Distinct Longitudinal Clinical‐Neuroanatomical Trajectories in Parkinson's Disease Clinical Subtypes: Insight toward Precision Medicine

ABSTRACT

Background

Parkinson's disease (PD) varies widely across individuals in clinical manifestations and course of progression. Identification of distinct biological subtypes could explain this heterogeneity, identify its pathophysiology, and predict disease progression.

Objectives

Our aim was to compare longitudinal clinical trajectories and brain atrophy patterns between clinical subtypes defined at the baseline de novo PD.

Methods

We analyzed data from 421 PD patients (mean follow‐up: 8.2 years) in the Parkinson's Progression Markers Initiative (PPMI). Using multi‐domain motor and non‐motor criteria, de novo patients were classified into “mild motor‐predominant” (n = 223), “intermediate” (n = 146), and “diffuse‐malignant” (n = 52) subtypes. Deformation‐based morphometry was performed on T1‐weighted magnetic resonance imaging (MRIs) from 128 PD patients with at least two MRIs (71 mild motor‐predominant, 42 intermediate, and 15 diffuse‐malignant) and 60 controls, with an average MRI follow‐up duration of 3.4 ± 1.1 in the PD cohort. Mixed‐effects models compared clinical progression and longitudinal pattern of regional atrophy across subtypes.

Results

The diffuse‐malignant subtype exhibited faster worsening of motor severity (P = 0.007), cognition (P < 0.0001), and activities of daily living (P < 0.0001) compared to mild motor‐predominant subtype over 8 years. These findings remained statistically significant after an age‐matched subgroup analysis and adjustment for the levodopa treatment. Accelerated atrophy was observed in the precuneus, temporal and fusiform gyri, cerebellum, and other regions (corrected‐P < 0.05).

Conclusions

Longitudinal analysis revealed distinct patterns of clinical progression and regional atrophy in PD subtypes, with the diffuse‐malignant subtype showing more severe neurodegeneration and clinical deterioration suggesting existence of diverse pathophysiological mechanisms in PD. © 2025 The Author(s). Movement Disorders published by Wiley Periodicals LLC on behalf of International Parkinson and Movement Disorder Society.

Parkinson's disease (PD) varies widely from person to person in terms of clinical manifestations and progression. Although the clinical and biological signature of PD may be unique to the individual, identification of commonalities characterizing disease subtypes can be one step forward toward precision medicine. ¹ , ² Biomarker characterization of distinct subtypes can parse this heterogeneity, shed light on the underlying pathophysiology, and predict disease progression across the PD spectrum. We previously introduced and validated a data‐driven approach to define three subtypes of idiopathic PD using multi‐domain clinical criteria. ³ , ⁴ According to this classification system, early development of three key non‐motor features—mild cognitive impairment, orthostatic hypotension, and rapid eye movement (REM) sleep behavior disorder (RBD)—at the drug‐naïve stage defines a distinct subtype of idiopathic PD called the “diffuse‐malignant” subtype with the most rapid disease progression. ⁴ Several independent studies also demonstrated faster disease progression, earlier development of dementia, and shorter survival in the diffuse‐malignant subtype. ⁵ , ⁶ Nevertheless, there is a dearth of knowledge on the long‐term clinical trajectories, as well as underlying pathophysiological causes of faster disease progression in a subgroup of people with PD. Using cross‐sectional data from diffusion‐weighted imaging, more prominent disruption in various cortico‐subcortical connectivity networks was found in the diffuse‐malignant subtype. ⁷ Another cross‐sectional study demonstrated further disease‐related brain atrophy at baseline in the diffuse‐malignant subtype using deformation‐based morphometry (DBM) of T1‐weighted magnetic resonance imaging (MRI). ⁸ A longitudinal clinical and brain imaging study can monitor clinical trajectories and pathological changes in brain structure over time, offering insights into the dynamics of atrophy for each PD subtype that cross‐sectional imaging cannot capture. Therefore, the aim of this study was to (1) compare long‐term clinical trajectories of the main motor and non‐motor features between PD subtypes defined at the early untreated stage; (2) use longitudinally acquired brain MRIs to investigate and compare different patterns of brain atrophy between the PD subtypes; and (3) to compare the within and between subtype neuroanatomical features to establish convergence and divergence across their trajectories.

Patients and Methods

Study Population

The Parkinson's Progression Markers Initiative (PPMI) is a multicenter longitudinal cohort of individuals with early idiopathic PD, untreated at recruitment ⁹ (detailed information at http://www.ppmi-info.org). The inclusion criteria are: age at least 30 years old, diagnosed with PD within the past 2 years, at least two signs or symptoms of parkinsonism (resting tremor, bradykinesia, and/or rigidity), a baseline Hoehn and Yahr stage of I or II, and does not require symptomatic treatment within 6 months of baseline. The PPMI protocol was approved by the institutional review board at all participating locations. Before including participants, written informed consent was obtained from all individuals. Scanning sites acquired high resolution 3D T1‐weighted sequences (magnetization prepared rapid gradient echo or spoiled gradient‐recalled echo) that were standardized based on harmonized Alzheimer's Disease Neuroimaging Initiative (ADNI) imaging protocols, with a field of view (FOV) covering the cortex, cerebellum, and pons. Scanning parameters remained consistent across longitudinal time points, https://www.ppmi-info.org/sites/default/files/docs/archives/PPMI-MRI-Operations-Manual-V7.pdf. For the purpose of this study, we obtained the T1‐weighted MRIs from participants in whom at least two MRIs were obtained.

