous system, including the spinal cord and the peripheral nervous system2,8,9. Spinal cord involvement contributes to the occurrence of motor and non-motor symptoms in PD such as autonomic symptoms, constipation, and pain3,4. The presence of α-synuclein inclusions has consistently been reported in the thoracic intermediolateral column and the sacral dorsal horns in PD2,8,9. In the present study, we did not find statistically significant between-group differences in spinal cord MRI metrics, and we do not provide direct evidence of spinal cord involvement. Nevertheless, in line with our hypothesis, the exploratory associations observed between MRI metrics and clinical features of cardiovascular dysautonomia within the PDRBD(+) subgroup were localized to the cervicothoracic junction, specifically the C6-T1 vertebral levels, corresponding to the C7-T2 spinal segments30,31. These spinal segments contain key cardiovascular autonomic centers involved in dysautonomia in PD7. This anatomical concordance provides biological plausability for the observed associations, although they should be interpreted cautiously due to the limited sample size and the absence of neuropathological confirmation.

Furthermore, in our study, no correlations were observed in the entire PD group or in the PDRBD(−) subgroup. This lack of association may align with the hypothesis that spinal cord damage is more prominent in PDRBD(+) patients. Indeed, PDRBD(+) patients are expected to follow an ascending model of disease propagation, characterized by earlier and more severe autonomic dysfunction10,11,12,13, which may mirror why MRI measures in the spinal cord were specifically associated with cardiovascular dysautonomia in this subgroup. Regarding disease propagation, a recent study reported that spinal pathology was only observed in patients already exhibiting Lewy pathology in the brain, with a strong correlation between the amount of spinal cord Lewy pathology and the severity of brain lesions9. Using unsupervised K-means analysis, the authors identified two cluster types of spinal and brain Lewy pathology: a caudo-rostral pattern (consistent with an ascending model of disease propagation) and an amygdala-based pattern (i.e., descending model) Lewy pathology types. Interestingly, the spinal cord Lewy pathology type was more strongly associated with the caudo-rostral-based type than the amygdala-based type, further supporting the hypothesis of two distinct propagation patterns of Lewy pathology9. While our findings remain exploratory and do not allow to infer about disease progression patterns, larger studies combining spinal cord MRI with longitudinal clinical assessments and, where possible, neuropathological data will be needed to validate these observations.

To our knowledge, only one study has investigated spinal cord structural abnormalities at the cervical level (C2–C5) in a cohort of PD patients (n = 68), stratified into early (n = 23), moderate (n = 22) and advanced (n = 23) stages, using diffusion, MTR and T2* metrics. Subtle but significant differences were observed between HC and the advanced PD group for FA in the white matter, as well as between HC and the moderate PD group for radial diffusivity in the white matter, based on average values across C2–C5. No significant associations were observed with UPDRS III scores33. Unlike our study, the authors did not stratify PD patients based on the presence of absence of RBD and restricted the field of view to the C2–C5 vertebral levels, which might not have captured alterations expected to occur preferentially in the upper thoracic cord and sacral regions2,8,9. A resting-state functional MRI (fMRI) study34 conducted on the same cohort of PD patients as in showed a decrease in functional connectivity in the cervical spinal cord, which was associated with upper limb motor symptoms severity between C4 and C6 spinal levels. However, these functional changes did not correlate with microstructural measures. Similarly, a study on transgenic M83 murine models of PD overexpressing the mutated A53T α-synuclein form (n = 22) did not reveal any structural spinal cord abnormalities in comparison with non-transgenic mice (n = 13) while oxygen saturation levels in the spinal cord measured with in vivo spiral volumetric optoacoustic tomography were shown to be reduced35.

Several factors may account for the absence of significant differences in MRI metrics between PD patients and HCs in our study. First, we lacked statistical power given the relatively small sample size of PD patients, further reduced after stratification. Second, spinal cord imaging is highly prone to motion artifacts from heart and respiratory activity or swallowing, despite the use of cardiac gating for diffusion imaging, as well as to susceptibility artefacts, particularly affecting the upper thoracic portion. As a result, almost one-quarter (23.3%) of HCs and one-fifth (20.6%) of PD patients were excluded due to insufficient image quality. To minimize the impact of such confounds, conservative quality control criteria were applied to ensure robust and reliable measurements. While these intrinsic technical challenges of imaging the upper thoracic spinal cord currently limit scalability, future technical improvements in spinal cord imaging will enable replication of our findings in larger cohorts. Third, the effect size of potential spinal cord alterations was small, with subtle microstructural changes that were hard to capture in PD in comparison with other conditions such as amyotrophic lateral sclerosis15 or spinal-muscular amyotrophy14. Stratifying PD patients based on the presence of orthostatic hypotension would have been interesting. However, the sample size of patients with this feature was too small (5/26). Finally, another limitation was the lack of objective assessment of cardiovascular autonomic function using formal autonomic testing.

To conclude, these preliminary findings suggest possible region-specific associations between structural metrics and cardiovascular dysautonomic features in PD patients with RBD. Correlations at the C6–T1 vertebral levels may support the potential role for cervicothoracic spinal cord alterations in the pathophysiology of autonomic failure in PD despite the lack of significant group-level structural differences. These results require validation and replication in larger study samples to determine whether spinal cord imaging markers can be used as surrogates for autonomic dysfunction in PD. Future studies will include individuals with isolated RBD and incorporate additional measures of cardiovascular function, such as the RR interval. Recent advances in analysis methods, notably rootlet-based instead of vertebral-based analyses30,36, might provide more sensitivity to the MRI metrics. Technological advances such as ultra–high-field MRI may improve spatial resolution and signal-to-noise ratio, enabling more sensitive detection of subtle spinal cord changes in PD. Furthermore, resting-state fMRI can reveal network-level alterations, with connectivity changes reflecting PD-related pathology. Combining structural and functional measures may enable the detection of subtle microstructural alterations associated with functional changes, offering a more comprehensive understanding of central nervous systems changes in PD.