Clinical Measures

The PPMI study conducted a thorough evaluation of motor and non‐motor manifestations at the screening, baseline, and follow‐up visit. For this study, we analyzed trajectories of the following clinical motor and non‐motor disease domains over time:

1. Motor: sum of the severity scores for motor signs on the Movement Disorder Society‐Unified Parkinson's Disease Rating Scale (MDS‐UPDRS) part III. ¹⁰

2. Cognition: global cognitive status measured by the Montreal Cognitive Assessment (MoCA) score. ¹¹

3. Daily activities and functioning: total score of the Schwab and England activities of daily living (SE‐ADL) score that measures functional independence in PD patients on a scale from 0 (complete dependence) to 100 (complete independence). ¹²

Subtyping Rules

Participants were classified into three clinical subtypes at the baseline early untreated stage using our previously validated multi‐domain subtyping criteria based on major motor (MDS‐UPDRS‐II and III) and non‐motor classifiers (early cognitive impairment, RBD, and dysautonomia). ⁴ The following definitions were used to assign individuals to their clinical subtype, subtype I (mild motor‐predominant): both composite motor score and all non‐motor summary scores (NMS) below the 75th percentile; subtype III (diffuse malignant): (composite motor score AND either >1 of 3 non‐motor scores >75th percentile, or all 3 non‐motor scores >75th percentile); and subtype II (intermediate): those individuals who do not meet criteria for subtype I or III. ³ , ⁴ In PPMI, RBD has been defined using the RBD screening questionnaire (RBDSQ) score. ¹³

Brain Imaging

We used T1‐weighted MRI to evaluate the atrophy pattern in PD and control subjects over time using DBM as a measure of local brain atrophy. Briefly, all MRI images underwent preprocessing steps including denoising, ¹⁴ non‐uniformity correction, ¹⁵ and intensity normalization using histogram matching followed by brain extraction step. ¹⁶ Longitudinal registration was then performed to maximize sensitivity to longitudinal atrophy. More specifically, for each participant, following iterative linear ¹⁷ and nonlinear ¹⁸ registrations, subject‐specific templates were created based on all available longitudinal visits. The subject‐specific templates were then nonlinearly registered to the MNI‐ICBM‐2009c symmetrical template. ¹⁹ Time point to subject‐specific template and subject‐specific template to MNI‐ICBM‐2009c nonlinear transformations were concatenated to generate time point to MNI‐ICBM‐2009c nonlinear transformations. Voxel‐wise DBM maps were generated by calculating the determinant of the Jacobian based on these nonlinear transformations. ²⁰ , ²¹ , ²² Our open source longitudinal image processing pipeline is available at https://github.com/VANDAlab/Preprocessing_Pipeline and has been used and validated in multiple studies. ²³ , ²⁴ , ²⁵ , ²⁶ All preprocessed MRIs, brain extraction, linear, and nonlinear registration steps were visually quality controlled. ¹⁷ , ²⁷ Because of sample size limitations, all MRI analyses were performed at a regional level to improve statistical power. Future studies with larger sample sizes and greater statistical power can explore voxel‐level analyses to potentially reveal more subtle and localized patterns of change. Voxel‐wise DBM maps were averaged across 282 cortical and subcortical regions using the neuroanatomically defined regions by the Allen Institute for brain sciences' atlas (https://community.brain-map.org/t/allen-human-reference-atlas-3d-2020-new/405) ²⁸ for each visit, providing us with the regional atrophy measures used in our statistical analysis.

Neuroanatomical Heterogeneity

To examine the degree of within group and between group neuroanatomical heterogeneity, we used the regional DBM values explained in the previous section (ie, 282 regions). We used the Spearman correlation between regional DBM values as a measure for similarity between two MRIs across all available subjects and visits with the baseline visits of participants (including controls) as the reference, excluding the same subject correlations over time to remove inflation because of dependency. For within‐group similarity, we calculated the mean Fisher transformed correlation values between each visit of a given subject and all subjects within the same subtype at baseline. Similarly, for between‐group similarity and to test whether the pattern of brain atrophy in mild motor‐predominant and intermediate subtypes would converge over time to that of the diffuse‐malignant subtype at baseline as well as the follow‐up visits (ie, significant increase in similarity to diffuse‐malignant baseline compared to similarity to other groups). To do so, we used the mean Fisher transformed correlation values between each visit of a given subject and all subjects across the diffuse‐malignant subtype at baseline and the follow‐up visits.

Statistical Analysis

For comparison of the demographics and baseline clinical features between the subtypes, we used one‐way analysis of variance or χ² test. Next, we applied mixed effect models to examine clinical trajectories between PD subtypes over 8 years, using the following model:

1. Clinical measure ~1 + age at baseline + sex + subtype * years follow‐up + levodopa (l‐dopa) equivalent dose + (1| ID)

• 2Similarity index ~1 + age at baseline + sex + subtype * years follow‐up + l‐dopa equivalent dose + (1| ID)

with subtype and its interaction with follow‐up time as the main effect of interest, controlling for sex and age. The model evaluated longitudinal changes in MDS‐UPDRS‐III, MoCA, and SE‐ ADL scores as our clinical measures of interest. Similarly, we used mixed effect models to examine (1) the atrophy pattern between each subtypes and healthy control subjects; and (2) the interaction between PD subtypes and the rate of atrophy across brain regions over time, controlling for sex, age, and medication as measured by the l‐dopa equivalent dose. Similarly, for neuroanatomical heterogeneity we used a mixed effect model using the following model:

Furthermore, to ensure that age differences at the baseline does not confound the results, we repeated the analyses using subsets of the participants from mild motor‐predominant and intermediate subtypes that were matched with the diffuse‐malignant subtype based on age at the baseline visit. Similarly, we repeated the analyses, adding scanner model as a categorical random effect to ensure potential scanner‐related variabilities do not impact the results.

We used MATLAB R2023b (The MathWorks, Natick, MA) for all statistical procedures and a P‐value less than 0.05 was considered statistically significant. In cases of multiple comparisons, the results are reported after false discovery rate (FDR) correction, with an FDR corrected P‐value of less than 0.05 considered as statistically significant.

Results

Demographics and Baseline Features

Data from 421 PD subjects (65.6% male) were used for clinical trajectory assessment. At baseline, 223 participants were categorized as mild motor‐predominant, 146 as intermediate, and 52 as diffuse‐malignant subtype. The average follow‐up duration was 8.2 ± 3.7 years, and 6481 total visits were included. Follow‐up duration was shorter in individuals with diffuse‐malignant subtype (6.7 ± 3.4 year vs. 9.0 ± 3.1 year in mild motor‐predominant and 7.8 ± 3.8 year in intermediate subtypes; P < 0.001). Table 1 summarizes demographics, motor, and non‐motor features within each PD subtype at baseline. As expected, and derived from the subtype definition, various motor and non‐motor features were significantly more impaired at baseline in the diffuse‐malignant subtype.

TABLE 1.

Demographics and baseline clinical features in individuals with different subtypes of Parkinson's disease in the PPMI cohort (n = 421)

Characteristic	Subtype I mild motor‐predominant (n = 223)	Subtype II intermediate (n = 146)	Subtype III diffuse malignant (n = 52)	ANOVA/χ2 P‐value
Age at onset (y)	60.0 (9.7)	62.2 (9.6)	62.8 (9.6)	0.038.
Sex, male (%)	140 (62.8%)	98 (67.1%)	38 (73.1%)	0.329
Race, White (%)	215 (96.4%)	135 (92.5%)	49 (94.2%)	0.331
Education history (y)	15.7 (3.0)	15.4 (2.9)	15.4 (3.2)	0.514
Symptoms duration (mo)	6.6 (6.7)	6.2 (6.2)	7.4 (7.0)	0.531
Positive family history (%)	29 (13.1%)	18 (12.5%)	8 (15.4%)	0.869
MDS‐UPDRS‐part I	4.0 (3.0)	6.5 (3.6)	9.9 (5.3)	<0.001
MDS‐UPDRS‐part II	4.0 (2.7)	6.9 (3.9)	11.8 (4.3)	<0.001
MDS‐UPDRS‐part III	18.3 (7.5)	21.9 (8.9)	30.0 (9.2)	<0.001
UPDRS‐total score	26.4 (9.4)	35.2 (11.8)	51.7 (11.3)	<0.001
Schwab & England score	94.7 (5.0)	92.2 (6.2)	88.9 (6.5)	<0.001
SCOPA‐AUT (total score)	6.6 (3.4)	11.7 (6.5)	16.5 (6.6)	<0.001
MoCA (adjusted score)	27.5 (2.1)	26.8 (2.5)	26.7 (2.5)	0.005

Abbreviations: PPMI, Parkinson's Progression Markers Initiative; ANOVA, analysis of variance; MDS‐UPDRS, Movement Disorder Society‐Unified Parkinson's Disease Rating Scale; UPDRS, Unified Parkinson's Disease Rating Scale; SCOPA‐AUT, Scale for Outcomes in Parkinson's disease for Autonomic symptoms; MOCA, Montreal Cognitive Assessment.

Clinical Trajectory

Mixed‐effects models demonstrated important differences in the longitudinal trajectories of various outcomes (Fig. 1). As listed in Table 2, compared to the mild motor‐predominant subtype, individuals who were initially categorized as intermediate or diffuse‐malignant subtypes at baseline experienced faster progression in motor severity (higher values indicate greater severity), with 0.37 (95% CI: 0.21–0.53, P < 0.0001) and 0.41 (95% CI: 0.11–0.71, P = 0.0070) units additional annual increase in MDS‐UPDRS‐III score. A more rapid cognitive decline was also demonstrated in the diffuse‐malignant and intermediate subtypes. After adjusting for baseline MoCA score and age, members of the intermediate and diffuse‐malignant subtype demonstrated −0.12 (95% CI: −0.17 to −0.07, P < 0.0001) and −0.41 (95% CI: −0.51 to −0.32, P < 0.0001) additional decline in MoCA score per year when compared to the mild motor‐predominant subtype (lower values indicate greater severity). A distinct trajectory was also noted in the overall impact of PD on independence in ADL. Individuals categorized as the intermediate or diffuse‐malignant subtypes at baseline had a significantly faster decline in ADL compared to the mild motor‐predominant subtype, with −0.59 (95% CI: −0.72 to −0.47, P < 0.0001) and −0.92 (95% CI: −1.14 to −0.70, P < 0.0001) percent further decline in SE‐ADL score per year, respectively (higher values indicate greater severity). As listed in Table 2, following adjustments for levodopa equivalent daily dose (LEDD), these differences in the trajectories of all clinical outcomes remained statistically significant, some even with an increased effect size (adjusted effect sizes).

FIG. 1.

Distinct clinical trajectories in (A) motor severity: Unified Parkinson's Disease Rating Scale (UPDRS)‐III score, (B) cognition: Montreal Cognitive Assessment (MOCA) score, (C) dysautonomia: SCOPA total score and (D) activities of daily living: Schwab and England activities of daily living (ADL) score between clinical subtypes of Parkinson's disease (PD) after 8 to 10 years of follow‐up (all mixed‐effect models P‐value > 0.001). [Color figure can be viewed at wileyonlinelibrary.com]

TABLE 2.

Mixed‐effects models to assess longitudinal trajectories of various clinical outcomes after >8 years of follow‐up in individuals with different subtypes of Parkinson's disease in the PPMI cohort (n = 412: mild motor‐predominant = 220, intermediate = 142, diffuse‐malignant = 50) adjusted for baseline age (Additional adjusted mixed‐effect models were performed by adding L‐dopa equivalent daily dose (LEDD) as another covariate)

	Age‐adjusted effect size	P‐value	LEDD‐adjusted effect size	LEDD‐adjusted P‐value
Outcome: motor severity (MDS‐UPDRS‐III)
Age effect	0.20 (0.12–0.29)	0.0007	0.20 (0.12–0.29)	0.0007
Group*time interaction
Subtype‐I: mild motor‐predominant	Reference	–	Reference	–
Subtype‐II: intermediate	0.37 (0.21–0.53)	<0.0001	0.38 (0.21–0.54)	<0.001
Subtype‐III: diffuse malignant	0.41 (0.11–0.71)	0.007	0.41 (0.12–0.71)	0.006
Outcome: cognition (MoCA)
Age effect	−0.11 (−0.13 to −0.08)	<0.0001	−0.08 (−0.10 to −0.06)	<0.001
Group*time interaction
Subtype‐I: mild motor‐predominant	Reference	–	Reference	–
Subtype‐II: intermediate	−0.12 (−0.17 to −0.07)	<0.0001	−0.12 (−0.17 to −0.07)	<0.0001
Subtype‐III: diffuse malignant	−0.41 (−0.51 to −0.32)	<0.0001	−0.41 (−0.50 to −0.31)	<0.0001
Outcome: daily activities (Schwab and England‐ADL)
Age effect	−0.12 (−0.18 to −0.06)	0.0002	−0.12 (−0.18 to −0.06)	<0.001
Group*time interaction
Subtype‐I: mild motor‐predominant	Reference	–	Reference	–
Subtype‐II: intermediate	−0.60 (−0.72 to −0.47)	<0.0001	−0.58 (−0.71 to −0.45)	<0.0001
Subtype‐III: diffuse malignant	−0.92 (−1.14 to −0.70)	<0.0001	−0.94 (−1.16 to −0.71)	<0.0001

Notes: Adjusted for baseline age (additional adjusted mixed‐effect models were performed by adding LEDD as another covariate)

Abbreviations: PPMI, Parkinson's Progression Markers Initiative; LEDD, levodopa equivalent daily dose; MDS‐UPDRS, Movement Disorder Society‐Unified Parkinson's Disease Rating Scale part III; MOCA, Montreal Cognitive Assessment ADL: activities of daily living.

Age‐Matched Subgroup Analysis

Other than statistical adjustment for age, we also performed a subgroup analysis by randomly selecting a subpopulation of mild motor‐predominant (n = 135, mean age = 62.9 ± 8.6 year) and intermediate (n = 108, mean age = 64.1 ± 8.4 year) subtypes age‐matched with the diffuse malignant group (n = 50, mean age = 63.7 ± 9.4 year) (P > 0.3). As shown in Table 3, individuals with diffuse‐malignant and intermediate subtypes revealed faster progression in all three outcomes of interest even after matching for age and adjustment for LEDD. Compared to the mild motor‐predominant subtype, the intermediate and diffuse‐malignant subtypes demonstrated significantly faster annual decline in motor severity with additional annual increases of 0.38 (95% CI: 0.19 to 0.58, P < 0.001) and 0.36 (95% CI: 0.05 to 0.67, P = 0.023) units in MDS‐UPDRS‐III score. Cognitive decline was also greater, with additional annual decreases in MoCA scores of −0.11 (95% CI: −0.18 to −0.05, P = 0.001) and −0.37 (95% CI: −0.47 to −0.27, P < 0.001). Similarly, SE‐ADL declined more rapidly, with additional annual reductions of −0.48 (95% CI: −0.64 to −0.32, P < 0.001) and −0.79 (95% CI: −1.03 to −0.55, P < 0.001) percent in the intermediate and diffuse‐malignant subtypes, respectively.

TABLE 3.

Mixed‐effects models to assess longitudinal trajectories of various clinical outcomes after >8 years of follow‐up in individuals with different subtypes of Parkinson's disease in the PPMI cohort age‐matched at baseline (n = 293: mild motor‐predominant = 135, intermediate = 108, diffuse‐malignant = 50) (Adjusted mixed‐effect models were performed by adding L‐dopa equivalent daily dose (LEDD) as a covariate)

	Age‐adjusted effect size	P‐value	LEDD‐adjusted effect size	P‐value
Outcome: motor severity (MDS‐UPDRS‐III)
Age effect	0.21 (0.09–0.32)	<0.001	0.21 (0.09–0.32)	<0.001
Group*Time interaction
Subtype‐I: mild motor‐predominant	Reference	–	Reference	–
Subtype‐II: intermediate	0.39 (0.19–0.59)	<0.001	0.38 (0.19–0.58)	<0.001
Subtype‐III: diffuse malignant	0.36 (0.05–0.67)	0.022	0.36 (0.05–0.67)	0.023
Outcome: cognition (MoCA)
Age Effect	−0.09 (−0.12 to −0.06)	<0.001	−0.09 (−0.12 to −0.06)	<0.001
Group*Time interaction
Subtype‐I: mild motor‐predominant	Reference	–	Reference	–
Subtype‐II: intermediate	−0.11 (−0.18 to −0.05)	0.001	−0.11 (−0.18 to −0.05)	0.001
Subtype‐III: diffuse malignant	−0.37 (−0.47 to −0.27)	<0.001	−0.37 (−0.47 to −0.27)	<0.001
Outcome: daily activities (Schwab and England‐ADL)
Age effect	−0.15 (−0.23 to −0.06)	0.001	−0.15 (−0.24 to −0.07)	0.001
Group*time interaction
Subtype‐I: mild motor‐predominant	Reference	–	Reference	–
Subtype‐II: intermediate	−0.50 (−0.66 to −0.34)	<0.001	−0.48 (−0.64 to −0.32)	<0.001
Subtype‐III: diffuse malignant	−0.77 (−1.01 to −0.53)	<0.001	−0.79 (−1.03 to −0.55)	<0.001

Notes: Adjusted mixed‐effect models were performed by adding LEDD as a covariate

Abbreviations: PPMI, Parkinson's Progression Markers Initiative; LEDD, levodopa equivalent daily dose; MDS‐UPDRS, Movement Disorder Society‐Unified Parkinson's Disease Rating Scale part III; MOCA, Montreal Cognitive Assessment ADL: activities of daily living.

Brain Atrophy Trajectory (Complementary Analysis)

For the neuroanatomical assessment, longitudinal MRI studies were available for a total of 194 participants, including 128 PD participants and 60 healthy controls comprising a total of 606 visits. Among the PPMI subpopulation included in the longitudinal MRI analysis, the sample size for each clinical subtype was as follows: mild motor‐predominant (n = 71), intermediate (n = 42), and diffuse‐malignant (n = 15). The average time between the first and the last brain MRIs was 3.4 ± 1.1 years in the PD cohort and 1.9 ± 1.3 years in healthy controls (see Supplementary Table S1 for demographic information).

All three groups showed significant atrophy over time in the midbrain including substantia nigra and red nucleus as well as the amygdala in comparison with healthy controls. In the direct comparison between subtypes, individuals with diffuse‐malignant PD showed a significantly higher rate of atrophy in comparison to mild motor‐predominant subtype across multiple brain regions (Fig. 2, blue colored), including precuneus as well as inferior temporal gyrus regions (ie, hippocampal area) implicated in cognitive decline and dementias. We also found accelerated atrophy in the occipital regions of the cortex and cerebellar hemispheres in the diffuse‐malignant subtype. Finally, we found accelerated enlargement of ventricles in the diffuse‐malignant subtype compared to mild motor‐predominant subtype. In contrast, we found bilateral further interval atrophy in the body of caudate in the mild motor‐predominant compared to the diffuse‐malignant group (Fig. 2, red colored). The intermediate subtype showed an accelerated atrophy only in the putamen region in comparison with the mild motor‐predominant subtype after correction for multiple comparisons. Similar results were obtained when adding the scanner model as a categorical random effect (Supplementary Fig. S1).

FIG. 2.

Significantly widespread accelerated atrophy was observed in the diffuse‐malignant subtype compared with the mild motor‐predominant subtype in cortical regions (blue) and less steep atrophy rate in the case of caudate, or a more positive ventricular enlargement (warm colors). These findings are based on the mixed effect modeling using interaction between subtype and time from baseline as main effect of interest with the following model. Regional atrophy ~1 + age at baseline + sex + subtype * years follow‐up + (1| ID). The results are reported after false discovery rate (FDR) correction with an FDR corrected P‐value less than 0.05 considered as statistically significant. It is worth noting that temporal gyrus will not reach the significant level after adding the levodopa equivalent dose as covariate to the model. [Color figure can be viewed at wileyonlinelibrary.com]

Within and between Subtypes Neuroanatomical Heterogeneity

As illustrated in Figure 3A, based on the linear mixed effect model, the largest within subtype similarity in the pattern of brain atrophy was seen in the diffuse‐malignant subtype with a statistically significant correlation between DBM values across brain regions (β = 0.07, t = 3.67, P = 0.00026). As illustrated in Figure 3B, after 3.4 years of follow‐up, the increase in the correlation coefficient between the overall pattern of longitudinal brain atrophy in the mild motor‐predominant and intermediate subtypes and the baseline pattern of brain atrophy in the diffuse‐malignant subtype was not significantly different from the increase observed in the correlation coefficient over time between healthy controls and the baseline pattern of brain atrophy in diffuse malignant subtype (P = 0.625 and P = 0.411, respectively). However, subjects with diffuse‐malignant subtype demonstrated an increase in overall correlation coefficient with their own baseline regional DBM values over time, which was significantly faster than that of the healthy controls (P = 0.0203).

FIG. 3.

Spearman correlations and mixed effect models assessment of within‐group (self‐similarity) and between‐group (similarity to baseline diffuse‐malignant pattern) neuroanatomical heterogeneity. In self‐similarity analysis, based on the mixed effect model, members of diffuse‐malignant subtype had the highest baseline within‐group similarity in regional DBM values of brain atrophy (β=0.07, t=3.67, p=0.00026). Over 4‐years of follow‐up, the progression of brain atrophy in subjects with mild motor‐predominant and intermediate subtypes though trending towards that of the diffuse‐malignant subtype at baseline, was not significantly different compared to healthy controls (‘normal aging’) (β=0.000, t=−0.48, p=0.62, and β=−0.001, t=−0.82, p=0.411, respectively). On the other hand, members of the diffuse‐malignant subtype showed a statistically significant increase in the correlation coefficient of diffuse malignant similarity over time when compared to the trend seen in healthy controls (p=0.0203). [Color figure can be viewed at wileyonlinelibrary.com]

Discussion

Investigating the longitudinal trajectories of clinical and neuroanatomical changes in a sample of de novo PD patients at baseline, our study revealed a distinct pattern of disease progression across subtypes of individuals with idiopathic PD. Participants initially subtyped as diffuse‐malignant at the early stage, when dopaminergic treatment was not yet deemed necessary, had a significantly faster clinical progression in motor and non‐motor domains during long‐term follow‐up. We discovered a hierarchical trend in the slope of progression in clinical measures between the subtypes after >8 years of follow‐up. Specifically, individuals classified under the diffuse‐malignant subtype displayed a faster decline in cognitive function and a more rapid progression of motor severity compared to those in the mild motor‐predominant subtype. These participants also exhibited a faster decline in activities of daily living. Similarly, members of the intermediate subtype demonstrated rates of decline falling between those of the “mild motor‐predominant” and diffuse‐malignant subtypes. Although differences in clinical measures exist at baseline, which are inherent to the multi‐domain definition of the subtypes, these differences are augmented in longitudinal trajectories over follow‐up with distinctive slopes of progression that remained statistically significant after adjustment for the effect of aging and baseline clinical differences. In our subsequent subgroup analysis, the divergence in clinical trajectories of all outcomes of interest remained robustly significant even after age‐matching of the clinical subtypes based on at baseline. In fact, the most eccentric variation in longitudinal trajectory was observed in SE‐ADL, a global measure that is not used to define subtyping membership at the early untreated stage. Of note, the follow‐up duration was shorter in the diffuse malignant subtype. Yet, this group still exhibited the worst outcomes and fastest progression. This further supports the rapidly progressive nature of this subtype, as significant progression was observed even over a relatively shorter follow‐up period. Our findings are consistent with those of a few other longitudinal studies reporting a more aggressive disease trajectory in individuals with PD with severe motor and non‐motor phenotypes at baseline and an older age of onset. ²⁹ , ³⁰

Using longitudinal brain MRI scans, we demonstrated an accelerated atrophy pattern within several cortical regions and cerebellum in participants who were initially classified as having the diffuse‐malignant PD subtype at the drug‐naïve stage. This faster rate of brain atrophy in the diffuse‐malignant subtype was evident even after a short follow‐up period (~3.5 years). In line with our findings, another study showed a more pronounced gray matter thinning in parahippocampal and inferior temporal gyri in individuals with early‐stage PD who concurrently had severe olfactory dysfunction and RBD. ³¹ Interestingly, some of the MRI regions with more atrophy in the diffuse‐malignant subtype of idiopathic PD overlap with those of regions with cholinergic terminal loss in dementia with Lewy bodies (DLB) as shown by positron emission tomography (PET) imaging ³² , ³³ These findings suggest the presence of a more diffuse and multi‐domain neurodegenerative process in a subgroup of people with PD, the diffuse‐malignant subtype, with more expanded underlying pathophysiology, probably with more prominent multi‐proteinopathies and earlier multi‐neurotransmitter system involvement. A postmortem study demonstrated evidence for different rates of progression of Lewy pathology and Alzheimer's‐related pathology between PD subtypes using our multi‐domain clinical criteria for subtyping, highlighting the role of multi‐proteinopathies in people with PD with a faster disease progression. ⁵ Notably, our analysis revealed significant longitudinal atrophy in the red nucleus in all PD subtypes. Although a recent cross‐sectional study did not find any evidence of synaptic loss in the red nucleus in early PD ³⁴ their findings are not necessarily contradictory, as they use a different imaging modality (PET), which offers lower resolution compared to MRI. Furthermore, longitudinal studies like ours are better equipped to detect subtle, gradual changes over time, particularly when supported by a larger sample size and greater statistical power.

We found that participants with mild motor‐predominant subtype showed further interval atrophy in the basal ganglia, likely indicating ongoing dopaminergic loss in the early stage, whereas in the diffuse‐malignant subtype, the majority of dopaminergic cells are already damaged even in the early years as previously shown by more severe changes in baseline DAT scans in this PD subtype, ⁴ suggesting a saturation effect. Nevertheless, the choice and sensitivity of the imaging modality are important determinants to depict between‐subtype or between‐stage differences in basal ganglia and/or substantia nigra integrity. Our group and others have previously demonstrated the sensitivity of T1‐weighted MRI for longitudinal disease progression, ²¹ , ³⁵ , ³⁶ , ³⁷ however, it was used to differentiate the progression patterns across PD subtypes. Using T1‐weighted MRI and neuromelanin‐sensitive imaging, a previous study showed ongoing longitudinal change in volume and signal intensity of substantia nigra in both early‐ and advanced‐PD patients after 2 to 3 years of follow‐up. ³⁸ In a review by Mitchell et al, ³⁹ the importance of selecting appropriate imaging biomarkers based on disease stage in clinical trials was emphasized, with dopaminergic imaging and other molecular modalities being effective for tracking progression in preclinical and prodromal stages, and advanced stages benefiting from MRI based imaging including structural T1‐weighted, diffusion based free‐water measure, ⁴⁰ and neuromelanin‐sensitive imaging. ⁴¹ Inguanzo et al ⁴² applied cross‐sectional clustering analysis on structural brain MRI to identify PD subtypes based on various patterns of brain atrophy in cortical and subcortical regions. Their MRI analysis emerged three distinct subtypes with a gradient of neurodegeneration and symptom severity. However, in our study, we defined the subtypes solely based on clinical features during the early stage of PD. Later on, we used longitudinal structural MRIs to compare the trajectories of brain atrophy as a post hoc comparison between the clinical subtypes. In line with these findings, the current study building on our previous findings, ⁴ , ⁸ , ²⁰ , ²¹ , ²² , ³⁵ demonstrates the applicability and sensitivity of DBM MRI analysis for not only tracking disease progression but also differentiating between subtypes of PD. It is worth mentioning that because of the limited MRI sample size especially in the case of the diffuse malignant subtype, the MRI‐based analysis was performed as a complementary analysis and future studies are needed to confirm these reported effects in larger cohorts.

We then performed a similarity analysis to test the hypothesis whether these subtypes are pathologically distinct, or whether they represent the same pathological spectrum with the main difference being only in the speed of progression. Our findings demonstrated that the trajectory in overall pattern of brain atrophy toward that of the diffuse‐malignant subtype was no greater in mild motor‐predominant and intermediate subtypes after 3.5‐years of follow‐up than in healthy controls. Interestingly, only subjects with the diffuse‐malignant subtype showed a significantly increasing within‐group convergence over time. Other subtypes' similarity with the diffuse malignant group did not significantly change over time, which suggests that baseline level differences also persist over time. Taken together, these results support the presence of a distinct neuropathological pattern of progression, at least in a subgroup of people with more aggressive PD.

In recent years, there has been a surge in attempts to biologically define PD, moving beyond traditional clinical definitions. As highlighted by advances in biomarker research, the detection of pathological α‐synuclein in cerebrospinal fluid (CSF) through seed amplification assays marks a pivotal step toward the understanding of the pathophysiology of PD and DLB. ⁴³ , ⁴⁴ , ⁴⁵ Our findings of clinical and neuroimaging‐based differences between subtypes provide complementary insights to further understand the heterogeneity in PD using a neuroanatomical biomarker. Evidence points to various pathological pathways, such as co‐proteinopathies, which contribute to the complexity of PD in different individuals. ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ Understanding both disease staging and subtyping, rooted in diverse biological mechanisms, is essential for advancing personalized medicine and targeted therapeutic approaches.

Our study has several strengths that contribute to its robustness and clinical relevance. First, our analysis features one of the longest follow‐up periods for comparing clinical trajectories between PD subtypes with >8 years of annual assessment, allowing for a greater understanding of disease progression within each subtype over time. Additionally, this is one the few studies to investigate longitudinal changes in brain atrophy using MRI, providing valuable insights into the structural alterations occurring in different PD subtypes. Methodologically, our study uses statistical models that enable us to adjust for confounding factors such as the aging effect and baseline clinical differences among participants with various PD subtypes. Moreover, our findings provide further evidence to support the biological and prognostic relevance of our multi‐domain clinical subtyping method for PD, offering a practical framework that clinicians can use in office settings to assign individuals to distinct subtypes with prognostic implications. This proof of concept enhances the clinical utility of our findings, paving the way for personalized management strategies tailored to the specific needs of PD patients. Our image processing pipeline has been developed and validated for use in multicenter and multi‐scanner studies, and has previously been used in numerous similar applications. ⁵⁰ , ⁵¹ , ⁵² , ⁵³ , ⁵⁴ , ⁵⁵ , ⁵⁶ Furthermore, in our previous work, we compared the DBM values across scanner types and found no significant scanner related differences. ⁸ , ²⁰ Similarly, we repeated the longitudinal MRI analyses with scanner model as a random effect, obtaining consistent results following FDR correction.

We acknowledge some study limitations. One of the main constraints of the study stems from the small sample size of participants with follow‐up MRI and the relatively short MRI follow‐up time, especially among those with the diffuse‐malignant subtype. Although our research question was hypothesis‐driven and the results are statistically significant, replication and external validation in future studies with larger sample sizes will increase the confidence and generalizability of our findings. Although the current study focused on longitudinal neuroanatomical changes across PD subtypes, future research should explore the direct associations between regional atrophy and progression in various clinical features to gain deeper insights into the disease pathophysiology and its clinical manifestations. This would require larger sample sizes as well as longer follow‐up durations for longitudinal MRI data to ensure sufficient statistical power for such analyses. Because of sample size limitations, all MRI analyses were performed at a regional level to improve statistical power. Future studies with larger sample sizes and greater statistical power can explore voxel‐level analyses to potentially reveal more subtle and localized patterns of change.

Conclusion

Our study provides further evidence that multi‐domain subtyping, based on initial motor and three key non‐motor features of PD (ie, RBD, autonomic disturbance, and early cognitive deficit), is a valid method to separate subgroups of PD. We now see evidence for more aggressive long‐term disease progression because of a distinct pattern of pathology. We have demonstrated an accelerated atrophy pattern within several brain regions in diffuse‐malignant PD subtype compared to the “mild motor‐predominant” group, distinguishable even after a rather short follow‐up period of 3.5 years. People with the diffuse‐malignant subtype showed the highest within‐group similarity in the overall pattern of brain MRI, whereas other subtypes failed to show a meaningful convergence toward the diffuse‐malignant subtype over time, compared to that of normal aging. These findings suggest the presence of a more diffuse multi‐domain neurodegenerative process in a subgroup of people with PD, favoring the existence of bioanatomical PD subtypes with diverse underlying pathophysiology. Our study has demonstrated brain MRI evidence to explain part of the diversity in people with idiopathic PD fostering the concept of precision medicine.

Author Roles

(1) Research project: A. Conception, B. Organization, C. Execution; (2) Statistical Analysis: A. Design, B. Execution, C. Review and Critique; (3) Manuscript: A. Writing of the First Draft, B. Review and Critique.

S.M.F: 1A, 1B, 1C, 2A, 2C, 3A, 3B

R.M: 1C, 2B, 3B

H.A: 1C, 2B, 3B

R.B.P: 2C, 3B

M.D: 1A, 2A, 2B, 2C, 3B

A.E.L: 2C, 3B

C.M: 1C, 2A, 2C, 3B

Y.Z: 1A, 1B, 1C, 2A, 2B, 2C, 3A, 3B

Financial Disclosures

S.M.F. received research grants from Parkinson Canada, Swedish Parkinson's Disease Foundation; is a consultant for Health Advances; and has received honoraria from European Science Foundation. R.M. has received a PhD scholarship from Fonds de la Recherche du Québec—Santé. H.A. has received a MSc scholarship from Fonds de la Recherche du Québec—Santé. R.B.P. has received grants and personal fees from Fonds de la Recherche du Québec—Santé, the Canadian Institute of Health Research, Parkinson Canada, the W. Garfield Weston Foundation, The Michael J. Fox Foundation (grant ID: 15989), the Webster Foundation, and personal fees from Takeda, Roche/Prothena, Biogen, Theranexus, Paladin, and Curasen outside the submitted work. M.D. has received research funding from the Healthy Brains for Healthy Lives, Fonds de recherche du Québec—Santé Chercheurs boursiers et chercheuses boursières, Natural Sciences and Engineering Research discovery grant, and Canadian Institutes of Health Research. A.E.L. has served as an advisor for AbbVie, AFFiRis, Alector, Amylyx, Aprinoia, Biogen, BioAdvance, BlueRock, Biovie, BMS, CoA Therapeutics, Denali, Janssen, Jazz, Lilly, Novartis, Paladin, Pharma 2B, PsychoGenetics, Retrophin, Roche, Sun Pharma, and UCB; received honoraria from Sun Pharma, AbbVie, and Sunovion; received grants from Brain Canada, Canadian Institutes of Health Research, Edmond J. Safra Philanthropic Foundation, The Michael J. Fox Foundation, Ontario Brain Institute, Parkinson Foundation, Parkinson Canada, and W. Garfield Weston Foundation; is serving as an expert witness in litigation related to paraquat and PD; and received publishing royalties from Elsevier, Saunders, Wiley‐Blackwell, Johns Hopkins Press, and Cambridge University Press. C.M. has received research grants from The Michael J. Fox Foundation, Canadian Institutes of Health Research, Parkinson's Foundation (US), National Institutes of Health (US), International Parkinson and Movement Disorders Society. Consultant for Grey Matter Technologies and receives financial compensation as a steering committee member from The Michael J. Fox Foundation. Y.Z. has received research funding from the Healthy Brains for Healthy Lives, Fonds de recherche du Québec—Santé Chercheurs boursiers et chercheuses boursières en Intelligence artificielle, Natural Sciences and Engineering Research discovery grant, and Canadian Institutes of Health Research.

Supporting information

FIGURE S1. (A) Accelerated atrophy, similar to Figure 2, with scanner model included as a random effect in the mixed‐effects model. (B) Scatter plot of T‐values for the interaction term reported in analyses with and without considering the scanner model. Red dots indicate regions that lose significance after FDR correction when accounting for the scanner model, while the green dot represents regions that remain statistically significant.

TABLE S1. Demographics features in individuals with longitudinal MRI included in the analysis with different subtypes of Parkinson's disease (N = 128).

Data Availability Statement

All data used in this study was from PPM) cohort. PPMI is a publicly available dataset that can be found at: http://www.ppmi-info.org.